diff --git a/.ci/windows_amd_base_files/run_amd_gpu_disable_smart_memory.bat b/.ci/windows_amd_base_files/run_amd_gpu_enable_dynamic_vram.bat similarity index 66% rename from .ci/windows_amd_base_files/run_amd_gpu_disable_smart_memory.bat rename to .ci/windows_amd_base_files/run_amd_gpu_enable_dynamic_vram.bat index cece0aeb2..94ad31942 100755 --- a/.ci/windows_amd_base_files/run_amd_gpu_disable_smart_memory.bat +++ b/.ci/windows_amd_base_files/run_amd_gpu_enable_dynamic_vram.bat @@ -1,2 +1,2 @@ -.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build --disable-smart-memory +.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build --enable-dynamic-vram pause diff --git a/.github/workflows/openapi-lint.yml b/.github/workflows/openapi-lint.yml new file mode 100644 index 000000000..be949de2a --- /dev/null +++ b/.github/workflows/openapi-lint.yml @@ -0,0 +1,31 @@ +name: OpenAPI Lint + +on: + pull_request: + paths: + - 'openapi.yaml' + - '.spectral.yaml' + - '.github/workflows/openapi-lint.yml' + +permissions: + contents: read + +jobs: + spectral: + name: Run Spectral + runs-on: ubuntu-latest + + steps: + - name: Checkout repository + uses: actions/checkout@v4 + + - name: Set up Node.js + uses: actions/setup-node@v4 + with: + node-version: '20' + + - name: Install Spectral + run: npm install -g @stoplight/spectral-cli@6 + + - name: Lint openapi.yaml + run: spectral lint openapi.yaml --ruleset .spectral.yaml --fail-severity=error diff --git a/.github/workflows/stable-release.yml b/.github/workflows/stable-release.yml index f501b7b31..bc64ed74d 100644 --- a/.github/workflows/stable-release.yml +++ b/.github/workflows/stable-release.yml @@ -145,6 +145,8 @@ jobs: cp -r ComfyUI/.ci/windows_${{ inputs.rel_name }}_base_files/* ./ cp ../update_comfyui_and_python_dependencies.bat ./update/ + echo 'local-portable' > ComfyUI/.comfy_environment + cd .. "C:\Program Files\7-Zip\7z.exe" a -t7z -m0=lzma2 -mx=9 -mfb=128 -md=768m -ms=on -mf=BCJ2 ComfyUI_windows_portable.7z ComfyUI_windows_portable diff --git a/.github/workflows/tag-dispatch-cloud.yml b/.github/workflows/tag-dispatch-cloud.yml new file mode 100644 index 000000000..53a0e91d6 --- /dev/null +++ b/.github/workflows/tag-dispatch-cloud.yml @@ -0,0 +1,45 @@ +name: Tag Dispatch to Cloud + +on: + push: + tags: + - 'v*' + +jobs: + dispatch-cloud: + runs-on: ubuntu-latest + steps: + - name: Send repository dispatch to cloud + env: + DISPATCH_TOKEN: ${{ secrets.CLOUD_REPO_DISPATCH_TOKEN }} + RELEASE_TAG: ${{ github.ref_name }} + run: | + set -euo pipefail + + if [ -z "${DISPATCH_TOKEN:-}" ]; then + echo "::error::CLOUD_REPO_DISPATCH_TOKEN is required but not set." + exit 1 + fi + + RELEASE_URL="https://github.com/${{ github.repository }}/releases/tag/${RELEASE_TAG}" + + PAYLOAD="$(jq -n \ + --arg release_tag "$RELEASE_TAG" \ + --arg release_url "$RELEASE_URL" \ + '{ + event_type: "comfyui_tag_pushed", + client_payload: { + release_tag: $release_tag, + release_url: $release_url + } + }')" + + curl -fsSL \ + -X POST \ + -H "Accept: application/vnd.github+json" \ + -H "Content-Type: application/json" \ + -H "Authorization: Bearer ${DISPATCH_TOKEN}" \ + https://api.github.com/repos/Comfy-Org/cloud/dispatches \ + -d "$PAYLOAD" + + echo "✅ Dispatched ComfyUI tag ${RELEASE_TAG} to Comfy-Org/cloud" diff --git a/.gitignore b/.gitignore index 2700ad5c2..fc426eda4 100644 --- a/.gitignore +++ b/.gitignore @@ -21,6 +21,6 @@ venv*/ *.log web_custom_versions/ .DS_Store -openapi.yaml filtered-openapi.yaml uv.lock +.comfy_environment diff --git a/.spectral.yaml b/.spectral.yaml new file mode 100644 index 000000000..a4b137628 --- /dev/null +++ b/.spectral.yaml @@ -0,0 +1,100 @@ +extends: + - spectral:oas + +# Severity levels: error, warn, info, hint, off +# Rules from the built-in "spectral:oas" ruleset are active by default. +# Below we tune severity and add custom rules for our conventions. +# +# This ruleset mirrors Comfy-Org/cloud/.spectral.yaml so specs across the +# organization are linted against a single consistent standard. + +rules: + # ----------------------------------------------------------------------- + # Built-in rule severity overrides + # ----------------------------------------------------------------------- + operation-operationId: error + operation-description: warn + operation-tag-defined: error + info-contact: off + info-description: warn + no-eval-in-markdown: error + no-$ref-siblings: error + + # ----------------------------------------------------------------------- + # Custom rules: naming conventions + # ----------------------------------------------------------------------- + + # Property names should be snake_case + property-name-snake-case: + description: Property names must be snake_case + severity: warn + given: "$.components.schemas.*.properties[*]~" + then: + function: pattern + functionOptions: + match: "^[a-z][a-z0-9]*(_[a-z0-9]+)*$" + + # Operation IDs should be camelCase + operation-id-camel-case: + description: Operation IDs must be camelCase + severity: warn + given: "$.paths.*.*.operationId" + then: + function: pattern + functionOptions: + match: "^[a-z][a-zA-Z0-9]*$" + + # ----------------------------------------------------------------------- + # Custom rules: response conventions + # ----------------------------------------------------------------------- + + # Error responses (4xx, 5xx) should use a consistent shape + error-response-schema: + description: Error responses should reference a standard error schema + severity: hint + given: "$.paths.*.*.responses[?(@property >= '400' && @property < '600')].content['application/json'].schema" + then: + field: "$ref" + function: truthy + + # All 2xx responses with JSON body should have a schema + response-schema-defined: + description: Success responses with JSON content should define a schema + severity: warn + given: "$.paths.*.*.responses[?(@property >= '200' && @property < '300')].content['application/json']" + then: + field: schema + function: truthy + + # ----------------------------------------------------------------------- + # Custom rules: best practices + # ----------------------------------------------------------------------- + + # Path parameters must have a description + path-param-description: + description: Path parameters should have a description + severity: warn + given: + - "$.paths.*.parameters[?(@.in == 'path')]" + - "$.paths.*.*.parameters[?(@.in == 'path')]" + then: + field: description + function: truthy + + # Schemas should have a description + schema-description: + description: Component schemas should have a description + severity: hint + given: "$.components.schemas.*" + then: + field: description + function: truthy + +overrides: + # /ws uses HTTP 101 (Switching Protocols) — a legitimate response for a + # WebSocket upgrade, but not a 2xx, so operation-success-response fires + # as a false positive. OpenAPI 3.x has no native WebSocket support. + - files: + - "openapi.yaml#/paths/~1ws" + rules: + operation-success-response: off diff --git a/CODEOWNERS b/CODEOWNERS index 4d5448636..946dbf946 100644 --- a/CODEOWNERS +++ b/CODEOWNERS @@ -1,2 +1,2 @@ # Admins -* @comfyanonymous @kosinkadink @guill +* @comfyanonymous @kosinkadink @guill @alexisrolland @rattus128 @kijai diff --git a/QUANTIZATION.md b/QUANTIZATION.md index 1693e13f3..300822029 100644 --- a/QUANTIZATION.md +++ b/QUANTIZATION.md @@ -139,9 +139,9 @@ Example: "_quantization_metadata": { "format_version": "1.0", "layers": { - "model.layers.0.mlp.up_proj": "float8_e4m3fn", - "model.layers.0.mlp.down_proj": "float8_e4m3fn", - "model.layers.1.mlp.up_proj": "float8_e4m3fn" + "model.layers.0.mlp.up_proj": {"format": "float8_e4m3fn"}, + "model.layers.0.mlp.down_proj": {"format": "float8_e4m3fn"}, + "model.layers.1.mlp.up_proj": {"format": "float8_e4m3fn"} } } } @@ -165,4 +165,4 @@ Activation quantization (e.g., for FP8 Tensor Core operations) requires `input_s 3. **Compute scales**: Derive `input_scale` from collected statistics 4. **Store in checkpoint**: Save `input_scale` parameters alongside weights -The calibration dataset should be representative of your target use case. For diffusion models, this typically means a diverse set of prompts and generation parameters. \ No newline at end of file +The calibration dataset should be representative of your target use case. For diffusion models, this typically means a diverse set of prompts and generation parameters. diff --git a/README.md b/README.md index 1eeb810de..0fd317d0a 100644 --- a/README.md +++ b/README.md @@ -1,7 +1,7 @@
# ComfyUI -**The most powerful and modular visual AI engine and application.** +**The most powerful and modular AI engine for content creation.** [![Website][website-shield]][website-url] @@ -31,10 +31,16 @@ [github-downloads-latest-shield]: https://img.shields.io/github/downloads/comfyanonymous/ComfyUI/latest/total?style=flat&label=downloads%40latest [github-downloads-link]: https://github.com/comfyanonymous/ComfyUI/releases -![ComfyUI Screenshot](https://github.com/user-attachments/assets/7ccaf2c1-9b72-41ae-9a89-5688c94b7abe) +ComfyUI Screenshot +
-ComfyUI lets you design and execute advanced stable diffusion pipelines using a graph/nodes/flowchart based interface. Available on Windows, Linux, and macOS. +ComfyUI is the AI creation engine for visual professionals who demand control over every model, every parameter, and every output. Its powerful and modular node graph interface empowers creatives to generate images, videos, 3D models, audio, and more... +- ComfyUI natively supports the latest open-source state of the art models. +- API nodes provide access to the best closed source models such as Nano Banana, Seedance, Hunyuan3D, etc. +- It is available on Windows, Linux, and macOS, locally with our desktop application or on our cloud. +- The most sophisticated workflows can be exposed through a simple UI thanks to App Mode. +- It integrates seamlessly into production pipelines with our API endpoints. ## Get Started @@ -77,6 +83,7 @@ See what ComfyUI can do with the [newer template workflows](https://comfy.org/wo - [Hunyuan Image 2.1](https://comfyanonymous.github.io/ComfyUI_examples/hunyuan_image/) - [Flux 2](https://comfyanonymous.github.io/ComfyUI_examples/flux2/) - [Z Image](https://comfyanonymous.github.io/ComfyUI_examples/z_image/) + - Ernie Image - Image Editing Models - [Omnigen 2](https://comfyanonymous.github.io/ComfyUI_examples/omnigen/) - [Flux Kontext](https://comfyanonymous.github.io/ComfyUI_examples/flux/#flux-kontext-image-editing-model) @@ -126,7 +133,7 @@ Workflow examples can be found on the [Examples page](https://comfyanonymous.git ComfyUI follows a weekly release cycle targeting Monday but this regularly changes because of model releases or large changes to the codebase. There are three interconnected repositories: 1. **[ComfyUI Core](https://github.com/comfyanonymous/ComfyUI)** - - Releases a new stable version (e.g., v0.7.0) roughly every week. + - Releases a new major stable version (e.g., v0.7.0) roughly every 2 weeks. - Starting from v0.4.0 patch versions will be used for fixes backported onto the current stable release. - Minor versions will be used for releases off the master branch. - Patch versions may still be used for releases on the master branch in cases where a backport would not make sense. @@ -193,11 +200,15 @@ If you have trouble extracting it, right click the file -> properties -> unblock The portable above currently comes with python 3.13 and pytorch cuda 13.0. Update your Nvidia drivers if it doesn't start. -#### Alternative Downloads: +#### All Official Portable Downloads: -[Experimental portable for AMD GPUs](https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_amd.7z) +[Portable for AMD GPUs](https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_amd.7z) -[Portable with pytorch cuda 12.6 and python 3.12](https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_nvidia_cu126.7z) (Supports Nvidia 10 series and older GPUs). +[Portable for Intel GPUs](https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_intel.7z) + +[Portable for Nvidia GPUs](https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_nvidia.7z) (supports 20 series and above). + +[Portable for Nvidia GPUs with pytorch cuda 12.6 and python 3.12](https://github.com/comfyanonymous/ComfyUI/releases/latest/download/ComfyUI_windows_portable_nvidia_cu126.7z) (Supports Nvidia 10 series and older GPUs). #### How do I share models between another UI and ComfyUI? diff --git a/api_server/routes/internal/internal_routes.py b/api_server/routes/internal/internal_routes.py index b224306da..1477afa01 100644 --- a/api_server/routes/internal/internal_routes.py +++ b/api_server/routes/internal/internal_routes.py @@ -67,7 +67,7 @@ class InternalRoutes: (entry for entry in os.scandir(directory) if is_visible_file(entry)), key=lambda entry: -entry.stat().st_mtime ) - return web.json_response([entry.name for entry in sorted_files], status=200) + return web.json_response([f"{entry.name} [{directory_type}]" for entry in sorted_files], status=200) def get_app(self): diff --git a/app/frontend_management.py b/app/frontend_management.py index f753ef0de..7108bd35a 100644 --- a/app/frontend_management.py +++ b/app/frontend_management.py @@ -27,7 +27,7 @@ def frontend_install_warning_message(): return f""" {get_missing_requirements_message()} -This error is happening because the ComfyUI frontend is no longer shipped as part of the main repo but as a pip package instead. +The ComfyUI frontend is shipped in a pip package so it needs to be updated separately from the ComfyUI code. """.strip() def parse_version(version: str) -> tuple[int, int, int]: diff --git a/app/node_replace_manager.py b/app/node_replace_manager.py index d9aab5b22..72e8ac2b1 100644 --- a/app/node_replace_manager.py +++ b/app/node_replace_manager.py @@ -1,5 +1,7 @@ from __future__ import annotations +import logging + from aiohttp import web from typing import TYPE_CHECKING, TypedDict @@ -31,8 +33,22 @@ class NodeReplaceManager: self._replacements: dict[str, list[NodeReplace]] = {} def register(self, node_replace: NodeReplace): - """Register a node replacement mapping.""" - self._replacements.setdefault(node_replace.old_node_id, []).append(node_replace) + """Register a node replacement mapping. + + Idempotent: if a replacement with the same (old_node_id, new_node_id) + is already registered, the duplicate is ignored. This prevents stale + entries from accumulating when custom nodes are reloaded in the same + process (e.g. via ComfyUI-Manager). + """ + existing = self._replacements.setdefault(node_replace.old_node_id, []) + for entry in existing: + if entry.new_node_id == node_replace.new_node_id: + logging.debug( + "Node replacement %s -> %s already registered, ignoring duplicate.", + node_replace.old_node_id, node_replace.new_node_id, + ) + return + existing.append(node_replace) def get_replacement(self, old_node_id: str) -> list[NodeReplace] | None: """Get replacements for an old node ID.""" diff --git a/app/user_manager.py b/app/user_manager.py index e18afb71b..0517b3344 100644 --- a/app/user_manager.py +++ b/app/user_manager.py @@ -28,8 +28,8 @@ def get_file_info(path: str, relative_to: str) -> FileInfo: return { "path": os.path.relpath(path, relative_to).replace(os.sep, '/'), "size": os.path.getsize(path), - "modified": os.path.getmtime(path), - "created": os.path.getctime(path) + "modified": int(os.path.getmtime(path) * 1000), + "created": int(os.path.getctime(path) * 1000), } diff --git a/blueprints/.glsl/Glow_30.frag b/blueprints/.glsl/Glow_30.frag index 0ee152628..f3c85a212 100644 --- a/blueprints/.glsl/Glow_30.frag +++ b/blueprints/.glsl/Glow_30.frag @@ -2,7 +2,6 @@ precision mediump float; uniform sampler2D u_image0; -uniform vec2 u_resolution; uniform int u_int0; // Blend mode uniform int u_int1; // Color tint uniform float u_float0; // Intensity @@ -75,7 +74,7 @@ void main() { float t0 = threshold - 0.15; float t1 = threshold + 0.15; - vec2 texelSize = 1.0 / u_resolution; + vec2 texelSize = 1.0 / vec2(textureSize(u_image0, 0)); float radius2 = radius * radius; float sampleScale = clamp(radius * 0.75, 0.35, 1.0); diff --git a/blueprints/.glsl/Image_Blur_1.frag b/blueprints/.glsl/Image_Blur_1.frag index 83238111d..1819e1695 100644 --- a/blueprints/.glsl/Image_Blur_1.frag +++ b/blueprints/.glsl/Image_Blur_1.frag @@ -12,7 +12,6 @@ const int RADIAL_SAMPLES = 12; const float RADIAL_STRENGTH = 0.0003; uniform sampler2D u_image0; -uniform vec2 u_resolution; uniform int u_int0; // Blur type (BLUR_GAUSSIAN, BLUR_BOX, BLUR_RADIAL) uniform float u_float0; // Blur radius/amount uniform int u_pass; // Pass index (0 = horizontal, 1 = vertical) @@ -25,7 +24,7 @@ float gaussian(float x, float sigma) { } void main() { - vec2 texelSize = 1.0 / u_resolution; + vec2 texelSize = 1.0 / vec2(textureSize(u_image0, 0)); float radius = max(u_float0, 0.0); // Radial (angular) blur - single pass, doesn't use separable diff --git a/blueprints/.glsl/Sharpen_23.frag b/blueprints/.glsl/Sharpen_23.frag index c03f94b66..e7463a329 100644 --- a/blueprints/.glsl/Sharpen_23.frag +++ b/blueprints/.glsl/Sharpen_23.frag @@ -2,14 +2,13 @@ precision highp float; uniform sampler2D u_image0; -uniform vec2 u_resolution; uniform float u_float0; // strength [0.0 – 2.0] typical: 0.3–1.0 in vec2 v_texCoord; layout(location = 0) out vec4 fragColor0; void main() { - vec2 texel = 1.0 / u_resolution; + vec2 texel = 1.0 / vec2(textureSize(u_image0, 0)); // Sample center and neighbors vec4 center = texture(u_image0, v_texCoord); diff --git a/blueprints/.glsl/Unsharp_Mask_26.frag b/blueprints/.glsl/Unsharp_Mask_26.frag index f5990cb4a..d968c9c03 100644 --- a/blueprints/.glsl/Unsharp_Mask_26.frag +++ b/blueprints/.glsl/Unsharp_Mask_26.frag @@ -2,7 +2,6 @@ precision highp float; uniform sampler2D u_image0; -uniform vec2 u_resolution; uniform float u_float0; // amount [0.0 - 3.0] typical: 0.5-1.5 uniform float u_float1; // radius [0.5 - 10.0] blur radius in pixels uniform float u_float2; // threshold [0.0 - 0.1] min difference to sharpen @@ -19,7 +18,7 @@ float getLuminance(vec3 color) { } void main() { - vec2 texel = 1.0 / u_resolution; + vec2 texel = 1.0 / vec2(textureSize(u_image0, 0)); float radius = max(u_float1, 0.5); float amount = u_float0; float threshold = u_float2; diff --git a/blueprints/Brightness and Contrast.json b/blueprints/Brightness and Contrast.json index 90bfe999d..78fc52f29 100644 --- a/blueprints/Brightness and Contrast.json +++ b/blueprints/Brightness and Contrast.json @@ -431,9 +431,10 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image Tools/Color adjust" + "category": "Image Tools/Color adjust", + "description": "Adjusts image brightness and contrast using a real-time GPU fragment shader." } ] }, "extra": {} -} +} \ No newline at end of file diff --git a/blueprints/Canny to Image (Z-Image-Turbo).json b/blueprints/Canny to Image (Z-Image-Turbo).json index ff9717308..14deb64cc 100644 --- a/blueprints/Canny to Image (Z-Image-Turbo).json +++ b/blueprints/Canny to Image (Z-Image-Turbo).json @@ -162,7 +162,7 @@ }, "revision": 0, "config": {}, - "name": "local-Canny to Image (Z-Image-Turbo)", + "name": "Canny to Image (Z-Image-Turbo)", "inputNode": { "id": -10, "bounding": [ @@ -1553,7 +1553,8 @@ "VHS_MetadataImage": true, "VHS_KeepIntermediate": true }, - "category": "Image generation and editing/Canny to image" + "category": "Image generation and editing/Canny to image", + "description": "Generates an image from a Canny edge map using Z-Image-Turbo, with text conditioning." } ] }, @@ -1574,4 +1575,4 @@ } }, "version": 0.4 -} +} \ No newline at end of file diff --git a/blueprints/Canny to Video (LTX 2.0).json b/blueprints/Canny to Video (LTX 2.0).json index fae8321b9..a9682c8a4 100644 --- a/blueprints/Canny to Video (LTX 2.0).json +++ b/blueprints/Canny to Video (LTX 2.0).json @@ -192,7 +192,7 @@ }, "revision": 0, "config": {}, - "name": "local-Canny to Video (LTX 2.0)", + "name": "Canny to Video (LTX 2.0)", "inputNode": { "id": -10, "bounding": [ @@ -3600,7 +3600,8 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Video generation and editing/Canny to video" + "category": "Video generation and editing/Canny to video", + "description": "Generates video from Canny edge maps using LTX-2, with optional synchronized audio." } ] }, @@ -3616,4 +3617,4 @@ } }, "version": 0.4 -} +} \ No newline at end of file diff --git a/blueprints/Chromatic Aberration.json b/blueprints/Chromatic Aberration.json index ae8037b1b..893fb1190 100644 --- a/blueprints/Chromatic Aberration.json +++ b/blueprints/Chromatic Aberration.json @@ -377,8 +377,9 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image Tools/Color adjust" + "category": "Image Tools/Color adjust", + "description": "Adds lens-style chromatic aberration (color fringing) using a real-time GPU fragment shader." } ] } -} +} \ No newline at end of file diff --git a/blueprints/Color Adjustment.json b/blueprints/Color Adjustment.json index 622bf28af..5abbf8baa 100644 --- a/blueprints/Color Adjustment.json +++ b/blueprints/Color Adjustment.json @@ -596,7 +596,8 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image Tools/Color adjust" + "category": "Image Tools/Color adjust", + "description": "Adjusts saturation, temperature, tint, and vibrance using a real-time GPU fragment shader." } ] } diff --git a/blueprints/Color Balance.json b/blueprints/Color Balance.json index 21d6319ed..d921eab37 100644 --- a/blueprints/Color Balance.json +++ b/blueprints/Color Balance.json @@ -1129,7 +1129,8 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image Tools/Color adjust" + "category": "Image Tools/Color adjust", + "description": "Balances colors across shadows, midtones, and highlights using a real-time GPU fragment shader." } ] } diff --git a/blueprints/Color Curves.json b/blueprints/Color Curves.json index 1461cf396..b9bfb7029 100644 --- a/blueprints/Color Curves.json +++ b/blueprints/Color Curves.json @@ -608,7 +608,8 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image Tools/Color adjust" + "category": "Image Tools/Color adjust", + "description": "Fine-tunes tone and color with per-channel curve adjustments using a real-time GPU fragment shader." } ] } diff --git a/blueprints/ControlNet (Z-Image-Turbo).json b/blueprints/ControlNet (Z-Image-Turbo).json new file mode 100644 index 000000000..fbec95a97 --- /dev/null +++ b/blueprints/ControlNet (Z-Image-Turbo).json @@ -0,0 +1,1412 @@ +{ + "revision": 0, + "last_node_id": 85, + "last_link_id": 0, + "nodes": [ + { + "id": 85, + "type": "d2e76ecf-6e84-4b8c-8913-48efc09ec1c4", + "pos": [ + 440, + 1220 + ], + "size": [ + 480, + 0 + ], + "flags": {}, + "order": 6, + "mode": 0, + "inputs": [ + { + "label": "control_image", + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": null + }, + { + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + }, + { + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": null + }, + { + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": null + }, + { + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": null + }, + { + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": null + }, + { + "label": "patch_model", + "name": "name", + "type": "COMBO", + "widget": { + "name": "name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [] + } + ], + "title": "ControlNet (Z-Image-Turbo)", + "properties": { + "proxyWidgets": [ + [ + "83", + "text" + ], + [ + "79", + "seed" + ], + [ + "74", + "unet_name" + ], + [ + "73", + "clip_name" + ], + [ + "75", + "vae_name" + ], + [ + "76", + "name" + ], + [ + "79", + "control_after_generate" + ] + ], + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [] + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "d2e76ecf-6e84-4b8c-8913-48efc09ec1c4", + "version": 1, + "state": { + "lastGroupId": 9, + "lastNodeId": 85, + "lastLinkId": 87, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "ControlNet (Z-Image-Turbo)", + "inputNode": { + "id": -10, + "bounding": [ + -500, + 620, + 120, + 180 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 1390, + 1100, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "fbbb968e-d3cf-40e4-b3ce-7abb074e5bd8", + "name": "image", + "type": "IMAGE", + "linkIds": [ + 65, + 80 + ], + "localized_name": "image", + "label": "control_image", + "pos": [ + -400, + 640 + ] + }, + { + "id": "c1b19877-5417-4580-aea1-44439c70c1dd", + "name": "text", + "type": "STRING", + "linkIds": [ + 81 + ], + "pos": [ + -400, + 660 + ] + }, + { + "id": "b5671515-bc7a-4be5-b1e7-d4f0f68907d6", + "name": "seed", + "type": "INT", + "linkIds": [ + 83 + ], + "pos": [ + -400, + 680 + ] + }, + { + "id": "2838be23-8034-4f16-87a5-d29d790e8391", + "name": "unet_name", + "type": "COMBO", + "linkIds": [ + 84 + ], + "pos": [ + -400, + 700 + ] + }, + { + "id": "8a6643b5-8f78-41ff-bbc6-e87b95459706", + "name": "clip_name", + "type": "COMBO", + "linkIds": [ + 85 + ], + "pos": [ + -400, + 720 + ] + }, + { + "id": "b103dc94-8ca7-456b-a809-414d7e341a1b", + "name": "vae_name", + "type": "COMBO", + "linkIds": [ + 86 + ], + "pos": [ + -400, + 740 + ] + }, + { + "id": "4a7d65af-f0fd-4a5c-832a-bdc0d15b1f30", + "name": "name", + "type": "COMBO", + "linkIds": [ + 87 + ], + "label": "patch_model", + "pos": [ + -400, + 760 + ] + } + ], + "outputs": [ + { + "id": "ccb7fa39-4a3d-4eb2-8fd2-91d08fad9570", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 45 + ], + "localized_name": "IMAGE", + "pos": [ + 1410, + 1120 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 73, + "type": "CLIPLoader", + "pos": [ + 20, + 500 + ], + "size": [ + 270, + 150 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "clip_name", + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": 85 + }, + { + "localized_name": "type", + "name": "type", + "type": "COMBO", + "widget": { + "name": "type" + }, + "link": null + }, + { + "localized_name": "device", + "name": "device", + "shape": 7, + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [ + 44 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "CLIPLoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "qwen_3_4b.safetensors", + "url": "https://huggingface.co/Comfy-Org/z_image_turbo/resolve/main/split_files/text_encoders/qwen_3_4b.safetensors", + "directory": "text_encoders" + } + ] + }, + "widgets_values": [ + "qwen_3_4b.safetensors", + "lumina2", + "default" + ] + }, + { + "id": 74, + "type": "UNETLoader", + "pos": [ + 20, + 320 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "unet_name", + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": 84 + }, + { + "localized_name": "weight_dtype", + "name": "weight_dtype", + "type": "COMBO", + "widget": { + "name": "weight_dtype" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 79 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "UNETLoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "z_image_turbo_bf16.safetensors", + "url": "https://huggingface.co/Comfy-Org/z_image_turbo/resolve/main/split_files/diffusion_models/z_image_turbo_bf16.safetensors", + "directory": "diffusion_models" + } + ] + }, + "widgets_values": [ + "z_image_turbo_bf16.safetensors", + "default" + ] + }, + { + "id": 75, + "type": "VAELoader", + "pos": [ + 20, + 760 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "vae_name", + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": 86 + } + ], + "outputs": [ + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "links": [ + 39, + 70 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "VAELoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "ae.safetensors", + "url": "https://huggingface.co/Comfy-Org/z_image_turbo/resolve/main/split_files/vae/ae.safetensors", + "directory": "vae" + } + ] + }, + "widgets_values": [ + "ae.safetensors" + ] + }, + { + "id": 76, + "type": "ModelPatchLoader", + "pos": [ + 20, + 940 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "name", + "name": "name", + "type": "COMBO", + "widget": { + "name": "name" + }, + "link": 87 + } + ], + "outputs": [ + { + "localized_name": "MODEL_PATCH", + "name": "MODEL_PATCH", + "type": "MODEL_PATCH", + "links": [ + 74 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.51", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ModelPatchLoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "Z-Image-Turbo-Fun-Controlnet-Union.safetensors", + "url": "https://huggingface.co/alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union/resolve/main/Z-Image-Turbo-Fun-Controlnet-Union.safetensors", + "directory": "model_patches" + } + ] + }, + "widgets_values": [ + "Z-Image-Turbo-Fun-Controlnet-Union.safetensors" + ] + }, + { + "id": 77, + "type": "VAEDecode", + "pos": [ + 940, + 1100 + ], + "size": [ + 230, + 100 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 38 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 39 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "slot_index": 0, + "links": [ + 45 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.64", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "VAEDecode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 78, + "type": "ModelSamplingAuraFlow", + "pos": [ + 910, + 270 + ], + "size": [ + 290, + 110 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 69 + }, + { + "localized_name": "shift", + "name": "shift", + "type": "FLOAT", + "widget": { + "name": "shift" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "slot_index": 0, + "links": [ + 40 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.64", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ModelSamplingAuraFlow", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 3 + ] + }, + { + "id": 79, + "type": "KSampler", + "pos": [ + 910, + 430 + ], + "size": [ + 300, + 570 + ], + "flags": {}, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 40 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 41 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 42 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 78 + }, + { + "localized_name": "seed", + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": 83 + }, + { + "localized_name": "steps", + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": null + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": null + }, + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + }, + { + "localized_name": "scheduler", + "name": "scheduler", + "type": "COMBO", + "widget": { + "name": "scheduler" + }, + "link": null + }, + { + "localized_name": "denoise", + "name": "denoise", + "type": "FLOAT", + "widget": { + "name": "denoise" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 38 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.64", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "KSampler", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 729703840979498, + "randomize", + 8, + 1, + "res_multistep", + "simple", + 1 + ] + }, + { + "id": 80, + "type": "ConditioningZeroOut", + "pos": [ + 610, + 830 + ], + "size": [ + 230, + 80 + ], + "flags": { + "collapsed": true + }, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "conditioning", + "name": "conditioning", + "type": "CONDITIONING", + "link": 36 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 42 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ConditioningZeroOut", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 81, + "type": "QwenImageDiffsynthControlnet", + "pos": [ + 490, + 970 + ], + "size": [ + 290, + 200 + ], + "flags": {}, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 79 + }, + { + "localized_name": "model_patch", + "name": "model_patch", + "type": "MODEL_PATCH", + "link": 74 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 70 + }, + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 65 + }, + { + "localized_name": "mask", + "name": "mask", + "shape": 7, + "type": "MASK", + "link": null + }, + { + "localized_name": "strength", + "name": "strength", + "type": "FLOAT", + "widget": { + "name": "strength" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 69 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.76", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "QwenImageDiffsynthControlnet", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1 + ] + }, + { + "id": 82, + "type": "EmptySD3LatentImage", + "pos": [ + 40, + 1200 + ], + "size": [ + 260, + 170 + ], + "flags": {}, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 76 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 77 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 78 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.64", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "EmptySD3LatentImage", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1024, + 1024, + 1 + ] + }, + { + "id": 83, + "type": "CLIPTextEncode", + "pos": [ + 430, + 310 + ], + "size": [ + 400, + 440 + ], + "flags": {}, + "order": 10, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 44 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 81 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 36, + 41 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "CLIPTextEncode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "" + ], + "color": "#232", + "bgcolor": "#353" + }, + { + "id": 84, + "type": "GetImageSize", + "pos": [ + 50, + 1410 + ], + "size": [ + 230, + 120 + ], + "flags": { + "collapsed": true + }, + "order": 11, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 80 + } + ], + "outputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "links": [ + 76 + ] + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "links": [ + 77 + ] + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "links": null + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.76", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "GetImageSize", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + } + ], + "groups": [ + { + "id": 3, + "title": "Prompt", + "bounding": [ + 410, + 230, + 440, + 630 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 4, + "title": "Model", + "bounding": [ + -50, + 230, + 430, + 840 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 8, + "title": "Apple ControlNet", + "bounding": [ + 410, + 890, + 440, + 330 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 9, + "title": "Image Size", + "bounding": [ + -50, + 1100, + 430, + 350 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 38, + "origin_id": 79, + "origin_slot": 0, + "target_id": 77, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 39, + "origin_id": 75, + "origin_slot": 0, + "target_id": 77, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 69, + "origin_id": 81, + "origin_slot": 0, + "target_id": 78, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 40, + "origin_id": 78, + "origin_slot": 0, + "target_id": 79, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 41, + "origin_id": 83, + "origin_slot": 0, + "target_id": 79, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 42, + "origin_id": 80, + "origin_slot": 0, + "target_id": 79, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 78, + "origin_id": 82, + "origin_slot": 0, + "target_id": 79, + "target_slot": 3, + "type": "LATENT" + }, + { + "id": 36, + "origin_id": 83, + "origin_slot": 0, + "target_id": 80, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 79, + "origin_id": 74, + "origin_slot": 0, + "target_id": 81, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 74, + "origin_id": 76, + "origin_slot": 0, + "target_id": 81, + "target_slot": 1, + "type": "MODEL_PATCH" + }, + { + "id": 70, + "origin_id": 75, + "origin_slot": 0, + "target_id": 81, + "target_slot": 2, + "type": "VAE" + }, + { + "id": 76, + "origin_id": 84, + "origin_slot": 0, + "target_id": 82, + "target_slot": 0, + "type": "INT" + }, + { + "id": 77, + "origin_id": 84, + "origin_slot": 1, + "target_id": 82, + "target_slot": 1, + "type": "INT" + }, + { + "id": 44, + "origin_id": 73, + "origin_slot": 0, + "target_id": 83, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 65, + "origin_id": -10, + "origin_slot": 0, + "target_id": 81, + "target_slot": 3, + "type": "IMAGE" + }, + { + "id": 80, + "origin_id": -10, + "origin_slot": 0, + "target_id": 84, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 45, + "origin_id": 77, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 81, + "origin_id": -10, + "origin_slot": 1, + "target_id": 83, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 83, + "origin_id": -10, + "origin_slot": 2, + "target_id": 79, + "target_slot": 4, + "type": "INT" + }, + { + "id": 84, + "origin_id": -10, + "origin_slot": 3, + "target_id": 74, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 85, + "origin_id": -10, + "origin_slot": 4, + "target_id": 73, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 86, + "origin_id": -10, + "origin_slot": 5, + "target_id": 75, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 87, + "origin_id": -10, + "origin_slot": 6, + "target_id": 76, + "target_slot": 0, + "type": "COMBO" + } + ], + "extra": { + "workflowRendererVersion": "LG" + }, + "category": "Image generation and editing/ControlNet", + "description": "Generates images from a text prompt and ControlNet conditioning (e.g. depth, canny) using Z-Image-Turbo." + } + ] + }, + "extra": { + "ue_links": [] + } +} \ No newline at end of file diff --git a/blueprints/Crop Images 2x2.json b/blueprints/Crop Images 2x2.json new file mode 100644 index 000000000..99b89b608 --- /dev/null +++ b/blueprints/Crop Images 2x2.json @@ -0,0 +1,1621 @@ +{ + "revision": 0, + "last_node_id": 139, + "last_link_id": 0, + "nodes": [ + { + "id": 135, + "type": "3b5ed000-6ab3-4458-91f7-8d6d366b0b40", + "pos": [ + -2479.9999801712506, + 2019.9999372732784 + ], + "size": [ + 230, + 170 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": null + } + ], + "outputs": [ + { + "label": "top_left", + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [] + }, + { + "label": "bottom_left", + "localized_name": "IMAGE_1", + "name": "IMAGE_1", + "type": "IMAGE", + "links": [] + }, + { + "label": "top_right", + "localized_name": "IMAGE_2", + "name": "IMAGE_2", + "type": "IMAGE", + "links": [] + }, + { + "label": "bottom_right", + "localized_name": "IMAGE_3", + "name": "IMAGE_3", + "type": "IMAGE", + "links": [] + }, + { + "label": "images", + "name": "IMAGE_4", + "type": "IMAGE", + "links": [] + } + ], + "properties": { + "proxyWidgets": [], + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1" + }, + "widgets_values": [], + "title": "Crop Images 2x2" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "3b5ed000-6ab3-4458-91f7-8d6d366b0b40", + "version": 1, + "state": { + "lastGroupId": 3, + "lastNodeId": 142, + "lastLinkId": 245, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Crop Images 2x2", + "inputNode": { + "id": -10, + "bounding": [ + -10, + 1570, + 120, + 60 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 2919.9998608196274, + 1435, + 120, + 140 + ] + }, + "inputs": [ + { + "id": "741854dd-bfb1-4700-ba8c-3b9dea59d021", + "name": "image", + "type": "IMAGE", + "linkIds": [ + 2, + 11, + 13, + 30, + 32 + ], + "localized_name": "image", + "pos": [ + 90, + 1590 + ] + } + ], + "outputs": [ + { + "id": "0eaca6d4-679a-433e-9703-bfa6dceacb18", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 41 + ], + "localized_name": "IMAGE", + "label": "top_left", + "pos": [ + 2939.9998608196274, + 1455 + ] + }, + { + "id": "fff5a1ad-3a74-4c87-938c-ee0fff55f840", + "name": "IMAGE_1", + "type": "IMAGE", + "linkIds": [ + 42 + ], + "localized_name": "IMAGE_1", + "label": "bottom_left", + "pos": [ + 2939.9998608196274, + 1475 + ] + }, + { + "id": "08f40978-fb25-4d98-b716-b61e43b16043", + "name": "IMAGE_2", + "type": "IMAGE", + "linkIds": [ + 43 + ], + "localized_name": "IMAGE_2", + "label": "top_right", + "pos": [ + 2939.9998608196274, + 1495 + ] + }, + { + "id": "17b9416f-3369-43c1-b62f-3e31fc2a7e32", + "name": "IMAGE_3", + "type": "IMAGE", + "linkIds": [ + 44 + ], + "localized_name": "IMAGE_3", + "label": "bottom_right", + "pos": [ + 2939.9998608196274, + 1515 + ] + }, + { + "id": "430e2f3b-c617-4549-9daf-3ebf5be423a3", + "name": "IMAGE_4", + "type": "IMAGE", + "linkIds": [ + 240 + ], + "label": "images", + "pos": [ + 2939.9998608196274, + 1535 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 7, + "type": "ComfyMathExpression", + "pos": [ + 740, + 1390 + ], + "size": [ + 370, + 190 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 3 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": 4 + }, + { + "label": "c", + "localized_name": "values.c", + "name": "values.c", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 7, + 14, + 28, + 40, + 242 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ComfyMathExpression" + }, + "widgets_values": [ + "max(1, int(a/b))" + ] + }, + { + "id": 8, + "type": "GetImageSize", + "pos": [ + 390, + 1450 + ], + "size": [ + 230, + 120 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 2 + } + ], + "outputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "links": [ + 3, + 241 + ] + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "links": [ + 5, + 245 + ] + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "links": null + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "GetImageSize" + } + }, + { + "id": 9, + "type": "PrimitiveInt", + "pos": [ + 390, + 1650 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 4, + 6 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "PrimitiveInt" + }, + "widgets_values": [ + 2, + "fixed" + ] + }, + { + "id": 10, + "type": "ImageCropV2", + "pos": [ + 1710, + 430 + ], + "size": [ + 300, + 480 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 11 + }, + { + "localized_name": "crop_region", + "name": "crop_region", + "type": "BOUNDING_BOX", + "widget": { + "name": "crop_region" + }, + "link": 9 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 41, + 236 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ImageCropV2" + }, + "widgets_values": [ + { + "x": 0, + "y": 0, + "width": 512, + "height": 512 + }, + 0, + 0, + 512, + 512 + ] + }, + { + "id": 12, + "type": "PrimitiveBoundingBox", + "pos": [ + 1370, + 570 + ], + "size": [ + 270, + 200 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "x", + "name": "x", + "type": "INT", + "widget": { + "name": "x" + }, + "link": null + }, + { + "localized_name": "y", + "name": "y", + "type": "INT", + "widget": { + "name": "y" + }, + "link": null + }, + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 7 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 8 + } + ], + "outputs": [ + { + "localized_name": "BOUNDING_BOX", + "name": "BOUNDING_BOX", + "type": "BOUNDING_BOX", + "links": [ + 9 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "PrimitiveBoundingBox" + }, + "widgets_values": [ + 0, + 0, + 512, + 512 + ] + }, + { + "id": 13, + "type": "ComfyMathExpression", + "pos": [ + 750, + 1650 + ], + "size": [ + 370, + 190 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 5 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": 6 + }, + { + "label": "c", + "localized_name": "values.c", + "name": "values.c", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 8, + 23, + 27, + 39, + 246 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ComfyMathExpression" + }, + "widgets_values": [ + "max(1, int(a/b))" + ] + }, + { + "id": 138, + "type": "ComfyMathExpression", + "pos": [ + 1170, + 1210 + ], + "size": [ + 420, + 190 + ], + "flags": {}, + "order": 13, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 241 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": 242 + }, + { + "label": "c", + "localized_name": "values.c", + "name": "values.c", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 243, + 244 + ] + } + ], + "title": "Math Expression (Right Width)", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ComfyMathExpression" + }, + "widgets_values": [ + "max(1, a - b)" + ] + }, + { + "id": 139, + "type": "ComfyMathExpression", + "pos": [ + 1170, + 1860 + ], + "size": [ + 420, + 190 + ], + "flags": {}, + "order": 14, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 245 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": 246 + }, + { + "label": "c", + "localized_name": "values.c", + "name": "values.c", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 247, + 248 + ] + } + ], + "title": "Math Expression (Bottom Height)", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ComfyMathExpression" + }, + "widgets_values": [ + "max(1, a - b)" + ] + }, + { + "id": 15, + "type": "ImageCropV2", + "pos": [ + 1740, + 1600 + ], + "size": [ + 300, + 480 + ], + "flags": {}, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 13 + }, + { + "localized_name": "crop_region", + "name": "crop_region", + "type": "BOUNDING_BOX", + "widget": { + "name": "crop_region" + }, + "link": 12 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 42, + 238 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ImageCropV2" + }, + "widgets_values": [ + { + "x": 0, + "y": 0, + "width": 512, + "height": 512 + }, + 0, + 0, + 512, + 512 + ] + }, + { + "id": 16, + "type": "PrimitiveBoundingBox", + "pos": [ + 1350, + 1780 + ], + "size": [ + 270, + 200 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "x", + "name": "x", + "type": "INT", + "widget": { + "name": "x" + }, + "link": null + }, + { + "localized_name": "y", + "name": "y", + "type": "INT", + "widget": { + "name": "y" + }, + "link": 23 + }, + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 14 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 247 + } + ], + "outputs": [ + { + "localized_name": "BOUNDING_BOX", + "name": "BOUNDING_BOX", + "type": "BOUNDING_BOX", + "links": [ + 12 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "PrimitiveBoundingBox" + }, + "widgets_values": [ + 0, + 0, + 512, + 512 + ] + }, + { + "id": 25, + "type": "PrimitiveBoundingBox", + "pos": [ + 1350, + 1200 + ], + "size": [ + 270, + 200 + ], + "flags": {}, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "x", + "name": "x", + "type": "INT", + "widget": { + "name": "x" + }, + "link": 28 + }, + { + "localized_name": "y", + "name": "y", + "type": "INT", + "widget": { + "name": "y" + }, + "link": null + }, + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 243 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 27 + } + ], + "outputs": [ + { + "localized_name": "BOUNDING_BOX", + "name": "BOUNDING_BOX", + "type": "BOUNDING_BOX", + "links": [ + 29 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "PrimitiveBoundingBox" + }, + "widgets_values": [ + 6, + 0, + 512, + 512 + ] + }, + { + "id": 26, + "type": "ImageCropV2", + "pos": [ + 1720, + 1050 + ], + "size": [ + 300, + 480 + ], + "flags": {}, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 30 + }, + { + "localized_name": "crop_region", + "name": "crop_region", + "type": "BOUNDING_BOX", + "widget": { + "name": "crop_region" + }, + "link": 29 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 43, + 237 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ImageCropV2" + }, + "widgets_values": [ + { + "x": 0, + "y": 0, + "width": 512, + "height": 512 + }, + 0, + 0, + 512, + 512 + ] + }, + { + "id": 30, + "type": "ImageCropV2", + "pos": [ + 1740, + 2130 + ], + "size": [ + 300, + 480 + ], + "flags": {}, + "order": 10, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 32 + }, + { + "localized_name": "crop_region", + "name": "crop_region", + "type": "BOUNDING_BOX", + "widget": { + "name": "crop_region" + }, + "link": 35 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 44, + 239 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ImageCropV2" + }, + "widgets_values": [ + { + "x": 0, + "y": 0, + "width": 512, + "height": 512 + }, + 0, + 0, + 512, + 512 + ] + }, + { + "id": 32, + "type": "PrimitiveBoundingBox", + "pos": [ + 1370, + 2280 + ], + "size": [ + 270, + 200 + ], + "flags": {}, + "order": 11, + "mode": 0, + "inputs": [ + { + "localized_name": "x", + "name": "x", + "type": "INT", + "widget": { + "name": "x" + }, + "link": 40 + }, + { + "localized_name": "y", + "name": "y", + "type": "INT", + "widget": { + "name": "y" + }, + "link": 39 + }, + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 244 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 248 + } + ], + "outputs": [ + { + "localized_name": "BOUNDING_BOX", + "name": "BOUNDING_BOX", + "type": "BOUNDING_BOX", + "links": [ + 35 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "PrimitiveBoundingBox" + }, + "widgets_values": [ + 6, + 0, + 512, + 512 + ] + }, + { + "id": 137, + "type": "BatchImagesNode", + "pos": [ + 2520, + 1540 + ], + "size": [ + 230, + 170 + ], + "flags": {}, + "order": 12, + "mode": 0, + "inputs": [ + { + "label": "image0", + "localized_name": "images.image0", + "name": "images.image0", + "type": "IMAGE", + "link": 236 + }, + { + "label": "image1", + "localized_name": "images.image1", + "name": "images.image1", + "type": "IMAGE", + "link": 237 + }, + { + "label": "image2", + "localized_name": "images.image2", + "name": "images.image2", + "shape": 7, + "type": "IMAGE", + "link": 238 + }, + { + "label": "image3", + "localized_name": "images.image3", + "name": "images.image3", + "shape": 7, + "type": "IMAGE", + "link": 239 + }, + { + "label": "image4", + "localized_name": "images.image4", + "name": "images.image4", + "shape": 7, + "type": "IMAGE", + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 240 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "BatchImagesNode" + } + } + ], + "groups": [ + { + "id": 1, + "title": "Crop Images 2x2", + "bounding": [ + 380, + 360, + 1710, + 2270 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 3, + "origin_id": 8, + "origin_slot": 0, + "target_id": 7, + "target_slot": 0, + "type": "INT" + }, + { + "id": 4, + "origin_id": 9, + "origin_slot": 0, + "target_id": 7, + "target_slot": 1, + "type": "INT" + }, + { + "id": 9, + "origin_id": 12, + "origin_slot": 0, + "target_id": 10, + "target_slot": 1, + "type": "BOUNDING_BOX" + }, + { + "id": 7, + "origin_id": 7, + "origin_slot": 1, + "target_id": 12, + "target_slot": 2, + "type": "INT" + }, + { + "id": 8, + "origin_id": 13, + "origin_slot": 1, + "target_id": 12, + "target_slot": 3, + "type": "INT" + }, + { + "id": 5, + "origin_id": 8, + "origin_slot": 1, + "target_id": 13, + "target_slot": 0, + "type": "INT" + }, + { + "id": 6, + "origin_id": 9, + "origin_slot": 0, + "target_id": 13, + "target_slot": 1, + "type": "INT" + }, + { + "id": 12, + "origin_id": 16, + "origin_slot": 0, + "target_id": 15, + "target_slot": 1, + "type": "BOUNDING_BOX" + }, + { + "id": 23, + "origin_id": 13, + "origin_slot": 1, + "target_id": 16, + "target_slot": 1, + "type": "INT" + }, + { + "id": 14, + "origin_id": 7, + "origin_slot": 1, + "target_id": 16, + "target_slot": 2, + "type": "INT" + }, + { + "id": 247, + "origin_id": 139, + "origin_slot": 1, + "target_id": 16, + "target_slot": 3, + "type": "INT" + }, + { + "id": 28, + "origin_id": 7, + "origin_slot": 1, + "target_id": 25, + "target_slot": 0, + "type": "INT" + }, + { + "id": 243, + "origin_id": 138, + "origin_slot": 1, + "target_id": 25, + "target_slot": 2, + "type": "INT" + }, + { + "id": 27, + "origin_id": 13, + "origin_slot": 1, + "target_id": 25, + "target_slot": 3, + "type": "INT" + }, + { + "id": 29, + "origin_id": 25, + "origin_slot": 0, + "target_id": 26, + "target_slot": 1, + "type": "BOUNDING_BOX" + }, + { + "id": 35, + "origin_id": 32, + "origin_slot": 0, + "target_id": 30, + "target_slot": 1, + "type": "BOUNDING_BOX" + }, + { + "id": 40, + "origin_id": 7, + "origin_slot": 1, + "target_id": 32, + "target_slot": 0, + "type": "INT" + }, + { + "id": 39, + "origin_id": 13, + "origin_slot": 1, + "target_id": 32, + "target_slot": 1, + "type": "INT" + }, + { + "id": 244, + "origin_id": 138, + "origin_slot": 1, + "target_id": 32, + "target_slot": 2, + "type": "INT" + }, + { + "id": 248, + "origin_id": 139, + "origin_slot": 1, + "target_id": 32, + "target_slot": 3, + "type": "INT" + }, + { + "id": 241, + "origin_id": 8, + "origin_slot": 0, + "target_id": 138, + "target_slot": 0, + "type": "INT" + }, + { + "id": 242, + "origin_id": 7, + "origin_slot": 1, + "target_id": 138, + "target_slot": 1, + "type": "INT" + }, + { + "id": 245, + "origin_id": 8, + "origin_slot": 1, + "target_id": 139, + "target_slot": 0, + "type": "INT" + }, + { + "id": 246, + "origin_id": 13, + "origin_slot": 1, + "target_id": 139, + "target_slot": 1, + "type": "INT" + }, + { + "id": 2, + "origin_id": -10, + "origin_slot": 0, + "target_id": 8, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 11, + "origin_id": -10, + "origin_slot": 0, + "target_id": 10, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 13, + "origin_id": -10, + "origin_slot": 0, + "target_id": 15, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 30, + "origin_id": -10, + "origin_slot": 0, + "target_id": 26, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 32, + "origin_id": -10, + "origin_slot": 0, + "target_id": 30, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 41, + "origin_id": 10, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 42, + "origin_id": 15, + "origin_slot": 0, + "target_id": -20, + "target_slot": 1, + "type": "IMAGE" + }, + { + "id": 43, + "origin_id": 26, + "origin_slot": 0, + "target_id": -20, + "target_slot": 2, + "type": "IMAGE" + }, + { + "id": 44, + "origin_id": 30, + "origin_slot": 0, + "target_id": -20, + "target_slot": 3, + "type": "IMAGE" + }, + { + "id": 236, + "origin_id": 10, + "origin_slot": 0, + "target_id": 137, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 237, + "origin_id": 26, + "origin_slot": 0, + "target_id": 137, + "target_slot": 1, + "type": "IMAGE" + }, + { + "id": 238, + "origin_id": 15, + "origin_slot": 0, + "target_id": 137, + "target_slot": 2, + "type": "IMAGE" + }, + { + "id": 239, + "origin_id": 30, + "origin_slot": 0, + "target_id": 137, + "target_slot": 3, + "type": "IMAGE" + }, + { + "id": 240, + "origin_id": 137, + "origin_slot": 0, + "target_id": -20, + "target_slot": 4, + "type": "IMAGE" + } + ], + "extra": {}, + "category": "Image Tools/Crop", + "description": "Splits an image into a 2×2 grid of four equal tiles." + } + ] + }, + "extra": { + "ue_links": [], + "links_added_by_ue": [] + } +} \ No newline at end of file diff --git a/blueprints/Crop Images 3x3.json b/blueprints/Crop Images 3x3.json new file mode 100644 index 000000000..6ac636da4 --- /dev/null +++ b/blueprints/Crop Images 3x3.json @@ -0,0 +1,2958 @@ +{ + "revision": 0, + "last_node_id": 141, + "last_link_id": 0, + "nodes": [ + { + "id": 134, + "type": "7fd47bca-ff89-476c-a98d-ca6f7cf756fe", + "pos": [ + -2620, + 1620 + ], + "size": [ + 230, + 290 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": null + } + ], + "outputs": [ + { + "label": "top_left", + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [] + }, + { + "label": "top_center", + "name": "IMAGE_1", + "type": "IMAGE", + "links": [] + }, + { + "label": "top_right", + "name": "IMAGE_2", + "type": "IMAGE", + "links": [] + }, + { + "label": "middle_left", + "name": "IMAGE_3", + "type": "IMAGE", + "links": [] + }, + { + "label": "middle_center", + "name": "IMAGE_4", + "type": "IMAGE", + "links": [] + }, + { + "label": "middle_right", + "name": "IMAGE_5", + "type": "IMAGE", + "links": [] + }, + { + "label": "bottom_left", + "name": "IMAGE_6", + "type": "IMAGE", + "links": [] + }, + { + "label": "bottom_center", + "name": "IMAGE_7", + "type": "IMAGE", + "links": [] + }, + { + "label": "bottom_right", + "name": "IMAGE_8", + "type": "IMAGE", + "links": [] + }, + { + "label": "images", + "name": "IMAGE_9", + "type": "IMAGE", + "links": [] + } + ], + "properties": { + "proxyWidgets": [], + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1" + }, + "widgets_values": [], + "title": "Crop Images 3x3" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "7fd47bca-ff89-476c-a98d-ca6f7cf756fe", + "version": 1, + "state": { + "lastGroupId": 3, + "lastNodeId": 142, + "lastLinkId": 245, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Crop Images 3x3", + "inputNode": { + "id": -10, + "bounding": [ + -710, + 5440, + 120, + 60 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 3430, + 5270, + 121.720703125, + 240 + ] + }, + "inputs": [ + { + "id": "e54e8e8b-6ce6-4f80-a38f-87a77d990efc", + "name": "image", + "type": "IMAGE", + "linkIds": [ + 74, + 75, + 82, + 91, + 94, + 117, + 129, + 137, + 148, + 157 + ], + "localized_name": "image", + "pos": [ + -610, + 5460 + ] + } + ], + "outputs": [ + { + "id": "3dd8abe2-a7da-4052-a556-9ae157ff3cf4", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 101 + ], + "localized_name": "IMAGE", + "label": "top_left", + "pos": [ + 3450, + 5290 + ] + }, + { + "id": "aa220733-759b-474e-9d29-634a3a23c5da", + "name": "IMAGE_1", + "type": "IMAGE", + "linkIds": [ + 192 + ], + "label": "top_center", + "pos": [ + 3450, + 5310 + ] + }, + { + "id": "f1911df1-d50c-4bf8-9623-5e581d2a8902", + "name": "IMAGE_2", + "type": "IMAGE", + "linkIds": [ + 193 + ], + "label": "top_right", + "pos": [ + 3450, + 5330 + ] + }, + { + "id": "71ebb807-e7e9-438f-990d-511e0745d10d", + "name": "IMAGE_3", + "type": "IMAGE", + "linkIds": [ + 194 + ], + "label": "middle_left", + "pos": [ + 3450, + 5350 + ] + }, + { + "id": "4fb9c99c-3340-4de5-ba2d-51a653aab0b3", + "name": "IMAGE_4", + "type": "IMAGE", + "linkIds": [ + 195 + ], + "label": "middle_center", + "pos": [ + 3450, + 5370 + ] + }, + { + "id": "398643e8-e349-4d59-9c68-6403b7a2772d", + "name": "IMAGE_5", + "type": "IMAGE", + "linkIds": [ + 196 + ], + "label": "middle_right", + "pos": [ + 3450, + 5390 + ] + }, + { + "id": "5b11949c-f4cc-4525-86ae-690e30d3dada", + "name": "IMAGE_6", + "type": "IMAGE", + "linkIds": [ + 197 + ], + "label": "bottom_left", + "pos": [ + 3450, + 5410 + ] + }, + { + "id": "82c69fd9-de36-4c8f-8311-a9e49159640b", + "name": "IMAGE_7", + "type": "IMAGE", + "linkIds": [ + 198 + ], + "label": "bottom_center", + "pos": [ + 3450, + 5430 + ] + }, + { + "id": "aef678db-20aa-47d4-be8a-978065f078c6", + "name": "IMAGE_8", + "type": "IMAGE", + "linkIds": [ + 199 + ], + "label": "bottom_right", + "pos": [ + 3450, + 5450 + ] + }, + { + "id": "77574277-edde-439c-8720-7daa849f4f27", + "name": "IMAGE_9", + "type": "IMAGE", + "linkIds": [ + 226 + ], + "label": "images", + "pos": [ + 3450, + 5470 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 50, + "type": "ComfyMathExpression", + "pos": [ + 770, + 5310 + ], + "size": [ + 370, + 190 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 73 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": 108 + }, + { + "label": "c", + "localized_name": "values.c", + "name": "values.c", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 77, + 85, + 89, + 97, + 99, + 127, + 142, + 146, + 152, + 300 + ] + } + ], + "title": "Math Expression （Width）", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ComfyMathExpression" + }, + "widgets_values": [ + "max(1, int(a/b))" + ] + }, + { + "id": 51, + "type": "GetImageSize", + "pos": [ + 440, + 5390 + ], + "size": [ + 230, + 120 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 74 + } + ], + "outputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "links": [ + 73, + 300 + ] + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "links": [ + 79, + 305 + ] + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "links": null + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "GetImageSize" + } + }, + { + "id": 52, + "type": "PrimitiveInt", + "pos": [ + 440, + 5590 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 80, + 108 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "PrimitiveInt" + }, + "widgets_values": [ + 3, + "fixed" + ] + }, + { + "id": 53, + "type": "ImageCropV2", + "pos": [ + 2080, + 3020 + ], + "size": [ + 300, + 480 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 75 + }, + { + "localized_name": "crop_region", + "name": "crop_region", + "type": "BOUNDING_BOX", + "widget": { + "name": "crop_region" + }, + "link": 76 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 101, + 227 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ImageCropV2" + }, + "widgets_values": [ + { + "x": 0, + "y": 0, + "width": 512, + "height": 512 + }, + 0, + 0, + 512, + 512 + ] + }, + { + "id": 54, + "type": "PrimitiveBoundingBox", + "pos": [ + 1740, + 3160 + ], + "size": [ + 270, + 200 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "x", + "name": "x", + "type": "INT", + "widget": { + "name": "x" + }, + "link": null + }, + { + "localized_name": "y", + "name": "y", + "type": "INT", + "widget": { + "name": "y" + }, + "link": null + }, + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 77 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 78 + } + ], + "outputs": [ + { + "localized_name": "BOUNDING_BOX", + "name": "BOUNDING_BOX", + "type": "BOUNDING_BOX", + "links": [ + 76 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "PrimitiveBoundingBox" + }, + "widgets_values": [ + 0, + 0, + 512, + 512 + ] + }, + { + "id": 55, + "type": "ComfyMathExpression", + "pos": [ + 780, + 5570 + ], + "size": [ + 370, + 190 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 79 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": 80 + }, + { + "label": "c", + "localized_name": "values.c", + "name": "values.c", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 78, + 84, + 86, + 88, + 90, + 98, + 100, + 121, + 123, + 126, + 161 + ] + } + ], + "title": "Math Expression(Height)", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ComfyMathExpression" + }, + "widgets_values": [ + "max(1, int(a/b))" + ] + }, + { + "id": 57, + "type": "ImageCropV2", + "pos": [ + 2080, + 4700 + ], + "size": [ + 300, + 480 + ], + "flags": {}, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 82 + }, + { + "localized_name": "crop_region", + "name": "crop_region", + "type": "BOUNDING_BOX", + "widget": { + "name": "crop_region" + }, + "link": 83 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 194, + 230 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ImageCropV2" + }, + "widgets_values": [ + { + "x": 0, + "y": 0, + "width": 512, + "height": 512 + }, + 0, + 0, + 512, + 512 + ] + }, + { + "id": 58, + "type": "PrimitiveBoundingBox", + "pos": [ + 1740, + 4830 + ], + "size": [ + 270, + 200 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "x", + "name": "x", + "type": "INT", + "widget": { + "name": "x" + }, + "link": null + }, + { + "localized_name": "y", + "name": "y", + "type": "INT", + "widget": { + "name": "y" + }, + "link": 84 + }, + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 85 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 86 + } + ], + "outputs": [ + { + "localized_name": "BOUNDING_BOX", + "name": "BOUNDING_BOX", + "type": "BOUNDING_BOX", + "links": [ + 83 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "PrimitiveBoundingBox" + }, + "widgets_values": [ + 0, + 0, + 512, + 512 + ] + }, + { + "id": 60, + "type": "PrimitiveBoundingBox", + "pos": [ + 1740, + 3700 + ], + "size": [ + 270, + 200 + ], + "flags": {}, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "x", + "name": "x", + "type": "INT", + "widget": { + "name": "x" + }, + "link": 88 + }, + { + "localized_name": "y", + "name": "y", + "type": "INT", + "widget": { + "name": "y" + }, + "link": null + }, + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 89 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 90 + } + ], + "outputs": [ + { + "localized_name": "BOUNDING_BOX", + "name": "BOUNDING_BOX", + "type": "BOUNDING_BOX", + "links": [ + 92 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "PrimitiveBoundingBox" + }, + "widgets_values": [ + 6, + 0, + 512, + 512 + ] + }, + { + "id": 61, + "type": "ImageCropV2", + "pos": [ + 2100, + 3570 + ], + "size": [ + 300, + 480 + ], + "flags": {}, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 91 + }, + { + "localized_name": "crop_region", + "name": "crop_region", + "type": "BOUNDING_BOX", + "widget": { + "name": "crop_region" + }, + "link": 92 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 192, + 228 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ImageCropV2" + }, + "widgets_values": [ + { + "x": 0, + "y": 0, + "width": 512, + "height": 512 + }, + 0, + 0, + 512, + 512 + ] + }, + { + "id": 63, + "type": "ImageCropV2", + "pos": [ + 2080, + 5310 + ], + "size": [ + 300, + 480 + ], + "flags": {}, + "order": 10, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 94 + }, + { + "localized_name": "crop_region", + "name": "crop_region", + "type": "BOUNDING_BOX", + "widget": { + "name": "crop_region" + }, + "link": 95 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 195, + 231 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ImageCropV2" + }, + "widgets_values": [ + { + "x": 0, + "y": 0, + "width": 512, + "height": 512 + }, + 0, + 0, + 512, + 512 + ] + }, + { + "id": 65, + "type": "PrimitiveBoundingBox", + "pos": [ + 1750, + 5330 + ], + "size": [ + 270, + 200 + ], + "flags": {}, + "order": 11, + "mode": 0, + "inputs": [ + { + "localized_name": "x", + "name": "x", + "type": "INT", + "widget": { + "name": "x" + }, + "link": 97 + }, + { + "localized_name": "y", + "name": "y", + "type": "INT", + "widget": { + "name": "y" + }, + "link": 98 + }, + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 99 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 100 + } + ], + "outputs": [ + { + "localized_name": "BOUNDING_BOX", + "name": "BOUNDING_BOX", + "type": "BOUNDING_BOX", + "links": [ + 95 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "PrimitiveBoundingBox" + }, + "widgets_values": [ + 6, + 0, + 512, + 512 + ] + }, + { + "id": 71, + "type": "ComfyMathExpression", + "pos": [ + 780, + 6090 + ], + "size": [ + 400, + 190 + ], + "flags": {}, + "order": 12, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 126 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 136, + 147, + 156, + 306 + ] + } + ], + "title": "Math Expression(height)", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ComfyMathExpression" + }, + "widgets_values": [ + "2 * a" + ] + }, + { + "id": 75, + "type": "ImageCropV2", + "pos": [ + 2100, + 5900 + ], + "size": [ + 300, + 480 + ], + "flags": {}, + "order": 13, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 117 + }, + { + "localized_name": "crop_region", + "name": "crop_region", + "type": "BOUNDING_BOX", + "widget": { + "name": "crop_region" + }, + "link": 118 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 196, + 232 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ImageCropV2" + }, + "widgets_values": [ + { + "x": 0, + "y": 0, + "width": 512, + "height": 512 + }, + 0, + 0, + 512, + 512 + ] + }, + { + "id": 77, + "type": "PrimitiveBoundingBox", + "pos": [ + 1750, + 5970 + ], + "size": [ + 270, + 200 + ], + "flags": {}, + "order": 14, + "mode": 0, + "inputs": [ + { + "localized_name": "x", + "name": "x", + "type": "INT", + "widget": { + "name": "x" + }, + "link": 128 + }, + { + "localized_name": "y", + "name": "y", + "type": "INT", + "widget": { + "name": "y" + }, + "link": 121 + }, + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 302 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 123 + } + ], + "outputs": [ + { + "localized_name": "BOUNDING_BOX", + "name": "BOUNDING_BOX", + "type": "BOUNDING_BOX", + "links": [ + 118 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "PrimitiveBoundingBox" + }, + "widgets_values": [ + 6, + 0, + 512, + 512 + ] + }, + { + "id": 78, + "type": "ComfyMathExpression", + "pos": [ + 780, + 5820 + ], + "size": [ + 400, + 200 + ], + "flags": {}, + "order": 15, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 127 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 128, + 132, + 163, + 301 + ] + } + ], + "title": "Math Expression(width)", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ComfyMathExpression" + }, + "widgets_values": [ + "2 * a" + ] + }, + { + "id": 140, + "type": "ComfyMathExpression", + "pos": [ + 1240, + 5640 + ], + "size": [ + 420, + 190 + ], + "flags": {}, + "order": 24, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 300 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": 301 + }, + { + "label": "c", + "localized_name": "values.c", + "name": "values.c", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 302, + 303, + 304 + ] + } + ], + "title": "Math Expression (Right Width)", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ComfyMathExpression" + }, + "widgets_values": [ + "max(1, a - b)" + ] + }, + { + "id": 141, + "type": "ComfyMathExpression", + "pos": [ + 1230, + 6340 + ], + "size": [ + 420, + 190 + ], + "flags": {}, + "order": 25, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 305 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": 306 + }, + { + "label": "c", + "localized_name": "values.c", + "name": "values.c", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 307, + 308, + 309 + ] + } + ], + "title": "Math Expression (Bottom Height)", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ComfyMathExpression" + }, + "widgets_values": [ + "max(1, a - b)" + ] + }, + { + "id": 79, + "type": "ImageCropV2", + "pos": [ + 2120, + 7580 + ], + "size": [ + 300, + 480 + ], + "flags": {}, + "order": 16, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 129 + }, + { + "localized_name": "crop_region", + "name": "crop_region", + "type": "BOUNDING_BOX", + "widget": { + "name": "crop_region" + }, + "link": 130 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 199, + 235 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ImageCropV2" + }, + "widgets_values": [ + { + "x": 0, + "y": 0, + "width": 512, + "height": 512 + }, + 0, + 0, + 512, + 512 + ] + }, + { + "id": 81, + "type": "PrimitiveBoundingBox", + "pos": [ + 1720, + 7620 + ], + "size": [ + 270, + 200 + ], + "flags": {}, + "order": 17, + "mode": 0, + "inputs": [ + { + "localized_name": "x", + "name": "x", + "type": "INT", + "widget": { + "name": "x" + }, + "link": 132 + }, + { + "localized_name": "y", + "name": "y", + "type": "INT", + "widget": { + "name": "y" + }, + "link": 136 + }, + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 303 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 307 + } + ], + "outputs": [ + { + "localized_name": "BOUNDING_BOX", + "name": "BOUNDING_BOX", + "type": "BOUNDING_BOX", + "links": [ + 130 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "PrimitiveBoundingBox" + }, + "widgets_values": [ + 6, + 0, + 512, + 512 + ] + }, + { + "id": 82, + "type": "ImageCropV2", + "pos": [ + 2120, + 7040 + ], + "size": [ + 300, + 480 + ], + "flags": {}, + "order": 18, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 137 + }, + { + "localized_name": "crop_region", + "name": "crop_region", + "type": "BOUNDING_BOX", + "widget": { + "name": "crop_region" + }, + "link": 138 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 198, + 234 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ImageCropV2" + }, + "widgets_values": [ + { + "x": 0, + "y": 0, + "width": 512, + "height": 512 + }, + 0, + 0, + 512, + 512 + ] + }, + { + "id": 84, + "type": "PrimitiveBoundingBox", + "pos": [ + 1720, + 7080 + ], + "size": [ + 270, + 200 + ], + "flags": {}, + "order": 19, + "mode": 0, + "inputs": [ + { + "localized_name": "x", + "name": "x", + "type": "INT", + "widget": { + "name": "x" + }, + "link": 146 + }, + { + "localized_name": "y", + "name": "y", + "type": "INT", + "widget": { + "name": "y" + }, + "link": 147 + }, + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 142 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 308 + } + ], + "outputs": [ + { + "localized_name": "BOUNDING_BOX", + "name": "BOUNDING_BOX", + "type": "BOUNDING_BOX", + "links": [ + 138 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "PrimitiveBoundingBox" + }, + "widgets_values": [ + 6, + 0, + 512, + 512 + ] + }, + { + "id": 85, + "type": "ImageCropV2", + "pos": [ + 2110, + 6480 + ], + "size": [ + 300, + 480 + ], + "flags": {}, + "order": 20, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 148 + }, + { + "localized_name": "crop_region", + "name": "crop_region", + "type": "BOUNDING_BOX", + "widget": { + "name": "crop_region" + }, + "link": 149 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 197, + 233 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ImageCropV2" + }, + "widgets_values": [ + { + "x": 0, + "y": 0, + "width": 512, + "height": 512 + }, + 0, + 0, + 512, + 512 + ] + }, + { + "id": 86, + "type": "PrimitiveBoundingBox", + "pos": [ + 1670, + 6570 + ], + "size": [ + 270, + 200 + ], + "flags": {}, + "order": 21, + "mode": 0, + "inputs": [ + { + "localized_name": "x", + "name": "x", + "type": "INT", + "widget": { + "name": "x" + }, + "link": null + }, + { + "localized_name": "y", + "name": "y", + "type": "INT", + "widget": { + "name": "y" + }, + "link": 156 + }, + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 152 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 309 + } + ], + "outputs": [ + { + "localized_name": "BOUNDING_BOX", + "name": "BOUNDING_BOX", + "type": "BOUNDING_BOX", + "links": [ + 149 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "PrimitiveBoundingBox" + }, + "widgets_values": [ + 0, + 0, + 512, + 512 + ] + }, + { + "id": 88, + "type": "ImageCropV2", + "pos": [ + 2060, + 4140 + ], + "size": [ + 300, + 480 + ], + "flags": {}, + "order": 22, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 157 + }, + { + "localized_name": "crop_region", + "name": "crop_region", + "type": "BOUNDING_BOX", + "widget": { + "name": "crop_region" + }, + "link": 158 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 193, + 229 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ImageCropV2" + }, + "widgets_values": [ + { + "x": 0, + "y": 0, + "width": 512, + "height": 512 + }, + 0, + 0, + 512, + 512 + ] + }, + { + "id": 89, + "type": "PrimitiveBoundingBox", + "pos": [ + 1720, + 4150 + ], + "size": [ + 270, + 200 + ], + "flags": {}, + "order": 23, + "mode": 0, + "inputs": [ + { + "localized_name": "x", + "name": "x", + "type": "INT", + "widget": { + "name": "x" + }, + "link": 163 + }, + { + "localized_name": "y", + "name": "y", + "type": "INT", + "widget": { + "name": "y" + }, + "link": null + }, + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 304 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 161 + } + ], + "outputs": [ + { + "localized_name": "BOUNDING_BOX", + "name": "BOUNDING_BOX", + "type": "BOUNDING_BOX", + "links": [ + 158 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "PrimitiveBoundingBox" + }, + "widgets_values": [ + 6, + 0, + 512, + 512 + ] + }, + { + "id": 136, + "type": "BatchImagesNode", + "pos": [ + 3170, + 5640 + ], + "size": [ + 230, + 290 + ], + "flags": {}, + "order": 24, + "mode": 0, + "inputs": [ + { + "label": "image0", + "localized_name": "images.image0", + "name": "images.image0", + "type": "IMAGE", + "link": 227 + }, + { + "label": "image1", + "localized_name": "images.image1", + "name": "images.image1", + "type": "IMAGE", + "link": 228 + }, + { + "label": "image2", + "localized_name": "images.image2", + "name": "images.image2", + "shape": 7, + "type": "IMAGE", + "link": 229 + }, + { + "label": "image3", + "localized_name": "images.image3", + "name": "images.image3", + "shape": 7, + "type": "IMAGE", + "link": 230 + }, + { + "label": "image4", + "localized_name": "images.image4", + "name": "images.image4", + "shape": 7, + "type": "IMAGE", + "link": 231 + }, + { + "label": "image5", + "localized_name": "images.image5", + "name": "images.image5", + "shape": 7, + "type": "IMAGE", + "link": 232 + }, + { + "label": "image6", + "localized_name": "images.image6", + "name": "images.image6", + "shape": 7, + "type": "IMAGE", + "link": 233 + }, + { + "label": "image7", + "localized_name": "images.image7", + "name": "images.image7", + "shape": 7, + "type": "IMAGE", + "link": 234 + }, + { + "label": "image8", + "localized_name": "images.image8", + "name": "images.image8", + "shape": 7, + "type": "IMAGE", + "link": 235 + }, + { + "label": "image9", + "localized_name": "images.image9", + "name": "images.image9", + "shape": 7, + "type": "IMAGE", + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 226 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "BatchImagesNode" + } + } + ], + "groups": [ + { + "id": 3, + "title": "Crop Images 3x3", + "bounding": [ + 100, + 2700, + 2640, + 5480 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 73, + "origin_id": 51, + "origin_slot": 0, + "target_id": 50, + "target_slot": 0, + "type": "INT" + }, + { + "id": 108, + "origin_id": 52, + "origin_slot": 0, + "target_id": 50, + "target_slot": 1, + "type": "INT" + }, + { + "id": 76, + "origin_id": 54, + "origin_slot": 0, + "target_id": 53, + "target_slot": 1, + "type": "BOUNDING_BOX" + }, + { + "id": 77, + "origin_id": 50, + "origin_slot": 1, + "target_id": 54, + "target_slot": 2, + "type": "INT" + }, + { + "id": 78, + "origin_id": 55, + "origin_slot": 1, + "target_id": 54, + "target_slot": 3, + "type": "INT" + }, + { + "id": 79, + "origin_id": 51, + "origin_slot": 1, + "target_id": 55, + "target_slot": 0, + "type": "INT" + }, + { + "id": 80, + "origin_id": 52, + "origin_slot": 0, + "target_id": 55, + "target_slot": 1, + "type": "INT" + }, + { + "id": 83, + "origin_id": 58, + "origin_slot": 0, + "target_id": 57, + "target_slot": 1, + "type": "BOUNDING_BOX" + }, + { + "id": 84, + "origin_id": 55, + "origin_slot": 1, + "target_id": 58, + "target_slot": 1, + "type": "INT" + }, + { + "id": 85, + "origin_id": 50, + "origin_slot": 1, + "target_id": 58, + "target_slot": 2, + "type": "INT" + }, + { + "id": 86, + "origin_id": 55, + "origin_slot": 1, + "target_id": 58, + "target_slot": 3, + "type": "INT" + }, + { + "id": 88, + "origin_id": 50, + "origin_slot": 1, + "target_id": 60, + "target_slot": 0, + "type": "INT" + }, + { + "id": 89, + "origin_id": 50, + "origin_slot": 1, + "target_id": 60, + "target_slot": 2, + "type": "INT" + }, + { + "id": 90, + "origin_id": 55, + "origin_slot": 1, + "target_id": 60, + "target_slot": 3, + "type": "INT" + }, + { + "id": 92, + "origin_id": 60, + "origin_slot": 0, + "target_id": 61, + "target_slot": 1, + "type": "BOUNDING_BOX" + }, + { + "id": 95, + "origin_id": 65, + "origin_slot": 0, + "target_id": 63, + "target_slot": 1, + "type": "BOUNDING_BOX" + }, + { + "id": 97, + "origin_id": 50, + "origin_slot": 1, + "target_id": 65, + "target_slot": 0, + "type": "INT" + }, + { + "id": 98, + "origin_id": 55, + "origin_slot": 1, + "target_id": 65, + "target_slot": 1, + "type": "INT" + }, + { + "id": 99, + "origin_id": 50, + "origin_slot": 1, + "target_id": 65, + "target_slot": 2, + "type": "INT" + }, + { + "id": 100, + "origin_id": 55, + "origin_slot": 1, + "target_id": 65, + "target_slot": 3, + "type": "INT" + }, + { + "id": 126, + "origin_id": 55, + "origin_slot": 1, + "target_id": 71, + "target_slot": 0, + "type": "INT" + }, + { + "id": 118, + "origin_id": 77, + "origin_slot": 0, + "target_id": 75, + "target_slot": 1, + "type": "BOUNDING_BOX" + }, + { + "id": 128, + "origin_id": 78, + "origin_slot": 1, + "target_id": 77, + "target_slot": 0, + "type": "INT" + }, + { + "id": 121, + "origin_id": 55, + "origin_slot": 1, + "target_id": 77, + "target_slot": 1, + "type": "INT" + }, + { + "id": 302, + "origin_id": 140, + "origin_slot": 1, + "target_id": 77, + "target_slot": 2, + "type": "INT" + }, + { + "id": 123, + "origin_id": 55, + "origin_slot": 1, + "target_id": 77, + "target_slot": 3, + "type": "INT" + }, + { + "id": 127, + "origin_id": 50, + "origin_slot": 1, + "target_id": 78, + "target_slot": 0, + "type": "INT" + }, + { + "id": 130, + "origin_id": 81, + "origin_slot": 0, + "target_id": 79, + "target_slot": 1, + "type": "BOUNDING_BOX" + }, + { + "id": 132, + "origin_id": 78, + "origin_slot": 1, + "target_id": 81, + "target_slot": 0, + "type": "INT" + }, + { + "id": 136, + "origin_id": 71, + "origin_slot": 1, + "target_id": 81, + "target_slot": 1, + "type": "INT" + }, + { + "id": 303, + "origin_id": 140, + "origin_slot": 1, + "target_id": 81, + "target_slot": 2, + "type": "INT" + }, + { + "id": 307, + "origin_id": 141, + "origin_slot": 1, + "target_id": 81, + "target_slot": 3, + "type": "INT" + }, + { + "id": 138, + "origin_id": 84, + "origin_slot": 0, + "target_id": 82, + "target_slot": 1, + "type": "BOUNDING_BOX" + }, + { + "id": 146, + "origin_id": 50, + "origin_slot": 1, + "target_id": 84, + "target_slot": 0, + "type": "INT" + }, + { + "id": 147, + "origin_id": 71, + "origin_slot": 1, + "target_id": 84, + "target_slot": 1, + "type": "INT" + }, + { + "id": 142, + "origin_id": 50, + "origin_slot": 1, + "target_id": 84, + "target_slot": 2, + "type": "INT" + }, + { + "id": 308, + "origin_id": 141, + "origin_slot": 1, + "target_id": 84, + "target_slot": 3, + "type": "INT" + }, + { + "id": 149, + "origin_id": 86, + "origin_slot": 0, + "target_id": 85, + "target_slot": 1, + "type": "BOUNDING_BOX" + }, + { + "id": 156, + "origin_id": 71, + "origin_slot": 1, + "target_id": 86, + "target_slot": 1, + "type": "INT" + }, + { + "id": 152, + "origin_id": 50, + "origin_slot": 1, + "target_id": 86, + "target_slot": 2, + "type": "INT" + }, + { + "id": 309, + "origin_id": 141, + "origin_slot": 1, + "target_id": 86, + "target_slot": 3, + "type": "INT" + }, + { + "id": 158, + "origin_id": 89, + "origin_slot": 0, + "target_id": 88, + "target_slot": 1, + "type": "BOUNDING_BOX" + }, + { + "id": 163, + "origin_id": 78, + "origin_slot": 1, + "target_id": 89, + "target_slot": 0, + "type": "INT" + }, + { + "id": 304, + "origin_id": 140, + "origin_slot": 1, + "target_id": 89, + "target_slot": 2, + "type": "INT" + }, + { + "id": 161, + "origin_id": 55, + "origin_slot": 1, + "target_id": 89, + "target_slot": 3, + "type": "INT" + }, + { + "id": 300, + "origin_id": 51, + "origin_slot": 0, + "target_id": 140, + "target_slot": 0, + "type": "INT" + }, + { + "id": 301, + "origin_id": 78, + "origin_slot": 1, + "target_id": 140, + "target_slot": 1, + "type": "INT" + }, + { + "id": 305, + "origin_id": 51, + "origin_slot": 1, + "target_id": 141, + "target_slot": 0, + "type": "INT" + }, + { + "id": 306, + "origin_id": 71, + "origin_slot": 1, + "target_id": 141, + "target_slot": 1, + "type": "INT" + }, + { + "id": 74, + "origin_id": -10, + "origin_slot": 0, + "target_id": 51, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 75, + "origin_id": -10, + "origin_slot": 0, + "target_id": 53, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 82, + "origin_id": -10, + "origin_slot": 0, + "target_id": 57, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 91, + "origin_id": -10, + "origin_slot": 0, + "target_id": 61, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 94, + "origin_id": -10, + "origin_slot": 0, + "target_id": 63, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 117, + "origin_id": -10, + "origin_slot": 0, + "target_id": 75, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 129, + "origin_id": -10, + "origin_slot": 0, + "target_id": 79, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 137, + "origin_id": -10, + "origin_slot": 0, + "target_id": 82, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 148, + "origin_id": -10, + "origin_slot": 0, + "target_id": 85, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 157, + "origin_id": -10, + "origin_slot": 0, + "target_id": 88, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 101, + "origin_id": 53, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 192, + "origin_id": 61, + "origin_slot": 0, + "target_id": -20, + "target_slot": 1, + "type": "IMAGE" + }, + { + "id": 193, + "origin_id": 88, + "origin_slot": 0, + "target_id": -20, + "target_slot": 2, + "type": "IMAGE" + }, + { + "id": 194, + "origin_id": 57, + "origin_slot": 0, + "target_id": -20, + "target_slot": 3, + "type": "IMAGE" + }, + { + "id": 195, + "origin_id": 63, + "origin_slot": 0, + "target_id": -20, + "target_slot": 4, + "type": "IMAGE" + }, + { + "id": 196, + "origin_id": 75, + "origin_slot": 0, + "target_id": -20, + "target_slot": 5, + "type": "IMAGE" + }, + { + "id": 197, + "origin_id": 85, + "origin_slot": 0, + "target_id": -20, + "target_slot": 6, + "type": "IMAGE" + }, + { + "id": 198, + "origin_id": 82, + "origin_slot": 0, + "target_id": -20, + "target_slot": 7, + "type": "IMAGE" + }, + { + "id": 199, + "origin_id": 79, + "origin_slot": 0, + "target_id": -20, + "target_slot": 8, + "type": "IMAGE" + }, + { + "id": 226, + "origin_id": 136, + "origin_slot": 0, + "target_id": -20, + "target_slot": 9, + "type": "IMAGE" + }, + { + "id": 227, + "origin_id": 53, + "origin_slot": 0, + "target_id": 136, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 228, + "origin_id": 61, + "origin_slot": 0, + "target_id": 136, + "target_slot": 1, + "type": "IMAGE" + }, + { + "id": 229, + "origin_id": 88, + "origin_slot": 0, + "target_id": 136, + "target_slot": 2, + "type": "IMAGE" + }, + { + "id": 230, + "origin_id": 57, + "origin_slot": 0, + "target_id": 136, + "target_slot": 3, + "type": "IMAGE" + }, + { + "id": 231, + "origin_id": 63, + "origin_slot": 0, + "target_id": 136, + "target_slot": 4, + "type": "IMAGE" + }, + { + "id": 232, + "origin_id": 75, + "origin_slot": 0, + "target_id": 136, + "target_slot": 5, + "type": "IMAGE" + }, + { + "id": 233, + "origin_id": 85, + "origin_slot": 0, + "target_id": 136, + "target_slot": 6, + "type": "IMAGE" + }, + { + "id": 234, + "origin_id": 82, + "origin_slot": 0, + "target_id": 136, + "target_slot": 7, + "type": "IMAGE" + }, + { + "id": 235, + "origin_id": 79, + "origin_slot": 0, + "target_id": 136, + "target_slot": 8, + "type": "IMAGE" + } + ], + "extra": {}, + "category": "Image Tools/Crop", + "description": "Splits an image into a 3×3 grid of nine equal tiles." + } + ] + }, + "extra": { + "ue_links": [], + "links_added_by_ue": [] + } +} \ No newline at end of file diff --git a/blueprints/Depth to Image (Z-Image-Turbo).json b/blueprints/Depth to Image (Z-Image-Turbo).json index 0b657534f..fe9ef0f72 100644 --- a/blueprints/Depth to Image (Z-Image-Turbo).json +++ b/blueprints/Depth to Image (Z-Image-Turbo).json @@ -160,7 +160,7 @@ }, "revision": 0, "config": {}, - "name": "local-Depth to Image (Z-Image-Turbo)", + "name": "Depth to Image (Z-Image-Turbo)", "inputNode": { "id": -10, "bounding": [ @@ -1579,7 +1579,8 @@ "VHS_MetadataImage": true, "VHS_KeepIntermediate": true }, - "category": "Image generation and editing/Depth to image" + "category": "Image generation and editing/Depth to image", + "description": "Generates an image from a depth map using Z-Image-Turbo with text conditioning." }, { "id": "458bdf3c-4b58-421c-af50-c9c663a4d74c", @@ -2461,7 +2462,8 @@ ] }, "workflowRendererVersion": "LG" - } + }, + "description": "Estimates a monocular depth map from an input image using the Lotus depth estimation model." } ] }, @@ -2482,4 +2484,4 @@ "VHS_KeepIntermediate": true }, "version": 0.4 -} +} \ No newline at end of file diff --git a/blueprints/Depth to Video (ltx 2.0).json b/blueprints/Depth to Video (ltx 2.0).json index 98c39eea5..bd51e4476 100644 --- a/blueprints/Depth to Video (ltx 2.0).json +++ b/blueprints/Depth to Video (ltx 2.0).json @@ -261,7 +261,7 @@ }, "revision": 0, "config": {}, - "name": "local-Depth to Video (LTX 2.0)", + "name": "Depth to Video (LTX 2.0)", "inputNode": { "id": -10, "bounding": [ @@ -4233,7 +4233,8 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Video generation and editing/Depth to video" + "category": "Video generation and editing/Depth to video", + "description": "Generates depth-controlled video with LTX-2: motion and structure follow a depth-reference video alongside text prompting, optional first-frame image conditioning, with optional synchronized audio." }, { "id": "38b60539-50a7-42f9-a5fe-bdeca26272e2", @@ -5192,7 +5193,8 @@ ], "extra": { "workflowRendererVersion": "LG" - } + }, + "description": "Estimates a monocular depth map from an input image using the Lotus depth estimation model." } ] }, @@ -5208,4 +5210,4 @@ "workflowRendererVersion": "LG" }, "version": 0.4 -} +} \ No newline at end of file diff --git a/blueprints/Edge-Preserving Blur.json b/blueprints/Edge-Preserving Blur.json index 18012beb1..fbda9f126 100644 --- a/blueprints/Edge-Preserving Blur.json +++ b/blueprints/Edge-Preserving Blur.json @@ -450,9 +450,10 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image Tools/Blur" + "category": "Image Tools/Blur", + "description": "Applies bilateral (edge-preserving) blur to soften images while retaining detail." } ] }, "extra": {} -} +} \ No newline at end of file diff --git a/blueprints/Film Grain.json b/blueprints/Film Grain.json index a680b3ece..3226ea9aa 100644 --- a/blueprints/Film Grain.json +++ b/blueprints/Film Grain.json @@ -580,8 +580,9 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image Tools/Color adjust" + "category": "Image Tools/Color adjust", + "description": "Adds procedural film grain texture for a cinematic look via GPU fragment shader." } ] } -} +} \ No newline at end of file diff --git a/blueprints/First-Last-Frame to Video (LTX-2.3).json b/blueprints/First-Last-Frame to Video (LTX-2.3).json new file mode 100644 index 000000000..f509aefe0 --- /dev/null +++ b/blueprints/First-Last-Frame to Video (LTX-2.3).json @@ -0,0 +1,3361 @@ +{ + "revision": 0, + "last_node_id": 228, + "last_link_id": 0, + "nodes": [ + { + "id": 228, + "type": "a5982aee-8136-4819-86a0-cf9d9e510ad6", + "pos": [ + 1490, + 4730 + ], + "size": [ + 274.8169921875, + 276 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "label": "first_frame", + "localized_name": "input", + "name": "input", + "type": "IMAGE,MASK", + "link": null + }, + { + "label": "last_frame", + "localized_name": "input_1", + "name": "input_1", + "type": "IMAGE,MASK", + "link": null + }, + { + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + }, + { + "label": "width", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": null + }, + { + "label": "height", + "name": "value_1", + "type": "INT", + "widget": { + "name": "value_1" + }, + "link": null + }, + { + "label": "duration", + "name": "value_2", + "type": "INT", + "widget": { + "name": "value_2" + }, + "link": null + }, + { + "label": "fps", + "name": "value_3", + "type": "INT", + "widget": { + "name": "value_3" + }, + "link": null + }, + { + "name": "noise_seed", + "type": "INT", + "widget": { + "name": "noise_seed" + }, + "link": null + }, + { + "label": "ckpt_name", + "name": "ckpt_name_1", + "type": "COMBO", + "widget": { + "name": "ckpt_name_1" + }, + "link": null + }, + { + "name": "text_encoder", + "type": "COMBO", + "widget": { + "name": "text_encoder" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "VIDEO", + "name": "VIDEO", + "type": "VIDEO", + "links": [] + } + ], + "properties": { + "proxyWidgets": [ + [ + "222", + "text" + ], + [ + "215", + "value" + ], + [ + "216", + "value" + ], + [ + "198", + "value" + ], + [ + "205", + "value" + ], + [ + "196", + "noise_seed" + ], + [ + "224", + "ckpt_name" + ], + [ + "225", + "text_encoder" + ] + ], + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1" + }, + "widgets_values": [], + "title": "First-Last-Frame to Video (LTX-2.3)" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "a5982aee-8136-4819-86a0-cf9d9e510ad6", + "version": 1, + "state": { + "lastGroupId": 22, + "lastNodeId": 228, + "lastLinkId": 276, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "First-Last-Frame to Video (LTX-2.3)", + "inputNode": { + "id": -10, + "bounding": [ + 270, + 3100, + 120, + 240 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 3620, + 3120, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "6fe179c4-d96f-4383-b202-844f6de4922e", + "name": "input", + "type": "IMAGE,MASK", + "linkIds": [ + 251 + ], + "localized_name": "input", + "label": "first_frame", + "pos": [ + 370, + 3120 + ] + }, + { + "id": "e80df1ae-5f39-4f86-91bd-0467635e2f2d", + "name": "input_1", + "type": "IMAGE,MASK", + "linkIds": [ + 253 + ], + "localized_name": "input_1", + "label": "last_frame", + "pos": [ + 370, + 3140 + ] + }, + { + "id": "433148fa-bf73-4ab1-81d9-09e2e38ed861", + "name": "text", + "type": "STRING", + "linkIds": [ + 265 + ], + "pos": [ + 370, + 3160 + ] + }, + { + "id": "36915bc8-a6ed-4d48-8619-e0e8723228e9", + "name": "value", + "type": "INT", + "linkIds": [ + 266 + ], + "label": "width", + "pos": [ + 370, + 3180 + ] + }, + { + "id": "425a36b8-91ab-41b7-81e9-496eba064ec8", + "name": "value_1", + "type": "INT", + "linkIds": [ + 267 + ], + "label": "height", + "pos": [ + 370, + 3200 + ] + }, + { + "id": "0c9e003b-bd07-4b7d-aa6d-789e138ed161", + "name": "value_2", + "type": "INT", + "linkIds": [ + 268 + ], + "label": "duration", + "pos": [ + 370, + 3220 + ] + }, + { + "id": "581b52ff-21c5-4774-ac2a-8f69a7e09e2e", + "name": "value_3", + "type": "INT", + "linkIds": [ + 269 + ], + "label": "fps", + "pos": [ + 370, + 3240 + ] + }, + { + "id": "d03cc171-45da-4658-99aa-77252bbcf522", + "name": "noise_seed", + "type": "INT", + "linkIds": [ + 270 + ], + "pos": [ + 370, + 3260 + ] + }, + { + "id": "e68e61c8-905e-43ac-8c76-65ac52270a08", + "name": "ckpt_name_1", + "type": "COMBO", + "linkIds": [ + 272, + 275, + 276 + ], + "label": "ckpt_name", + "pos": [ + 370, + 3280 + ] + }, + { + "id": "5d065f3b-891b-499f-950b-c2df0be24536", + "name": "text_encoder", + "type": "COMBO", + "linkIds": [ + 273 + ], + "pos": [ + 370, + 3300 + ] + } + ], + "outputs": [ + { + "id": "0c8c2dc0-c67c-4bc2-9e57-6aa00db2e3a9", + "name": "VIDEO", + "type": "VIDEO", + "linkIds": [ + 252 + ], + "localized_name": "VIDEO", + "pos": [ + 3640, + 3140 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 195, + "type": "LTXVPreprocess", + "pos": [ + 1480, + 3780 + ], + "size": [ + 230, + 110 + ], + "flags": { + "collapsed": false + }, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 203 + }, + { + "localized_name": "img_compression", + "name": "img_compression", + "type": "INT", + "widget": { + "name": "img_compression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "output_image", + "name": "output_image", + "type": "IMAGE", + "links": [ + 229 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "LTXVPreprocess", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 25 + ] + }, + { + "id": 196, + "type": "RandomNoise", + "pos": [ + 1990, + 2320 + ], + "size": [ + 280, + 110 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "noise_seed", + "name": "noise_seed", + "type": "INT", + "widget": { + "name": "noise_seed" + }, + "link": 270 + } + ], + "outputs": [ + { + "localized_name": "NOISE", + "name": "NOISE", + "type": "NOISE", + "links": [ + 246 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": { + "noise_seed": true + }, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.14.1", + "Node name for S&R": "RandomNoise", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 315253765879496, + "randomize" + ] + }, + { + "id": 197, + "type": "LTXVEmptyLatentAudio", + "pos": [ + 2090, + 3820 + ], + "size": [ + 280, + 170 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "audio_vae", + "name": "audio_vae", + "type": "VAE", + "link": 205 + }, + { + "localized_name": "frames_number", + "name": "frames_number", + "type": "INT", + "widget": { + "name": "frames_number" + }, + "link": 262 + }, + { + "localized_name": "frame_rate", + "name": "frame_rate", + "type": "INT", + "widget": { + "name": "frame_rate" + }, + "link": 207 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "Latent", + "name": "Latent", + "type": "LATENT", + "links": [ + 245 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.68", + "Node name for S&R": "LTXVEmptyLatentAudio", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 97, + 25, + 1 + ] + }, + { + "id": 198, + "type": "PrimitiveInt", + "pos": [ + 760, + 3650 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": 268 + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 260 + ] + } + ], + "title": "Duration", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 5, + "fixed" + ] + }, + { + "id": 199, + "type": "LTXVPreprocess", + "pos": [ + 1480, + 3340 + ], + "size": [ + 230, + 110 + ], + "flags": { + "collapsed": false + }, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 210 + }, + { + "localized_name": "img_compression", + "name": "img_compression", + "type": "INT", + "widget": { + "name": "img_compression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "output_image", + "name": "output_image", + "type": "IMAGE", + "links": [ + 240 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "LTXVPreprocess", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 25 + ] + }, + { + "id": 200, + "type": "LTXVCropGuides", + "pos": [ + 2820, + 2450 + ], + "size": [ + 280, + 120 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 213 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 214 + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "link": 215 + } + ], + "outputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "links": [] + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "links": [] + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "links": [ + 211 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.5.2" + }, + "cnr_id": "comfy-core", + "ver": "0.8.2", + "Node name for S&R": "LTXVCropGuides", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 201, + "type": "EmptyLTXVLatentVideo", + "pos": [ + 2090, + 3580 + ], + "size": [ + 280, + 200 + ], + "flags": {}, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 218 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 219 + }, + { + "localized_name": "length", + "name": "length", + "type": "INT", + "widget": { + "name": "length" + }, + "link": 263 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "links": [ + 239 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.60", + "Node name for S&R": "EmptyLTXVLatentVideo", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 768, + 512, + 97, + 1 + ] + }, + { + "id": 202, + "type": "LTXVConditioning", + "pos": [ + 2090, + 3400 + ], + "size": [ + 280, + 130 + ], + "flags": {}, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 221 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 222 + }, + { + "localized_name": "frame_rate", + "name": "frame_rate", + "type": "FLOAT", + "widget": { + "name": "frame_rate" + }, + "link": 223 + } + ], + "outputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "links": [ + 236 + ] + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "links": [ + 237 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.56", + "Node name for S&R": "LTXVConditioning", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 25 + ] + }, + { + "id": 203, + "type": "GetImageSize", + "pos": [ + 1480, + 3500 + ], + "size": [ + 230, + 130 + ], + "flags": {}, + "order": 10, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 224 + } + ], + "outputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "links": [ + 218 + ] + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "links": [ + 219 + ] + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "links": [] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.14.1", + "Node name for S&R": "GetImageSize", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 204, + "type": "LTXVAddGuide", + "pos": [ + 2750, + 3700 + ], + "size": [ + 280, + 240 + ], + "flags": {}, + "order": 11, + "mode": 0, + "inputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 225 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 226 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 227 + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "link": 228 + }, + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 229 + }, + { + "localized_name": "frame_idx", + "name": "frame_idx", + "type": "INT", + "widget": { + "name": "frame_idx" + }, + "link": null + }, + { + "localized_name": "strength", + "name": "strength", + "type": "FLOAT", + "widget": { + "name": "strength" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "links": [ + 213, + 242 + ] + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "links": [ + 214, + 243 + ] + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "links": [ + 244 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.12.3", + "Node name for S&R": "LTXVAddGuide", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + -1, + 0.7 + ] + }, + { + "id": 205, + "type": "PrimitiveInt", + "pos": [ + 760, + 3800 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 12, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": 269 + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 207, + 235, + 261 + ] + } + ], + "title": "Frame Rate(int)", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 25, + "fixed" + ] + }, + { + "id": 206, + "type": "LTXVAddGuide", + "pos": [ + 2750, + 3430 + ], + "size": [ + 280, + 240 + ], + "flags": {}, + "order": 13, + "mode": 0, + "inputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 236 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 237 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 238 + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "link": 239 + }, + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 240 + }, + { + "localized_name": "frame_idx", + "name": "frame_idx", + "type": "INT", + "widget": { + "name": "frame_idx" + }, + "link": null + }, + { + "localized_name": "strength", + "name": "strength", + "type": "FLOAT", + "widget": { + "name": "strength" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "links": [ + 225 + ] + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "links": [ + 226 + ] + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "links": [ + 228 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.12.3", + "Node name for S&R": "LTXVAddGuide", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 0, + 0.7 + ] + }, + { + "id": 207, + "type": "CFGGuider", + "pos": [ + 1990, + 2500 + ], + "size": [ + 280, + 160 + ], + "flags": {}, + "order": 14, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 241 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 242 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 243 + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "GUIDER", + "name": "GUIDER", + "type": "GUIDER", + "links": [ + 247 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.14.1", + "Node name for S&R": "CFGGuider", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1 + ] + }, + { + "id": 208, + "type": "SamplerEulerAncestral", + "pos": [ + 1990, + 2720 + ], + "size": [ + 280, + 120 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "eta", + "name": "eta", + "type": "FLOAT", + "widget": { + "name": "eta" + }, + "link": null + }, + { + "localized_name": "s_noise", + "name": "s_noise", + "type": "FLOAT", + "widget": { + "name": "s_noise" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "SAMPLER", + "name": "SAMPLER", + "type": "SAMPLER", + "links": [ + 248 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.14.1", + "Node name for S&R": "SamplerEulerAncestral", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 0, + 1 + ] + }, + { + "id": 209, + "type": "ManualSigmas", + "pos": [ + 1990, + 2910 + ], + "size": [ + 280, + 110 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "sigmas", + "name": "sigmas", + "type": "STRING", + "widget": { + "name": "sigmas" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "SIGMAS", + "name": "SIGMAS", + "type": "SIGMAS", + "links": [ + 249 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.14.1", + "Node name for S&R": "ManualSigmas", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "1., 0.99375, 0.9875, 0.98125, 0.975, 0.909375, 0.725, 0.421875, 0.0" + ] + }, + { + "id": 210, + "type": "LTXVConcatAVLatent", + "pos": [ + 1990, + 3090 + ], + "size": [ + 280, + 100 + ], + "flags": {}, + "order": 15, + "mode": 0, + "inputs": [ + { + "localized_name": "video_latent", + "name": "video_latent", + "type": "LATENT", + "link": 244 + }, + { + "localized_name": "audio_latent", + "name": "audio_latent", + "type": "LATENT", + "link": 245 + } + ], + "outputs": [ + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "links": [ + 250 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "LTXVConcatAVLatent", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 211, + "type": "SamplerCustomAdvanced", + "pos": [ + 2460, + 2330 + ], + "size": [ + 230, + 170 + ], + "flags": {}, + "order": 16, + "mode": 0, + "inputs": [ + { + "localized_name": "noise", + "name": "noise", + "type": "NOISE", + "link": 246 + }, + { + "localized_name": "guider", + "name": "guider", + "type": "GUIDER", + "link": 247 + }, + { + "localized_name": "sampler", + "name": "sampler", + "type": "SAMPLER", + "link": 248 + }, + { + "localized_name": "sigmas", + "name": "sigmas", + "type": "SIGMAS", + "link": 249 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 250 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "LATENT", + "links": [] + }, + { + "localized_name": "denoised_output", + "name": "denoised_output", + "type": "LATENT", + "links": [ + 204 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.14.1", + "Node name for S&R": "SamplerCustomAdvanced", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 212, + "type": "ComfyMathExpression", + "pos": [ + 760, + 3970 + ], + "size": [ + 230, + 170 + ], + "flags": { + "collapsed": true + }, + "order": 17, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 235 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": [ + 223, + 234 + ] + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.17.0", + "Node name for S&R": "ComfyMathExpression", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "a" + ] + }, + { + "id": 213, + "type": "ResizeImageMaskNode", + "pos": [ + 1130, + 3340 + ], + "size": [ + 280, + 160 + ], + "flags": {}, + "order": 18, + "mode": 0, + "inputs": [ + { + "localized_name": "input", + "name": "input", + "type": "IMAGE,MASK", + "link": 251 + }, + { + "localized_name": "resize_type", + "name": "resize_type", + "type": "COMFY_DYNAMICCOMBO_V3", + "widget": { + "name": "resize_type" + }, + "link": null + }, + { + "localized_name": "width", + "name": "resize_type.width", + "type": "INT", + "widget": { + "name": "resize_type.width" + }, + "link": 208 + }, + { + "localized_name": "height", + "name": "resize_type.height", + "type": "INT", + "widget": { + "name": "resize_type.height" + }, + "link": 209 + }, + { + "localized_name": "crop", + "name": "resize_type.crop", + "type": "COMBO", + "widget": { + "name": "resize_type.crop" + }, + "link": null + }, + { + "localized_name": "scale_method", + "name": "scale_method", + "type": "COMBO", + "widget": { + "name": "scale_method" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "resized", + "name": "resized", + "type": "*", + "links": [ + 210, + 224 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": { + "resize_type.width": true, + "resize_type.height": true + }, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.14.1", + "Node name for S&R": "ResizeImageMaskNode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "scale dimensions", + 640, + 360, + "center", + "nearest-exact" + ] + }, + { + "id": 214, + "type": "ResizeImageMaskNode", + "pos": [ + 1130, + 3780 + ], + "size": [ + 280, + 160 + ], + "flags": {}, + "order": 19, + "mode": 0, + "inputs": [ + { + "localized_name": "input", + "name": "input", + "type": "IMAGE,MASK", + "link": 253 + }, + { + "localized_name": "resize_type", + "name": "resize_type", + "type": "COMFY_DYNAMICCOMBO_V3", + "widget": { + "name": "resize_type" + }, + "link": null + }, + { + "localized_name": "width", + "name": "resize_type.width", + "type": "INT", + "widget": { + "name": "resize_type.width" + }, + "link": 201 + }, + { + "localized_name": "height", + "name": "resize_type.height", + "type": "INT", + "widget": { + "name": "resize_type.height" + }, + "link": 202 + }, + { + "localized_name": "crop", + "name": "resize_type.crop", + "type": "COMBO", + "widget": { + "name": "resize_type.crop" + }, + "link": null + }, + { + "localized_name": "scale_method", + "name": "scale_method", + "type": "COMBO", + "widget": { + "name": "scale_method" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "resized", + "name": "resized", + "type": "*", + "links": [ + 203 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": { + "resize_type.width": true, + "resize_type.height": true + }, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.14.1", + "Node name for S&R": "ResizeImageMaskNode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "scale dimensions", + 640, + 360, + "center", + "nearest-exact" + ] + }, + { + "id": 215, + "type": "PrimitiveInt", + "pos": [ + 760, + 3340 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 20, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": 266 + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 201, + 208 + ] + } + ], + "title": "Width", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1280, + "fixed" + ] + }, + { + "id": 216, + "type": "PrimitiveInt", + "pos": [ + 760, + 3490 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 21, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": 267 + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 202, + 209 + ] + } + ], + "title": "height", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 720, + "fixed" + ] + }, + { + "id": 217, + "type": "CLIPTextEncode", + "pos": [ + 1320, + 2870 + ], + "size": [ + 590, + 200 + ], + "flags": { + "collapsed": false + }, + "order": 22, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 230 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 222 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.56", + "Node name for S&R": "CLIPTextEncode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "blurry, out of focus, overexposed, underexposed, low contrast, washed out colors, excessive noise, grainy texture, poor lighting, flickering, motion blur, distorted proportions, unnatural skin tones, deformed facial features, asymmetrical face, missing facial features, extra limbs, disfigured hands, wrong hand count, artifacts around text, unreadable text on shirt or hat, incorrect lettering on cap (“PNTR”), incorrect t-shirt slogan (“JUST DO IT”), missing microphone, misplaced microphone, inconsistent perspective, camera shake, incorrect depth of field, background too sharp, background clutter, distracting reflections, harsh shadows, inconsistent lighting direction, color banding, cartoonish rendering, 3D CGI look, unrealistic materials, uncanny valley effect, incorrect ethnicity, wrong gender, exaggerated expressions, smiling, laughing, exaggerated sadness, wrong gaze direction, eyes looking at camera, mismatched lip sync, silent or muted audio, distorted voice, robotic voice, echo, background noise, off-sync audio, missing sniff sounds, incorrect dialogue, added dialogue, repetitive speech, jittery movement, awkward pauses, incorrect timing, unnatural transitions, inconsistent framing, tilted camera, missing door or shelves, missing shallow depth of field, flat lighting, inconsistent tone, cinematic oversaturation, stylized filters, or AI artifacts." + ], + "color": "#323", + "bgcolor": "#535" + }, + { + "id": 218, + "type": "CreateVideo", + "pos": [ + 3280, + 2320 + ], + "size": [ + 280, + 130 + ], + "flags": {}, + "order": 23, + "mode": 0, + "inputs": [ + { + "localized_name": "images", + "name": "images", + "type": "IMAGE", + "link": 232 + }, + { + "localized_name": "audio", + "name": "audio", + "shape": 7, + "type": "AUDIO", + "link": 233 + }, + { + "localized_name": "fps", + "name": "fps", + "type": "FLOAT", + "widget": { + "name": "fps" + }, + "link": 234 + } + ], + "outputs": [ + { + "localized_name": "VIDEO", + "name": "VIDEO", + "type": "VIDEO", + "links": [ + 252 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.14.1", + "Node name for S&R": "CreateVideo", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 24 + ] + }, + { + "id": 219, + "type": "VAEDecodeTiled", + "pos": [ + 2820, + 2630 + ], + "size": [ + 280, + 200 + ], + "flags": {}, + "order": 24, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 211 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 212 + }, + { + "localized_name": "tile_size", + "name": "tile_size", + "type": "INT", + "widget": { + "name": "tile_size" + }, + "link": null + }, + { + "localized_name": "overlap", + "name": "overlap", + "type": "INT", + "widget": { + "name": "overlap" + }, + "link": null + }, + { + "localized_name": "temporal_size", + "name": "temporal_size", + "type": "INT", + "widget": { + "name": "temporal_size" + }, + "link": null + }, + { + "localized_name": "temporal_overlap", + "name": "temporal_overlap", + "type": "INT", + "widget": { + "name": "temporal_overlap" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 232 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "VAEDecodeTiled", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 768, + 64, + 4096, + 64 + ] + }, + { + "id": 220, + "type": "LTXVAudioVAEDecode", + "pos": [ + 2820, + 2920 + ], + "size": [ + 280, + 100 + ], + "flags": {}, + "order": 25, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 216 + }, + { + "label": "Audio VAE", + "localized_name": "audio_vae", + "name": "audio_vae", + "type": "VAE", + "link": 217 + } + ], + "outputs": [ + { + "localized_name": "Audio", + "name": "Audio", + "type": "AUDIO", + "links": [ + 233 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "LTXVAudioVAEDecode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 221, + "type": "LTXVSeparateAVLatent", + "pos": [ + 2460, + 2580 + ], + "size": [ + 250, + 100 + ], + "flags": {}, + "order": 26, + "mode": 0, + "inputs": [ + { + "localized_name": "av_latent", + "name": "av_latent", + "type": "LATENT", + "link": 204 + } + ], + "outputs": [ + { + "localized_name": "video_latent", + "name": "video_latent", + "type": "LATENT", + "links": [ + 215 + ] + }, + { + "localized_name": "audio_latent", + "name": "audio_latent", + "type": "LATENT", + "links": [ + 216 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.5.1", + "Node name for S&R": "LTXVSeparateAVLatent", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 222, + "type": "CLIPTextEncode", + "pos": [ + 1310, + 2380 + ], + "size": [ + 620, + 420 + ], + "flags": {}, + "order": 27, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 231 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 265 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 221 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.56", + "Node name for S&R": "CLIPTextEncode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "" + ], + "color": "#232", + "bgcolor": "#353" + }, + { + "id": 223, + "type": "CheckpointLoaderSimple", + "pos": [ + 770, + 2380 + ], + "size": [ + 420, + 160 + ], + "flags": {}, + "order": 28, + "mode": 0, + "inputs": [ + { + "localized_name": "ckpt_name", + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": 276 + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 241 + ] + }, + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [] + }, + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "links": [ + 212, + 227, + 238 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.5.2" + }, + "cnr_id": "comfy-core", + "ver": "0.10.0", + "Node name for S&R": "CheckpointLoaderSimple", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "ltx-2.3-22b-distilled-fp8.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2.3-fp8/resolve/main/ltx-2.3-22b-distilled-fp8.safetensors", + "directory": "checkpoints" + } + ] + }, + "widgets_values": [ + "ltx-2.3-22b-distilled-fp8.safetensors" + ] + }, + { + "id": 224, + "type": "LTXVAudioVAELoader", + "pos": [ + 770, + 2660 + ], + "size": [ + 420, + 110 + ], + "flags": {}, + "order": 29, + "mode": 0, + "inputs": [ + { + "localized_name": "ckpt_name", + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": 272 + } + ], + "outputs": [ + { + "localized_name": "Audio VAE", + "name": "Audio VAE", + "type": "VAE", + "links": [ + 205, + 217 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.5.2" + }, + "cnr_id": "comfy-core", + "ver": "0.10.0", + "Node name for S&R": "LTXVAudioVAELoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "ltx-2.3-22b-distilled-fp8.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2.3-fp8/resolve/main/ltx-2.3-22b-distilled-fp8.safetensors", + "directory": "checkpoints" + } + ] + }, + "widgets_values": [ + "ltx-2.3-22b-distilled-fp8.safetensors" + ] + }, + { + "id": 225, + "type": "LTXAVTextEncoderLoader", + "pos": [ + 770, + 2890 + ], + "size": [ + 410, + 160 + ], + "flags": {}, + "order": 30, + "mode": 0, + "inputs": [ + { + "localized_name": "text_encoder", + "name": "text_encoder", + "type": "COMBO", + "widget": { + "name": "text_encoder" + }, + "link": 273 + }, + { + "localized_name": "ckpt_name", + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": 275 + }, + { + "localized_name": "device", + "name": "device", + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [ + 230, + 231 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.5.2" + }, + "cnr_id": "comfy-core", + "ver": "0.10.0", + "Node name for S&R": "LTXAVTextEncoderLoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "gemma_3_12B_it_fp4_mixed.safetensors", + "url": "https://huggingface.co/Comfy-Org/ltx-2/resolve/main/split_files/text_encoders/gemma_3_12B_it_fp4_mixed.safetensors", + "directory": "text_encoders" + }, + { + "name": "ltx-2.3-22b-distilled-fp8.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2.3-fp8/resolve/main/ltx-2.3-22b-distilled-fp8.safetensors", + "directory": "checkpoints" + } + ] + }, + "widgets_values": [ + "gemma_3_12B_it_fp4_mixed.safetensors", + "ltx-2.3-22b-distilled-fp8.safetensors", + "default" + ] + }, + { + "id": 226, + "type": "ComfyMathExpression", + "pos": [ + 760, + 4020 + ], + "size": [ + 400, + 200 + ], + "flags": { + "collapsed": true + }, + "order": 31, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 260 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": 261 + }, + { + "label": "c", + "localized_name": "values.c", + "name": "values.c", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 262, + 263 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ComfyMathExpression" + }, + "widgets_values": [ + "a * b + 1" + ] + } + ], + "groups": [ + { + "id": 1, + "title": "Conditioning", + "bounding": [ + 1850, + 3250, + 1370, + 800 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 2, + "title": "Settings", + "bounding": [ + 730, + 3250, + 290, + 800 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 3, + "title": "FIrst Frame", + "bounding": [ + 1050, + 3250, + 770, + 400 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 4, + "title": "Last Frame", + "bounding": [ + 1050, + 3680, + 770, + 370 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 5, + "title": "Model", + "bounding": [ + 730, + 2240, + 500, + 980 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 6, + "title": "Prompt", + "bounding": [ + 1260, + 2240, + 680, + 980 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 7, + "title": "Sampling", + "bounding": [ + 1970, + 2240, + 770, + 980 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 8, + "title": "Decoding", + "bounding": [ + 2770, + 2240, + 450, + 980 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 203, + "origin_id": 214, + "origin_slot": 0, + "target_id": 195, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 205, + "origin_id": 224, + "origin_slot": 0, + "target_id": 197, + "target_slot": 0, + "type": "VAE" + }, + { + "id": 207, + "origin_id": 205, + "origin_slot": 0, + "target_id": 197, + "target_slot": 2, + "type": "INT" + }, + { + "id": 210, + "origin_id": 213, + "origin_slot": 0, + "target_id": 199, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 213, + "origin_id": 204, + "origin_slot": 0, + "target_id": 200, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 214, + "origin_id": 204, + "origin_slot": 1, + "target_id": 200, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 215, + "origin_id": 221, + "origin_slot": 0, + "target_id": 200, + "target_slot": 2, + "type": "LATENT" + }, + { + "id": 218, + "origin_id": 203, + "origin_slot": 0, + "target_id": 201, + "target_slot": 0, + "type": "INT" + }, + { + "id": 219, + "origin_id": 203, + "origin_slot": 1, + "target_id": 201, + "target_slot": 1, + "type": "INT" + }, + { + "id": 221, + "origin_id": 222, + "origin_slot": 0, + "target_id": 202, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 222, + "origin_id": 217, + "origin_slot": 0, + "target_id": 202, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 223, + "origin_id": 212, + "origin_slot": 0, + "target_id": 202, + "target_slot": 2, + "type": "FLOAT" + }, + { + "id": 224, + "origin_id": 213, + "origin_slot": 0, + "target_id": 203, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 225, + "origin_id": 206, + "origin_slot": 0, + "target_id": 204, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 226, + "origin_id": 206, + "origin_slot": 1, + "target_id": 204, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 227, + "origin_id": 223, + "origin_slot": 2, + "target_id": 204, + "target_slot": 2, + "type": "VAE" + }, + { + "id": 228, + "origin_id": 206, + "origin_slot": 2, + "target_id": 204, + "target_slot": 3, + "type": "LATENT" + }, + { + "id": 229, + "origin_id": 195, + "origin_slot": 0, + "target_id": 204, + "target_slot": 4, + "type": "IMAGE" + }, + { + "id": 236, + "origin_id": 202, + "origin_slot": 0, + "target_id": 206, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 237, + "origin_id": 202, + "origin_slot": 1, + "target_id": 206, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 238, + "origin_id": 223, + "origin_slot": 2, + "target_id": 206, + "target_slot": 2, + "type": "VAE" + }, + { + "id": 239, + "origin_id": 201, + "origin_slot": 0, + "target_id": 206, + "target_slot": 3, + "type": "LATENT" + }, + { + "id": 240, + "origin_id": 199, + "origin_slot": 0, + "target_id": 206, + "target_slot": 4, + "type": "IMAGE" + }, + { + "id": 241, + "origin_id": 223, + "origin_slot": 0, + "target_id": 207, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 242, + "origin_id": 204, + "origin_slot": 0, + "target_id": 207, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 243, + "origin_id": 204, + "origin_slot": 1, + "target_id": 207, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 244, + "origin_id": 204, + "origin_slot": 2, + "target_id": 210, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 245, + "origin_id": 197, + "origin_slot": 0, + "target_id": 210, + "target_slot": 1, + "type": "LATENT" + }, + { + "id": 246, + "origin_id": 196, + "origin_slot": 0, + "target_id": 211, + "target_slot": 0, + "type": "NOISE" + }, + { + "id": 247, + "origin_id": 207, + "origin_slot": 0, + "target_id": 211, + "target_slot": 1, + "type": "GUIDER" + }, + { + "id": 248, + "origin_id": 208, + "origin_slot": 0, + "target_id": 211, + "target_slot": 2, + "type": "SAMPLER" + }, + { + "id": 249, + "origin_id": 209, + "origin_slot": 0, + "target_id": 211, + "target_slot": 3, + "type": "SIGMAS" + }, + { + "id": 250, + "origin_id": 210, + "origin_slot": 0, + "target_id": 211, + "target_slot": 4, + "type": "LATENT" + }, + { + "id": 235, + "origin_id": 205, + "origin_slot": 0, + "target_id": 212, + "target_slot": 0, + "type": "INT" + }, + { + "id": 208, + "origin_id": 215, + "origin_slot": 0, + "target_id": 213, + "target_slot": 2, + "type": "INT" + }, + { + "id": 209, + "origin_id": 216, + "origin_slot": 0, + "target_id": 213, + "target_slot": 3, + "type": "INT" + }, + { + "id": 201, + "origin_id": 215, + "origin_slot": 0, + "target_id": 214, + "target_slot": 2, + "type": "INT" + }, + { + "id": 202, + "origin_id": 216, + "origin_slot": 0, + "target_id": 214, + "target_slot": 3, + "type": "INT" + }, + { + "id": 230, + "origin_id": 225, + "origin_slot": 0, + "target_id": 217, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 232, + "origin_id": 219, + "origin_slot": 0, + "target_id": 218, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 233, + "origin_id": 220, + "origin_slot": 0, + "target_id": 218, + "target_slot": 1, + "type": "AUDIO" + }, + { + "id": 234, + "origin_id": 212, + "origin_slot": 0, + "target_id": 218, + "target_slot": 2, + "type": "FLOAT" + }, + { + "id": 211, + "origin_id": 200, + "origin_slot": 2, + "target_id": 219, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 212, + "origin_id": 223, + "origin_slot": 2, + "target_id": 219, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 216, + "origin_id": 221, + "origin_slot": 1, + "target_id": 220, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 217, + "origin_id": 224, + "origin_slot": 0, + "target_id": 220, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 204, + "origin_id": 211, + "origin_slot": 1, + "target_id": 221, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 231, + "origin_id": 225, + "origin_slot": 0, + "target_id": 222, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 251, + "origin_id": -10, + "origin_slot": 0, + "target_id": 213, + "target_slot": 0, + "type": "IMAGE,MASK" + }, + { + "id": 253, + "origin_id": -10, + "origin_slot": 1, + "target_id": 214, + "target_slot": 0, + "type": "IMAGE,MASK" + }, + { + "id": 252, + "origin_id": 218, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "VIDEO" + }, + { + "id": 260, + "origin_id": 198, + "origin_slot": 0, + "target_id": 226, + "target_slot": 0, + "type": "INT" + }, + { + "id": 261, + "origin_id": 205, + "origin_slot": 0, + "target_id": 226, + "target_slot": 1, + "type": "INT" + }, + { + "id": 262, + "origin_id": 226, + "origin_slot": 1, + "target_id": 197, + "target_slot": 1, + "type": "INT" + }, + { + "id": 263, + "origin_id": 226, + "origin_slot": 1, + "target_id": 201, + "target_slot": 2, + "type": "INT" + }, + { + "id": 265, + "origin_id": -10, + "origin_slot": 2, + "target_id": 222, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 266, + "origin_id": -10, + "origin_slot": 3, + "target_id": 215, + "target_slot": 0, + "type": "INT" + }, + { + "id": 267, + "origin_id": -10, + "origin_slot": 4, + "target_id": 216, + "target_slot": 0, + "type": "INT" + }, + { + "id": 268, + "origin_id": -10, + "origin_slot": 5, + "target_id": 198, + "target_slot": 0, + "type": "INT" + }, + { + "id": 269, + "origin_id": -10, + "origin_slot": 6, + "target_id": 205, + "target_slot": 0, + "type": "INT" + }, + { + "id": 270, + "origin_id": -10, + "origin_slot": 7, + "target_id": 196, + "target_slot": 0, + "type": "INT" + }, + { + "id": 272, + "origin_id": -10, + "origin_slot": 8, + "target_id": 224, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 273, + "origin_id": -10, + "origin_slot": 9, + "target_id": 225, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 275, + "origin_id": -10, + "origin_slot": 8, + "target_id": 225, + "target_slot": 1, + "type": "COMBO" + }, + { + "id": 276, + "origin_id": -10, + "origin_slot": 8, + "target_id": 223, + "target_slot": 0, + "type": "COMBO" + } + ], + "extra": {}, + "category": "Video generation and editing/First-Last-Frame to Video", + "description": "Generates a video interpolating between first and last keyframes using LTX-2.3." + } + ] + }, + "extra": { + "ue_links": [] + } +} \ No newline at end of file diff --git a/blueprints/First-Last-Frame to Video.json b/blueprints/First-Last-Frame to Video.json new file mode 100644 index 000000000..84dfafbcd --- /dev/null +++ b/blueprints/First-Last-Frame to Video.json @@ -0,0 +1,3361 @@ +{ + "revision": 0, + "last_node_id": 227, + "last_link_id": 0, + "nodes": [ + { + "id": 227, + "type": "283e4561-61a2-4538-b960-265736eb041f", + "pos": [ + 620, + 3140 + ], + "size": [ + 540, + 0 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "label": "first_frame", + "localized_name": "input", + "name": "input", + "type": "IMAGE,MASK", + "link": null + }, + { + "label": "last_frame", + "localized_name": "input_1", + "name": "input_1", + "type": "IMAGE,MASK", + "link": null + }, + { + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + }, + { + "label": "width", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": null + }, + { + "label": "height", + "name": "value_1", + "type": "INT", + "widget": { + "name": "value_1" + }, + "link": null + }, + { + "label": "duration", + "name": "value_2", + "type": "INT", + "widget": { + "name": "value_2" + }, + "link": null + }, + { + "label": "fps", + "name": "value_3", + "type": "INT", + "widget": { + "name": "value_3" + }, + "link": null + }, + { + "name": "noise_seed", + "type": "INT", + "widget": { + "name": "noise_seed" + }, + "link": null + }, + { + "label": "ckpt_name", + "name": "ckpt_name_1", + "type": "COMBO", + "widget": { + "name": "ckpt_name_1" + }, + "link": null + }, + { + "name": "text_encoder", + "type": "COMBO", + "widget": { + "name": "text_encoder" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "VIDEO", + "name": "VIDEO", + "type": "VIDEO", + "links": [] + } + ], + "title": "First-Last-Frame to Video", + "properties": { + "proxyWidgets": [ + [ + "222", + "text" + ], + [ + "215", + "value" + ], + [ + "216", + "value" + ], + [ + "198", + "value" + ], + [ + "205", + "value" + ], + [ + "196", + "noise_seed" + ], + [ + "224", + "ckpt_name" + ], + [ + "225", + "text_encoder" + ] + ], + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + } + }, + "widgets_values": [] + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "283e4561-61a2-4538-b960-265736eb041f", + "version": 1, + "state": { + "lastGroupId": 22, + "lastNodeId": 227, + "lastLinkId": 276, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "First-Last-Frame to Video", + "inputNode": { + "id": -10, + "bounding": [ + 270, + 3100, + 120, + 240 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 3620, + 3120, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "6fe179c4-d96f-4383-b202-844f6de4922e", + "name": "input", + "type": "IMAGE,MASK", + "linkIds": [ + 251 + ], + "localized_name": "input", + "label": "first_frame", + "pos": [ + 370, + 3120 + ] + }, + { + "id": "e80df1ae-5f39-4f86-91bd-0467635e2f2d", + "name": "input_1", + "type": "IMAGE,MASK", + "linkIds": [ + 253 + ], + "localized_name": "input_1", + "label": "last_frame", + "pos": [ + 370, + 3140 + ] + }, + { + "id": "433148fa-bf73-4ab1-81d9-09e2e38ed861", + "name": "text", + "type": "STRING", + "linkIds": [ + 265 + ], + "pos": [ + 370, + 3160 + ] + }, + { + "id": "36915bc8-a6ed-4d48-8619-e0e8723228e9", + "name": "value", + "type": "INT", + "linkIds": [ + 266 + ], + "label": "width", + "pos": [ + 370, + 3180 + ] + }, + { + "id": "425a36b8-91ab-41b7-81e9-496eba064ec8", + "name": "value_1", + "type": "INT", + "linkIds": [ + 267 + ], + "label": "height", + "pos": [ + 370, + 3200 + ] + }, + { + "id": "0c9e003b-bd07-4b7d-aa6d-789e138ed161", + "name": "value_2", + "type": "INT", + "linkIds": [ + 268 + ], + "label": "duration", + "pos": [ + 370, + 3220 + ] + }, + { + "id": "581b52ff-21c5-4774-ac2a-8f69a7e09e2e", + "name": "value_3", + "type": "INT", + "linkIds": [ + 269 + ], + "label": "fps", + "pos": [ + 370, + 3240 + ] + }, + { + "id": "d03cc171-45da-4658-99aa-77252bbcf522", + "name": "noise_seed", + "type": "INT", + "linkIds": [ + 270 + ], + "pos": [ + 370, + 3260 + ] + }, + { + "id": "e68e61c8-905e-43ac-8c76-65ac52270a08", + "name": "ckpt_name_1", + "type": "COMBO", + "linkIds": [ + 272, + 275, + 276 + ], + "label": "ckpt_name", + "pos": [ + 370, + 3280 + ] + }, + { + "id": "5d065f3b-891b-499f-950b-c2df0be24536", + "name": "text_encoder", + "type": "COMBO", + "linkIds": [ + 273 + ], + "pos": [ + 370, + 3300 + ] + } + ], + "outputs": [ + { + "id": "0c8c2dc0-c67c-4bc2-9e57-6aa00db2e3a9", + "name": "VIDEO", + "type": "VIDEO", + "linkIds": [ + 252 + ], + "localized_name": "VIDEO", + "pos": [ + 3640, + 3140 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 195, + "type": "LTXVPreprocess", + "pos": [ + 1480, + 3780 + ], + "size": [ + 230, + 110 + ], + "flags": { + "collapsed": false + }, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 203 + }, + { + "localized_name": "img_compression", + "name": "img_compression", + "type": "INT", + "widget": { + "name": "img_compression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "output_image", + "name": "output_image", + "type": "IMAGE", + "links": [ + 229 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "LTXVPreprocess", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 25 + ] + }, + { + "id": 196, + "type": "RandomNoise", + "pos": [ + 1990, + 2320 + ], + "size": [ + 280, + 110 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "noise_seed", + "name": "noise_seed", + "type": "INT", + "widget": { + "name": "noise_seed" + }, + "link": 270 + } + ], + "outputs": [ + { + "localized_name": "NOISE", + "name": "NOISE", + "type": "NOISE", + "links": [ + 246 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.14.1", + "ue_properties": { + "widget_ue_connectable": { + "noise_seed": true + }, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "RandomNoise", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 315253765879496, + "randomize" + ] + }, + { + "id": 197, + "type": "LTXVEmptyLatentAudio", + "pos": [ + 2090, + 3820 + ], + "size": [ + 280, + 170 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "audio_vae", + "name": "audio_vae", + "type": "VAE", + "link": 205 + }, + { + "localized_name": "frames_number", + "name": "frames_number", + "type": "INT", + "widget": { + "name": "frames_number" + }, + "link": 262 + }, + { + "localized_name": "frame_rate", + "name": "frame_rate", + "type": "INT", + "widget": { + "name": "frame_rate" + }, + "link": 207 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "Latent", + "name": "Latent", + "type": "LATENT", + "links": [ + 245 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.68", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "LTXVEmptyLatentAudio", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 97, + 25, + 1 + ] + }, + { + "id": 198, + "type": "PrimitiveInt", + "pos": [ + 760, + 3650 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": 268 + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 260 + ] + } + ], + "title": "Duration", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 5, + "fixed" + ] + }, + { + "id": 199, + "type": "LTXVPreprocess", + "pos": [ + 1480, + 3340 + ], + "size": [ + 230, + 110 + ], + "flags": { + "collapsed": false + }, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 210 + }, + { + "localized_name": "img_compression", + "name": "img_compression", + "type": "INT", + "widget": { + "name": "img_compression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "output_image", + "name": "output_image", + "type": "IMAGE", + "links": [ + 240 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "LTXVPreprocess", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 25 + ] + }, + { + "id": 200, + "type": "LTXVCropGuides", + "pos": [ + 2820, + 2450 + ], + "size": [ + 280, + 120 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 213 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 214 + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "link": 215 + } + ], + "outputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "links": [] + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "links": [] + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "links": [ + 211 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.8.2", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.5.2" + }, + "Node name for S&R": "LTXVCropGuides", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 201, + "type": "EmptyLTXVLatentVideo", + "pos": [ + 2090, + 3580 + ], + "size": [ + 280, + 200 + ], + "flags": {}, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 218 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 219 + }, + { + "localized_name": "length", + "name": "length", + "type": "INT", + "widget": { + "name": "length" + }, + "link": 263 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "links": [ + 239 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.60", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "EmptyLTXVLatentVideo", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 768, + 512, + 97, + 1 + ] + }, + { + "id": 202, + "type": "LTXVConditioning", + "pos": [ + 2090, + 3400 + ], + "size": [ + 280, + 130 + ], + "flags": {}, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 221 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 222 + }, + { + "localized_name": "frame_rate", + "name": "frame_rate", + "type": "FLOAT", + "widget": { + "name": "frame_rate" + }, + "link": 223 + } + ], + "outputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "links": [ + 236 + ] + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "links": [ + 237 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.56", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "LTXVConditioning", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 25 + ] + }, + { + "id": 203, + "type": "GetImageSize", + "pos": [ + 1480, + 3500 + ], + "size": [ + 230, + 130 + ], + "flags": {}, + "order": 10, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 224 + } + ], + "outputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "links": [ + 218 + ] + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "links": [ + 219 + ] + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "links": [] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.14.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "GetImageSize", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 204, + "type": "LTXVAddGuide", + "pos": [ + 2750, + 3700 + ], + "size": [ + 280, + 240 + ], + "flags": {}, + "order": 11, + "mode": 0, + "inputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 225 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 226 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 227 + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "link": 228 + }, + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 229 + }, + { + "localized_name": "frame_idx", + "name": "frame_idx", + "type": "INT", + "widget": { + "name": "frame_idx" + }, + "link": null + }, + { + "localized_name": "strength", + "name": "strength", + "type": "FLOAT", + "widget": { + "name": "strength" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "links": [ + 213, + 242 + ] + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "links": [ + 214, + 243 + ] + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "links": [ + 244 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.12.3", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "LTXVAddGuide", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + -1, + 0.7 + ] + }, + { + "id": 205, + "type": "PrimitiveInt", + "pos": [ + 760, + 3800 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 12, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": 269 + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 207, + 235, + 261 + ] + } + ], + "title": "Frame Rate(int)", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 25, + "fixed" + ] + }, + { + "id": 206, + "type": "LTXVAddGuide", + "pos": [ + 2750, + 3430 + ], + "size": [ + 280, + 240 + ], + "flags": {}, + "order": 13, + "mode": 0, + "inputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 236 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 237 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 238 + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "link": 239 + }, + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 240 + }, + { + "localized_name": "frame_idx", + "name": "frame_idx", + "type": "INT", + "widget": { + "name": "frame_idx" + }, + "link": null + }, + { + "localized_name": "strength", + "name": "strength", + "type": "FLOAT", + "widget": { + "name": "strength" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "links": [ + 225 + ] + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "links": [ + 226 + ] + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "links": [ + 228 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.12.3", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "LTXVAddGuide", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 0, + 0.7 + ] + }, + { + "id": 207, + "type": "CFGGuider", + "pos": [ + 1990, + 2500 + ], + "size": [ + 280, + 160 + ], + "flags": {}, + "order": 14, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 241 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 242 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 243 + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "GUIDER", + "name": "GUIDER", + "type": "GUIDER", + "links": [ + 247 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.14.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "CFGGuider", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1 + ] + }, + { + "id": 208, + "type": "SamplerEulerAncestral", + "pos": [ + 1990, + 2720 + ], + "size": [ + 280, + 120 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "eta", + "name": "eta", + "type": "FLOAT", + "widget": { + "name": "eta" + }, + "link": null + }, + { + "localized_name": "s_noise", + "name": "s_noise", + "type": "FLOAT", + "widget": { + "name": "s_noise" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "SAMPLER", + "name": "SAMPLER", + "type": "SAMPLER", + "links": [ + 248 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.14.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "SamplerEulerAncestral", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 0, + 1 + ] + }, + { + "id": 209, + "type": "ManualSigmas", + "pos": [ + 1990, + 2910 + ], + "size": [ + 280, + 110 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "sigmas", + "name": "sigmas", + "type": "STRING", + "widget": { + "name": "sigmas" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "SIGMAS", + "name": "SIGMAS", + "type": "SIGMAS", + "links": [ + 249 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.14.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ManualSigmas", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "1., 0.99375, 0.9875, 0.98125, 0.975, 0.909375, 0.725, 0.421875, 0.0" + ] + }, + { + "id": 210, + "type": "LTXVConcatAVLatent", + "pos": [ + 1990, + 3090 + ], + "size": [ + 280, + 100 + ], + "flags": {}, + "order": 15, + "mode": 0, + "inputs": [ + { + "localized_name": "video_latent", + "name": "video_latent", + "type": "LATENT", + "link": 244 + }, + { + "localized_name": "audio_latent", + "name": "audio_latent", + "type": "LATENT", + "link": 245 + } + ], + "outputs": [ + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "links": [ + 250 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "LTXVConcatAVLatent", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 211, + "type": "SamplerCustomAdvanced", + "pos": [ + 2460, + 2330 + ], + "size": [ + 230, + 170 + ], + "flags": {}, + "order": 16, + "mode": 0, + "inputs": [ + { + "localized_name": "noise", + "name": "noise", + "type": "NOISE", + "link": 246 + }, + { + "localized_name": "guider", + "name": "guider", + "type": "GUIDER", + "link": 247 + }, + { + "localized_name": "sampler", + "name": "sampler", + "type": "SAMPLER", + "link": 248 + }, + { + "localized_name": "sigmas", + "name": "sigmas", + "type": "SIGMAS", + "link": 249 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 250 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "LATENT", + "links": [] + }, + { + "localized_name": "denoised_output", + "name": "denoised_output", + "type": "LATENT", + "links": [ + 204 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.14.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "SamplerCustomAdvanced", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 212, + "type": "ComfyMathExpression", + "pos": [ + 760, + 3970 + ], + "size": [ + 230, + 170 + ], + "flags": { + "collapsed": true + }, + "order": 17, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 235 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": [ + 223, + 234 + ] + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.17.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ComfyMathExpression", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "a" + ] + }, + { + "id": 213, + "type": "ResizeImageMaskNode", + "pos": [ + 1130, + 3340 + ], + "size": [ + 280, + 160 + ], + "flags": {}, + "order": 18, + "mode": 0, + "inputs": [ + { + "localized_name": "input", + "name": "input", + "type": "IMAGE,MASK", + "link": 251 + }, + { + "localized_name": "resize_type", + "name": "resize_type", + "type": "COMFY_DYNAMICCOMBO_V3", + "widget": { + "name": "resize_type" + }, + "link": null + }, + { + "localized_name": "width", + "name": "resize_type.width", + "type": "INT", + "widget": { + "name": "resize_type.width" + }, + "link": 208 + }, + { + "localized_name": "height", + "name": "resize_type.height", + "type": "INT", + "widget": { + "name": "resize_type.height" + }, + "link": 209 + }, + { + "localized_name": "crop", + "name": "resize_type.crop", + "type": "COMBO", + "widget": { + "name": "resize_type.crop" + }, + "link": null + }, + { + "localized_name": "scale_method", + "name": "scale_method", + "type": "COMBO", + "widget": { + "name": "scale_method" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "resized", + "name": "resized", + "type": "*", + "links": [ + 210, + 224 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.14.1", + "ue_properties": { + "widget_ue_connectable": { + "resize_type.width": true, + "resize_type.height": true + }, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ResizeImageMaskNode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "scale dimensions", + 640, + 360, + "center", + "nearest-exact" + ] + }, + { + "id": 214, + "type": "ResizeImageMaskNode", + "pos": [ + 1130, + 3780 + ], + "size": [ + 280, + 160 + ], + "flags": {}, + "order": 19, + "mode": 0, + "inputs": [ + { + "localized_name": "input", + "name": "input", + "type": "IMAGE,MASK", + "link": 253 + }, + { + "localized_name": "resize_type", + "name": "resize_type", + "type": "COMFY_DYNAMICCOMBO_V3", + "widget": { + "name": "resize_type" + }, + "link": null + }, + { + "localized_name": "width", + "name": "resize_type.width", + "type": "INT", + "widget": { + "name": "resize_type.width" + }, + "link": 201 + }, + { + "localized_name": "height", + "name": "resize_type.height", + "type": "INT", + "widget": { + "name": "resize_type.height" + }, + "link": 202 + }, + { + "localized_name": "crop", + "name": "resize_type.crop", + "type": "COMBO", + "widget": { + "name": "resize_type.crop" + }, + "link": null + }, + { + "localized_name": "scale_method", + "name": "scale_method", + "type": "COMBO", + "widget": { + "name": "scale_method" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "resized", + "name": "resized", + "type": "*", + "links": [ + 203 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.14.1", + "ue_properties": { + "widget_ue_connectable": { + "resize_type.width": true, + "resize_type.height": true + }, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ResizeImageMaskNode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "scale dimensions", + 640, + 360, + "center", + "nearest-exact" + ] + }, + { + "id": 215, + "type": "PrimitiveInt", + "pos": [ + 760, + 3340 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 20, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": 266 + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 201, + 208 + ] + } + ], + "title": "Width", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1280, + "fixed" + ] + }, + { + "id": 216, + "type": "PrimitiveInt", + "pos": [ + 760, + 3490 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 21, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": 267 + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 202, + 209 + ] + } + ], + "title": "height", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 720, + "fixed" + ] + }, + { + "id": 217, + "type": "CLIPTextEncode", + "pos": [ + 1320, + 2870 + ], + "size": [ + 590, + 200 + ], + "flags": { + "collapsed": false + }, + "order": 22, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 230 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 222 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.56", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "CLIPTextEncode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "blurry, out of focus, overexposed, underexposed, low contrast, washed out colors, excessive noise, grainy texture, poor lighting, flickering, motion blur, distorted proportions, unnatural skin tones, deformed facial features, asymmetrical face, missing facial features, extra limbs, disfigured hands, wrong hand count, artifacts around text, unreadable text on shirt or hat, incorrect lettering on cap (“PNTR”), incorrect t-shirt slogan (“JUST DO IT”), missing microphone, misplaced microphone, inconsistent perspective, camera shake, incorrect depth of field, background too sharp, background clutter, distracting reflections, harsh shadows, inconsistent lighting direction, color banding, cartoonish rendering, 3D CGI look, unrealistic materials, uncanny valley effect, incorrect ethnicity, wrong gender, exaggerated expressions, smiling, laughing, exaggerated sadness, wrong gaze direction, eyes looking at camera, mismatched lip sync, silent or muted audio, distorted voice, robotic voice, echo, background noise, off-sync audio, missing sniff sounds, incorrect dialogue, added dialogue, repetitive speech, jittery movement, awkward pauses, incorrect timing, unnatural transitions, inconsistent framing, tilted camera, missing door or shelves, missing shallow depth of field, flat lighting, inconsistent tone, cinematic oversaturation, stylized filters, or AI artifacts." + ], + "color": "#323", + "bgcolor": "#535" + }, + { + "id": 218, + "type": "CreateVideo", + "pos": [ + 3280, + 2320 + ], + "size": [ + 280, + 130 + ], + "flags": {}, + "order": 23, + "mode": 0, + "inputs": [ + { + "localized_name": "images", + "name": "images", + "type": "IMAGE", + "link": 232 + }, + { + "localized_name": "audio", + "name": "audio", + "shape": 7, + "type": "AUDIO", + "link": 233 + }, + { + "localized_name": "fps", + "name": "fps", + "type": "FLOAT", + "widget": { + "name": "fps" + }, + "link": 234 + } + ], + "outputs": [ + { + "localized_name": "VIDEO", + "name": "VIDEO", + "type": "VIDEO", + "links": [ + 252 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.14.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "CreateVideo", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 24 + ] + }, + { + "id": 219, + "type": "VAEDecodeTiled", + "pos": [ + 2820, + 2630 + ], + "size": [ + 280, + 200 + ], + "flags": {}, + "order": 24, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 211 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 212 + }, + { + "localized_name": "tile_size", + "name": "tile_size", + "type": "INT", + "widget": { + "name": "tile_size" + }, + "link": null + }, + { + "localized_name": "overlap", + "name": "overlap", + "type": "INT", + "widget": { + "name": "overlap" + }, + "link": null + }, + { + "localized_name": "temporal_size", + "name": "temporal_size", + "type": "INT", + "widget": { + "name": "temporal_size" + }, + "link": null + }, + { + "localized_name": "temporal_overlap", + "name": "temporal_overlap", + "type": "INT", + "widget": { + "name": "temporal_overlap" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 232 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "VAEDecodeTiled", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 768, + 64, + 4096, + 64 + ] + }, + { + "id": 220, + "type": "LTXVAudioVAEDecode", + "pos": [ + 2820, + 2920 + ], + "size": [ + 280, + 100 + ], + "flags": {}, + "order": 25, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 216 + }, + { + "label": "Audio VAE", + "localized_name": "audio_vae", + "name": "audio_vae", + "type": "VAE", + "link": 217 + } + ], + "outputs": [ + { + "localized_name": "Audio", + "name": "Audio", + "type": "AUDIO", + "links": [ + 233 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "LTXVAudioVAEDecode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 221, + "type": "LTXVSeparateAVLatent", + "pos": [ + 2460, + 2580 + ], + "size": [ + 250, + 100 + ], + "flags": {}, + "order": 26, + "mode": 0, + "inputs": [ + { + "localized_name": "av_latent", + "name": "av_latent", + "type": "LATENT", + "link": 204 + } + ], + "outputs": [ + { + "localized_name": "video_latent", + "name": "video_latent", + "type": "LATENT", + "links": [ + 215 + ] + }, + { + "localized_name": "audio_latent", + "name": "audio_latent", + "type": "LATENT", + "links": [ + 216 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "LTXVSeparateAVLatent", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 222, + "type": "CLIPTextEncode", + "pos": [ + 1310, + 2380 + ], + "size": [ + 620, + 420 + ], + "flags": {}, + "order": 27, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 231 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 265 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 221 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.56", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.5.2", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "CLIPTextEncode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "" + ], + "color": "#232", + "bgcolor": "#353" + }, + { + "id": 223, + "type": "CheckpointLoaderSimple", + "pos": [ + 770, + 2380 + ], + "size": [ + 420, + 160 + ], + "flags": {}, + "order": 28, + "mode": 0, + "inputs": [ + { + "localized_name": "ckpt_name", + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": 276 + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 241 + ] + }, + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [] + }, + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "links": [ + 212, + 227, + 238 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.10.0", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.5.2" + }, + "Node name for S&R": "CheckpointLoaderSimple", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "ltx-2.3-22b-distilled-fp8.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2.3-fp8/resolve/main/ltx-2.3-22b-distilled-fp8.safetensors", + "directory": "checkpoints" + } + ] + }, + "widgets_values": [ + "ltx-2.3-22b-distilled-fp8.safetensors" + ] + }, + { + "id": 224, + "type": "LTXVAudioVAELoader", + "pos": [ + 770, + 2660 + ], + "size": [ + 420, + 110 + ], + "flags": {}, + "order": 29, + "mode": 0, + "inputs": [ + { + "localized_name": "ckpt_name", + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": 272 + } + ], + "outputs": [ + { + "localized_name": "Audio VAE", + "name": "Audio VAE", + "type": "VAE", + "links": [ + 205, + 217 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.10.0", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.5.2" + }, + "Node name for S&R": "LTXVAudioVAELoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "ltx-2.3-22b-distilled-fp8.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2.3-fp8/resolve/main/ltx-2.3-22b-distilled-fp8.safetensors", + "directory": "checkpoints" + } + ] + }, + "widgets_values": [ + "ltx-2.3-22b-distilled-fp8.safetensors" + ] + }, + { + "id": 225, + "type": "LTXAVTextEncoderLoader", + "pos": [ + 770, + 2890 + ], + "size": [ + 410, + 160 + ], + "flags": {}, + "order": 30, + "mode": 0, + "inputs": [ + { + "localized_name": "text_encoder", + "name": "text_encoder", + "type": "COMBO", + "widget": { + "name": "text_encoder" + }, + "link": 273 + }, + { + "localized_name": "ckpt_name", + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": 275 + }, + { + "localized_name": "device", + "name": "device", + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [ + 230, + 231 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.10.0", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.5.2" + }, + "Node name for S&R": "LTXAVTextEncoderLoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "gemma_3_12B_it_fp4_mixed.safetensors", + "url": "https://huggingface.co/Comfy-Org/ltx-2/resolve/main/split_files/text_encoders/gemma_3_12B_it_fp4_mixed.safetensors", + "directory": "text_encoders" + }, + { + "name": "ltx-2.3-22b-distilled-fp8.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2.3-fp8/resolve/main/ltx-2.3-22b-distilled-fp8.safetensors", + "directory": "checkpoints" + } + ] + }, + "widgets_values": [ + "gemma_3_12B_it_fp4_mixed.safetensors", + "ltx-2.3-22b-distilled-fp8.safetensors", + "default" + ] + }, + { + "id": 226, + "type": "ComfyMathExpression", + "pos": [ + 760, + 4020 + ], + "size": [ + 400, + 200 + ], + "flags": { + "collapsed": true + }, + "order": 31, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 260 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": 261 + }, + { + "label": "c", + "localized_name": "values.c", + "name": "values.c", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 262, + 263 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "Node name for S&R": "ComfyMathExpression" + }, + "widgets_values": [ + "a * b + 1" + ] + } + ], + "groups": [ + { + "id": 1, + "title": "Conditioning", + "bounding": [ + 1850, + 3250, + 1370, + 800 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 2, + "title": "Settings", + "bounding": [ + 730, + 3250, + 290, + 800 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 3, + "title": "FIrst Frame", + "bounding": [ + 1050, + 3250, + 770, + 400 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 4, + "title": "Last Frame", + "bounding": [ + 1050, + 3680, + 770, + 370 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 5, + "title": "Model", + "bounding": [ + 730, + 2240, + 500, + 980 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 6, + "title": "Prompt", + "bounding": [ + 1260, + 2240, + 680, + 980 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 7, + "title": "Sampling", + "bounding": [ + 1970, + 2240, + 770, + 980 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 8, + "title": "Decoding", + "bounding": [ + 2770, + 2240, + 450, + 980 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 203, + "origin_id": 214, + "origin_slot": 0, + "target_id": 195, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 205, + "origin_id": 224, + "origin_slot": 0, + "target_id": 197, + "target_slot": 0, + "type": "VAE" + }, + { + "id": 207, + "origin_id": 205, + "origin_slot": 0, + "target_id": 197, + "target_slot": 2, + "type": "INT" + }, + { + "id": 210, + "origin_id": 213, + "origin_slot": 0, + "target_id": 199, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 213, + "origin_id": 204, + "origin_slot": 0, + "target_id": 200, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 214, + "origin_id": 204, + "origin_slot": 1, + "target_id": 200, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 215, + "origin_id": 221, + "origin_slot": 0, + "target_id": 200, + "target_slot": 2, + "type": "LATENT" + }, + { + "id": 218, + "origin_id": 203, + "origin_slot": 0, + "target_id": 201, + "target_slot": 0, + "type": "INT" + }, + { + "id": 219, + "origin_id": 203, + "origin_slot": 1, + "target_id": 201, + "target_slot": 1, + "type": "INT" + }, + { + "id": 221, + "origin_id": 222, + "origin_slot": 0, + "target_id": 202, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 222, + "origin_id": 217, + "origin_slot": 0, + "target_id": 202, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 223, + "origin_id": 212, + "origin_slot": 0, + "target_id": 202, + "target_slot": 2, + "type": "FLOAT" + }, + { + "id": 224, + "origin_id": 213, + "origin_slot": 0, + "target_id": 203, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 225, + "origin_id": 206, + "origin_slot": 0, + "target_id": 204, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 226, + "origin_id": 206, + "origin_slot": 1, + "target_id": 204, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 227, + "origin_id": 223, + "origin_slot": 2, + "target_id": 204, + "target_slot": 2, + "type": "VAE" + }, + { + "id": 228, + "origin_id": 206, + "origin_slot": 2, + "target_id": 204, + "target_slot": 3, + "type": "LATENT" + }, + { + "id": 229, + "origin_id": 195, + "origin_slot": 0, + "target_id": 204, + "target_slot": 4, + "type": "IMAGE" + }, + { + "id": 236, + "origin_id": 202, + "origin_slot": 0, + "target_id": 206, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 237, + "origin_id": 202, + "origin_slot": 1, + "target_id": 206, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 238, + "origin_id": 223, + "origin_slot": 2, + "target_id": 206, + "target_slot": 2, + "type": "VAE" + }, + { + "id": 239, + "origin_id": 201, + "origin_slot": 0, + "target_id": 206, + "target_slot": 3, + "type": "LATENT" + }, + { + "id": 240, + "origin_id": 199, + "origin_slot": 0, + "target_id": 206, + "target_slot": 4, + "type": "IMAGE" + }, + { + "id": 241, + "origin_id": 223, + "origin_slot": 0, + "target_id": 207, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 242, + "origin_id": 204, + "origin_slot": 0, + "target_id": 207, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 243, + "origin_id": 204, + "origin_slot": 1, + "target_id": 207, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 244, + "origin_id": 204, + "origin_slot": 2, + "target_id": 210, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 245, + "origin_id": 197, + "origin_slot": 0, + "target_id": 210, + "target_slot": 1, + "type": "LATENT" + }, + { + "id": 246, + "origin_id": 196, + "origin_slot": 0, + "target_id": 211, + "target_slot": 0, + "type": "NOISE" + }, + { + "id": 247, + "origin_id": 207, + "origin_slot": 0, + "target_id": 211, + "target_slot": 1, + "type": "GUIDER" + }, + { + "id": 248, + "origin_id": 208, + "origin_slot": 0, + "target_id": 211, + "target_slot": 2, + "type": "SAMPLER" + }, + { + "id": 249, + "origin_id": 209, + "origin_slot": 0, + "target_id": 211, + "target_slot": 3, + "type": "SIGMAS" + }, + { + "id": 250, + "origin_id": 210, + "origin_slot": 0, + "target_id": 211, + "target_slot": 4, + "type": "LATENT" + }, + { + "id": 235, + "origin_id": 205, + "origin_slot": 0, + "target_id": 212, + "target_slot": 0, + "type": "INT" + }, + { + "id": 208, + "origin_id": 215, + "origin_slot": 0, + "target_id": 213, + "target_slot": 2, + "type": "INT" + }, + { + "id": 209, + "origin_id": 216, + "origin_slot": 0, + "target_id": 213, + "target_slot": 3, + "type": "INT" + }, + { + "id": 201, + "origin_id": 215, + "origin_slot": 0, + "target_id": 214, + "target_slot": 2, + "type": "INT" + }, + { + "id": 202, + "origin_id": 216, + "origin_slot": 0, + "target_id": 214, + "target_slot": 3, + "type": "INT" + }, + { + "id": 230, + "origin_id": 225, + "origin_slot": 0, + "target_id": 217, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 232, + "origin_id": 219, + "origin_slot": 0, + "target_id": 218, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 233, + "origin_id": 220, + "origin_slot": 0, + "target_id": 218, + "target_slot": 1, + "type": "AUDIO" + }, + { + "id": 234, + "origin_id": 212, + "origin_slot": 0, + "target_id": 218, + "target_slot": 2, + "type": "FLOAT" + }, + { + "id": 211, + "origin_id": 200, + "origin_slot": 2, + "target_id": 219, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 212, + "origin_id": 223, + "origin_slot": 2, + "target_id": 219, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 216, + "origin_id": 221, + "origin_slot": 1, + "target_id": 220, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 217, + "origin_id": 224, + "origin_slot": 0, + "target_id": 220, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 204, + "origin_id": 211, + "origin_slot": 1, + "target_id": 221, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 231, + "origin_id": 225, + "origin_slot": 0, + "target_id": 222, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 251, + "origin_id": -10, + "origin_slot": 0, + "target_id": 213, + "target_slot": 0, + "type": "IMAGE,MASK" + }, + { + "id": 253, + "origin_id": -10, + "origin_slot": 1, + "target_id": 214, + "target_slot": 0, + "type": "IMAGE,MASK" + }, + { + "id": 252, + "origin_id": 218, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "VIDEO" + }, + { + "id": 260, + "origin_id": 198, + "origin_slot": 0, + "target_id": 226, + "target_slot": 0, + "type": "INT" + }, + { + "id": 261, + "origin_id": 205, + "origin_slot": 0, + "target_id": 226, + "target_slot": 1, + "type": "INT" + }, + { + "id": 262, + "origin_id": 226, + "origin_slot": 1, + "target_id": 197, + "target_slot": 1, + "type": "INT" + }, + { + "id": 263, + "origin_id": 226, + "origin_slot": 1, + "target_id": 201, + "target_slot": 2, + "type": "INT" + }, + { + "id": 265, + "origin_id": -10, + "origin_slot": 2, + "target_id": 222, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 266, + "origin_id": -10, + "origin_slot": 3, + "target_id": 215, + "target_slot": 0, + "type": "INT" + }, + { + "id": 267, + "origin_id": -10, + "origin_slot": 4, + "target_id": 216, + "target_slot": 0, + "type": "INT" + }, + { + "id": 268, + "origin_id": -10, + "origin_slot": 5, + "target_id": 198, + "target_slot": 0, + "type": "INT" + }, + { + "id": 269, + "origin_id": -10, + "origin_slot": 6, + "target_id": 205, + "target_slot": 0, + "type": "INT" + }, + { + "id": 270, + "origin_id": -10, + "origin_slot": 7, + "target_id": 196, + "target_slot": 0, + "type": "INT" + }, + { + "id": 272, + "origin_id": -10, + "origin_slot": 8, + "target_id": 224, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 273, + "origin_id": -10, + "origin_slot": 9, + "target_id": 225, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 275, + "origin_id": -10, + "origin_slot": 8, + "target_id": 225, + "target_slot": 1, + "type": "COMBO" + }, + { + "id": 276, + "origin_id": -10, + "origin_slot": 8, + "target_id": 223, + "target_slot": 0, + "type": "COMBO" + } + ], + "extra": {}, + "category": "Video generation and editing/First-Last-Frame to Video", + "description": "Generates a video that interpolates between the first and last keyframes using LTX-2.3, including optional audio." + } + ] + }, + "extra": { + "ue_links": [] + } +} \ No newline at end of file diff --git a/blueprints/Frame Interpolation.json b/blueprints/Frame Interpolation.json new file mode 100644 index 000000000..8e183de7e --- /dev/null +++ b/blueprints/Frame Interpolation.json @@ -0,0 +1,858 @@ +{ + "revision": 0, + "last_node_id": 16, + "last_link_id": 0, + "nodes": [ + { + "id": 16, + "type": "022693be-2baa-4009-870a-28921508a7ef", + "pos": [ + -2990, + -3240 + ], + "size": [ + 410, + 200 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "video", + "name": "video", + "type": "VIDEO", + "link": null + }, + { + "label": "multiplier", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": null + }, + { + "label": "enable_fps_multiplier", + "name": "value_1", + "type": "BOOLEAN", + "widget": { + "name": "value_1" + }, + "link": null + }, + { + "name": "model_name", + "type": "COMBO", + "widget": { + "name": "model_name" + }, + "link": null + } + ], + "outputs": [ + { + "label": "VIDEO", + "name": "VIDEO_1", + "type": "VIDEO", + "links": [] + }, + { + "name": "IMAGE", + "type": "IMAGE", + "links": null + } + ], + "properties": { + "proxyWidgets": [ + [ + "9", + "value" + ], + [ + "13", + "value" + ], + [ + "1", + "model_name" + ] + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "cnr_id": "comfy-core", + "ver": "0.19.3" + }, + "widgets_values": [], + "title": "Frame Interpolation" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "022693be-2baa-4009-870a-28921508a7ef", + "version": 1, + "state": { + "lastGroupId": 0, + "lastNodeId": 17, + "lastLinkId": 28, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Frame Interpolation", + "inputNode": { + "id": -10, + "bounding": [ + -2810, + -3070, + 159.7421875, + 120 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + -1270, + -3075, + 120, + 80 + ] + }, + "inputs": [ + { + "id": "05e31c51-dcb6-4a1e-9651-1b9ad4f7a287", + "name": "video", + "type": "VIDEO", + "linkIds": [ + 2 + ], + "localized_name": "video", + "pos": [ + -2670.2578125, + -3050 + ] + }, + { + "id": "feecb409-7d1c-4a99-9c63-50c5fecdd3c9", + "name": "value", + "type": "INT", + "linkIds": [ + 22 + ], + "label": "multiplier", + "pos": [ + -2670.2578125, + -3030 + ] + }, + { + "id": "0b8a861b-b581-4068-9e8c-f8d15daf1ca6", + "name": "value_1", + "type": "BOOLEAN", + "linkIds": [ + 23 + ], + "label": "enable_fps_multiplier", + "pos": [ + -2670.2578125, + -3010 + ] + }, + { + "id": "a22b101e-8773-4e17-a297-7ee3aae09162", + "name": "model_name", + "type": "COMBO", + "linkIds": [ + 24 + ], + "pos": [ + -2670.2578125, + -2990 + ] + } + ], + "outputs": [ + { + "id": "ef2ada05-d5aa-492a-9394-6c3e71e39ebb", + "name": "VIDEO_1", + "type": "VIDEO", + "linkIds": [ + 26 + ], + "label": "VIDEO", + "pos": [ + -1250, + -3055 + ] + }, + { + "id": "5aacc622-2a07-4983-b31c-e04461f7f953", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 28 + ], + "pos": [ + -1250, + -3035 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 1, + "type": "FrameInterpolationModelLoader", + "pos": [ + -2510, + -3370 + ], + "size": [ + 370, + 90 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "model_name", + "name": "model_name", + "type": "COMBO", + "widget": { + "name": "model_name" + }, + "link": 24 + } + ], + "outputs": [ + { + "localized_name": "INTERP_MODEL", + "name": "INTERP_MODEL", + "type": "INTERP_MODEL", + "links": [ + 1 + ] + } + ], + "properties": { + "Node name for S&R": "FrameInterpolationModelLoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "cnr_id": "comfy-core", + "ver": "0.19.3", + "models": [ + { + "name": "film_net_fp16.safetensors", + "url": "https://huggingface.co/Comfy-Org/frame_interpolation/resolve/main/frame_interpolation/film_net_fp16.safetensors", + "directory": "frame_interpolation" + } + ] + }, + "widgets_values": [ + "film_net_fp16.safetensors" + ] + }, + { + "id": 2, + "type": "FrameInterpolate", + "pos": [ + -2040, + -3370 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "interp_model", + "name": "interp_model", + "type": "INTERP_MODEL", + "link": 1 + }, + { + "localized_name": "images", + "name": "images", + "type": "IMAGE", + "link": 3 + }, + { + "localized_name": "multiplier", + "name": "multiplier", + "type": "INT", + "widget": { + "name": "multiplier" + }, + "link": 8 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 4, + 28 + ] + } + ], + "properties": { + "Node name for S&R": "FrameInterpolate", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "cnr_id": "comfy-core", + "ver": "0.19.3" + }, + "widgets_values": [ + 2 + ] + }, + { + "id": 5, + "type": "CreateVideo", + "pos": [ + -1600, + -3370 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "images", + "name": "images", + "type": "IMAGE", + "link": 4 + }, + { + "localized_name": "audio", + "name": "audio", + "shape": 7, + "type": "AUDIO", + "link": 5 + }, + { + "localized_name": "fps", + "name": "fps", + "type": "FLOAT", + "widget": { + "name": "fps" + }, + "link": 12 + } + ], + "outputs": [ + { + "localized_name": "VIDEO", + "name": "VIDEO", + "type": "VIDEO", + "links": [ + 26 + ] + } + ], + "properties": { + "Node name for S&R": "CreateVideo", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "cnr_id": "comfy-core", + "ver": "0.19.3" + }, + "widgets_values": [ + 30 + ] + }, + { + "id": 9, + "type": "PrimitiveInt", + "pos": [ + -2500, + -2970 + ], + "size": [ + 270, + 90 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": 22 + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 8, + 19 + ] + } + ], + "title": "Int (Multiplier)", + "properties": { + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "cnr_id": "comfy-core", + "ver": "0.19.3" + }, + "widgets_values": [ + 2, + "fixed" + ] + }, + { + "id": 10, + "type": "ComfySwitchNode", + "pos": [ + -1610, + -3120 + ], + "size": [ + 270, + 130 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "on_false", + "name": "on_false", + "type": "*", + "link": 11 + }, + { + "localized_name": "on_true", + "name": "on_true", + "type": "*", + "link": 13 + }, + { + "localized_name": "switch", + "name": "switch", + "type": "BOOLEAN", + "widget": { + "name": "switch" + }, + "link": 15 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "*", + "links": [ + 12 + ] + } + ], + "properties": { + "Node name for S&R": "ComfySwitchNode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "cnr_id": "comfy-core", + "ver": "0.19.3" + }, + "widgets_values": [ + true + ] + }, + { + "id": 13, + "type": "PrimitiveBoolean", + "pos": [ + -2500, + -2770 + ], + "size": [ + 310, + 90 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "BOOLEAN", + "widget": { + "name": "value" + }, + "link": 23 + } + ], + "outputs": [ + { + "localized_name": "BOOLEAN", + "name": "BOOLEAN", + "type": "BOOLEAN", + "links": [ + 15 + ] + } + ], + "title": "Boolean (Apply multiplier to FPS?)", + "properties": { + "Node name for S&R": "PrimitiveBoolean", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "cnr_id": "comfy-core", + "ver": "0.19.3" + }, + "widgets_values": [ + true + ] + }, + { + "id": 3, + "type": "GetVideoComponents", + "pos": [ + -2500, + -3170 + ], + "size": [ + 230, + 100 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "video", + "name": "video", + "type": "VIDEO", + "link": 2 + } + ], + "outputs": [ + { + "localized_name": "images", + "name": "images", + "type": "IMAGE", + "links": [ + 3 + ] + }, + { + "localized_name": "audio", + "name": "audio", + "type": "AUDIO", + "links": [ + 5 + ] + }, + { + "localized_name": "fps", + "name": "fps", + "type": "FLOAT", + "links": [ + 11, + 18 + ] + } + ], + "properties": { + "Node name for S&R": "GetVideoComponents", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "cnr_id": "comfy-core", + "ver": "0.19.3" + } + }, + { + "id": 11, + "type": "ComfyMathExpression", + "pos": [ + -2090, + -3070 + ], + "size": [ + 400, + 210 + ], + "flags": { + "collapsed": false + }, + "order": 6, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 18 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": 19 + }, + { + "label": "c", + "localized_name": "values.c", + "name": "values.c", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": [ + 13 + ] + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": null + } + ], + "properties": { + "Node name for S&R": "ComfyMathExpression", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "cnr_id": "comfy-core", + "ver": "0.19.3" + }, + "widgets_values": [ + "min(abs(b), 16) * a" + ] + } + ], + "groups": [], + "links": [ + { + "id": 1, + "origin_id": 1, + "origin_slot": 0, + "target_id": 2, + "target_slot": 0, + "type": "INTERP_MODEL" + }, + { + "id": 3, + "origin_id": 3, + "origin_slot": 0, + "target_id": 2, + "target_slot": 1, + "type": "IMAGE" + }, + { + "id": 8, + "origin_id": 9, + "origin_slot": 0, + "target_id": 2, + "target_slot": 2, + "type": "INT" + }, + { + "id": 4, + "origin_id": 2, + "origin_slot": 0, + "target_id": 5, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 5, + "origin_id": 3, + "origin_slot": 1, + "target_id": 5, + "target_slot": 1, + "type": "AUDIO" + }, + { + "id": 12, + "origin_id": 10, + "origin_slot": 0, + "target_id": 5, + "target_slot": 2, + "type": "FLOAT" + }, + { + "id": 11, + "origin_id": 3, + "origin_slot": 2, + "target_id": 10, + "target_slot": 0, + "type": "FLOAT" + }, + { + "id": 13, + "origin_id": 11, + "origin_slot": 0, + "target_id": 10, + "target_slot": 1, + "type": "FLOAT" + }, + { + "id": 15, + "origin_id": 13, + "origin_slot": 0, + "target_id": 10, + "target_slot": 2, + "type": "BOOLEAN" + }, + { + "id": 18, + "origin_id": 3, + "origin_slot": 2, + "target_id": 11, + "target_slot": 0, + "type": "FLOAT" + }, + { + "id": 19, + "origin_id": 9, + "origin_slot": 0, + "target_id": 11, + "target_slot": 1, + "type": "INT" + }, + { + "id": 2, + "origin_id": -10, + "origin_slot": 0, + "target_id": 3, + "target_slot": 0, + "type": "VIDEO" + }, + { + "id": 22, + "origin_id": -10, + "origin_slot": 1, + "target_id": 9, + "target_slot": 0, + "type": "INT" + }, + { + "id": 23, + "origin_id": -10, + "origin_slot": 2, + "target_id": 13, + "target_slot": 0, + "type": "BOOLEAN" + }, + { + "id": 24, + "origin_id": -10, + "origin_slot": 3, + "target_id": 1, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 26, + "origin_id": 5, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "VIDEO" + }, + { + "id": 28, + "origin_id": 2, + "origin_slot": 0, + "target_id": -20, + "target_slot": 1, + "type": "IMAGE" + } + ], + "extra": {}, + "category": "Video Tools", + "description": "Increases video frame rate by synthesizing intermediate frames with a frame interpolation model." + } + ] + }, + "extra": {} +} \ No newline at end of file diff --git a/blueprints/Get Any Video Frame.json b/blueprints/Get Any Video Frame.json new file mode 100644 index 000000000..9ff0f8e6e --- /dev/null +++ b/blueprints/Get Any Video Frame.json @@ -0,0 +1,485 @@ +{ + "revision": 0, + "last_node_id": 98, + "last_link_id": 0, + "nodes": [ + { + "id": 98, + "type": "dca6e78d-fb06-421e-97f7-6ce17a665260", + "pos": [ + -410, + -2230 + ], + "size": [ + 270, + 104 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "name": "video", + "type": "VIDEO", + "link": null + }, + { + "label": "frame_index", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "name": "IMAGE", + "type": "IMAGE", + "links": [] + } + ], + "title": "Get Any Video Frame", + "properties": { + "proxyWidgets": [ + [ + "100", + "value" + ] + ] + }, + "widgets_values": [] + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "dca6e78d-fb06-421e-97f7-6ce17a665260", + "version": 1, + "state": { + "lastGroupId": 1, + "lastNodeId": 136, + "lastLinkId": 302, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Get Any Video Frame", + "inputNode": { + "id": -10, + "bounding": [ + 380, + -57, + 120, + 80 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 1460, + -57, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "2ceec378-8dcf-4340-8570-155967f59a93", + "name": "video", + "type": "VIDEO", + "linkIds": [ + 4 + ], + "pos": [ + 480, + -37 + ] + }, + { + "id": "819955f6-c686-4896-8032-ff2d0059109a", + "name": "value", + "type": "INT", + "linkIds": [ + 283 + ], + "label": "frame_index", + "pos": [ + 480, + -17 + ] + } + ], + "outputs": [ + { + "id": "1ab0684d-6a44-45b6-8aa4-a0b971a1d41e", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 5 + ], + "pos": [ + 1480, + -37 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 1, + "type": "GetVideoComponents", + "pos": [ + 560, + -150 + ], + "size": [ + 230, + 120 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "video", + "name": "video", + "type": "VIDEO", + "link": 4 + } + ], + "outputs": [ + { + "localized_name": "images", + "name": "images", + "type": "IMAGE", + "links": [ + 1, + 2 + ] + }, + { + "localized_name": "audio", + "name": "audio", + "type": "AUDIO", + "links": null + }, + { + "localized_name": "fps", + "name": "fps", + "type": "FLOAT", + "links": null + } + ], + "properties": { + "Node name for S&R": "GetVideoComponents" + } + }, + { + "id": 2, + "type": "GetImageSize", + "pos": [ + 560, + 50 + ], + "size": [ + 230, + 120 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 1 + } + ], + "outputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "links": null + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "links": null + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "links": [ + 285 + ] + } + ], + "properties": { + "Node name for S&R": "GetImageSize" + } + }, + { + "id": 3, + "type": "ImageFromBatch", + "pos": [ + 1130, + -150 + ], + "size": [ + 270, + 140 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 2 + }, + { + "localized_name": "batch_index", + "name": "batch_index", + "type": "INT", + "widget": { + "name": "batch_index" + }, + "link": 286 + }, + { + "localized_name": "length", + "name": "length", + "type": "INT", + "widget": { + "name": "length" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 5 + ] + } + ], + "properties": { + "Node name for S&R": "ImageFromBatch" + }, + "widgets_values": [ + 0, + 1 + ] + }, + { + "id": 99, + "type": "ComfyMathExpression", + "pos": [ + 910, + 100 + ], + "size": [ + 400, + 200 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 284 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": 285 + }, + { + "label": "c", + "localized_name": "values.c", + "name": "values.c", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 286 + ] + } + ], + "properties": { + "Node name for S&R": "ComfyMathExpression" + }, + "widgets_values": [ + "min(max(int(a if a >= 0 else b + a), 0), b - 1)" + ] + }, + { + "id": 100, + "type": "PrimitiveInt", + "pos": [ + 560, + 250 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": 283 + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 284 + ] + } + ], + "properties": { + "Node name for S&R": "PrimitiveInt" + }, + "widgets_values": [ + 0, + "fixed" + ] + } + ], + "groups": [], + "links": [ + { + "id": 1, + "origin_id": 1, + "origin_slot": 0, + "target_id": 2, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 2, + "origin_id": 1, + "origin_slot": 0, + "target_id": 3, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 4, + "origin_id": -10, + "origin_slot": 0, + "target_id": 1, + "target_slot": 0, + "type": "VIDEO" + }, + { + "id": 5, + "origin_id": 3, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 283, + "origin_id": -10, + "origin_slot": 1, + "target_id": 100, + "target_slot": 0, + "type": "INT" + }, + { + "id": 284, + "origin_id": 100, + "origin_slot": 0, + "target_id": 99, + "target_slot": 0, + "type": "INT" + }, + { + "id": 285, + "origin_id": 2, + "origin_slot": 2, + "target_id": 99, + "target_slot": 1, + "type": "INT" + }, + { + "id": 286, + "origin_id": 99, + "origin_slot": 1, + "target_id": 3, + "target_slot": 1, + "type": "INT" + } + ], + "extra": {}, + "category": "Video Tools", + "description": "Extracts one image frame from a video at a chosen index, with optional trim and FPS control." + } + ] + }, + "extra": { + "ds": { + "scale": 1.197015527856339, + "offset": [ + -168.76833554248222, + 540.6638955283997 + ] + }, + "frontendVersion": "1.42.8" + } +} \ No newline at end of file diff --git a/blueprints/Glow.json b/blueprints/Glow.json index 8c690fc68..2bbfdee51 100644 --- a/blueprints/Glow.json +++ b/blueprints/Glow.json @@ -268,7 +268,7 @@ "Node name for S&R": "GLSLShader" }, "widgets_values": [ - "#version 300 es\nprecision mediump float;\n\nuniform sampler2D u_image0;\nuniform vec2 u_resolution;\nuniform int u_int0; // Blend mode\nuniform int u_int1; // Color tint\nuniform float u_float0; // Intensity\nuniform float u_float1; // Radius\nuniform float u_float2; // Threshold\n\nin vec2 v_texCoord;\nout vec4 fragColor;\n\nconst int BLEND_ADD = 0;\nconst int BLEND_SCREEN = 1;\nconst int BLEND_SOFT = 2;\nconst int BLEND_OVERLAY = 3;\nconst int BLEND_LIGHTEN = 4;\n\nconst float GOLDEN_ANGLE = 2.39996323;\nconst int MAX_SAMPLES = 48;\nconst vec3 LUMA = vec3(0.299, 0.587, 0.114);\n\nfloat hash(vec2 p) {\n p = fract(p * vec2(123.34, 456.21));\n p += dot(p, p + 45.32);\n return fract(p.x * p.y);\n}\n\nvec3 hexToRgb(int h) {\n return vec3(\n float((h >> 16) & 255),\n float((h >> 8) & 255),\n float(h & 255)\n ) * (1.0 / 255.0);\n}\n\nvec3 blend(vec3 base, vec3 glow, int mode) {\n if (mode == BLEND_SCREEN) {\n return 1.0 - (1.0 - base) * (1.0 - glow);\n }\n if (mode == BLEND_SOFT) {\n return mix(\n base - (1.0 - 2.0 * glow) * base * (1.0 - base),\n base + (2.0 * glow - 1.0) * (sqrt(base) - base),\n step(0.5, glow)\n );\n }\n if (mode == BLEND_OVERLAY) {\n return mix(\n 2.0 * base * glow,\n 1.0 - 2.0 * (1.0 - base) * (1.0 - glow),\n step(0.5, base)\n );\n }\n if (mode == BLEND_LIGHTEN) {\n return max(base, glow);\n }\n return base + glow;\n}\n\nvoid main() {\n vec4 original = texture(u_image0, v_texCoord);\n \n float intensity = u_float0 * 0.05;\n float radius = u_float1 * u_float1 * 0.012;\n \n if (intensity < 0.001 || radius < 0.1) {\n fragColor = original;\n return;\n }\n \n float threshold = 1.0 - u_float2 * 0.01;\n float t0 = threshold - 0.15;\n float t1 = threshold + 0.15;\n \n vec2 texelSize = 1.0 / u_resolution;\n float radius2 = radius * radius;\n \n float sampleScale = clamp(radius * 0.75, 0.35, 1.0);\n int samples = int(float(MAX_SAMPLES) * sampleScale);\n \n float noise = hash(gl_FragCoord.xy);\n float angleOffset = noise * GOLDEN_ANGLE;\n float radiusJitter = 0.85 + noise * 0.3;\n \n float ca = cos(GOLDEN_ANGLE);\n float sa = sin(GOLDEN_ANGLE);\n vec2 dir = vec2(cos(angleOffset), sin(angleOffset));\n \n vec3 glow = vec3(0.0);\n float totalWeight = 0.0;\n \n // Center tap\n float centerMask = smoothstep(t0, t1, dot(original.rgb, LUMA));\n glow += original.rgb * centerMask * 2.0;\n totalWeight += 2.0;\n \n for (int i = 1; i < MAX_SAMPLES; i++) {\n if (i >= samples) break;\n \n float fi = float(i);\n float dist = sqrt(fi / float(samples)) * radius * radiusJitter;\n \n vec2 offset = dir * dist * texelSize;\n vec3 c = texture(u_image0, v_texCoord + offset).rgb;\n float mask = smoothstep(t0, t1, dot(c, LUMA));\n \n float w = 1.0 - (dist * dist) / (radius2 * 1.5);\n w = max(w, 0.0);\n w *= w;\n \n glow += c * mask * w;\n totalWeight += w;\n \n dir = vec2(\n dir.x * ca - dir.y * sa,\n dir.x * sa + dir.y * ca\n );\n }\n \n glow *= intensity / max(totalWeight, 0.001);\n \n if (u_int1 > 0) {\n glow *= hexToRgb(u_int1);\n }\n \n vec3 result = blend(original.rgb, glow, u_int0);\n result += (noise - 0.5) * (1.0 / 255.0);\n \n fragColor = vec4(clamp(result, 0.0, 1.0), original.a);\n}", + "#version 300 es\nprecision mediump float;\n\nuniform sampler2D u_image0;\nuniform int u_int0; // Blend mode\nuniform int u_int1; // Color tint\nuniform float u_float0; // Intensity\nuniform float u_float1; // Radius\nuniform float u_float2; // Threshold\n\nin vec2 v_texCoord;\nout vec4 fragColor;\n\nconst int BLEND_ADD = 0;\nconst int BLEND_SCREEN = 1;\nconst int BLEND_SOFT = 2;\nconst int BLEND_OVERLAY = 3;\nconst int BLEND_LIGHTEN = 4;\n\nconst float GOLDEN_ANGLE = 2.39996323;\nconst int MAX_SAMPLES = 48;\nconst vec3 LUMA = vec3(0.299, 0.587, 0.114);\n\nfloat hash(vec2 p) {\n p = fract(p * vec2(123.34, 456.21));\n p += dot(p, p + 45.32);\n return fract(p.x * p.y);\n}\n\nvec3 hexToRgb(int h) {\n return vec3(\n float((h >> 16) & 255),\n float((h >> 8) & 255),\n float(h & 255)\n ) * (1.0 / 255.0);\n}\n\nvec3 blend(vec3 base, vec3 glow, int mode) {\n if (mode == BLEND_SCREEN) {\n return 1.0 - (1.0 - base) * (1.0 - glow);\n }\n if (mode == BLEND_SOFT) {\n return mix(\n base - (1.0 - 2.0 * glow) * base * (1.0 - base),\n base + (2.0 * glow - 1.0) * (sqrt(base) - base),\n step(0.5, glow)\n );\n }\n if (mode == BLEND_OVERLAY) {\n return mix(\n 2.0 * base * glow,\n 1.0 - 2.0 * (1.0 - base) * (1.0 - glow),\n step(0.5, base)\n );\n }\n if (mode == BLEND_LIGHTEN) {\n return max(base, glow);\n }\n return base + glow;\n}\n\nvoid main() {\n vec4 original = texture(u_image0, v_texCoord);\n \n float intensity = u_float0 * 0.05;\n float radius = u_float1 * u_float1 * 0.012;\n \n if (intensity < 0.001 || radius < 0.1) {\n fragColor = original;\n return;\n }\n \n float threshold = 1.0 - u_float2 * 0.01;\n float t0 = threshold - 0.15;\n float t1 = threshold + 0.15;\n \n vec2 texelSize = 1.0 / vec2(textureSize(u_image0, 0));\n float radius2 = radius * radius;\n \n float sampleScale = clamp(radius * 0.75, 0.35, 1.0);\n int samples = int(float(MAX_SAMPLES) * sampleScale);\n \n float noise = hash(gl_FragCoord.xy);\n float angleOffset = noise * GOLDEN_ANGLE;\n float radiusJitter = 0.85 + noise * 0.3;\n \n float ca = cos(GOLDEN_ANGLE);\n float sa = sin(GOLDEN_ANGLE);\n vec2 dir = vec2(cos(angleOffset), sin(angleOffset));\n \n vec3 glow = vec3(0.0);\n float totalWeight = 0.0;\n \n // Center tap\n float centerMask = smoothstep(t0, t1, dot(original.rgb, LUMA));\n glow += original.rgb * centerMask * 2.0;\n totalWeight += 2.0;\n \n for (int i = 1; i < MAX_SAMPLES; i++) {\n if (i >= samples) break;\n \n float fi = float(i);\n float dist = sqrt(fi / float(samples)) * radius * radiusJitter;\n \n vec2 offset = dir * dist * texelSize;\n vec3 c = texture(u_image0, v_texCoord + offset).rgb;\n float mask = smoothstep(t0, t1, dot(c, LUMA));\n \n float w = 1.0 - (dist * dist) / (radius2 * 1.5);\n w = max(w, 0.0);\n w *= w;\n \n glow += c * mask * w;\n totalWeight += w;\n \n dir = vec2(\n dir.x * ca - dir.y * sa,\n dir.x * sa + dir.y * ca\n );\n }\n \n glow *= intensity / max(totalWeight, 0.001);\n \n if (u_int1 > 0) {\n glow *= hexToRgb(u_int1);\n }\n \n vec3 result = blend(original.rgb, glow, u_int0);\n result += (noise - 0.5) * (1.0 / 255.0);\n \n fragColor = vec4(clamp(result, 0.0, 1.0), original.a);\n}", "from_input" ] }, @@ -575,8 +575,9 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image Tools/Color adjust" + "category": "Image Tools/Color adjust", + "description": "Adds a glow/bloom effect around bright image areas via GPU fragment shader." } ] } -} +} \ No newline at end of file diff --git a/blueprints/Hue and Saturation.json b/blueprints/Hue and Saturation.json index 1a2df8937..cddf0154a 100644 --- a/blueprints/Hue and Saturation.json +++ b/blueprints/Hue and Saturation.json @@ -752,8 +752,9 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image Tools/Color adjust" + "category": "Image Tools/Color adjust", + "description": "Adjusts hue, saturation, and lightness of an image using a real-time GPU fragment shader." } ] } -} +} \ No newline at end of file diff --git a/blueprints/Image Blur.json b/blueprints/Image Blur.json index b1d449e32..0ca8d9931 100644 --- a/blueprints/Image Blur.json +++ b/blueprints/Image Blur.json @@ -331,7 +331,7 @@ "Node name for S&R": "GLSLShader" }, "widgets_values": [ - "#version 300 es\n#pragma passes 2\nprecision highp float;\n\n// Blur type constants\nconst int BLUR_GAUSSIAN = 0;\nconst int BLUR_BOX = 1;\nconst int BLUR_RADIAL = 2;\n\n// Radial blur config\nconst int RADIAL_SAMPLES = 12;\nconst float RADIAL_STRENGTH = 0.0003;\n\nuniform sampler2D u_image0;\nuniform vec2 u_resolution;\nuniform int u_int0; // Blur type (BLUR_GAUSSIAN, BLUR_BOX, BLUR_RADIAL)\nuniform float u_float0; // Blur radius/amount\nuniform int u_pass; // Pass index (0 = horizontal, 1 = vertical)\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\n\nfloat gaussian(float x, float sigma) {\n return exp(-(x * x) / (2.0 * sigma * sigma));\n}\n\nvoid main() {\n vec2 texelSize = 1.0 / u_resolution;\n float radius = max(u_float0, 0.0);\n\n // Radial (angular) blur - single pass, doesn't use separable\n if (u_int0 == BLUR_RADIAL) {\n // Only execute on first pass\n if (u_pass > 0) {\n fragColor0 = texture(u_image0, v_texCoord);\n return;\n }\n\n vec2 center = vec2(0.5);\n vec2 dir = v_texCoord - center;\n float dist = length(dir);\n\n if (dist < 1e-4) {\n fragColor0 = texture(u_image0, v_texCoord);\n return;\n }\n\n vec4 sum = vec4(0.0);\n float totalWeight = 0.0;\n float angleStep = radius * RADIAL_STRENGTH;\n\n dir /= dist;\n\n float cosStep = cos(angleStep);\n float sinStep = sin(angleStep);\n\n float negAngle = -float(RADIAL_SAMPLES) * angleStep;\n vec2 rotDir = vec2(\n dir.x * cos(negAngle) - dir.y * sin(negAngle),\n dir.x * sin(negAngle) + dir.y * cos(negAngle)\n );\n\n for (int i = -RADIAL_SAMPLES; i <= RADIAL_SAMPLES; i++) {\n vec2 uv = center + rotDir * dist;\n float w = 1.0 - abs(float(i)) / float(RADIAL_SAMPLES);\n sum += texture(u_image0, uv) * w;\n totalWeight += w;\n\n rotDir = vec2(\n rotDir.x * cosStep - rotDir.y * sinStep,\n rotDir.x * sinStep + rotDir.y * cosStep\n );\n }\n\n fragColor0 = sum / max(totalWeight, 0.001);\n return;\n }\n\n // Separable Gaussian / Box blur\n int samples = int(ceil(radius));\n\n if (samples == 0) {\n fragColor0 = texture(u_image0, v_texCoord);\n return;\n }\n\n // Direction: pass 0 = horizontal, pass 1 = vertical\n vec2 dir = (u_pass == 0) ? vec2(1.0, 0.0) : vec2(0.0, 1.0);\n\n vec4 color = vec4(0.0);\n float totalWeight = 0.0;\n float sigma = radius / 2.0;\n\n for (int i = -samples; i <= samples; i++) {\n vec2 offset = dir * float(i) * texelSize;\n vec4 sample_color = texture(u_image0, v_texCoord + offset);\n\n float weight;\n if (u_int0 == BLUR_GAUSSIAN) {\n weight = gaussian(float(i), sigma);\n } else {\n // BLUR_BOX\n weight = 1.0;\n }\n\n color += sample_color * weight;\n totalWeight += weight;\n }\n\n fragColor0 = color / totalWeight;\n}\n", + "#version 300 es\n#pragma passes 2\nprecision highp float;\n\n// Blur type constants\nconst int BLUR_GAUSSIAN = 0;\nconst int BLUR_BOX = 1;\nconst int BLUR_RADIAL = 2;\n\n// Radial blur config\nconst int RADIAL_SAMPLES = 12;\nconst float RADIAL_STRENGTH = 0.0003;\n\nuniform sampler2D u_image0;\nuniform int u_int0; // Blur type (BLUR_GAUSSIAN, BLUR_BOX, BLUR_RADIAL)\nuniform float u_float0; // Blur radius/amount\nuniform int u_pass; // Pass index (0 = horizontal, 1 = vertical)\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\n\nfloat gaussian(float x, float sigma) {\n return exp(-(x * x) / (2.0 * sigma * sigma));\n}\n\nvoid main() {\n vec2 texelSize = 1.0 / vec2(textureSize(u_image0, 0));\n float radius = max(u_float0, 0.0);\n\n // Radial (angular) blur - single pass, doesn't use separable\n if (u_int0 == BLUR_RADIAL) {\n // Only execute on first pass\n if (u_pass > 0) {\n fragColor0 = texture(u_image0, v_texCoord);\n return;\n }\n\n vec2 center = vec2(0.5);\n vec2 dir = v_texCoord - center;\n float dist = length(dir);\n\n if (dist < 1e-4) {\n fragColor0 = texture(u_image0, v_texCoord);\n return;\n }\n\n vec4 sum = vec4(0.0);\n float totalWeight = 0.0;\n float angleStep = radius * RADIAL_STRENGTH;\n\n dir /= dist;\n\n float cosStep = cos(angleStep);\n float sinStep = sin(angleStep);\n\n float negAngle = -float(RADIAL_SAMPLES) * angleStep;\n vec2 rotDir = vec2(\n dir.x * cos(negAngle) - dir.y * sin(negAngle),\n dir.x * sin(negAngle) + dir.y * cos(negAngle)\n );\n\n for (int i = -RADIAL_SAMPLES; i <= RADIAL_SAMPLES; i++) {\n vec2 uv = center + rotDir * dist;\n float w = 1.0 - abs(float(i)) / float(RADIAL_SAMPLES);\n sum += texture(u_image0, uv) * w;\n totalWeight += w;\n\n rotDir = vec2(\n rotDir.x * cosStep - rotDir.y * sinStep,\n rotDir.x * sinStep + rotDir.y * cosStep\n );\n }\n\n fragColor0 = sum / max(totalWeight, 0.001);\n return;\n }\n\n // Separable Gaussian / Box blur\n int samples = int(ceil(radius));\n\n if (samples == 0) {\n fragColor0 = texture(u_image0, v_texCoord);\n return;\n }\n\n // Direction: pass 0 = horizontal, pass 1 = vertical\n vec2 dir = (u_pass == 0) ? vec2(1.0, 0.0) : vec2(0.0, 1.0);\n\n vec4 color = vec4(0.0);\n float totalWeight = 0.0;\n float sigma = radius / 2.0;\n\n for (int i = -samples; i <= samples; i++) {\n vec2 offset = dir * float(i) * texelSize;\n vec4 sample_color = texture(u_image0, v_texCoord + offset);\n\n float weight;\n if (u_int0 == BLUR_GAUSSIAN) {\n weight = gaussian(float(i), sigma);\n } else {\n // BLUR_BOX\n weight = 1.0;\n }\n\n color += sample_color * weight;\n totalWeight += weight;\n }\n\n fragColor0 = color / totalWeight;\n}\n", "from_input" ] } @@ -374,7 +374,8 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image Tools/Blur" + "category": "Image Tools/Blur", + "description": "Applies Gaussian, Box, or Radial blur to soften images and create stylized depth or motion effects." } ] } diff --git a/blueprints/Image Captioning (gemini).json b/blueprints/Image Captioning (gemini).json index 98cfb8999..2fc5d6746 100644 --- a/blueprints/Image Captioning (gemini).json +++ b/blueprints/Image Captioning (gemini).json @@ -310,7 +310,8 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Text generation/Image Captioning" + "category": "Text generation/Image Captioning", + "description": "Generates descriptive captions for images using Google's Gemini multimodal LLM." } ] } diff --git a/blueprints/Image Channels.json b/blueprints/Image Channels.json index 9c7b675b2..b6fdff5be 100644 --- a/blueprints/Image Channels.json +++ b/blueprints/Image Channels.json @@ -315,8 +315,9 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image Tools/Color adjust" + "category": "Image Tools/Color adjust", + "description": "Manipulates individual RGBA channels for masking, compositing, and channel effects." } ] } -} +} \ No newline at end of file diff --git a/blueprints/Image Edit (FireRed Image Edit 1.1).json b/blueprints/Image Edit (FireRed Image Edit 1.1).json new file mode 100644 index 000000000..b82c7d18b --- /dev/null +++ b/blueprints/Image Edit (FireRed Image Edit 1.1).json @@ -0,0 +1,2149 @@ +{ + "revision": 0, + "last_node_id": 213, + "last_link_id": 0, + "nodes": [ + { + "id": 213, + "type": "e35fbbeb-d7b1-46d1-a74e-959517d0fb1a", + "pos": [ + -700, + -470 + ], + "size": [ + 500, + 0 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": null + }, + { + "label": "image2 (optional)", + "name": "image2_1", + "type": "IMAGE", + "link": null + }, + { + "label": "image3 (optional)", + "name": "image3_1", + "type": "IMAGE", + "link": null + }, + { + "name": "prompt", + "type": "STRING", + "widget": { + "name": "prompt" + }, + "link": null + }, + { + "label": "enable_turbo_mode", + "name": "value", + "type": "BOOLEAN", + "widget": { + "name": "value" + }, + "link": null + }, + { + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": null + }, + { + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": null + }, + { + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": null + }, + { + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": null + }, + { + "name": "lora_name", + "type": "COMBO", + "widget": { + "name": "lora_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [] + } + ], + "properties": { + "proxyWidgets": [ + [ + "208", + "prompt" + ], + [ + "207", + "value" + ], + [ + "210", + "seed" + ], + [ + "205", + "unet_name" + ], + [ + "203", + "clip_name" + ], + [ + "202", + "vae_name" + ], + [ + "204", + "lora_name" + ], + [ + "210", + "control_after_generate" + ] + ], + "cnr_id": "comfy-core", + "ver": "0.15.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [], + "title": "Image Edit (FireRed Image Edit 1.1)" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "e35fbbeb-d7b1-46d1-a74e-959517d0fb1a", + "version": 1, + "state": { + "lastGroupId": 8, + "lastNodeId": 213, + "lastLinkId": 378, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Image Edit (FireRed Image Edit 1.1)", + "inputNode": { + "id": -10, + "bounding": [ + -1670, + -1370, + 151.744140625, + 240 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 1860, + -1340, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "1d810e30-f1fb-4d10-95f8-3c5f7db2c8b7", + "name": "image", + "type": "IMAGE", + "linkIds": [ + 371 + ], + "localized_name": "image", + "pos": [ + -1538.255859375, + -1350 + ] + }, + { + "id": "a8decf32-2262-4cdd-9e6b-c0ca7d4cdebe", + "name": "image2_1", + "type": "IMAGE", + "linkIds": [ + 355, + 356 + ], + "label": "image2 (optional)", + "pos": [ + -1538.255859375, + -1330 + ] + }, + { + "id": "3ff7a4ed-8e3d-45d4-b1d8-40ed88a6def6", + "name": "image3_1", + "type": "IMAGE", + "linkIds": [ + 357, + 358 + ], + "label": "image3 (optional)", + "pos": [ + -1538.255859375, + -1310 + ] + }, + { + "id": "01d9e68c-c664-4584-9cde-66f60e54eb3c", + "name": "prompt", + "type": "STRING", + "linkIds": [ + 359 + ], + "pos": [ + -1538.255859375, + -1290 + ] + }, + { + "id": "97d24b10-6540-48c4-81eb-a432832f5729", + "name": "value", + "type": "BOOLEAN", + "linkIds": [ + 364 + ], + "label": "enable_turbo_mode", + "pos": [ + -1538.255859375, + -1270 + ] + }, + { + "id": "15890efb-ba15-41cd-91ef-5adad7a52167", + "name": "seed", + "type": "INT", + "linkIds": [ + 372 + ], + "pos": [ + -1538.255859375, + -1250 + ] + }, + { + "id": "43f22fe2-6836-4f75-8146-04c84fbba75d", + "name": "unet_name", + "type": "COMBO", + "linkIds": [ + 373 + ], + "pos": [ + -1538.255859375, + -1230 + ] + }, + { + "id": "cd5e4502-2aca-4645-9e2e-ca8719f05bf6", + "name": "clip_name", + "type": "COMBO", + "linkIds": [ + 374 + ], + "pos": [ + -1538.255859375, + -1210 + ] + }, + { + "id": "f6ae73dc-39e8-44b2-958d-705ae159ea86", + "name": "vae_name", + "type": "COMBO", + "linkIds": [ + 375 + ], + "pos": [ + -1538.255859375, + -1190 + ] + }, + { + "id": "66dc179d-e6c9-4485-a2db-a47d25b44363", + "name": "lora_name", + "type": "COMBO", + "linkIds": [ + 376 + ], + "pos": [ + -1538.255859375, + -1170 + ] + } + ], + "outputs": [ + { + "id": "712c5c76-8620-44e1-9c9d-0798b6cdb77a", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 292 + ], + "localized_name": "IMAGE", + "pos": [ + 1880, + -1320 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 193, + "type": "ModelSamplingAuraFlow", + "pos": [ + 1010, + -1680 + ], + "size": [ + 290, + 110 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 326 + }, + { + "localized_name": "shift", + "name": "shift", + "type": "FLOAT", + "widget": { + "name": "shift" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 294 + ] + } + ], + "properties": { + "Node name for S&R": "ModelSamplingAuraFlow", + "cnr_id": "comfy-core", + "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 3.1 + ] + }, + { + "id": 194, + "type": "ComfySwitchNode", + "pos": [ + 680, + -1690 + ], + "size": [ + 260, + 140 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "on_false", + "name": "on_false", + "type": "*", + "link": 324 + }, + { + "localized_name": "on_true", + "name": "on_true", + "type": "*", + "link": 325 + }, + { + "localized_name": "switch", + "name": "switch", + "type": "BOOLEAN", + "widget": { + "name": "switch" + }, + "link": 323 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "*", + "links": [ + 326 + ] + } + ], + "title": "Switch (Model)", + "properties": { + "Node name for S&R": "ComfySwitchNode", + "cnr_id": "comfy-core", + "ver": "0.15.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + false + ] + }, + { + "id": 195, + "type": "PrimitiveInt", + "pos": [ + 190, + -1680 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 329 + ] + } + ], + "title": "Int (Steps)", + "properties": { + "Node name for S&R": "PrimitiveInt", + "cnr_id": "comfy-core", + "ver": "0.15.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 40, + "fixed" + ] + }, + { + "id": 196, + "type": "CFGNorm", + "pos": [ + 1010, + -1510 + ], + "size": [ + 290, + 110 + ], + "flags": {}, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 294 + }, + { + "localized_name": "strength", + "name": "strength", + "type": "FLOAT", + "widget": { + "name": "strength" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "patched_model", + "name": "patched_model", + "type": "MODEL", + "links": [ + 295 + ] + } + ], + "properties": { + "Node name for S&R": "CFGNorm", + "cnr_id": "comfy-core", + "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1 + ] + }, + { + "id": 197, + "type": "ComfySwitchNode", + "pos": [ + 680, + -1250 + ], + "size": [ + 230, + 130 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "on_false", + "name": "on_false", + "type": "*", + "link": 333 + }, + { + "localized_name": "on_true", + "name": "on_true", + "type": "*", + "link": 334 + }, + { + "localized_name": "switch", + "name": "switch", + "type": "BOOLEAN", + "widget": { + "name": "switch" + }, + "link": 336 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "*", + "links": [ + 335 + ] + } + ], + "title": "Switch (CFG)", + "properties": { + "Node name for S&R": "ComfySwitchNode", + "cnr_id": "comfy-core", + "ver": "0.15.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + false + ] + }, + { + "id": 198, + "type": "PrimitiveInt", + "pos": [ + 190, + -1060 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 337 + ] + } + ], + "title": "Float (Steps)", + "properties": { + "Node name for S&R": "PrimitiveInt", + "cnr_id": "comfy-core", + "ver": "0.15.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 8, + "fixed" + ] + }, + { + "id": 199, + "type": "PrimitiveFloat", + "pos": [ + 190, + -1500 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "FLOAT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": [ + 333 + ] + } + ], + "title": "Float (CFG)", + "properties": { + "Node name for S&R": "PrimitiveFloat", + "cnr_id": "comfy-core", + "ver": "0.15.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 4 + ] + }, + { + "id": 200, + "type": "PrimitiveFloat", + "pos": [ + 190, + -1230 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "FLOAT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": [ + 334 + ] + } + ], + "title": "Float (CFG)", + "properties": { + "Node name for S&R": "PrimitiveFloat", + "cnr_id": "comfy-core", + "ver": "0.15.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1 + ] + }, + { + "id": 201, + "type": "ComfySwitchNode", + "pos": [ + 680, + -1470 + ], + "size": [ + 230, + 130 + ], + "flags": {}, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "on_false", + "name": "on_false", + "type": "*", + "link": 329 + }, + { + "localized_name": "on_true", + "name": "on_true", + "type": "*", + "link": 337 + }, + { + "localized_name": "switch", + "name": "switch", + "type": "BOOLEAN", + "widget": { + "name": "switch" + }, + "link": 330 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "*", + "links": [ + 345 + ] + } + ], + "title": "Switch (Steps)", + "properties": { + "Node name for S&R": "ComfySwitchNode", + "cnr_id": "comfy-core", + "ver": "0.15.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + false + ] + }, + { + "id": 202, + "type": "VAELoader", + "pos": [ + -960, + -1100 + ], + "size": [ + 400, + 110 + ], + "flags": { + "collapsed": false + }, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "vae_name", + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": 375 + } + ], + "outputs": [ + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "slot_index": 0, + "links": [ + 298, + 299, + 300, + 314 + ] + } + ], + "properties": { + "Node name for S&R": "VAELoader", + "cnr_id": "comfy-core", + "ver": "0.5.1", + "models": [ + { + "name": "qwen_image_vae.safetensors", + "url": "https://huggingface.co/FireRedTeam/FireRed-Image-Edit-1.0-ComfyUI/resolve/main/qwen_image_vae.safetensors", + "directory": "vae" + } + ], + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "qwen_image_vae.safetensors" + ] + }, + { + "id": 203, + "type": "CLIPLoader", + "pos": [ + -960, + -1400 + ], + "size": [ + 400, + 150 + ], + "flags": {}, + "order": 10, + "mode": 0, + "inputs": [ + { + "localized_name": "clip_name", + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": 374 + }, + { + "localized_name": "type", + "name": "type", + "type": "COMBO", + "widget": { + "name": "type" + }, + "link": null + }, + { + "localized_name": "device", + "name": "device", + "shape": 7, + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [ + 296, + 297 + ] + } + ], + "properties": { + "Node name for S&R": "CLIPLoader", + "cnr_id": "comfy-core", + "ver": "0.5.1", + "models": [ + { + "name": "qwen_2.5_vl_7b_fp8_scaled.safetensors", + "url": "https://huggingface.co/Comfy-Org/HunyuanVideo_1.5_repackaged/resolve/main/split_files/text_encoders/qwen_2.5_vl_7b_fp8_scaled.safetensors", + "directory": "text_encoders" + } + ], + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "qwen_2.5_vl_7b_fp8_scaled.safetensors", + "qwen_image", + "default" + ] + }, + { + "id": 204, + "type": "LoraLoaderModelOnly", + "pos": [ + 100, + -900 + ], + "size": [ + 400, + 140 + ], + "flags": {}, + "order": 11, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 316 + }, + { + "localized_name": "lora_name", + "name": "lora_name", + "type": "COMBO", + "widget": { + "name": "lora_name" + }, + "link": 376 + }, + { + "localized_name": "strength_model", + "name": "strength_model", + "type": "FLOAT", + "widget": { + "name": "strength_model" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 325 + ] + } + ], + "properties": { + "Node name for S&R": "LoraLoaderModelOnly", + "cnr_id": "comfy-core", + "ver": "0.15.1", + "models": [ + { + "name": "FireRed-Image-Edit-1.0-Lightning-8steps-v1.0.safetensors", + "url": "https://huggingface.co/FireRedTeam/FireRed-Image-Edit-1.0-ComfyUI/resolve/main/FireRed-Image-Edit-1.0-Lightning-8steps-v1.0.safetensors", + "directory": "loras" + } + ], + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "FireRed-Image-Edit-1.0-Lightning-8steps-v1.0.safetensors", + 1 + ] + }, + { + "id": 205, + "type": "UNETLoader", + "pos": [ + -960, + -1670 + ], + "size": [ + 400, + 110 + ], + "flags": {}, + "order": 12, + "mode": 0, + "inputs": [ + { + "localized_name": "unet_name", + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": 373 + }, + { + "localized_name": "weight_dtype", + "name": "weight_dtype", + "type": "COMBO", + "widget": { + "name": "weight_dtype" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "slot_index": 0, + "links": [ + 316, + 324 + ] + } + ], + "properties": { + "Node name for S&R": "UNETLoader", + "cnr_id": "comfy-core", + "ver": "0.5.1", + "models": [ + { + "name": "FireRed-Image-Edit-1.1-transformer.safetensors", + "url": "https://huggingface.co/FireRedTeam/FireRed-Image-Edit-1.1-ComfyUI/resolve/main/FireRed-Image-Edit-1.1-transformer.safetensors", + "directory": "diffusion_models" + } + ], + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "FireRed-Image-Edit-1.1-transformer.safetensors", + "default" + ] + }, + { + "id": 206, + "type": "VAEEncode", + "pos": [ + -390, + -810 + ], + "size": [ + 390, + 100 + ], + "flags": {}, + "order": 13, + "mode": 0, + "inputs": [ + { + "localized_name": "pixels", + "name": "pixels", + "type": "IMAGE", + "link": 368 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 300 + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "links": [ + 303 + ] + } + ], + "properties": { + "Node name for S&R": "VAEEncode", + "cnr_id": "comfy-core", + "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 207, + "type": "PrimitiveBoolean", + "pos": [ + 160, + -650 + ], + "size": [ + 400, + 100 + ], + "flags": {}, + "order": 14, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "BOOLEAN", + "widget": { + "name": "value" + }, + "link": 364 + } + ], + "outputs": [ + { + "localized_name": "BOOLEAN", + "name": "BOOLEAN", + "type": "BOOLEAN", + "links": [ + 323, + 330, + 336 + ] + } + ], + "title": "Enable Lightning LoRA?", + "properties": { + "Node name for S&R": "PrimitiveBoolean", + "cnr_id": "comfy-core", + "ver": "0.15.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + false + ] + }, + { + "id": 208, + "type": "TextEncodeQwenImageEditPlus", + "pos": [ + -480, + -1690 + ], + "size": [ + 470, + 370 + ], + "flags": {}, + "order": 15, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 296 + }, + { + "localized_name": "vae", + "name": "vae", + "shape": 7, + "type": "VAE", + "link": 298 + }, + { + "localized_name": "image1", + "name": "image1", + "shape": 7, + "type": "IMAGE", + "link": 369 + }, + { + "localized_name": "image2", + "name": "image2", + "shape": 7, + "type": "IMAGE", + "link": 355 + }, + { + "localized_name": "image3", + "name": "image3", + "shape": 7, + "type": "IMAGE", + "link": 357 + }, + { + "localized_name": "prompt", + "name": "prompt", + "type": "STRING", + "widget": { + "name": "prompt" + }, + "link": 359 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 312 + ] + } + ], + "title": "TextEncodeQwenImageEditPlus (Positive)", + "properties": { + "Node name for S&R": "TextEncodeQwenImageEditPlus", + "cnr_id": "comfy-core", + "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "" + ], + "color": "#232", + "bgcolor": "#353" + }, + { + "id": 209, + "type": "TextEncodeQwenImageEditPlus", + "pos": [ + -470, + -1240 + ], + "size": [ + 460, + 290 + ], + "flags": {}, + "order": 16, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 297 + }, + { + "localized_name": "vae", + "name": "vae", + "shape": 7, + "type": "VAE", + "link": 299 + }, + { + "localized_name": "image1", + "name": "image1", + "shape": 7, + "type": "IMAGE", + "link": 370 + }, + { + "localized_name": "image2", + "name": "image2", + "shape": 7, + "type": "IMAGE", + "link": 356 + }, + { + "localized_name": "image3", + "name": "image3", + "shape": 7, + "type": "IMAGE", + "link": 358 + }, + { + "localized_name": "prompt", + "name": "prompt", + "type": "STRING", + "widget": { + "name": "prompt" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 313 + ] + } + ], + "properties": { + "Node name for S&R": "TextEncodeQwenImageEditPlus", + "cnr_id": "comfy-core", + "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "" + ], + "color": "#323", + "bgcolor": "#535" + }, + { + "id": 210, + "type": "KSampler", + "pos": [ + 1010, + -1340 + ], + "size": [ + 270, + 480 + ], + "flags": {}, + "order": 17, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 295 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 312 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 313 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 303 + }, + { + "localized_name": "seed", + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": 372 + }, + { + "localized_name": "steps", + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": 345 + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": 335 + }, + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + }, + { + "localized_name": "scheduler", + "name": "scheduler", + "type": "COMBO", + "widget": { + "name": "scheduler" + }, + "link": null + }, + { + "localized_name": "denoise", + "name": "denoise", + "type": "FLOAT", + "widget": { + "name": "denoise" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "links": [ + 273 + ] + } + ], + "properties": { + "Node name for S&R": "KSampler", + "cnr_id": "comfy-core", + "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 43, + "fixed", + 40, + 4, + "euler", + "simple", + 1 + ] + }, + { + "id": 211, + "type": "VAEDecode", + "pos": [ + 1440, + -1340 + ], + "size": [ + 230, + 100 + ], + "flags": { + "collapsed": false + }, + "order": 18, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 273 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 314 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "slot_index": 0, + "links": [ + 292 + ] + } + ], + "properties": { + "Node name for S&R": "VAEDecode", + "cnr_id": "comfy-core", + "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 212, + "type": "ResizeImageMaskNode", + "pos": [ + -900, + -810 + ], + "size": [ + 280, + 110 + ], + "flags": {}, + "order": 19, + "mode": 0, + "inputs": [ + { + "localized_name": "input", + "name": "input", + "type": "IMAGE,MASK", + "link": 371 + }, + { + "localized_name": "resize_type", + "name": "resize_type", + "type": "COMFY_DYNAMICCOMBO_V3", + "widget": { + "name": "resize_type" + }, + "link": null + }, + { + "localized_name": "resize_type.megapixels", + "name": "resize_type.megapixels", + "type": "FLOAT", + "widget": { + "name": "resize_type.megapixels" + }, + "link": null + }, + { + "localized_name": "scale_method", + "name": "scale_method", + "type": "COMBO", + "widget": { + "name": "scale_method" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "resized", + "name": "resized", + "type": "*", + "links": [ + 368, + 369, + 370 + ] + } + ], + "properties": { + "Node name for S&R": "ResizeImageMaskNode", + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "scale total pixels", + 1, + "lanczos" + ] + } + ], + "groups": [ + { + "id": 1, + "title": "Model", + "bounding": [ + -990, + -1770, + 460, + 870 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 2, + "title": "Prompt", + "bounding": [ + -500, + -1770, + 510, + 870 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 7, + "title": "Original", + "bounding": [ + 40, + -1770, + 530, + 410 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 8, + "title": "Lightning LoRA", + "bounding": [ + 40, + -1330, + 560, + 610 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 326, + "origin_id": 194, + "origin_slot": 0, + "target_id": 193, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 324, + "origin_id": 205, + "origin_slot": 0, + "target_id": 194, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 325, + "origin_id": 204, + "origin_slot": 0, + "target_id": 194, + "target_slot": 1, + "type": "MODEL" + }, + { + "id": 323, + "origin_id": 207, + "origin_slot": 0, + "target_id": 194, + "target_slot": 2, + "type": "BOOLEAN" + }, + { + "id": 294, + "origin_id": 193, + "origin_slot": 0, + "target_id": 196, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 333, + "origin_id": 199, + "origin_slot": 0, + "target_id": 197, + "target_slot": 0, + "type": "FLOAT" + }, + { + "id": 334, + "origin_id": 200, + "origin_slot": 0, + "target_id": 197, + "target_slot": 1, + "type": "FLOAT" + }, + { + "id": 336, + "origin_id": 207, + "origin_slot": 0, + "target_id": 197, + "target_slot": 2, + "type": "BOOLEAN" + }, + { + "id": 329, + "origin_id": 195, + "origin_slot": 0, + "target_id": 201, + "target_slot": 0, + "type": "INT" + }, + { + "id": 337, + "origin_id": 198, + "origin_slot": 0, + "target_id": 201, + "target_slot": 1, + "type": "INT" + }, + { + "id": 330, + "origin_id": 207, + "origin_slot": 0, + "target_id": 201, + "target_slot": 2, + "type": "BOOLEAN" + }, + { + "id": 297, + "origin_id": 203, + "origin_slot": 0, + "target_id": 209, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 299, + "origin_id": 202, + "origin_slot": 0, + "target_id": 209, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 316, + "origin_id": 205, + "origin_slot": 0, + "target_id": 204, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 296, + "origin_id": 203, + "origin_slot": 0, + "target_id": 208, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 298, + "origin_id": 202, + "origin_slot": 0, + "target_id": 208, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 300, + "origin_id": 202, + "origin_slot": 0, + "target_id": 206, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 295, + "origin_id": 196, + "origin_slot": 0, + "target_id": 210, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 312, + "origin_id": 208, + "origin_slot": 0, + "target_id": 210, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 313, + "origin_id": 209, + "origin_slot": 0, + "target_id": 210, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 303, + "origin_id": 206, + "origin_slot": 0, + "target_id": 210, + "target_slot": 3, + "type": "LATENT" + }, + { + "id": 345, + "origin_id": 201, + "origin_slot": 0, + "target_id": 210, + "target_slot": 5, + "type": "INT" + }, + { + "id": 335, + "origin_id": 197, + "origin_slot": 0, + "target_id": 210, + "target_slot": 6, + "type": "FLOAT" + }, + { + "id": 273, + "origin_id": 210, + "origin_slot": 0, + "target_id": 211, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 314, + "origin_id": 202, + "origin_slot": 0, + "target_id": 211, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 292, + "origin_id": 211, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 355, + "origin_id": -10, + "origin_slot": 1, + "target_id": 208, + "target_slot": 3, + "type": "IMAGE" + }, + { + "id": 356, + "origin_id": -10, + "origin_slot": 1, + "target_id": 209, + "target_slot": 3, + "type": "IMAGE" + }, + { + "id": 357, + "origin_id": -10, + "origin_slot": 2, + "target_id": 208, + "target_slot": 4, + "type": "IMAGE" + }, + { + "id": 358, + "origin_id": -10, + "origin_slot": 2, + "target_id": 209, + "target_slot": 4, + "type": "IMAGE" + }, + { + "id": 359, + "origin_id": -10, + "origin_slot": 3, + "target_id": 208, + "target_slot": 5, + "type": "STRING" + }, + { + "id": 364, + "origin_id": -10, + "origin_slot": 4, + "target_id": 207, + "target_slot": 0, + "type": "BOOLEAN" + }, + { + "id": 368, + "origin_id": 212, + "origin_slot": 0, + "target_id": 206, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 369, + "origin_id": 212, + "origin_slot": 0, + "target_id": 208, + "target_slot": 2, + "type": "IMAGE" + }, + { + "id": 370, + "origin_id": 212, + "origin_slot": 0, + "target_id": 209, + "target_slot": 2, + "type": "IMAGE" + }, + { + "id": 371, + "origin_id": -10, + "origin_slot": 0, + "target_id": 212, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 372, + "origin_id": -10, + "origin_slot": 5, + "target_id": 210, + "target_slot": 4, + "type": "INT" + }, + { + "id": 373, + "origin_id": -10, + "origin_slot": 6, + "target_id": 205, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 374, + "origin_id": -10, + "origin_slot": 7, + "target_id": 203, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 375, + "origin_id": -10, + "origin_slot": 8, + "target_id": 202, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 376, + "origin_id": -10, + "origin_slot": 9, + "target_id": 204, + "target_slot": 1, + "type": "COMBO" + } + ], + "extra": { + "workflowRendererVersion": "LG" + }, + "category": "Image generation and editing/Edit image", + "description": "Edits images via text instructions using FireRed Image Edit 1.1, a diffusion-based instruction-following editing model." + } + ] + }, + "extra": { + "ue_links": [] + } +} \ No newline at end of file diff --git a/blueprints/Image Edit (Flux.2 Dev).json b/blueprints/Image Edit (Flux.2 Dev).json new file mode 100644 index 000000000..92827bf17 --- /dev/null +++ b/blueprints/Image Edit (Flux.2 Dev).json @@ -0,0 +1,2050 @@ +{ + "revision": 0, + "last_node_id": 139, + "last_link_id": 0, + "nodes": [ + { + "id": 139, + "type": "41b0c117-7470-454c-914e-b8742dc06d62", + "pos": [ + -650, + 570 + ], + "size": [ + 400, + 0 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "label": "image", + "localized_name": "pixels", + "name": "pixels", + "type": "IMAGE", + "link": null + }, + { + "label": "prompt", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + }, + { + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": null + }, + { + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": null + }, + { + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": null + }, + { + "label": "enable_turbo_mode", + "name": "value", + "type": "BOOLEAN", + "widget": { + "name": "value" + }, + "link": null + }, + { + "label": "turbo_lora", + "name": "lora_name", + "type": "COMBO", + "widget": { + "name": "lora_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [] + } + ], + "properties": { + "proxyWidgets": [ + [ + "123", + "text" + ], + [ + "129", + "unet_name" + ], + [ + "124", + "clip_name" + ], + [ + "121", + "vae_name" + ], + [ + "138", + "value" + ], + [ + "128", + "lora_name" + ], + [ + "125", + "noise_seed" + ], + [ + "125", + "control_after_generate" + ] + ], + "cnr_id": "comfy-core", + "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": { + "text": true, + "value": true, + "lora_name": true + }, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [], + "title": "Image Edit (Flux.2 Dev)" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "41b0c117-7470-454c-914e-b8742dc06d62", + "version": 1, + "state": { + "lastGroupId": 8, + "lastNodeId": 139, + "lastLinkId": 194, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Image Edit (Flux.2 Dev)", + "inputNode": { + "id": -10, + "bounding": [ + -1520, + 400, + 151.744140625, + 180 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 1240, + 420, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "fc74acd5-30a9-410b-abb5-4a4171ba3d25", + "name": "pixels", + "type": "IMAGE", + "linkIds": [ + 126, + 169 + ], + "localized_name": "pixels", + "label": "image", + "pos": [ + -1388.255859375, + 420 + ] + }, + { + "id": "3e69affa-397b-4d52-82d7-68dfcef9e761", + "name": "text", + "type": "STRING", + "linkIds": [ + 168 + ], + "label": "prompt", + "pos": [ + -1388.255859375, + 440 + ] + }, + { + "id": "2f016a8a-fb3e-4cb9-97f2-a991defe4fa2", + "name": "unet_name", + "type": "COMBO", + "linkIds": [ + 177 + ], + "pos": [ + -1388.255859375, + 460 + ] + }, + { + "id": "799b9dc7-0c90-4b19-9a13-e01d896bea1f", + "name": "clip_name", + "type": "COMBO", + "linkIds": [ + 178 + ], + "pos": [ + -1388.255859375, + 480 + ] + }, + { + "id": "e58a83c9-1b93-4378-9598-f24068820313", + "name": "vae_name", + "type": "COMBO", + "linkIds": [ + 179 + ], + "pos": [ + -1388.255859375, + 500 + ] + }, + { + "id": "8335a4a9-0ce4-4e67-a641-1c9d7a762977", + "name": "value", + "type": "BOOLEAN", + "linkIds": [ + 191 + ], + "label": "enable_turbo_mode", + "pos": [ + -1388.255859375, + 520 + ] + }, + { + "id": "890b22b4-44a7-4707-912a-ca8b4ee7b7c9", + "name": "lora_name", + "type": "COMBO", + "linkIds": [ + 192 + ], + "label": "turbo_lora", + "pos": [ + -1388.255859375, + 540 + ] + } + ], + "outputs": [ + { + "id": "3eaa05d6-4960-4a7c-bf2a-8b585fbb7c9c", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 9 + ], + "localized_name": "IMAGE", + "pos": [ + 1260, + 440 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 118, + "type": "Flux2Scheduler", + "pos": [ + 540, + 430 + ], + "size": [ + 230, + 170 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "steps", + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": 188 + }, + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 170 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 172 + } + ], + "outputs": [ + { + "localized_name": "SIGMAS", + "name": "SIGMAS", + "type": "SIGMAS", + "links": [ + 132 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "Flux2Scheduler", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 20, + 1248, + 832 + ] + }, + { + "id": 119, + "type": "BasicGuider", + "pos": [ + 530, + 120 + ], + "size": [ + 230, + 100 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 185 + }, + { + "localized_name": "conditioning", + "name": "conditioning", + "type": "CONDITIONING", + "link": 166 + } + ], + "outputs": [ + { + "localized_name": "GUIDER", + "name": "GUIDER", + "type": "GUIDER", + "slot_index": 0, + "links": [ + 30 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "BasicGuider", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 120, + "type": "KSamplerSelect", + "pos": [ + 530, + 270 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "SAMPLER", + "name": "SAMPLER", + "type": "SAMPLER", + "links": [ + 19 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "KSamplerSelect", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "euler" + ] + }, + { + "id": 121, + "type": "VAELoader", + "pos": [ + -970, + 390 + ], + "size": [ + 300, + 110 + ], + "flags": {}, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "vae_name", + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": 179 + } + ], + "outputs": [ + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "slot_index": 0, + "links": [ + 127, + 159 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "VAELoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "full_encoder_small_decoder.safetensors", + "url": "https://huggingface.co/black-forest-labs/FLUX.2-small-decoder/resolve/main/full_encoder_small_decoder.safetensors", + "directory": "vae" + } + ] + }, + "widgets_values": [ + "full_encoder_small_decoder.safetensors" + ] + }, + { + "id": 122, + "type": "SamplerCustomAdvanced", + "pos": [ + 790, + -50 + ], + "size": [ + 280, + 170 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "noise", + "name": "noise", + "type": "NOISE", + "link": 37 + }, + { + "localized_name": "guider", + "name": "guider", + "type": "GUIDER", + "link": 30 + }, + { + "localized_name": "sampler", + "name": "sampler", + "type": "SAMPLER", + "link": 19 + }, + { + "localized_name": "sigmas", + "name": "sigmas", + "type": "SIGMAS", + "link": 132 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 161 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "LATENT", + "slot_index": 0, + "links": [ + 24 + ] + }, + { + "localized_name": "denoised_output", + "name": "denoised_output", + "type": "LATENT", + "links": null + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "SamplerCustomAdvanced", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 123, + "type": "CLIPTextEncode", + "pos": [ + -630, + -50 + ], + "size": [ + 430, + 360 + ], + "flags": {}, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 117 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 168 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "slot_index": 0, + "links": [ + 41 + ] + } + ], + "title": "CLIP Text Encode (Positive Prompt)", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "CLIPTextEncode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "" + ], + "color": "#232", + "bgcolor": "#353" + }, + { + "id": 124, + "type": "CLIPLoader", + "pos": [ + -970, + 160 + ], + "size": [ + 300, + 150 + ], + "flags": {}, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "clip_name", + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": 178 + }, + { + "localized_name": "type", + "name": "type", + "type": "COMBO", + "widget": { + "name": "type" + }, + "link": null + }, + { + "localized_name": "device", + "name": "device", + "shape": 7, + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [ + 117 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "CLIPLoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "mistral_3_small_flux2_bf16.safetensors", + "url": "https://huggingface.co/Comfy-Org/flux2-dev/resolve/main/split_files/text_encoders/mistral_3_small_flux2_bf16.safetensors", + "directory": "text_encoders" + } + ] + }, + "widgets_values": [ + "mistral_3_small_flux2_bf16.safetensors", + "flux2", + "default" + ] + }, + { + "id": 125, + "type": "RandomNoise", + "pos": [ + 530, + -50 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "noise_seed", + "name": "noise_seed", + "type": "INT", + "widget": { + "name": "noise_seed" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "NOISE", + "name": "NOISE", + "type": "NOISE", + "links": [ + 37 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "RandomNoise", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 342971778941390, + "randomize" + ] + }, + { + "id": 126, + "type": "VAEDecode", + "pos": [ + 830, + 410 + ], + "size": [ + 230, + 100 + ], + "flags": {}, + "order": 10, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 24 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 159 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "slot_index": 0, + "links": [ + 9 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "VAEDecode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 127, + "type": "FluxGuidance", + "pos": [ + -520, + 390 + ], + "size": [ + 320, + 110 + ], + "flags": {}, + "order": 11, + "mode": 0, + "inputs": [ + { + "localized_name": "conditioning", + "name": "conditioning", + "type": "CONDITIONING", + "link": 41 + }, + { + "localized_name": "guidance", + "name": "guidance", + "type": "FLOAT", + "widget": { + "name": "guidance" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "slot_index": 0, + "links": [ + 144 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "FluxGuidance", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 4 + ], + "color": "#233", + "bgcolor": "#355" + }, + { + "id": 128, + "type": "LoraLoaderModelOnly", + "pos": [ + -150, + 200 + ], + "size": [ + 300, + 140 + ], + "flags": {}, + "order": 12, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 181 + }, + { + "localized_name": "lora_name", + "name": "lora_name", + "type": "COMBO", + "widget": { + "name": "lora_name" + }, + "link": 192 + }, + { + "localized_name": "strength_model", + "name": "strength_model", + "type": "FLOAT", + "widget": { + "name": "strength_model" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 183 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "LoraLoaderModelOnly", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "Flux_2-Turbo-LoRA_comfyui.safetensors", + "url": "https://huggingface.co/ByteZSzn/Flux.2-Turbo-ComfyUI/resolve/main/Flux_2-Turbo-LoRA_comfyui.safetensors", + "directory": "loras" + } + ] + }, + "widgets_values": [ + "Flux_2-Turbo-LoRA_comfyui.safetensors", + 1 + ] + }, + { + "id": 129, + "type": "UNETLoader", + "pos": [ + -970, + -40 + ], + "size": [ + 300, + 110 + ], + "flags": {}, + "order": 13, + "mode": 0, + "inputs": [ + { + "localized_name": "unet_name", + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": 177 + }, + { + "localized_name": "weight_dtype", + "name": "weight_dtype", + "type": "COMBO", + "widget": { + "name": "weight_dtype" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "slot_index": 0, + "links": [ + 181, + 184 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "UNETLoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "flux2_dev_fp8mixed.safetensors", + "url": "https://huggingface.co/Comfy-Org/flux2-dev/resolve/main/split_files/diffusion_models/flux2_dev_fp8mixed.safetensors", + "directory": "diffusion_models" + } + ] + }, + "widgets_values": [ + "flux2_dev_fp8mixed.safetensors", + "default" + ] + }, + { + "id": 130, + "type": "ComfySwitchNode", + "pos": [ + 220, + 10 + ], + "size": [ + 270, + 130 + ], + "flags": {}, + "order": 14, + "mode": 0, + "inputs": [ + { + "localized_name": "on_false", + "name": "on_false", + "type": "*", + "link": 184 + }, + { + "localized_name": "on_true", + "name": "on_true", + "type": "*", + "link": 183 + }, + { + "localized_name": "switch", + "name": "switch", + "type": "BOOLEAN", + "widget": { + "name": "switch" + }, + "link": 190 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "*", + "links": [ + 185 + ] + } + ], + "title": "Switch(model)", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ComfySwitchNode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + false + ] + }, + { + "id": 131, + "type": "PrimitiveInt", + "pos": [ + -150, + 430 + ], + "size": [ + 300, + 110 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 186 + ] + } + ], + "title": "Steps", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 8, + "fixed" + ] + }, + { + "id": 132, + "type": "PrimitiveInt", + "pos": [ + -150, + -50 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 187 + ] + } + ], + "title": "Steps", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 20, + "fixed" + ] + }, + { + "id": 133, + "type": "ComfySwitchNode", + "pos": [ + 220, + 280 + ], + "size": [ + 270, + 130 + ], + "flags": {}, + "order": 15, + "mode": 0, + "inputs": [ + { + "localized_name": "on_false", + "name": "on_false", + "type": "*", + "link": 187 + }, + { + "localized_name": "on_true", + "name": "on_true", + "type": "*", + "link": 186 + }, + { + "localized_name": "switch", + "name": "switch", + "type": "BOOLEAN", + "widget": { + "name": "switch" + }, + "link": 189 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "*", + "links": [ + 188 + ] + } + ], + "title": "Switch(steps)", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ComfySwitchNode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + false + ] + }, + { + "id": 134, + "type": "EmptyFlux2LatentImage", + "pos": [ + 530, + 790 + ], + "size": [ + 270, + 170 + ], + "flags": {}, + "order": 16, + "mode": 0, + "inputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 171 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 173 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "links": [ + 161 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "EmptyFlux2LatentImage", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1248, + 832, + 1 + ] + }, + { + "id": 135, + "type": "GetImageSize", + "pos": [ + -100, + 810 + ], + "size": [ + 230, + 120 + ], + "flags": {}, + "order": 17, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 169 + } + ], + "outputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "links": [ + 170, + 171 + ] + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "links": [ + 172, + 173 + ] + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "links": null + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "GetImageSize", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 136, + "type": "VAEEncode", + "pos": [ + -910, + 600 + ], + "size": [ + 230, + 100 + ], + "flags": { + "collapsed": true + }, + "order": 18, + "mode": 0, + "inputs": [ + { + "localized_name": "pixels", + "name": "pixels", + "type": "IMAGE", + "link": 126 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 127 + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "links": [ + 125 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "VAEEncode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 137, + "type": "ReferenceLatent", + "pos": [ + -470, + 580 + ], + "size": [ + 230, + 100 + ], + "flags": {}, + "order": 19, + "mode": 0, + "inputs": [ + { + "localized_name": "conditioning", + "name": "conditioning", + "type": "CONDITIONING", + "link": 144 + }, + { + "localized_name": "latent", + "name": "latent", + "shape": 7, + "type": "LATENT", + "link": 125 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 166 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ReferenceLatent", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 138, + "type": "PrimitiveBoolean", + "pos": [ + -130, + 640 + ], + "size": [ + 270, + 100 + ], + "flags": {}, + "order": 20, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "BOOLEAN", + "widget": { + "name": "value" + }, + "link": 191 + } + ], + "outputs": [ + { + "localized_name": "BOOLEAN", + "name": "BOOLEAN", + "type": "BOOLEAN", + "links": [ + 189, + 190 + ] + } + ], + "title": "Enable 8 steps lora", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "PrimitiveBoolean", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + false + ] + } + ], + "groups": [ + { + "id": 1, + "title": "Models", + "bounding": [ + -980, + -120, + 320, + 640 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 2, + "title": "Custom sampler", + "bounding": [ + 520, + -120, + 590, + 740 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 3, + "title": "Image size", + "bounding": [ + 510, + 690, + 590, + 290 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 4, + "title": "Prompt", + "bounding": [ + -640, + -120, + 450, + 640 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 7, + "title": "Original", + "bounding": [ + -160, + -120, + 340, + 230 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 8, + "title": "8 Steps LoRA", + "bounding": [ + -160, + 130, + 340, + 430 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 41, + "origin_id": 123, + "origin_slot": 0, + "target_id": 127, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 144, + "origin_id": 127, + "origin_slot": 0, + "target_id": 137, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 125, + "origin_id": 136, + "origin_slot": 0, + "target_id": 137, + "target_slot": 1, + "type": "LATENT" + }, + { + "id": 37, + "origin_id": 125, + "origin_slot": 0, + "target_id": 122, + "target_slot": 0, + "type": "NOISE" + }, + { + "id": 30, + "origin_id": 119, + "origin_slot": 0, + "target_id": 122, + "target_slot": 1, + "type": "GUIDER" + }, + { + "id": 19, + "origin_id": 120, + "origin_slot": 0, + "target_id": 122, + "target_slot": 2, + "type": "SAMPLER" + }, + { + "id": 132, + "origin_id": 118, + "origin_slot": 0, + "target_id": 122, + "target_slot": 3, + "type": "SIGMAS" + }, + { + "id": 161, + "origin_id": 134, + "origin_slot": 0, + "target_id": 122, + "target_slot": 4, + "type": "LATENT" + }, + { + "id": 24, + "origin_id": 122, + "origin_slot": 0, + "target_id": 126, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 159, + "origin_id": 121, + "origin_slot": 0, + "target_id": 126, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 117, + "origin_id": 124, + "origin_slot": 0, + "target_id": 123, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 127, + "origin_id": 121, + "origin_slot": 0, + "target_id": 136, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 126, + "origin_id": -10, + "origin_slot": 0, + "target_id": 136, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 9, + "origin_id": 126, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 166, + "origin_id": 137, + "origin_slot": 0, + "target_id": 119, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 168, + "origin_id": -10, + "origin_slot": 1, + "target_id": 123, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 169, + "origin_id": -10, + "origin_slot": 0, + "target_id": 135, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 170, + "origin_id": 135, + "origin_slot": 0, + "target_id": 118, + "target_slot": 1, + "type": "INT" + }, + { + "id": 171, + "origin_id": 135, + "origin_slot": 0, + "target_id": 134, + "target_slot": 0, + "type": "INT" + }, + { + "id": 172, + "origin_id": 135, + "origin_slot": 1, + "target_id": 118, + "target_slot": 2, + "type": "INT" + }, + { + "id": 173, + "origin_id": 135, + "origin_slot": 1, + "target_id": 134, + "target_slot": 1, + "type": "INT" + }, + { + "id": 177, + "origin_id": -10, + "origin_slot": 2, + "target_id": 129, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 178, + "origin_id": -10, + "origin_slot": 3, + "target_id": 124, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 179, + "origin_id": -10, + "origin_slot": 4, + "target_id": 121, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 181, + "origin_id": 129, + "origin_slot": 0, + "target_id": 128, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 183, + "origin_id": 128, + "origin_slot": 0, + "target_id": 130, + "target_slot": 1, + "type": "MODEL" + }, + { + "id": 184, + "origin_id": 129, + "origin_slot": 0, + "target_id": 130, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 185, + "origin_id": 130, + "origin_slot": 0, + "target_id": 119, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 186, + "origin_id": 131, + "origin_slot": 0, + "target_id": 133, + "target_slot": 1, + "type": "INT" + }, + { + "id": 187, + "origin_id": 132, + "origin_slot": 0, + "target_id": 133, + "target_slot": 0, + "type": "INT" + }, + { + "id": 188, + "origin_id": 133, + "origin_slot": 0, + "target_id": 118, + "target_slot": 0, + "type": "INT" + }, + { + "id": 189, + "origin_id": 138, + "origin_slot": 0, + "target_id": 133, + "target_slot": 2, + "type": "BOOLEAN" + }, + { + "id": 190, + "origin_id": 138, + "origin_slot": 0, + "target_id": 130, + "target_slot": 2, + "type": "BOOLEAN" + }, + { + "id": 191, + "origin_id": -10, + "origin_slot": 5, + "target_id": 138, + "target_slot": 0, + "type": "BOOLEAN" + }, + { + "id": 192, + "origin_id": -10, + "origin_slot": 6, + "target_id": 128, + "target_slot": 1, + "type": "COMBO" + } + ], + "extra": { + "workflowRendererVersion": "LG" + }, + "category": "Image generation and editing/Edit image", + "description": "Edits an image from text instructions using Flux.2 [dev], with guidance, schedulers, and optional Turbo LoRAs." + } + ] + }, + "extra": { + "ue_links": [] + } +} \ No newline at end of file diff --git a/blueprints/Image Edit (Flux.2 Klein 4B).json b/blueprints/Image Edit (Flux.2 Klein 4B).json index 78bbb7414..7f6fa7a4b 100644 --- a/blueprints/Image Edit (Flux.2 Klein 4B).json +++ b/blueprints/Image Edit (Flux.2 Klein 4B).json @@ -128,7 +128,7 @@ }, "revision": 0, "config": {}, - "name": "local-Image Edit (Flux.2 Klein 4B)", + "name": "Image Edit (Flux.2 Klein 4B)", "inputNode": { "id": -10, "bounding": [ @@ -1472,7 +1472,8 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image generation and editing/Edit image" + "category": "Image generation and editing/Edit image", + "description": "Edits an input image via text instructions using FLUX.2 [klein] 4B." }, { "id": "6007e698-2ebd-4917-84d8-299b35d7b7ab", @@ -1821,7 +1822,8 @@ ], "extra": { "workflowRendererVersion": "LG" - } + }, + "description": "Applies reference image conditioning for style/identity transfer (Flux.2 Klein 4B)." } ] }, diff --git a/blueprints/Image Edit (LongCat Image Edit).json b/blueprints/Image Edit (LongCat Image Edit).json new file mode 100644 index 000000000..de1c155a2 --- /dev/null +++ b/blueprints/Image Edit (LongCat Image Edit).json @@ -0,0 +1,1428 @@ +{ + "revision": 0, + "last_node_id": 176, + "last_link_id": 0, + "nodes": [ + { + "id": 176, + "type": "372a02a0-a79c-40b4-84a9-34f246fe0e9c", + "pos": [ + 967.0861152473078, + 4977.534165136897 + ], + "size": [ + 330, + 380 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": null + }, + { + "name": "prompt", + "type": "STRING", + "widget": { + "name": "prompt" + }, + "link": null + }, + { + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": null + }, + { + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": null + }, + { + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": null + }, + { + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": null + }, + { + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": null + }, + { + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [] + } + ], + "properties": { + "proxyWidgets": [ + [ + "27", + "prompt" + ], + [ + "33", + "steps" + ], + [ + "33", + "cfg" + ], + [ + "33", + "seed" + ], + [ + "34", + "unet_name" + ], + [ + "38", + "clip_name" + ], + [ + "26", + "vae_name" + ] + ], + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [], + "title": "Image Edit (LongCat Image Edit)" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "372a02a0-a79c-40b4-84a9-34f246fe0e9c", + "version": 1, + "state": { + "lastGroupId": 8, + "lastNodeId": 176, + "lastLinkId": 376, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Image Edit (LongCat Image Edit)", + "inputNode": { + "id": -10, + "bounding": [ + -750, + 380, + 120, + 200 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 1680, + 340, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "616c4f3e-8b64-4711-bee2-5ecbe1814fe4", + "name": "image", + "type": "IMAGE", + "linkIds": [ + 14 + ], + "localized_name": "image", + "pos": [ + -650, + 400 + ] + }, + { + "id": "d39759fc-a5a9-4b82-a88f-df9b953f1d98", + "name": "prompt", + "type": "STRING", + "linkIds": [ + 36 + ], + "pos": [ + -650, + 420 + ] + }, + { + "id": "48627f43-cdf1-4ea9-9e11-ec13451a7323", + "name": "steps", + "type": "INT", + "linkIds": [ + 37 + ], + "pos": [ + -650, + 440 + ] + }, + { + "id": "2213f872-d40f-4fc3-be01-b8fc73f1d92c", + "name": "cfg", + "type": "FLOAT", + "linkIds": [ + 42 + ], + "pos": [ + -650, + 460 + ] + }, + { + "id": "2c7b3e65-e71e-4a9b-a9f8-d2e814ccb6af", + "name": "seed", + "type": "INT", + "linkIds": [ + 43 + ], + "pos": [ + -650, + 480 + ] + }, + { + "id": "bddb2317-7210-48d5-81fd-6b2d6fac33f4", + "name": "unet_name", + "type": "COMBO", + "linkIds": [ + 44 + ], + "pos": [ + -650, + 500 + ] + }, + { + "id": "a283167b-6d7f-4d19-ad86-1fff2335c08d", + "name": "clip_name", + "type": "COMBO", + "linkIds": [ + 45 + ], + "pos": [ + -650, + 520 + ] + }, + { + "id": "e033047f-cc37-4043-b4a0-25d7bab661af", + "name": "vae_name", + "type": "COMBO", + "linkIds": [ + 46 + ], + "pos": [ + -650, + 540 + ] + } + ], + "outputs": [ + { + "id": "0a288e93-c03f-4805-80f3-4e320a6a492e", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 20 + ], + "localized_name": "IMAGE", + "pos": [ + 1700, + 360 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 26, + "type": "VAELoader", + "pos": [ + -360, + 590 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "vae_name", + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": 46 + } + ], + "outputs": [ + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "slot_index": 0, + "links": [ + 4, + 5, + 6, + 7 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "VAELoader", + "models": [ + { + "name": "ae.safetensors", + "url": "https://huggingface.co/Comfy-Org/z_image_turbo/resolve/main/split_files/vae/ae.safetensors", + "directory": "vae" + } + ] + }, + "widgets_values": [ + "ae.safetensors" + ] + }, + { + "id": 27, + "type": "TextEncodeQwenImageEdit", + "pos": [ + 10, + 200 + ], + "size": [ + 280, + 190 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 2 + }, + { + "localized_name": "vae", + "name": "vae", + "shape": 7, + "type": "VAE", + "link": 4 + }, + { + "localized_name": "image", + "name": "image", + "shape": 7, + "type": "IMAGE", + "link": 15 + }, + { + "localized_name": "prompt", + "name": "prompt", + "type": "STRING", + "widget": { + "name": "prompt" + }, + "link": 36 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "slot_index": 0, + "links": [ + 8 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "TextEncodeQwenImageEdit" + }, + "widgets_values": [ + "" + ] + }, + { + "id": 28, + "type": "TextEncodeQwenImageEdit", + "pos": [ + 10, + 440 + ], + "size": [ + 280, + 190 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 3 + }, + { + "localized_name": "vae", + "name": "vae", + "shape": 7, + "type": "VAE", + "link": 5 + }, + { + "localized_name": "image", + "name": "image", + "shape": 7, + "type": "IMAGE", + "link": 16 + }, + { + "localized_name": "prompt", + "name": "prompt", + "type": "STRING", + "widget": { + "name": "prompt" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "slot_index": 0, + "links": [ + 9 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "TextEncodeQwenImageEdit" + }, + "widgets_values": [ + "" + ] + }, + { + "id": 29, + "type": "FluxKontextMultiReferenceLatentMethod", + "pos": [ + 660, + 200 + ], + "size": [ + 270, + 80 + ], + "flags": {}, + "order": 3, + "mode": 0, + "showAdvanced": false, + "inputs": [ + { + "localized_name": "conditioning", + "name": "conditioning", + "type": "CONDITIONING", + "link": 10 + }, + { + "localized_name": "reference_latents_method", + "name": "reference_latents_method", + "type": "COMBO", + "widget": { + "name": "reference_latents_method" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "slot_index": 0, + "links": [ + 12 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "FluxKontextMultiReferenceLatentMethod" + }, + "widgets_values": [ + "index" + ] + }, + { + "id": 30, + "type": "FluxGuidance", + "pos": [ + 330, + 440 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "conditioning", + "name": "conditioning", + "type": "CONDITIONING", + "link": 9 + }, + { + "localized_name": "guidance", + "name": "guidance", + "type": "FLOAT", + "widget": { + "name": "guidance" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "slot_index": 0, + "links": [ + 11 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "FluxGuidance" + }, + "widgets_values": [ + 4.5 + ] + }, + { + "id": 31, + "type": "FluxGuidance", + "pos": [ + 330, + 200 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "conditioning", + "name": "conditioning", + "type": "CONDITIONING", + "link": 8 + }, + { + "localized_name": "guidance", + "name": "guidance", + "type": "FLOAT", + "widget": { + "name": "guidance" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "slot_index": 0, + "links": [ + 10 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "FluxGuidance" + }, + "widgets_values": [ + 4.5 + ] + }, + { + "id": 32, + "type": "FluxKontextMultiReferenceLatentMethod", + "pos": [ + 660, + 440 + ], + "size": [ + 270, + 80 + ], + "flags": {}, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "conditioning", + "name": "conditioning", + "type": "CONDITIONING", + "link": 11 + }, + { + "localized_name": "reference_latents_method", + "name": "reference_latents_method", + "type": "COMBO", + "widget": { + "name": "reference_latents_method" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "slot_index": 0, + "links": [ + 13 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "FluxKontextMultiReferenceLatentMethod" + }, + "widgets_values": [ + "index" + ] + }, + { + "id": 33, + "type": "KSampler", + "pos": [ + 1080, + 210 + ], + "size": [ + 270, + 460 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 1 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 12 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 13 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 18 + }, + { + "localized_name": "seed", + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": 43 + }, + { + "localized_name": "steps", + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": 37 + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": 42 + }, + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + }, + { + "localized_name": "scheduler", + "name": "scheduler", + "type": "COMBO", + "widget": { + "name": "scheduler" + }, + "link": null + }, + { + "localized_name": "denoise", + "name": "denoise", + "type": "FLOAT", + "widget": { + "name": "denoise" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 19 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "KSampler" + }, + "widgets_values": [ + 43, + "fixed", + 50, + 4.5, + "euler", + "simple", + 1 + ] + }, + { + "id": 34, + "type": "UNETLoader", + "pos": [ + -360, + 170 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "unet_name", + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": 44 + }, + { + "localized_name": "weight_dtype", + "name": "weight_dtype", + "type": "COMBO", + "widget": { + "name": "weight_dtype" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "slot_index": 0, + "links": [ + 1 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "UNETLoader", + "models": [ + { + "name": "longcat_image_edit_bf16.safetensors", + "url": "https://huggingface.co/TalmajM/LongCat-Image-Edit_ComfyUI_repackaged/resolve/main/split_files/diffusion_models/longcat_image_edit_bf16.safetensors", + "directory": "diffusion_models" + } + ] + }, + "widgets_values": [ + "longcat_image_edit_bf16.safetensors", + "default" + ] + }, + { + "id": 35, + "type": "VAEEncode", + "pos": [ + 710, + 790 + ], + "size": [ + 260, + 100 + ], + "flags": {}, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "pixels", + "name": "pixels", + "type": "IMAGE", + "link": 17 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 6 + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 18 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "VAEEncode" + } + }, + { + "id": 36, + "type": "VAEDecode", + "pos": [ + 1100, + 800 + ], + "size": [ + 230, + 100 + ], + "flags": {}, + "order": 10, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 19 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 7 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "slot_index": 0, + "links": [ + 20 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "VAEDecode" + } + }, + { + "id": 37, + "type": "ImageScaleToTotalPixels", + "pos": [ + -370, + 790 + ], + "size": [ + 270, + 140 + ], + "flags": {}, + "order": 11, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 14 + }, + { + "localized_name": "upscale_method", + "name": "upscale_method", + "type": "COMBO", + "widget": { + "name": "upscale_method" + }, + "link": null + }, + { + "localized_name": "megapixels", + "name": "megapixels", + "type": "FLOAT", + "widget": { + "name": "megapixels" + }, + "link": null + }, + { + "localized_name": "resolution_steps", + "name": "resolution_steps", + "type": "INT", + "widget": { + "name": "resolution_steps" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "slot_index": 0, + "links": [ + 15, + 16, + 17 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ImageScaleToTotalPixels" + }, + "widgets_values": [ + "lanczos", + 1, + 16 + ] + }, + { + "id": 38, + "type": "CLIPLoader", + "pos": [ + -360, + 360 + ], + "size": [ + 270, + 150 + ], + "flags": {}, + "order": 12, + "mode": 0, + "inputs": [ + { + "localized_name": "clip_name", + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": 45 + }, + { + "localized_name": "type", + "name": "type", + "type": "COMBO", + "widget": { + "name": "type" + }, + "link": null + }, + { + "localized_name": "device", + "name": "device", + "shape": 7, + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "slot_index": 0, + "links": [ + 2, + 3 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "CLIPLoader", + "models": [ + { + "name": "qwen_2.5_vl_7b_fp8_scaled.safetensors", + "url": "https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/resolve/main/split_files/text_encoders/qwen_2.5_vl_7b_fp8_scaled.safetensors", + "directory": "text_encoders" + } + ] + }, + "widgets_values": [ + "qwen_2.5_vl_7b_fp8_scaled.safetensors", + "longcat_image", + "default" + ] + } + ], + "groups": [ + { + "id": 1, + "title": "Models", + "bounding": [ + -380, + 100, + 320, + 630 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 3, + "title": "Conditioning", + "bounding": [ + -30, + 100, + 1030, + 630 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 4, + "title": "Sample", + "bounding": [ + 1030, + 100, + 360, + 630 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 2, + "origin_id": 38, + "origin_slot": 0, + "target_id": 27, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 4, + "origin_id": 26, + "origin_slot": 0, + "target_id": 27, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 15, + "origin_id": 37, + "origin_slot": 0, + "target_id": 27, + "target_slot": 2, + "type": "IMAGE" + }, + { + "id": 3, + "origin_id": 38, + "origin_slot": 0, + "target_id": 28, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 5, + "origin_id": 26, + "origin_slot": 0, + "target_id": 28, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 16, + "origin_id": 37, + "origin_slot": 0, + "target_id": 28, + "target_slot": 2, + "type": "IMAGE" + }, + { + "id": 10, + "origin_id": 31, + "origin_slot": 0, + "target_id": 29, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 9, + "origin_id": 28, + "origin_slot": 0, + "target_id": 30, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 8, + "origin_id": 27, + "origin_slot": 0, + "target_id": 31, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 11, + "origin_id": 30, + "origin_slot": 0, + "target_id": 32, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 1, + "origin_id": 34, + "origin_slot": 0, + "target_id": 33, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 12, + "origin_id": 29, + "origin_slot": 0, + "target_id": 33, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 13, + "origin_id": 32, + "origin_slot": 0, + "target_id": 33, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 18, + "origin_id": 35, + "origin_slot": 0, + "target_id": 33, + "target_slot": 3, + "type": "LATENT" + }, + { + "id": 17, + "origin_id": 37, + "origin_slot": 0, + "target_id": 35, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 6, + "origin_id": 26, + "origin_slot": 0, + "target_id": 35, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 19, + "origin_id": 33, + "origin_slot": 0, + "target_id": 36, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 7, + "origin_id": 26, + "origin_slot": 0, + "target_id": 36, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 14, + "origin_id": -10, + "origin_slot": 0, + "target_id": 37, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 20, + "origin_id": 36, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 36, + "origin_id": -10, + "origin_slot": 1, + "target_id": 27, + "target_slot": 3, + "type": "STRING" + }, + { + "id": 37, + "origin_id": -10, + "origin_slot": 2, + "target_id": 33, + "target_slot": 5, + "type": "INT" + }, + { + "id": 42, + "origin_id": -10, + "origin_slot": 3, + "target_id": 33, + "target_slot": 6, + "type": "FLOAT" + }, + { + "id": 43, + "origin_id": -10, + "origin_slot": 4, + "target_id": 33, + "target_slot": 4, + "type": "INT" + }, + { + "id": 44, + "origin_id": -10, + "origin_slot": 5, + "target_id": 34, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 45, + "origin_id": -10, + "origin_slot": 6, + "target_id": 38, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 46, + "origin_id": -10, + "origin_slot": 7, + "target_id": 26, + "target_slot": 0, + "type": "COMBO" + } + ], + "extra": {}, + "category": "Image generation and editing/Edit image", + "description": "Edits images via text instructions using LongCat Image Edit, an instruction-following image editing diffusion model." + } + ] + }, + "extra": { + "ue_links": [] + } +} \ No newline at end of file diff --git a/blueprints/Image Edit (Qwen 2509).json b/blueprints/Image Edit (Qwen 2509).json new file mode 100644 index 000000000..f7be322a0 --- /dev/null +++ b/blueprints/Image Edit (Qwen 2509).json @@ -0,0 +1,1947 @@ +{ + "revision": 0, + "last_node_id": 433, + "last_link_id": 0, + "nodes": [ + { + "id": 433, + "type": "eba40a3a-f6c5-48ac-b58e-55525d06b373", + "pos": [ + 90, + -160 + ], + "size": [ + 390, + 610 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "label": "image", + "name": "image", + "type": "IMAGE", + "link": null + }, + { + "label": "image2 (optional)", + "name": "image2", + "type": "IMAGE", + "link": null + }, + { + "label": "image3 (optional)", + "name": "image3", + "type": "IMAGE", + "link": null + }, + { + "name": "prompt", + "type": "STRING", + "widget": { + "name": "prompt" + }, + "link": null + }, + { + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": null + }, + { + "label": "enable_turbo_mode", + "name": "value", + "type": "BOOLEAN", + "widget": { + "name": "value" + }, + "link": null + }, + { + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": null + }, + { + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": null + }, + { + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [] + } + ], + "properties": { + "proxyWidgets": [ + [ + "111", + "prompt" + ], + [ + "3", + "seed" + ], + [ + "443", + "value" + ], + [ + "37", + "unet_name" + ], + [ + "38", + "clip_name" + ], + [ + "39", + "vae_name" + ], + [ + "3", + "control_after_generate" + ] + ], + "cnr_id": "comfy-core", + "ver": "0.3.62" + }, + "widgets_values": [], + "title": "Image Edit (Qwen 2509)" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "eba40a3a-f6c5-48ac-b58e-55525d06b373", + "version": 1, + "state": { + "lastGroupId": 51, + "lastNodeId": 468, + "lastLinkId": 731, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Image Edit (Qwen 2509)", + "inputNode": { + "id": -10, + "bounding": [ + -1160, + 280, + 151.744140625, + 220 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 2030, + -20, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "d5089bd3-63bc-4a24-b478-6565ed2364e3", + "name": "image", + "type": "IMAGE", + "linkIds": [ + 248 + ], + "label": "image", + "pos": [ + -1028.255859375, + 300 + ] + }, + { + "id": "9e80fff0-ed0a-439f-a16e-a4a6cc1eb601", + "name": "image2", + "type": "IMAGE", + "linkIds": [ + 235, + 236 + ], + "label": "image2 (optional)", + "pos": [ + -1028.255859375, + 320 + ] + }, + { + "id": "49d98fd6-01b5-440b-8603-579252fd7fef", + "name": "image3", + "type": "IMAGE", + "linkIds": [ + 237, + 238 + ], + "label": "image3 (optional)", + "pos": [ + -1028.255859375, + 340 + ] + }, + { + "id": "5de32f24-a7b5-4423-b772-72824005f585", + "name": "prompt", + "type": "STRING", + "linkIds": [ + 244 + ], + "pos": [ + -1028.255859375, + 360 + ] + }, + { + "id": "85fb3d74-7881-4c71-bc8c-624be5eedc3d", + "name": "seed", + "type": "INT", + "linkIds": [ + 718 + ], + "pos": [ + -1028.255859375, + 380 + ] + }, + { + "id": "b0c828de-d7eb-42a3-8dfb-4f53360d4fc9", + "name": "value", + "type": "BOOLEAN", + "linkIds": [ + 719 + ], + "label": "enable_turbo_mode", + "pos": [ + -1028.255859375, + 400 + ] + }, + { + "id": "072baa05-5551-4a98-bd66-015a36833ac2", + "name": "unet_name", + "type": "COMBO", + "linkIds": [ + 720 + ], + "pos": [ + -1028.255859375, + 420 + ] + }, + { + "id": "d2891d11-b336-4750-9742-b93717c9ae39", + "name": "clip_name", + "type": "COMBO", + "linkIds": [ + 721 + ], + "pos": [ + -1028.255859375, + 440 + ] + }, + { + "id": "4218135f-5128-4b7e-8572-92cc55615793", + "name": "vae_name", + "type": "COMBO", + "linkIds": [ + 722 + ], + "pos": [ + -1028.255859375, + 460 + ] + } + ], + "outputs": [ + { + "id": "c4ebfc18-de83-4361-8e42-767c3c8c25c0", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 110 + ], + "localized_name": "IMAGE", + "pos": [ + 2050, + 0 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 75, + "type": "CFGNorm", + "pos": [ + 1080, + 30 + ], + "size": [ + 290, + 110 + ], + "flags": {}, + "order": 11, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 141 + }, + { + "localized_name": "strength", + "name": "strength", + "type": "FLOAT", + "widget": { + "name": "strength" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "patched_model", + "name": "patched_model", + "type": "MODEL", + "links": [ + 186 + ] + } + ], + "properties": { + "Node name for S&R": "CFGNorm", + "cnr_id": "comfy-core", + "ver": "0.3.50", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": { + "strength": true + } + } + }, + "widgets_values": [ + 1 + ] + }, + { + "id": 39, + "type": "VAELoader", + "pos": [ + -730, + 410 + ], + "size": [ + 330, + 110 + ], + "flags": {}, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "vae_name", + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": 722 + } + ], + "outputs": [ + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "slot_index": 0, + "links": [ + 76, + 168, + 206, + 207 + ] + } + ], + "properties": { + "Node name for S&R": "VAELoader", + "cnr_id": "comfy-core", + "ver": "0.3.48", + "models": [ + { + "name": "qwen_image_vae.safetensors", + "url": "https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/resolve/main/split_files/vae/qwen_image_vae.safetensors", + "directory": "vae" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "widget_ue_connectable": {} + }, + "widgets_values": [ + "qwen_image_vae.safetensors" + ] + }, + { + "id": 38, + "type": "CLIPLoader", + "pos": [ + -730, + 150 + ], + "size": [ + 330, + 150 + ], + "flags": {}, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "clip_name", + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": 721 + }, + { + "localized_name": "type", + "name": "type", + "type": "COMBO", + "widget": { + "name": "type" + }, + "link": null + }, + { + "localized_name": "device", + "name": "device", + "shape": 7, + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "slot_index": 0, + "links": [ + 204, + 205 + ] + } + ], + "properties": { + "Node name for S&R": "CLIPLoader", + "cnr_id": "comfy-core", + "ver": "0.3.48", + "models": [ + { + "name": "qwen_2.5_vl_7b_fp8_scaled.safetensors", + "url": "https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/resolve/main/split_files/text_encoders/qwen_2.5_vl_7b_fp8_scaled.safetensors", + "directory": "text_encoders" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "widget_ue_connectable": {} + }, + "widgets_values": [ + "qwen_2.5_vl_7b_fp8_scaled.safetensors", + "qwen_image", + "default" + ] + }, + { + "id": 37, + "type": "UNETLoader", + "pos": [ + -730, + -60 + ], + "size": [ + 330, + 110 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "unet_name", + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": 720 + }, + { + "localized_name": "weight_dtype", + "name": "weight_dtype", + "type": "COMBO", + "widget": { + "name": "weight_dtype" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "slot_index": 0, + "links": [ + 184, + 710 + ] + } + ], + "properties": { + "Node name for S&R": "UNETLoader", + "cnr_id": "comfy-core", + "ver": "0.3.48", + "models": [ + { + "name": "qwen_image_edit_2509_fp8_e4m3fn.safetensors", + "url": "https://huggingface.co/Comfy-Org/Qwen-Image-Edit_ComfyUI/resolve/main/split_files/diffusion_models/qwen_image_edit_2509_fp8_e4m3fn.safetensors", + "directory": "diffusion_models" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "widget_ue_connectable": {} + }, + "widgets_values": [ + "qwen_image_edit_2509_fp8_e4m3fn.safetensors", + "default" + ] + }, + { + "id": 110, + "type": "TextEncodeQwenImageEditPlus", + "pos": [ + -240, + 320 + ], + "size": [ + 400, + 240 + ], + "flags": {}, + "order": 14, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 204 + }, + { + "localized_name": "vae", + "name": "vae", + "shape": 7, + "type": "VAE", + "link": 206 + }, + { + "localized_name": "image1", + "name": "image1", + "shape": 7, + "type": "IMAGE", + "link": 251 + }, + { + "localized_name": "image2", + "name": "image2", + "shape": 7, + "type": "IMAGE", + "link": 236 + }, + { + "localized_name": "image3", + "name": "image3", + "shape": 7, + "type": "IMAGE", + "link": 238 + }, + { + "localized_name": "prompt", + "name": "prompt", + "type": "STRING", + "widget": { + "name": "prompt" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 210 + ] + } + ], + "properties": { + "Node name for S&R": "TextEncodeQwenImageEditPlus", + "cnr_id": "comfy-core", + "ver": "0.3.59" + }, + "widgets_values": [ + "" + ], + "color": "#223", + "bgcolor": "#335" + }, + { + "id": 66, + "type": "ModelSamplingAuraFlow", + "pos": [ + 1070, + -120 + ], + "size": [ + 290, + 110 + ], + "flags": {}, + "order": 10, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 708 + }, + { + "localized_name": "shift", + "name": "shift", + "type": "FLOAT", + "widget": { + "name": "shift" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 141 + ] + } + ], + "properties": { + "Node name for S&R": "ModelSamplingAuraFlow", + "cnr_id": "comfy-core", + "ver": "0.3.48", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "widget_ue_connectable": {} + }, + "widgets_values": [ + 3 + ] + }, + { + "id": 111, + "type": "TextEncodeQwenImageEditPlus", + "pos": [ + -250, + -70 + ], + "size": [ + 410, + 330 + ], + "flags": {}, + "order": 15, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 205 + }, + { + "localized_name": "vae", + "name": "vae", + "shape": 7, + "type": "VAE", + "link": 207 + }, + { + "localized_name": "image1", + "name": "image1", + "shape": 7, + "type": "IMAGE", + "link": 250 + }, + { + "localized_name": "image2", + "name": "image2", + "shape": 7, + "type": "IMAGE", + "link": 235 + }, + { + "localized_name": "image3", + "name": "image3", + "shape": 7, + "type": "IMAGE", + "link": 237 + }, + { + "localized_name": "prompt", + "name": "prompt", + "type": "STRING", + "widget": { + "name": "prompt" + }, + "link": 244 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 211 + ] + } + ], + "properties": { + "Node name for S&R": "TextEncodeQwenImageEditPlus", + "cnr_id": "comfy-core", + "ver": "0.3.59" + }, + "widgets_values": [ + "" + ], + "color": "#232", + "bgcolor": "#353" + }, + { + "id": 88, + "type": "VAEEncode", + "pos": [ + -70, + 640 + ], + "size": [ + 230, + 100 + ], + "flags": {}, + "order": 12, + "mode": 0, + "inputs": [ + { + "localized_name": "pixels", + "name": "pixels", + "type": "IMAGE", + "link": 249 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 168 + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "links": [ + 246 + ] + } + ], + "properties": { + "Node name for S&R": "VAEEncode", + "cnr_id": "comfy-core", + "ver": "0.3.50", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {} + } + } + }, + { + "id": 8, + "type": "VAEDecode", + "pos": [ + 1590, + -60 + ], + "size": [ + 230, + 100 + ], + "flags": { + "collapsed": false + }, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 128 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 76 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "slot_index": 0, + "links": [ + 110 + ] + } + ], + "properties": { + "Node name for S&R": "VAEDecode", + "cnr_id": "comfy-core", + "ver": "0.3.48", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "widget_ue_connectable": {} + } + }, + { + "id": 89, + "type": "LoraLoaderModelOnly", + "pos": [ + 320, + 300 + ], + "size": [ + 300, + 140 + ], + "flags": {}, + "order": 13, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 184 + }, + { + "localized_name": "lora_name", + "name": "lora_name", + "type": "COMBO", + "widget": { + "name": "lora_name" + }, + "link": null + }, + { + "localized_name": "strength_model", + "name": "strength_model", + "type": "FLOAT", + "widget": { + "name": "strength_model" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 709 + ] + } + ], + "properties": { + "Node name for S&R": "LoraLoaderModelOnly", + "cnr_id": "comfy-core", + "ver": "0.3.50", + "models": [ + { + "name": "Qwen-Image-Edit-2509-Lightning-4steps-V1.0-bf16.safetensors", + "url": "https://huggingface.co/lightx2v/Qwen-Image-Lightning/resolve/main/Qwen-Image-Edit-2509/Qwen-Image-Edit-2509-Lightning-4steps-V1.0-bf16.safetensors", + "directory": "loras" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": { + "lora_name": true, + "strength_model": true + } + } + }, + "widgets_values": [ + "Qwen-Image-Edit-2509-Lightning-4steps-V1.0-bf16.safetensors", + 1 + ] + }, + { + "id": 117, + "type": "FluxKontextImageScale", + "pos": [ + -680, + 630 + ], + "size": [ + 230, + 80 + ], + "flags": {}, + "order": 16, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 248 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 249, + 250, + 251 + ] + } + ], + "properties": { + "Node name for S&R": "FluxKontextImageScale" + } + }, + { + "id": 3, + "type": "KSampler", + "pos": [ + 1070, + 210 + ], + "size": [ + 300, + 590 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 186 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 211 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 210 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 246 + }, + { + "localized_name": "seed", + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": 718 + }, + { + "localized_name": "steps", + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": 707 + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": 706 + }, + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + }, + { + "localized_name": "scheduler", + "name": "scheduler", + "type": "COMBO", + "widget": { + "name": "scheduler" + }, + "link": null + }, + { + "localized_name": "denoise", + "name": "denoise", + "type": "FLOAT", + "widget": { + "name": "denoise" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 128 + ] + } + ], + "properties": { + "Node name for S&R": "KSampler", + "cnr_id": "comfy-core", + "ver": "0.3.48", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "widget_ue_connectable": {} + }, + "widgets_values": [ + 973414316252139, + "randomize", + 4, + 1, + "euler", + "simple", + 1 + ] + }, + { + "id": 436, + "type": "PrimitiveInt", + "pos": [ + 320, + 500 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 713 + ] + } + ], + "title": "Steps", + "properties": { + "Node name for S&R": "PrimitiveInt" + }, + "widgets_values": [ + 4, + "fixed" + ] + }, + { + "id": 437, + "type": "PrimitiveFloat", + "pos": [ + 320, + 670 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "FLOAT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": [ + 714 + ] + } + ], + "title": "CFG", + "properties": { + "Node name for S&R": "PrimitiveFloat" + }, + "widgets_values": [ + 1 + ] + }, + { + "id": 438, + "type": "PrimitiveInt", + "pos": [ + 320, + -100 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 711 + ] + } + ], + "title": "Steps", + "properties": { + "Node name for S&R": "PrimitiveInt" + }, + "widgets_values": [ + 20, + "fixed" + ] + }, + { + "id": 439, + "type": "PrimitiveFloat", + "pos": [ + 320, + 70 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "FLOAT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": [ + 712 + ] + } + ], + "title": "CFG", + "properties": { + "Node name for S&R": "PrimitiveFloat" + }, + "widgets_values": [ + 4 + ] + }, + { + "id": 440, + "type": "ComfySwitchNode", + "pos": [ + 750, + -80 + ], + "size": [ + 270, + 130 + ], + "flags": {}, + "order": 17, + "mode": 0, + "inputs": [ + { + "localized_name": "on_false", + "name": "on_false", + "type": "*", + "link": 710 + }, + { + "localized_name": "on_true", + "name": "on_true", + "type": "*", + "link": 709 + }, + { + "localized_name": "switch", + "name": "switch", + "type": "BOOLEAN", + "widget": { + "name": "switch" + }, + "link": 715 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "*", + "links": [ + 708 + ] + } + ], + "title": "Switch (Model)", + "properties": { + "Node name for S&R": "ComfySwitchNode" + }, + "widgets_values": [ + false + ] + }, + { + "id": 441, + "type": "ComfySwitchNode", + "pos": [ + 730, + 340 + ], + "size": [ + 270, + 130 + ], + "flags": {}, + "order": 18, + "mode": 0, + "inputs": [ + { + "localized_name": "on_false", + "name": "on_false", + "type": "*", + "link": 711 + }, + { + "localized_name": "on_true", + "name": "on_true", + "type": "*", + "link": 713 + }, + { + "localized_name": "switch", + "name": "switch", + "type": "BOOLEAN", + "widget": { + "name": "switch" + }, + "link": 716 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "*", + "links": [ + 707 + ] + } + ], + "title": "Switch (Steps)", + "properties": { + "Node name for S&R": "ComfySwitchNode" + }, + "widgets_values": [ + false + ] + }, + { + "id": 442, + "type": "ComfySwitchNode", + "pos": [ + 730, + 520 + ], + "size": [ + 270, + 130 + ], + "flags": {}, + "order": 19, + "mode": 0, + "inputs": [ + { + "localized_name": "on_false", + "name": "on_false", + "type": "*", + "link": 712 + }, + { + "localized_name": "on_true", + "name": "on_true", + "type": "*", + "link": 714 + }, + { + "localized_name": "switch", + "name": "switch", + "type": "BOOLEAN", + "widget": { + "name": "switch" + }, + "link": 717 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "*", + "links": [ + 706 + ] + } + ], + "title": "Switch (CFG)", + "properties": { + "Node name for S&R": "ComfySwitchNode" + }, + "widgets_values": [ + false + ] + }, + { + "id": 443, + "type": "PrimitiveBoolean", + "pos": [ + 330, + 850 + ], + "size": [ + 270, + 100 + ], + "flags": {}, + "order": 20, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "BOOLEAN", + "widget": { + "name": "value" + }, + "link": 719 + } + ], + "outputs": [ + { + "localized_name": "BOOLEAN", + "name": "BOOLEAN", + "type": "BOOLEAN", + "links": [ + 715, + 716, + 717 + ] + } + ], + "title": "Enable Lightning LoRA", + "properties": { + "Node name for S&R": "PrimitiveBoolean" + }, + "widgets_values": [ + true + ] + }, + { + "id": 444, + "type": "MarkdownNote", + "pos": [ + 240, + -500 + ], + "size": [ + 450, + 310 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [], + "outputs": [], + "title": "Note: KSampler settings", + "properties": {}, + "widgets_values": [ + "You can test and find the best setting by yourself. The following table is for reference.\n| Parameters | Qwen Team | Comfy Original | with 4steps LoRA |\n|--------|---------|------------|---------------------------|\n| Steps | 50 | 20 | 4 |\n| CFG | 4.0 | 2.5 | 1.0 |" + ], + "color": "#432", + "bgcolor": "#000" + } + ], + "groups": [ + { + "id": 1, + "title": "Step1 - Load models", + "bounding": [ + -770, + -170, + 410, + 750 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 3, + "title": "Step 4 - Prompt", + "bounding": [ + -330, + -170, + 570, + 750 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 50, + "title": "Lightning LoRA", + "bounding": [ + 270, + 220, + 390, + 570 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 51, + "title": "Original Settings", + "bounding": [ + 270, + -170, + 390, + 360 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 141, + "origin_id": 66, + "origin_slot": 0, + "target_id": 75, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 128, + "origin_id": 3, + "origin_slot": 0, + "target_id": 8, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 76, + "origin_id": 39, + "origin_slot": 0, + "target_id": 8, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 184, + "origin_id": 37, + "origin_slot": 0, + "target_id": 89, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 186, + "origin_id": 75, + "origin_slot": 0, + "target_id": 3, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 211, + "origin_id": 111, + "origin_slot": 0, + "target_id": 3, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 210, + "origin_id": 110, + "origin_slot": 0, + "target_id": 3, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 168, + "origin_id": 39, + "origin_slot": 0, + "target_id": 88, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 204, + "origin_id": 38, + "origin_slot": 0, + "target_id": 110, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 206, + "origin_id": 39, + "origin_slot": 0, + "target_id": 110, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 205, + "origin_id": 38, + "origin_slot": 0, + "target_id": 111, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 207, + "origin_id": 39, + "origin_slot": 0, + "target_id": 111, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 110, + "origin_id": 8, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 235, + "origin_id": -10, + "origin_slot": 1, + "target_id": 111, + "target_slot": 3, + "type": "IMAGE" + }, + { + "id": 236, + "origin_id": -10, + "origin_slot": 1, + "target_id": 110, + "target_slot": 3, + "type": "IMAGE" + }, + { + "id": 237, + "origin_id": -10, + "origin_slot": 2, + "target_id": 111, + "target_slot": 4, + "type": "IMAGE" + }, + { + "id": 238, + "origin_id": -10, + "origin_slot": 2, + "target_id": 110, + "target_slot": 4, + "type": "IMAGE" + }, + { + "id": 244, + "origin_id": -10, + "origin_slot": 3, + "target_id": 111, + "target_slot": 5, + "type": "STRING" + }, + { + "id": 246, + "origin_id": 88, + "origin_slot": 0, + "target_id": 3, + "target_slot": 3, + "type": "LATENT" + }, + { + "id": 248, + "origin_id": -10, + "origin_slot": 0, + "target_id": 117, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 249, + "origin_id": 117, + "origin_slot": 0, + "target_id": 88, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 250, + "origin_id": 117, + "origin_slot": 0, + "target_id": 111, + "target_slot": 2, + "type": "IMAGE" + }, + { + "id": 251, + "origin_id": 117, + "origin_slot": 0, + "target_id": 110, + "target_slot": 2, + "type": "IMAGE" + }, + { + "id": 706, + "origin_id": 442, + "origin_slot": 0, + "target_id": 3, + "target_slot": 6, + "type": "FLOAT" + }, + { + "id": 707, + "origin_id": 441, + "origin_slot": 0, + "target_id": 3, + "target_slot": 5, + "type": "INT" + }, + { + "id": 708, + "origin_id": 440, + "origin_slot": 0, + "target_id": 66, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 709, + "origin_id": 89, + "origin_slot": 0, + "target_id": 440, + "target_slot": 1, + "type": "MODEL" + }, + { + "id": 710, + "origin_id": 37, + "origin_slot": 0, + "target_id": 440, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 711, + "origin_id": 438, + "origin_slot": 0, + "target_id": 441, + "target_slot": 0, + "type": "INT" + }, + { + "id": 712, + "origin_id": 439, + "origin_slot": 0, + "target_id": 442, + "target_slot": 0, + "type": "FLOAT" + }, + { + "id": 713, + "origin_id": 436, + "origin_slot": 0, + "target_id": 441, + "target_slot": 1, + "type": "INT" + }, + { + "id": 714, + "origin_id": 437, + "origin_slot": 0, + "target_id": 442, + "target_slot": 1, + "type": "FLOAT" + }, + { + "id": 715, + "origin_id": 443, + "origin_slot": 0, + "target_id": 440, + "target_slot": 2, + "type": "BOOLEAN" + }, + { + "id": 716, + "origin_id": 443, + "origin_slot": 0, + "target_id": 441, + "target_slot": 2, + "type": "BOOLEAN" + }, + { + "id": 717, + "origin_id": 443, + "origin_slot": 0, + "target_id": 442, + "target_slot": 2, + "type": "BOOLEAN" + }, + { + "id": 718, + "origin_id": -10, + "origin_slot": 4, + "target_id": 3, + "target_slot": 4, + "type": "INT" + }, + { + "id": 719, + "origin_id": -10, + "origin_slot": 5, + "target_id": 443, + "target_slot": 0, + "type": "BOOLEAN" + }, + { + "id": 720, + "origin_id": -10, + "origin_slot": 6, + "target_id": 37, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 721, + "origin_id": -10, + "origin_slot": 7, + "target_id": 38, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 722, + "origin_id": -10, + "origin_slot": 8, + "target_id": 39, + "target_slot": 0, + "type": "COMBO" + } + ], + "extra": { + "workflowRendererVersion": "LG" + }, + "category": "Image generation and editing/Edit image", + "description": "Edits images from text instructions using Qwen-Image-Edit-2509 with optional Lightning LoRA for few-step sampling." + } + ] + }, + "extra": {} +} diff --git a/blueprints/Image Edit (Qwen 2511).json b/blueprints/Image Edit (Qwen 2511).json index 582171fa0..1aa7e5765 100644 --- a/blueprints/Image Edit (Qwen 2511).json +++ b/blueprints/Image Edit (Qwen 2511).json @@ -132,7 +132,7 @@ }, "revision": 0, "config": {}, - "name": "local-Image Edit (Qwen 2511)", + "name": "Image Edit (Qwen 2511)", "inputNode": { "id": -10, "bounding": [ @@ -1468,7 +1468,8 @@ "VHS_MetadataImage": true, "VHS_KeepIntermediate": true }, - "category": "Image generation and editing/Edit image" + "category": "Image generation and editing/Edit image", + "description": "Edits images via text instructions using Qwen-Image-Edit-2511 with improved character consistency and integrated LoRA." } ] }, @@ -1489,4 +1490,4 @@ } }, "version": 0.4 -} +} \ No newline at end of file diff --git a/blueprints/Image Inpainting (Flux.1 Fill Dev).json b/blueprints/Image Inpainting (Flux.1 Fill Dev).json new file mode 100644 index 000000000..c1326ed3d --- /dev/null +++ b/blueprints/Image Inpainting (Flux.1 Fill Dev).json @@ -0,0 +1,1206 @@ +{ + "revision": 0, + "last_node_id": 232, + "last_link_id": 0, + "nodes": [ + { + "id": 232, + "type": "6e8d6e38-bdc3-436c-be85-ef9e67e70e07", + "pos": [ + 1270, + 4640 + ], + "size": [ + 400, + 470 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "label": "image", + "localized_name": "pixels", + "name": "pixels", + "type": "IMAGE", + "link": null + }, + { + "localized_name": "mask", + "name": "mask", + "type": "MASK", + "link": null + }, + { + "label": "prompt", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + }, + { + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": null + }, + { + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": null + }, + { + "name": "clip_name1", + "type": "COMBO", + "widget": { + "name": "clip_name1" + }, + "link": null + }, + { + "name": "clip_name2", + "type": "COMBO", + "widget": { + "name": "clip_name2" + }, + "link": null + }, + { + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [] + } + ], + "properties": { + "proxyWidgets": [ + [ + "23", + "text" + ], + [ + "3", + "seed" + ], + [ + "31", + "unet_name" + ], + [ + "34", + "clip_name1" + ], + [ + "34", + "clip_name2" + ], + [ + "230", + "vae_name" + ] + ], + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.18.1" + }, + "widgets_values": [], + "title": "Image Inpainting (Flux.1 Fill Dev)" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "6e8d6e38-bdc3-436c-be85-ef9e67e70e07", + "version": 1, + "state": { + "lastGroupId": 22, + "lastNodeId": 232, + "lastLinkId": 286, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Image Inpainting (Flux.1 Fill Dev)", + "inputNode": { + "id": -10, + "bounding": [ + -850, + 164, + 120, + 200 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 1230, + 140, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "65727ee9-09d0-40c9-bd86-11e0823eb676", + "name": "pixels", + "type": "IMAGE", + "linkIds": [ + 99 + ], + "localized_name": "pixels", + "label": "image", + "pos": [ + -750, + 184 + ] + }, + { + "id": "28424f77-56c5-49c1-ba41-6bd78287c186", + "name": "mask", + "type": "MASK", + "linkIds": [ + 100 + ], + "localized_name": "mask", + "pos": [ + -750, + 204 + ] + }, + { + "id": "2339e5e0-8f8d-4600-b158-7d7dae5f0535", + "name": "text", + "type": "STRING", + "linkIds": [ + 277 + ], + "label": "prompt", + "pos": [ + -750, + 224 + ] + }, + { + "id": "5f433d9b-b97e-4bac-bb88-eb668de2d5a7", + "name": "seed", + "type": "INT", + "linkIds": [ + 282 + ], + "pos": [ + -750, + 244 + ] + }, + { + "id": "35a8b6c1-c92c-4c1a-9b24-2e9bae7808f6", + "name": "unet_name", + "type": "COMBO", + "linkIds": [ + 283 + ], + "pos": [ + -750, + 264 + ] + }, + { + "id": "3af8f8be-bce8-4ba0-aea0-ccf6b377d5f6", + "name": "clip_name1", + "type": "COMBO", + "linkIds": [ + 284 + ], + "pos": [ + -750, + 284 + ] + }, + { + "id": "d9a4af80-4fa1-4792-b955-78bdaef4596e", + "name": "clip_name2", + "type": "COMBO", + "linkIds": [ + 285 + ], + "pos": [ + -750, + 304 + ] + }, + { + "id": "d59398cf-7e9c-4dae-8c5a-08c4756f256a", + "name": "vae_name", + "type": "COMBO", + "linkIds": [ + 286 + ], + "pos": [ + -750, + 324 + ] + } + ], + "outputs": [ + { + "id": "1dee24ec-54a8-41be-aa30-a8fb797d3d23", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 95 + ], + "localized_name": "IMAGE", + "pos": [ + 1250, + 160 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 34, + "type": "DualCLIPLoader", + "pos": [ + -590, + 150 + ], + "size": [ + 320, + 180 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "clip_name1", + "name": "clip_name1", + "type": "COMBO", + "widget": { + "name": "clip_name1" + }, + "link": 284 + }, + { + "localized_name": "clip_name2", + "name": "clip_name2", + "type": "COMBO", + "widget": { + "name": "clip_name2" + }, + "link": 285 + }, + { + "localized_name": "type", + "name": "type", + "type": "COMBO", + "widget": { + "name": "type" + }, + "link": null + }, + { + "localized_name": "device", + "name": "device", + "shape": 7, + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [ + 62 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.64", + "Node name for S&R": "DualCLIPLoader", + "models": [ + { + "name": "clip_l.safetensors", + "url": "https://huggingface.co/comfyanonymous/flux_text_encoders/resolve/main/clip_l.safetensors", + "directory": "text_encoders" + }, + { + "name": "t5xxl_fp16.safetensors", + "url": "https://huggingface.co/comfyanonymous/flux_text_encoders/resolve/main/t5xxl_fp16.safetensors", + "directory": "text_encoders" + } + ] + }, + "widgets_values": [ + "clip_l.safetensors", + "t5xxl_fp16.safetensors", + "flux", + "default" + ] + }, + { + "id": 229, + "type": "FluxGuidance", + "pos": [ + 410, + -40 + ], + "size": [ + 320, + 110 + ], + "flags": { + "collapsed": false + }, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "conditioning", + "name": "conditioning", + "type": "CONDITIONING", + "link": 41 + }, + { + "localized_name": "guidance", + "name": "guidance", + "type": "FLOAT", + "widget": { + "name": "guidance" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "slot_index": 0, + "links": [ + 80 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.64", + "Node name for S&R": "FluxGuidance" + }, + "widgets_values": [ + 30 + ] + }, + { + "id": 230, + "type": "VAELoader", + "pos": [ + -590, + 450 + ], + "size": [ + 320, + 110 + ], + "flags": {}, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "vae_name", + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": 286 + } + ], + "outputs": [ + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "slot_index": 0, + "links": [ + 60, + 82 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.64", + "Node name for S&R": "VAELoader", + "models": [ + { + "name": "ae.safetensors", + "url": "https://huggingface.co/Comfy-Org/Lumina_Image_2.0_Repackaged/resolve/main/split_files/vae/ae.safetensors", + "directory": "vae" + } + ] + }, + "widgets_values": [ + "ae.safetensors" + ] + }, + { + "id": 31, + "type": "UNETLoader", + "pos": [ + -590, + -90 + ], + "size": [ + 320, + 110 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "unet_name", + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": 283 + }, + { + "localized_name": "weight_dtype", + "name": "weight_dtype", + "type": "COMBO", + "widget": { + "name": "weight_dtype" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "slot_index": 0, + "links": [ + 85 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.64", + "Node name for S&R": "UNETLoader", + "models": [ + { + "name": "flux1-fill-dev.safetensors", + "url": "https://huggingface.co/Comfy-Org/flux1-dev/resolve/main/split_files/diffusion_models/flux1-fill-dev.safetensors", + "directory": "diffusion_models" + } + ] + }, + "widgets_values": [ + "flux1-fill-dev.safetensors", + "default" + ] + }, + { + "id": 46, + "type": "ConditioningZeroOut", + "pos": [ + 90, + 420 + ], + "size": [ + 230, + 80 + ], + "flags": {}, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "conditioning", + "name": "conditioning", + "type": "CONDITIONING", + "link": 101 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 102 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.64", + "Node name for S&R": "ConditioningZeroOut" + } + }, + { + "id": 23, + "type": "CLIPTextEncode", + "pos": [ + -160, + -70 + ], + "size": [ + 480, + 410 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 62 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 277 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "slot_index": 0, + "links": [ + 41, + 101 + ] + } + ], + "title": "CLIP Text Encode (Positive Prompt)", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.64", + "Node name for S&R": "CLIPTextEncode" + }, + "widgets_values": [ + "" + ], + "color": "#232", + "bgcolor": "#353" + }, + { + "id": 39, + "type": "DifferentialDiffusion", + "pos": [ + 780, + -110 + ], + "size": [ + 280, + 110 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 85 + }, + { + "localized_name": "strength", + "name": "strength", + "shape": 7, + "type": "FLOAT", + "widget": { + "name": "strength" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "slot_index": 0, + "links": [ + 86 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.64", + "Node name for S&R": "DifferentialDiffusion" + }, + "widgets_values": [ + 1 + ] + }, + { + "id": 231, + "type": "VAEDecode", + "pos": [ + 780, + 590 + ], + "size": [ + 230, + 100 + ], + "flags": {}, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 7 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 60 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "slot_index": 0, + "links": [ + 95 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.64", + "Node name for S&R": "VAEDecode" + } + }, + { + "id": 38, + "type": "InpaintModelConditioning", + "pos": [ + 420, + 120 + ], + "size": [ + 310, + 200 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 80 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 102 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 82 + }, + { + "localized_name": "pixels", + "name": "pixels", + "type": "IMAGE", + "link": 99 + }, + { + "localized_name": "mask", + "name": "mask", + "type": "MASK", + "link": 100 + }, + { + "localized_name": "noise_mask", + "name": "noise_mask", + "type": "BOOLEAN", + "widget": { + "name": "noise_mask" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "slot_index": 0, + "links": [ + 77 + ] + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "slot_index": 1, + "links": [ + 78 + ] + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "slot_index": 2, + "links": [ + 88 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.64", + "Node name for S&R": "InpaintModelConditioning" + }, + "widgets_values": [ + true + ] + }, + { + "id": 3, + "type": "KSampler", + "pos": [ + 770, + 40 + ], + "size": [ + 290, + 470 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 86 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 77 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 78 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 88 + }, + { + "localized_name": "seed", + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": 282 + }, + { + "localized_name": "steps", + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": null + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": null + }, + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + }, + { + "localized_name": "scheduler", + "name": "scheduler", + "type": "COMBO", + "widget": { + "name": "scheduler" + }, + "link": null + }, + { + "localized_name": "denoise", + "name": "denoise", + "type": "FLOAT", + "widget": { + "name": "denoise" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 7 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.64", + "Node name for S&R": "KSampler" + }, + "widgets_values": [ + 0, + "randomize", + 20, + 1, + "euler", + "normal", + 1 + ] + } + ], + "groups": [ + { + "id": 1, + "title": "Load models", + "bounding": [ + -620, + -160, + 410, + 790 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 3, + "title": "Prompt", + "bounding": [ + -180, + -160, + 520, + 670 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 41, + "origin_id": 23, + "origin_slot": 0, + "target_id": 229, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 101, + "origin_id": 23, + "origin_slot": 0, + "target_id": 46, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 62, + "origin_id": 34, + "origin_slot": 0, + "target_id": 23, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 85, + "origin_id": 31, + "origin_slot": 0, + "target_id": 39, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 86, + "origin_id": 39, + "origin_slot": 0, + "target_id": 3, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 77, + "origin_id": 38, + "origin_slot": 0, + "target_id": 3, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 78, + "origin_id": 38, + "origin_slot": 1, + "target_id": 3, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 88, + "origin_id": 38, + "origin_slot": 2, + "target_id": 3, + "target_slot": 3, + "type": "LATENT" + }, + { + "id": 7, + "origin_id": 3, + "origin_slot": 0, + "target_id": 231, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 60, + "origin_id": 230, + "origin_slot": 0, + "target_id": 231, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 80, + "origin_id": 229, + "origin_slot": 0, + "target_id": 38, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 102, + "origin_id": 46, + "origin_slot": 0, + "target_id": 38, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 82, + "origin_id": 230, + "origin_slot": 0, + "target_id": 38, + "target_slot": 2, + "type": "VAE" + }, + { + "id": 99, + "origin_id": -10, + "origin_slot": 0, + "target_id": 38, + "target_slot": 3, + "type": "IMAGE" + }, + { + "id": 100, + "origin_id": -10, + "origin_slot": 1, + "target_id": 38, + "target_slot": 4, + "type": "MASK" + }, + { + "id": 95, + "origin_id": 231, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 277, + "origin_id": -10, + "origin_slot": 2, + "target_id": 23, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 282, + "origin_id": -10, + "origin_slot": 3, + "target_id": 3, + "target_slot": 4, + "type": "INT" + }, + { + "id": 283, + "origin_id": -10, + "origin_slot": 4, + "target_id": 31, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 284, + "origin_id": -10, + "origin_slot": 5, + "target_id": 34, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 285, + "origin_id": -10, + "origin_slot": 6, + "target_id": 34, + "target_slot": 1, + "type": "COMBO" + }, + { + "id": 286, + "origin_id": -10, + "origin_slot": 7, + "target_id": 230, + "target_slot": 0, + "type": "COMBO" + } + ], + "extra": { + "workflowRendererVersion": "LG" + }, + "category": "Image generation and editing/Inpaint image", + "description": "Inpaints masked image regions using Flux.1 fill [dev], Black Forest Labs' inpainting/outpainting model." + } + ] + }, + "extra": { + "ds": { + "scale": 0.8480949417360862, + "offset": [ + 833.9510730024642, + 210.32152847588895 + ] + }, + "ue_links": [] + } +} diff --git a/blueprints/Image Inpainting (Qwen-image).json b/blueprints/Image Inpainting (Qwen-image).json index d06f31dd2..a06d57e19 100644 --- a/blueprints/Image Inpainting (Qwen-image).json +++ b/blueprints/Image Inpainting (Qwen-image).json @@ -124,7 +124,7 @@ }, "revision": 0, "config": {}, - "name": "local-Image Inpainting (Qwen-image)", + "name": "Image Inpainting (Qwen-image)", "inputNode": { "id": -10, "bounding": [ @@ -1548,7 +1548,8 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image generation and editing/Inpaint image" + "category": "Image generation and editing/Inpaint image", + "description": "Inpaints masked regions using Qwen-Image, extending its multilingual text rendering to inpainting tasks." }, { "id": "56a1f603-fbd2-40ed-94ef-c9ecbd96aca8", @@ -1907,7 +1908,8 @@ ], "extra": { "workflowRendererVersion": "LG" - } + }, + "description": "Expands and softens mask edges to reduce visible seams after image processing." } ] }, @@ -1923,4 +1925,4 @@ "workflowRendererVersion": "LG" }, "version": 0.4 -} +} \ No newline at end of file diff --git a/blueprints/Image Levels.json b/blueprints/Image Levels.json index ef256a1aa..1a1b18932 100644 --- a/blueprints/Image Levels.json +++ b/blueprints/Image Levels.json @@ -742,9 +742,10 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image Tools/Color adjust" + "category": "Image Tools/Color adjust", + "description": "Adjusts black point, white point, and gamma for tonal range control via GPU shader." } ] }, "extra": {} -} +} \ No newline at end of file diff --git a/blueprints/Image Outpainting (Qwen-Image).json b/blueprints/Image Outpainting (Qwen-Image).json index bf2c4241a..6c07227c0 100644 --- a/blueprints/Image Outpainting (Qwen-Image).json +++ b/blueprints/Image Outpainting (Qwen-Image).json @@ -204,7 +204,7 @@ }, "revision": 0, "config": {}, - "name": "local-Image Outpainting (Qwen-Image)", + "name": "Image Outpainting (Qwen-Image)", "inputNode": { "id": -10, "bounding": [ @@ -1919,7 +1919,8 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image generation and editing/Outpaint image" + "category": "Image generation and editing/Outpaint image", + "description": "Outpaints beyond image boundaries using Qwen-Image's outpainting capabilities." }, { "id": "f93c215e-c393-460e-9534-ed2c3d8a652e", @@ -2278,7 +2279,8 @@ ], "extra": { "workflowRendererVersion": "LG" - } + }, + "description": "Expands and softens mask edges to reduce visible seams after image processing." }, { "id": "2a4b2cc0-db37-4302-a067-da392f38f06b", @@ -2733,7 +2735,8 @@ ], "extra": { "workflowRendererVersion": "LG" - } + }, + "description": "Scales both image and mask together while preserving alignment for editing workflows." } ] }, @@ -2749,4 +2752,4 @@ } }, "version": 0.4 -} +} \ No newline at end of file diff --git a/blueprints/Image Segmentation (SAM3).json b/blueprints/Image Segmentation (SAM3).json new file mode 100644 index 000000000..b405bf623 --- /dev/null +++ b/blueprints/Image Segmentation (SAM3).json @@ -0,0 +1,714 @@ +{ + "revision": 0, + "last_node_id": 99, + "last_link_id": 0, + "nodes": [ + { + "id": 99, + "type": "6e7ab3ea-96aa-470f-9b94-3d9d0e01f481", + "pos": [ + -1630, + -3270 + ], + "size": [ + 290, + 370 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "label": "image", + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": null + }, + { + "label": "object", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + }, + { + "name": "bboxes", + "type": "BOUNDING_BOX", + "link": null + }, + { + "name": "positive_coords", + "type": "STRING", + "link": null + }, + { + "name": "negative_coords", + "type": "STRING", + "link": null + }, + { + "name": "threshold", + "type": "FLOAT", + "widget": { + "name": "threshold" + }, + "link": null + }, + { + "name": "refine_iterations", + "type": "INT", + "widget": { + "name": "refine_iterations" + }, + "link": null + }, + { + "name": "individual_masks", + "type": "BOOLEAN", + "widget": { + "name": "individual_masks" + }, + "link": null + }, + { + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "masks", + "name": "masks", + "type": "MASK", + "links": [] + }, + { + "localized_name": "bboxes", + "name": "bboxes", + "type": "BOUNDING_BOX", + "links": [] + } + ], + "properties": { + "proxyWidgets": [ + [ + "78", + "text" + ], + [ + "75", + "threshold" + ], + [ + "75", + "refine_iterations" + ], + [ + "75", + "individual_masks" + ], + [ + "77", + "ckpt_name" + ] + ], + "ue_properties": { + "widget_ue_connectable": { + "text": true + }, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.19.3", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [], + "title": "Image Segmentation (SAM3)" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "6e7ab3ea-96aa-470f-9b94-3d9d0e01f481", + "version": 1, + "state": { + "lastGroupId": 0, + "lastNodeId": 113, + "lastLinkId": 283, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Image Segmentation (SAM3)", + "inputNode": { + "id": -10, + "bounding": [ + -2260, + -3450, + 136.369140625, + 220 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + -1130, + -3305, + 120, + 80 + ] + }, + "inputs": [ + { + "id": "a6e75fa2-162a-4af0-a2fd-1e9c899a5ab6", + "name": "image", + "type": "IMAGE", + "linkIds": [ + 264 + ], + "localized_name": "image", + "label": "image", + "pos": [ + -2143.630859375, + -3430 + ] + }, + { + "id": "3cefd304-7631-4ff6-a5a0-5a0ffb120745", + "name": "text", + "type": "STRING", + "linkIds": [ + 265 + ], + "label": "object", + "pos": [ + -2143.630859375, + -3410 + ] + }, + { + "id": "1aec91c5-d8d2-441c-928c-49c14e7e80ed", + "name": "bboxes", + "type": "BOUNDING_BOX", + "linkIds": [ + 266 + ], + "pos": [ + -2143.630859375, + -3390 + ] + }, + { + "id": "1ec7ce1a-8257-4719-8a81-60ebc8a98899", + "name": "positive_coords", + "type": "STRING", + "linkIds": [ + 267 + ], + "pos": [ + -2143.630859375, + -3370 + ] + }, + { + "id": "c65f8b87-9bd7-48be-9fc2-823431e95019", + "name": "negative_coords", + "type": "STRING", + "linkIds": [ + 268 + ], + "pos": [ + -2143.630859375, + -3350 + ] + }, + { + "id": "bb4ba35a-ccfe-4c37-98e5-d9b0d69585fb", + "name": "threshold", + "type": "FLOAT", + "linkIds": [ + 269 + ], + "pos": [ + -2143.630859375, + -3330 + ] + }, + { + "id": "b1439668-b050-490b-a5dc-fc4052c55666", + "name": "refine_iterations", + "type": "INT", + "linkIds": [ + 270 + ], + "pos": [ + -2143.630859375, + -3310 + ] + }, + { + "id": "86e239e5-c098-4302-b54d-d42a38bc0f89", + "name": "individual_masks", + "type": "BOOLEAN", + "linkIds": [ + 271 + ], + "pos": [ + -2143.630859375, + -3290 + ] + }, + { + "id": "f9e0b9d4-b2f1-4907-a4a5-305656576706", + "name": "ckpt_name", + "type": "COMBO", + "linkIds": [ + 272 + ], + "pos": [ + -2143.630859375, + -3270 + ] + } + ], + "outputs": [ + { + "id": "ff50da09-1e59-4a58-9b7f-be1a00aa5913", + "name": "masks", + "type": "MASK", + "linkIds": [ + 231 + ], + "localized_name": "masks", + "pos": [ + -1110, + -3285 + ] + }, + { + "id": "8f622e40-8528-4078-b7d3-147e9f872194", + "name": "bboxes", + "type": "BOUNDING_BOX", + "linkIds": [ + 232 + ], + "localized_name": "bboxes", + "pos": [ + -1110, + -3265 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 75, + "type": "SAM3_Detect", + "pos": [ + -1470, + -3460 + ], + "size": [ + 270, + 260 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "label": "model", + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 237 + }, + { + "label": "image", + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 264 + }, + { + "label": "conditioning", + "localized_name": "conditioning", + "name": "conditioning", + "shape": 7, + "type": "CONDITIONING", + "link": 200 + }, + { + "label": "bboxes", + "localized_name": "bboxes", + "name": "bboxes", + "shape": 7, + "type": "BOUNDING_BOX", + "link": 266 + }, + { + "label": "positive_coords", + "localized_name": "positive_coords", + "name": "positive_coords", + "shape": 7, + "type": "STRING", + "link": 267 + }, + { + "label": "negative_coords", + "localized_name": "negative_coords", + "name": "negative_coords", + "shape": 7, + "type": "STRING", + "link": 268 + }, + { + "localized_name": "threshold", + "name": "threshold", + "type": "FLOAT", + "widget": { + "name": "threshold" + }, + "link": 269 + }, + { + "localized_name": "refine_iterations", + "name": "refine_iterations", + "type": "INT", + "widget": { + "name": "refine_iterations" + }, + "link": 270 + }, + { + "localized_name": "individual_masks", + "name": "individual_masks", + "type": "BOOLEAN", + "widget": { + "name": "individual_masks" + }, + "link": 271 + } + ], + "outputs": [ + { + "localized_name": "masks", + "name": "masks", + "type": "MASK", + "links": [ + 231 + ] + }, + { + "localized_name": "bboxes", + "name": "bboxes", + "type": "BOUNDING_BOX", + "links": [ + 232 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.19.3", + "Node name for S&R": "SAM3_Detect", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 0.5, + 2, + false + ] + }, + { + "id": 77, + "type": "CheckpointLoaderSimple", + "pos": [ + -1970, + -3200 + ], + "size": [ + 330, + 140 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "ckpt_name", + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": 272 + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 237 + ] + }, + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [ + 240 + ] + }, + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "links": null + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.19.3", + "Node name for S&R": "CheckpointLoaderSimple", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "sam3.1_multiplex_fp16.safetensors", + "url": "https://huggingface.co/Comfy-Org/sam3.1/resolve/main/checkpoints/sam3.1_multiplex_fp16.safetensors", + "directory": "checkpoints" + } + ] + }, + "widgets_values": [ + "sam3.1_multiplex_fp16.safetensors" + ] + }, + { + "id": 78, + "type": "CLIPTextEncode", + "pos": [ + -2000, + -3000 + ], + "size": [ + 400, + 200 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 240 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 265 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 200 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.19.3", + "Node name for S&R": "CLIPTextEncode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "" + ] + } + ], + "groups": [], + "links": [ + { + "id": 237, + "origin_id": 77, + "origin_slot": 0, + "target_id": 75, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 200, + "origin_id": 78, + "origin_slot": 0, + "target_id": 75, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 240, + "origin_id": 77, + "origin_slot": 1, + "target_id": 78, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 231, + "origin_id": 75, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "MASK" + }, + { + "id": 232, + "origin_id": 75, + "origin_slot": 1, + "target_id": -20, + "target_slot": 1, + "type": "BOUNDING_BOX" + }, + { + "id": 264, + "origin_id": -10, + "origin_slot": 0, + "target_id": 75, + "target_slot": 1, + "type": "IMAGE" + }, + { + "id": 265, + "origin_id": -10, + "origin_slot": 1, + "target_id": 78, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 266, + "origin_id": -10, + "origin_slot": 2, + "target_id": 75, + "target_slot": 3, + "type": "BOUNDING_BOX" + }, + { + "id": 267, + "origin_id": -10, + "origin_slot": 3, + "target_id": 75, + "target_slot": 4, + "type": "STRING" + }, + { + "id": 268, + "origin_id": -10, + "origin_slot": 4, + "target_id": 75, + "target_slot": 5, + "type": "STRING" + }, + { + "id": 269, + "origin_id": -10, + "origin_slot": 5, + "target_id": 75, + "target_slot": 6, + "type": "FLOAT" + }, + { + "id": 270, + "origin_id": -10, + "origin_slot": 6, + "target_id": 75, + "target_slot": 7, + "type": "INT" + }, + { + "id": 271, + "origin_id": -10, + "origin_slot": 7, + "target_id": 75, + "target_slot": 8, + "type": "BOOLEAN" + }, + { + "id": 272, + "origin_id": -10, + "origin_slot": 8, + "target_id": 77, + "target_slot": 0, + "type": "COMBO" + } + ], + "extra": {}, + "category": "Image Tools/Image Segmentation", + "description": "Segments images into masks using Meta SAM3 from text prompts, points, or boxes." + } + ] + }, + "extra": { + "ue_links": [] + } +} diff --git a/blueprints/Image Upscale(Z-image-Turbo).json b/blueprints/Image Upscale(Z-image-Turbo).json index 0d2b6e240..bd803a0b1 100644 --- a/blueprints/Image Upscale(Z-image-Turbo).json +++ b/blueprints/Image Upscale(Z-image-Turbo).json @@ -141,7 +141,7 @@ }, "revision": 0, "config": {}, - "name": "local-Image Upscale(Z-image-Turbo)", + "name": "Image Upscale (Z-image-Turbo)", "inputNode": { "id": -10, "bounding": [ @@ -1302,7 +1302,8 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image generation and editing/Enhance" + "category": "Image generation and editing/Enhance", + "description": "Upscales images to higher resolution using Z-Image-Turbo." } ] }, diff --git a/blueprints/Image to Depth Map (Lotus).json b/blueprints/Image to Depth Map (Lotus).json index 089f2cd42..12f10ba5b 100644 --- a/blueprints/Image to Depth Map (Lotus).json +++ b/blueprints/Image to Depth Map (Lotus).json @@ -99,7 +99,7 @@ }, "revision": 0, "config": {}, - "name": "local-Image to Depth Map (Lotus)", + "name": "Image to Depth Map (Lotus)", "inputNode": { "id": -10, "bounding": [ @@ -948,7 +948,8 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image generation and editing/Depth to image" + "category": "Image generation and editing/Depth to image", + "description": "Estimates a monocular depth map from an input image using the Lotus depth estimation model." } ] }, @@ -964,4 +965,4 @@ "workflowRendererVersion": "LG" }, "version": 0.4 -} +} \ No newline at end of file diff --git a/blueprints/Image to Layers(Qwen-Image Layered).json b/blueprints/Image to Layers(Qwen-Image-Layered).json similarity index 82% rename from blueprints/Image to Layers(Qwen-Image Layered).json rename to blueprints/Image to Layers(Qwen-Image-Layered).json index 164ffbd8d..7b44f0563 100644 --- a/blueprints/Image to Layers(Qwen-Image Layered).json +++ b/blueprints/Image to Layers(Qwen-Image-Layered).json @@ -1,15 +1,14 @@ { - "id": "1a761372-7c82-4016-b9bf-fa285967e1e9", "revision": 0, - "last_node_id": 83, + "last_node_id": 176, "last_link_id": 0, "nodes": [ { - "id": 83, - "type": "f754a936-daaf-4b6e-9658-41fdc54d301d", + "id": 176, + "type": "2d2e3c8e-53b3-4618-be52-6d1d99382f0e", "pos": [ - 61.999827823554256, - 153.3332507624185 + -1150, + 200 ], "size": [ 400, @@ -56,6 +55,38 @@ "name": "layers" }, "link": null + }, + { + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": null + }, + { + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": null + }, + { + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": null + }, + { + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": null } ], "outputs": [ @@ -66,28 +97,41 @@ "links": [] } ], + "title": "Image to Layers (Qwen-Image-Layered)", "properties": { "proxyWidgets": [ [ - "-1", + "6", "text" ], [ - "-1", + "3", "steps" ], [ - "-1", + "3", "cfg" ], [ - "-1", + "83", "layers" ], [ "3", "seed" ], + [ + "37", + "unet_name" + ], + [ + "38", + "clip_name" + ], + [ + "39", + "vae_name" + ], [ "3", "control_after_generate" @@ -95,6 +139,11 @@ ], "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -103,25 +152,20 @@ "secondTabOffset": 80, "secondTabWidth": 65 }, - "widgets_values": [ - "", - 20, - 2.5, - 2 - ] + "widgets_values": [] } ], "links": [], - "groups": [], + "version": 0.4, "definitions": { "subgraphs": [ { - "id": "f754a936-daaf-4b6e-9658-41fdc54d301d", + "id": "2d2e3c8e-53b3-4618-be52-6d1d99382f0e", "version": 1, "state": { - "lastGroupId": 3, - "lastNodeId": 83, - "lastLinkId": 159, + "lastGroupId": 8, + "lastNodeId": 176, + "lastLinkId": 380, "lastRerouteId": 0 }, "revision": 0, @@ -130,10 +174,10 @@ "inputNode": { "id": -10, "bounding": [ - -510, - 523, + -720, + 720, 120, - 140 + 220 ] }, "outputNode": { @@ -156,8 +200,8 @@ ], "localized_name": "image", "pos": [ - -410, - 543 + -620, + 740 ] }, { @@ -168,8 +212,8 @@ 150 ], "pos": [ - -410, - 563 + -620, + 760 ] }, { @@ -180,8 +224,8 @@ 153 ], "pos": [ - -410, - 583 + -620, + 780 ] }, { @@ -192,8 +236,8 @@ 154 ], "pos": [ - -410, - 603 + -620, + 800 ] }, { @@ -204,8 +248,56 @@ 159 ], "pos": [ - -410, - 623 + -620, + 820 + ] + }, + { + "id": "9f76338b-f4ca-4bb3-b61a-57b3f233061e", + "name": "seed", + "type": "INT", + "linkIds": [ + 377 + ], + "pos": [ + -620, + 840 + ] + }, + { + "id": "8d0422d5-5eee-4f7e-9817-dc613cc62eca", + "name": "unet_name", + "type": "COMBO", + "linkIds": [ + 378 + ], + "pos": [ + -620, + 860 + ] + }, + { + "id": "552eece2-a735-4d00-ae78-ded454622bc1", + "name": "clip_name", + "type": "COMBO", + "linkIds": [ + 379 + ], + "pos": [ + -620, + 880 + ] + }, + { + "id": "1e6d141c-d0f9-4a2b-895c-b6780e57cfa0", + "name": "vae_name", + "type": "COMBO", + "linkIds": [ + 380 + ], + "pos": [ + -620, + 900 ] } ], @@ -231,14 +323,14 @@ "type": "CLIPLoader", "pos": [ -320, - 310 + 360 ], "size": [ - 346.7470703125, - 106 + 350, + 150 ], "flags": {}, - "order": 0, + "order": 5, "mode": 0, "inputs": [ { @@ -248,7 +340,7 @@ "widget": { "name": "clip_name" }, - "link": null + "link": 379 }, { "localized_name": "type", @@ -283,9 +375,14 @@ } ], "properties": { - "Node name for S&R": "CLIPLoader", "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "Node name for S&R": "CLIPLoader", "models": [ { "name": "qwen_2.5_vl_7b_fp8_scaled.safetensors", @@ -312,14 +409,14 @@ "type": "VAELoader", "pos": [ -320, - 460 + 580 ], "size": [ - 346.7470703125, - 58 + 350, + 110 ], "flags": {}, - "order": 1, + "order": 6, "mode": 0, "inputs": [ { @@ -329,7 +426,7 @@ "widget": { "name": "vae_name" }, - "link": null + "link": 380 } ], "outputs": [ @@ -345,9 +442,14 @@ } ], "properties": { - "Node name for S&R": "VAELoader", "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "Node name for S&R": "VAELoader", "models": [ { "name": "qwen_image_layered_vae.safetensors", @@ -375,11 +477,11 @@ 420 ], "size": [ - 425.27801513671875, - 180.6060791015625 + 430, + 190 ], "flags": {}, - "order": 3, + "order": 2, "mode": 0, "inputs": [ { @@ -411,9 +513,14 @@ ], "title": "CLIP Text Encode (Negative Prompt)", "properties": { - "Node name for S&R": "CLIPTextEncode", "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "Node name for S&R": "CLIPTextEncode", "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -432,12 +539,12 @@ "id": 70, "type": "ReferenceLatent", "pos": [ - 330, - 670 + 140, + 700 ], "size": [ - 204.1666717529297, - 46 + 210, + 50 ], "flags": { "collapsed": true @@ -470,9 +577,14 @@ } ], "properties": { - "Node name for S&R": "ReferenceLatent", "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "Node name for S&R": "ReferenceLatent", "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -480,19 +592,18 @@ "secondTabText": "Send Back", "secondTabOffset": 80, "secondTabWidth": 65 - }, - "widgets_values": [] + } }, { "id": 69, "type": "ReferenceLatent", "pos": [ - 330, - 710 + 160, + 820 ], "size": [ - 204.1666717529297, - 46 + 210, + 50 ], "flags": { "collapsed": true @@ -525,9 +636,14 @@ } ], "properties": { - "Node name for S&R": "ReferenceLatent", "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "Node name for S&R": "ReferenceLatent", "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -535,8 +651,7 @@ "secondTabText": "Send Back", "secondTabOffset": 80, "secondTabWidth": 65 - }, - "widgets_values": [] + } }, { "id": 66, @@ -547,10 +662,10 @@ ], "size": [ 270, - 58 + 110 ], "flags": {}, - "order": 4, + "order": 7, "mode": 0, "inputs": [ { @@ -580,9 +695,14 @@ } ], "properties": { - "Node name for S&R": "ModelSamplingAuraFlow", "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "Node name for S&R": "ModelSamplingAuraFlow", "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -600,11 +720,11 @@ "type": "LatentCutToBatch", "pos": [ 830, - 160 + 140 ], "size": [ 270, - 82 + 140 ], "flags": {}, "order": 11, @@ -646,9 +766,14 @@ } ], "properties": { - "Node name for S&R": "LatentCutToBatch", "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "Node name for S&R": "LatentCutToBatch", "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -666,12 +791,12 @@ "id": 71, "type": "VAEEncode", "pos": [ - 100, - 690 + -280, + 780 ], "size": [ - 140, - 46 + 230, + 100 ], "flags": { "collapsed": false @@ -704,9 +829,14 @@ } ], "properties": { - "Node name for S&R": "VAEEncode", "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "Node name for S&R": "VAEEncode", "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -714,24 +844,23 @@ "secondTabText": "Send Back", "secondTabOffset": 80, "secondTabWidth": 65 - }, - "widgets_values": [] + } }, { "id": 8, "type": "VAEDecode", "pos": [ 850, - 310 + 370 ], "size": [ 210, - 46 + 50 ], "flags": { "collapsed": true }, - "order": 7, + "order": 3, "mode": 0, "inputs": [ { @@ -759,9 +888,14 @@ } ], "properties": { - "Node name for S&R": "VAEDecode", "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "Node name for S&R": "VAEDecode", "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -769,8 +903,7 @@ "secondTabText": "Send Back", "secondTabOffset": 80, "secondTabWidth": 65 - }, - "widgets_values": [] + } }, { "id": 6, @@ -780,11 +913,11 @@ 180 ], "size": [ - 422.84503173828125, - 164.31304931640625 + 430, + 170 ], "flags": {}, - "order": 6, + "order": 1, "mode": 0, "inputs": [ { @@ -816,9 +949,14 @@ ], "title": "CLIP Text Encode (Positive Prompt)", "properties": { - "Node name for S&R": "CLIPTextEncode", "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "Node name for S&R": "CLIPTextEncode", "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -838,14 +976,14 @@ "type": "KSampler", "pos": [ 530, - 280 + 340 ], "size": [ 270, 400 ], "flags": {}, - "order": 5, + "order": 0, "mode": 0, "inputs": [ { @@ -879,7 +1017,7 @@ "widget": { "name": "seed" }, - "link": null + "link": 377 }, { "localized_name": "steps", @@ -939,9 +1077,14 @@ } ], "properties": { - "Node name for S&R": "KSampler", "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "Node name for S&R": "KSampler", "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -964,12 +1107,12 @@ "id": 78, "type": "GetImageSize", "pos": [ - 80, - 790 + -280, + 930 ], "size": [ - 210, - 136 + 230, + 140 ], "flags": {}, "order": 12, @@ -1007,9 +1150,14 @@ } ], "properties": { - "Node name for S&R": "GetImageSize", "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "Node name for S&R": "GetImageSize", "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -1017,23 +1165,23 @@ "secondTabText": "Send Back", "secondTabOffset": 80, "secondTabWidth": 65 - }, - "widgets_values": [] + } }, { "id": 83, "type": "EmptyQwenImageLayeredLatentImage", "pos": [ - 320, - 790 + -280, + 1120 ], "size": [ - 330.9341796875, - 130 + 340, + 200 ], "flags": {}, "order": 13, "mode": 0, + "showAdvanced": true, "inputs": [ { "localized_name": "width", @@ -1083,9 +1231,14 @@ } ], "properties": { - "Node name for S&R": "EmptyQwenImageLayeredLatentImage", "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "Node name for S&R": "EmptyQwenImageLayeredLatentImage", "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -1109,11 +1262,11 @@ 180 ], "size": [ - 346.7470703125, - 82 + 350, + 110 ], "flags": {}, - "order": 2, + "order": 4, "mode": 0, "inputs": [ { @@ -1123,7 +1276,7 @@ "widget": { "name": "unet_name" }, - "link": null + "link": 378 }, { "localized_name": "weight_dtype", @@ -1147,9 +1300,14 @@ } ], "properties": { - "Node name for S&R": "UNETLoader", "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {}, + "version": "7.7" + }, + "Node name for S&R": "UNETLoader", "models": [ { "name": "qwen_image_layered_bf16.safetensors", @@ -1191,8 +1349,8 @@ "bounding": [ -330, 110, - 366.7470703125, - 421.6 + 370, + 610 ], "color": "#3f789e", "font_size": 24, @@ -1391,16 +1549,48 @@ "target_id": 83, "target_slot": 2, "type": "INT" + }, + { + "id": 377, + "origin_id": -10, + "origin_slot": 5, + "target_id": 3, + "target_slot": 4, + "type": "INT" + }, + { + "id": 378, + "origin_id": -10, + "origin_slot": 6, + "target_id": 37, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 379, + "origin_id": -10, + "origin_slot": 7, + "target_id": 38, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 380, + "origin_id": -10, + "origin_slot": 8, + "target_id": 39, + "target_slot": 0, + "type": "COMBO" } ], "extra": { "workflowRendererVersion": "LG" }, - "category": "Image generation and editing/Image to layers" + "category": "Image generation and editing/Image to layers", + "description": "Decomposes an image into variable-resolution RGBA layers for independent editing using Qwen-Image-Layered." } ] }, - "config": {}, "extra": { "ds": { "scale": 1.14, @@ -1409,7 +1599,6 @@ 6.855893974423647 ] }, - "workflowRendererVersion": "LG" - }, - "version": 0.4 -} + "ue_links": [] + } +} \ No newline at end of file diff --git a/blueprints/Image to Model (Hunyuan3d 2.1).json b/blueprints/Image to Model (Hunyuan3d 2.1).json index 4705603a8..ee5552656 100644 --- a/blueprints/Image to Model (Hunyuan3d 2.1).json +++ b/blueprints/Image to Model (Hunyuan3d 2.1).json @@ -72,7 +72,7 @@ }, "revision": 0, "config": {}, - "name": "local-Image to Model (Hunyuan3d 2.1)", + "name": "Image to 3D Model (Hunyuan3d 2.1)", "inputNode": { "id": -10, "bounding": [ @@ -765,7 +765,8 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "3D/Image to 3D Model" + "category": "3D/Image to 3D Model", + "description": "Generates 3D mesh models from a single input image using Hunyuan3D 2.0/2.1." } ] }, diff --git a/blueprints/Image to Video (LTX-2.3).json b/blueprints/Image to Video (LTX-2.3).json new file mode 100644 index 000000000..3db524ea0 --- /dev/null +++ b/blueprints/Image to Video (LTX-2.3).json @@ -0,0 +1,4234 @@ +{ + "revision": 0, + "last_node_id": 320, + "last_link_id": 0, + "nodes": [ + { + "id": 320, + "type": "2454ad83-157c-40dd-9f19-5daaf4041ce0", + "pos": [ + 30, + 4150 + ], + "size": [ + 390, + 466.625 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "label": "first_frame", + "localized_name": "input", + "name": "input", + "type": "IMAGE,MASK", + "link": null + }, + { + "name": "value", + "type": "STRING", + "widget": { + "name": "value" + }, + "link": null + }, + { + "label": "width", + "name": "value_2", + "type": "INT", + "widget": { + "name": "value_2" + }, + "link": null + }, + { + "label": "height", + "name": "value_3", + "type": "INT", + "widget": { + "name": "value_3" + }, + "link": null + }, + { + "label": "duration", + "name": "value_4", + "type": "INT", + "widget": { + "name": "value_4" + }, + "link": null + }, + { + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": null + }, + { + "label": "distilled_lora", + "name": "lora_name", + "type": "COMBO", + "widget": { + "name": "lora_name" + }, + "link": null + }, + { + "name": "text_encoder", + "type": "COMBO", + "widget": { + "name": "text_encoder" + }, + "link": null + }, + { + "label": "latent_upscale_model", + "name": "model_name", + "type": "COMBO", + "widget": { + "name": "model_name" + }, + "link": null + }, + { + "label": "fps", + "name": "value_5", + "type": "INT", + "widget": { + "name": "value_5" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "VIDEO", + "name": "VIDEO", + "type": "VIDEO", + "links": [] + } + ], + "title": "Image to Video (LTX-2.3)", + "properties": { + "proxyWidgets": [ + [ + "319", + "value" + ], + [ + "312", + "value" + ], + [ + "299", + "value" + ], + [ + "301", + "value" + ], + [ + "300", + "value" + ], + [ + "316", + "ckpt_name" + ], + [ + "277", + "control_after_generate" + ], + [ + "277", + "noise_seed" + ], + [ + "285", + "lora_name" + ], + [ + "317", + "text_encoder" + ], + [ + "311", + "model_name" + ] + ], + "ue_properties": { + "widget_ue_connectable": { + "value_1": true, + "value_2": true, + "value_3": true, + "value_4": true, + "lora_name": true, + "model_name": true, + "value_5": true + }, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.3", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [] + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "2454ad83-157c-40dd-9f19-5daaf4041ce0", + "version": 1, + "state": { + "lastGroupId": 25, + "lastNodeId": 323, + "lastLinkId": 631, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Image to Video (LTX-2.3)", + "inputNode": { + "id": -10, + "bounding": [ + 730, + 4110, + 162.162109375, + 240 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 6590, + 4360, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "7afd6ea8-c738-4fd9-97b8-66fa905cd381", + "name": "input", + "type": "IMAGE,MASK", + "linkIds": [ + 535 + ], + "localized_name": "input", + "label": "first_frame", + "pos": [ + 872.162109375, + 4130 + ] + }, + { + "id": "9494c550-4172-49c6-930e-5b508f775e77", + "name": "value", + "type": "STRING", + "linkIds": [ + 595 + ], + "pos": [ + 872.162109375, + 4150 + ] + }, + { + "id": "58dbb3f6-f924-4548-96ef-e0e34610bd4e", + "name": "value_2", + "type": "INT", + "linkIds": [ + 597 + ], + "label": "width", + "pos": [ + 872.162109375, + 4170 + ] + }, + { + "id": "6086d5b8-2586-448c-a641-dd14d76dd102", + "name": "value_3", + "type": "INT", + "linkIds": [ + 598 + ], + "label": "height", + "pos": [ + 872.162109375, + 4190 + ] + }, + { + "id": "feb8c2eb-ae48-4fa8-bc24-929552d656c3", + "name": "value_4", + "type": "INT", + "linkIds": [ + 599 + ], + "label": "duration", + "pos": [ + 872.162109375, + 4210 + ] + }, + { + "id": "d7255058-319a-4880-8f9a-7e542c8f3c3c", + "name": "ckpt_name", + "type": "COMBO", + "linkIds": [ + 601, + 604, + 605 + ], + "pos": [ + 872.162109375, + 4230 + ] + }, + { + "id": "4afce68d-8f65-4342-9d6d-ae0a7688c3e3", + "name": "lora_name", + "type": "COMBO", + "linkIds": [ + 602 + ], + "label": "distilled_lora", + "pos": [ + 872.162109375, + 4250 + ] + }, + { + "id": "ab842b4b-c977-4679-b421-424722785b57", + "name": "text_encoder", + "type": "COMBO", + "linkIds": [ + 606 + ], + "pos": [ + 872.162109375, + 4270 + ] + }, + { + "id": "9e47372d-28d9-4311-91e9-e90d03f4eb43", + "name": "model_name", + "type": "COMBO", + "linkIds": [ + 607 + ], + "label": "latent_upscale_model", + "pos": [ + 872.162109375, + 4290 + ] + }, + { + "id": "3e32ce15-0ae7-4cd0-909f-a354e8e9c4c9", + "name": "value_5", + "type": "INT", + "linkIds": [ + 624 + ], + "label": "fps", + "pos": [ + 872.162109375, + 4310 + ] + } + ], + "outputs": [ + { + "id": "954ef307-c897-4eea-8b5c-5c6ce15a5357", + "name": "VIDEO", + "type": "VIDEO", + "linkIds": [ + 536 + ], + "localized_name": "VIDEO", + "pos": [ + 6610, + 4380 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 276, + "type": "RandomNoise", + "pos": [ + 4700, + 3650 + ], + "size": [ + 280, + 110 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "noise_seed", + "name": "noise_seed", + "type": "INT", + "widget": { + "name": "noise_seed" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "NOISE", + "name": "NOISE", + "type": "NOISE", + "links": [ + 490 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.75", + "Node name for S&R": "RandomNoise", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 42, + "fixed" + ] + }, + { + "id": 277, + "type": "RandomNoise", + "pos": [ + 3160, + 3630 + ], + "size": [ + 280, + 110 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "noise_seed", + "name": "noise_seed", + "type": "INT", + "widget": { + "name": "noise_seed" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "NOISE", + "name": "NOISE", + "type": "NOISE", + "links": [ + 483 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.56", + "Node name for S&R": "RandomNoise", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 519681071352364, + "randomize" + ] + }, + { + "id": 278, + "type": "LTXVConcatAVLatent", + "pos": [ + 4710, + 4490 + ], + "size": [ + 280, + 100 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "video_latent", + "name": "video_latent", + "type": "LATENT", + "link": 512 + }, + { + "localized_name": "audio_latent", + "name": "audio_latent", + "type": "LATENT", + "link": 513 + } + ], + "outputs": [ + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "links": [ + 494 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.5.1", + "Node name for S&R": "LTXVConcatAVLatent", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 279, + "type": "LTXVAudioVAELoader", + "pos": [ + 1660, + 4100 + ], + "size": [ + 430, + 110 + ], + "flags": {}, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "ckpt_name", + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": 604 + } + ], + "outputs": [ + { + "localized_name": "Audio VAE", + "name": "Audio VAE", + "type": "VAE", + "links": [ + 481, + 496 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.68", + "Node name for S&R": "LTXVAudioVAELoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "ltx-2.3-22b-dev-fp8.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2.3-fp8/resolve/main/ltx-2.3-22b-dev-fp8.safetensors", + "directory": "checkpoints" + } + ] + }, + "widgets_values": [ + "ltx-2.3-22b-dev-fp8.safetensors" + ] + }, + { + "id": 280, + "type": "KSamplerSelect", + "pos": [ + 4700, + 4100 + ], + "size": [ + 280, + 110 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "SAMPLER", + "name": "SAMPLER", + "type": "SAMPLER", + "links": [ + 492 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.75", + "Node name for S&R": "KSamplerSelect", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "euler_cfg_pp" + ] + }, + { + "id": 281, + "type": "ManualSigmas", + "pos": [ + 4700, + 4290 + ], + "size": [ + 280, + 110 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "sigmas", + "name": "sigmas", + "type": "STRING", + "widget": { + "name": "sigmas" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "SIGMAS", + "name": "SIGMAS", + "type": "SIGMAS", + "links": [ + 493 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.5.1", + "Node name for S&R": "ManualSigmas", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "0.85, 0.7250, 0.4219, 0.0" + ] + }, + { + "id": 282, + "type": "CFGGuider", + "pos": [ + 4700, + 3850 + ], + "size": [ + 280, + 160 + ], + "flags": {}, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 478 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 479 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 480 + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "GUIDER", + "name": "GUIDER", + "type": "GUIDER", + "links": [ + 491 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.71", + "Node name for S&R": "CFGGuider", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1 + ] + }, + { + "id": 283, + "type": "SamplerCustomAdvanced", + "pos": [ + 3550, + 3630 + ], + "size": [ + 230, + 170 + ], + "flags": {}, + "order": 10, + "mode": 0, + "inputs": [ + { + "localized_name": "noise", + "name": "noise", + "type": "NOISE", + "link": 483 + }, + { + "localized_name": "guider", + "name": "guider", + "type": "GUIDER", + "link": 484 + }, + { + "localized_name": "sampler", + "name": "sampler", + "type": "SAMPLER", + "link": 485 + }, + { + "localized_name": "sigmas", + "name": "sigmas", + "type": "SIGMAS", + "link": 544 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 487 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "LATENT", + "links": [ + 488 + ] + }, + { + "localized_name": "denoised_output", + "name": "denoised_output", + "type": "LATENT", + "links": [] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.60", + "Node name for S&R": "SamplerCustomAdvanced", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 284, + "type": "LTXVCropGuides", + "pos": [ + 3830, + 3810 + ], + "size": [ + 250, + 120 + ], + "flags": {}, + "order": 11, + "mode": 0, + "inputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 475 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 476 + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "link": 477 + } + ], + "outputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "links": [ + 479 + ] + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "links": [ + 480 + ] + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "slot_index": 2, + "links": [] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.68", + "Node name for S&R": "LTXVCropGuides", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 285, + "type": "LoraLoaderModelOnly", + "pos": [ + 1660, + 3890 + ], + "size": [ + 430, + 140 + ], + "flags": {}, + "order": 12, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 520 + }, + { + "localized_name": "lora_name", + "name": "lora_name", + "type": "COMBO", + "widget": { + "name": "lora_name" + }, + "link": 602 + }, + { + "localized_name": "strength_model", + "name": "strength_model", + "type": "FLOAT", + "widget": { + "name": "strength_model" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 478, + 541 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.75", + "Node name for S&R": "LoraLoaderModelOnly", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "ltx-2.3-22b-distilled-lora-384.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2.3/resolve/main/ltx-2.3-22b-distilled-lora-384.safetensors", + "directory": "loras" + } + ] + }, + "widgets_values": [ + "ltx-2.3-22b-distilled-lora-384.safetensors", + 0.5 + ] + }, + { + "id": 286, + "type": "ResizeImagesByLongerEdge", + "pos": [ + 2070, + 4810 + ], + "size": [ + 310, + 110 + ], + "flags": { + "collapsed": false + }, + "order": 13, + "mode": 0, + "inputs": [ + { + "localized_name": "images", + "name": "images", + "type": "IMAGE", + "link": 523 + }, + { + "localized_name": "longer_edge", + "name": "longer_edge", + "type": "INT", + "widget": { + "name": "longer_edge" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "images", + "name": "images", + "type": "IMAGE", + "links": [ + 505 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.5.1", + "Node name for S&R": "ResizeImagesByLongerEdge", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1536 + ] + }, + { + "id": 287, + "type": "LTXVLatentUpsampler", + "pos": [ + 4250, + 3760 + ], + "size": [ + 330, + 120 + ], + "flags": {}, + "order": 14, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 547 + }, + { + "localized_name": "upscale_model", + "name": "upscale_model", + "type": "LATENT_UPSCALE_MODEL", + "link": 545 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 554 + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "links": [ + 548 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.14.1", + "Node name for S&R": "LTXVLatentUpsampler", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 288, + "type": "LTXVImgToVideoInplace", + "pos": [ + 4230, + 4100 + ], + "size": [ + 300, + 180 + ], + "flags": {}, + "order": 15, + "mode": 0, + "inputs": [ + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 552 + }, + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 515 + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "link": 548 + }, + { + "localized_name": "strength", + "name": "strength", + "type": "FLOAT", + "widget": { + "name": "strength" + }, + "link": null + }, + { + "localized_name": "bypass", + "name": "bypass", + "type": "BOOLEAN", + "widget": { + "name": "bypass" + }, + "link": 543 + } + ], + "outputs": [ + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "links": [ + 512 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "LTXVImgToVideoInplace", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1, + false + ] + }, + { + "id": 289, + "type": "LTXVPreprocess", + "pos": [ + 2100, + 5010 + ], + "size": [ + 290, + 110 + ], + "flags": {}, + "order": 16, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 505 + }, + { + "localized_name": "img_compression", + "name": "img_compression", + "type": "INT", + "widget": { + "name": "img_compression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "output_image", + "name": "output_image", + "type": "IMAGE", + "links": [ + 510, + 515 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "LTXVPreprocess", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 18 + ] + }, + { + "id": 290, + "type": "ResizeImageMaskNode", + "pos": [ + 1660, + 4810 + ], + "size": [ + 300, + 160 + ], + "flags": {}, + "order": 17, + "mode": 0, + "inputs": [ + { + "localized_name": "input", + "name": "input", + "type": "IMAGE,MASK", + "link": 535 + }, + { + "localized_name": "resize_type", + "name": "resize_type", + "type": "COMFY_DYNAMICCOMBO_V3", + "widget": { + "name": "resize_type" + }, + "link": null + }, + { + "localized_name": "width", + "name": "resize_type.width", + "type": "INT", + "widget": { + "name": "resize_type.width" + }, + "link": 558 + }, + { + "localized_name": "height", + "name": "resize_type.height", + "type": "INT", + "widget": { + "name": "resize_type.height" + }, + "link": 559 + }, + { + "localized_name": "crop", + "name": "resize_type.crop", + "type": "COMBO", + "widget": { + "name": "resize_type.crop" + }, + "link": null + }, + { + "localized_name": "scale_method", + "name": "scale_method", + "type": "COMBO", + "widget": { + "name": "scale_method" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "resized", + "name": "resized", + "type": "*", + "links": [ + 523 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "ResizeImageMaskNode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "scale dimensions", + 1920, + 1088, + "center", + "lanczos" + ] + }, + { + "id": 291, + "type": "KSamplerSelect", + "pos": [ + 3160, + 4040 + ], + "size": [ + 280, + 110 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "SAMPLER", + "name": "SAMPLER", + "type": "SAMPLER", + "links": [ + 485 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.56", + "Node name for S&R": "KSamplerSelect", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "euler_ancestral_cfg_pp" + ] + }, + { + "id": 292, + "type": "ComfyMathExpression", + "pos": [ + 2540, + 4830 + ], + "size": [ + 210, + 80 + ], + "flags": { + "collapsed": true + }, + "order": 18, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 560 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 561 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.3", + "Node name for S&R": "ComfyMathExpression", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "a/2" + ] + }, + { + "id": 293, + "type": "Reroute", + "pos": [ + 3850, + 4050 + ], + "size": [ + 230, + 40 + ], + "flags": {}, + "order": 19, + "mode": 0, + "inputs": [ + { + "name": "", + "type": "*", + "link": 557 + } + ], + "outputs": [ + { + "name": "", + "type": "VAE", + "links": [ + 552, + 553, + 554 + ] + } + ], + "properties": { + "showOutputText": false, + "horizontal": false, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + } + }, + { + "id": 294, + "type": "ComfyMathExpression", + "pos": [ + 2550, + 4890 + ], + "size": [ + 210, + 80 + ], + "flags": { + "collapsed": true + }, + "order": 20, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 562 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 563 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.3", + "Node name for S&R": "ComfyMathExpression", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "a/2" + ] + }, + { + "id": 295, + "type": "EmptyLTXVLatentVideo", + "pos": [ + 2870, + 4940 + ], + "size": [ + 280, + 200 + ], + "flags": {}, + "order": 21, + "mode": 0, + "inputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 561 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 563 + }, + { + "localized_name": "length", + "name": "length", + "type": "INT", + "widget": { + "name": "length" + }, + "link": 631 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "links": [ + 511 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.60", + "Node name for S&R": "EmptyLTXVLatentVideo", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 768, + 512, + 97, + 1 + ] + }, + { + "id": 296, + "type": "LTXVImgToVideoInplace", + "pos": [ + 3230, + 4810 + ], + "size": [ + 280, + 180 + ], + "flags": {}, + "order": 22, + "mode": 0, + "inputs": [ + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 556 + }, + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 510 + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "link": 511 + }, + { + "localized_name": "strength", + "name": "strength", + "type": "FLOAT", + "widget": { + "name": "strength" + }, + "link": null + }, + { + "localized_name": "bypass", + "name": "bypass", + "type": "BOOLEAN", + "widget": { + "name": "bypass" + }, + "link": 542 + } + ], + "outputs": [ + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "links": [ + 497 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "LTXVImgToVideoInplace", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 0.7, + false + ] + }, + { + "id": 297, + "type": "LTXVAudioVAEDecode", + "pos": [ + 5760, + 3970 + ], + "size": [ + 270, + 100 + ], + "flags": {}, + "order": 23, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 495 + }, + { + "label": "Audio VAE", + "localized_name": "audio_vae", + "name": "audio_vae", + "type": "VAE", + "link": 496 + } + ], + "outputs": [ + { + "localized_name": "Audio", + "name": "Audio", + "type": "AUDIO", + "links": [ + 534 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "LTXVAudioVAEDecode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 298, + "type": "ComfyMathExpression", + "pos": [ + 2540, + 5030 + ], + "size": [ + 210, + 80 + ], + "flags": { + "collapsed": true + }, + "order": 24, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 564 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": [ + 566, + 591 + ] + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 565 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.3", + "Node name for S&R": "ComfyMathExpression", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "a" + ] + }, + { + "id": 299, + "type": "PrimitiveInt", + "pos": [ + 1190, + 4650 + ], + "size": [ + 370, + 110 + ], + "flags": {}, + "order": 25, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": 598 + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 559, + 562 + ] + } + ], + "title": "Height", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.3", + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 720, + "fixed" + ] + }, + { + "id": 300, + "type": "PrimitiveInt", + "pos": [ + 1190, + 4840 + ], + "size": [ + 370, + 110 + ], + "flags": {}, + "order": 26, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": 624 + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 564, + 629 + ] + } + ], + "title": "Frame Rate", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.3", + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 25, + "fixed" + ] + }, + { + "id": 301, + "type": "PrimitiveInt", + "pos": [ + 1190, + 4280 + ], + "size": [ + 370, + 110 + ], + "flags": {}, + "order": 27, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": 599 + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 628 + ] + } + ], + "title": "Duration", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 5, + "fixed" + ] + }, + { + "id": 302, + "type": "PrimitiveBoolean", + "pos": [ + 1190, + 4110 + ], + "size": [ + 370, + 100 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "BOOLEAN", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "BOOLEAN", + "name": "BOOLEAN", + "type": "BOOLEAN", + "links": [ + 542, + 543 + ] + } + ], + "title": "Switch to Text to Video?", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.0", + "Node name for S&R": "PrimitiveBoolean", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + false + ] + }, + { + "id": 303, + "type": "CLIPTextEncode", + "pos": [ + 2170, + 3640 + ], + "size": [ + 600, + 390 + ], + "flags": {}, + "order": 28, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 615 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 625 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 526 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.56", + "Node name for S&R": "CLIPTextEncode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "" + ], + "color": "#232", + "bgcolor": "#353" + }, + { + "id": 304, + "type": "LTXVConditioning", + "pos": [ + 2800, + 3810 + ], + "size": [ + 280, + 130 + ], + "flags": {}, + "order": 29, + "mode": 0, + "inputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 526 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 527 + }, + { + "localized_name": "frame_rate", + "name": "frame_rate", + "type": "FLOAT", + "widget": { + "name": "frame_rate" + }, + "link": 566 + } + ], + "outputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "links": [ + 475, + 518 + ] + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "links": [ + 476, + 519 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.56", + "Node name for S&R": "LTXVConditioning", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 24 + ] + }, + { + "id": 305, + "type": "LTXVEmptyLatentAudio", + "pos": [ + 3540, + 4960 + ], + "size": [ + 280, + 170 + ], + "flags": {}, + "order": 30, + "mode": 0, + "inputs": [ + { + "localized_name": "audio_vae", + "name": "audio_vae", + "type": "VAE", + "link": 481 + }, + { + "localized_name": "frames_number", + "name": "frames_number", + "type": "INT", + "widget": { + "name": "frames_number" + }, + "link": 630 + }, + { + "localized_name": "frame_rate", + "name": "frame_rate", + "type": "INT", + "widget": { + "name": "frame_rate" + }, + "link": 565 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "Latent", + "name": "Latent", + "type": "LATENT", + "links": [ + 498 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.68", + "Node name for S&R": "LTXVEmptyLatentAudio", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 97, + 25, + 1 + ] + }, + { + "id": 306, + "type": "ManualSigmas", + "pos": [ + 3160, + 4220 + ], + "size": [ + 500, + 110 + ], + "flags": {}, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "sigmas", + "name": "sigmas", + "type": "STRING", + "widget": { + "name": "sigmas" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "SIGMAS", + "name": "SIGMAS", + "type": "SIGMAS", + "links": [ + 544 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.14.1", + "Node name for S&R": "ManualSigmas", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "1.0, 0.99375, 0.9875, 0.98125, 0.975, 0.909375, 0.725, 0.421875, 0.0" + ] + }, + { + "id": 307, + "type": "LTXVSeparateAVLatent", + "pos": [ + 3820, + 3630 + ], + "size": [ + 250, + 100 + ], + "flags": {}, + "order": 31, + "mode": 0, + "inputs": [ + { + "localized_name": "av_latent", + "name": "av_latent", + "type": "LATENT", + "link": 488 + } + ], + "outputs": [ + { + "localized_name": "video_latent", + "name": "video_latent", + "type": "LATENT", + "links": [ + 477, + 547 + ] + }, + { + "localized_name": "audio_latent", + "name": "audio_latent", + "type": "LATENT", + "links": [ + 513 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.5.1", + "Node name for S&R": "LTXVSeparateAVLatent", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 308, + "type": "SamplerCustomAdvanced", + "pos": [ + 5050, + 3650 + ], + "size": [ + 230, + 170 + ], + "flags": {}, + "order": 32, + "mode": 0, + "inputs": [ + { + "localized_name": "noise", + "name": "noise", + "type": "NOISE", + "link": 490 + }, + { + "localized_name": "guider", + "name": "guider", + "type": "GUIDER", + "link": 491 + }, + { + "localized_name": "sampler", + "name": "sampler", + "type": "SAMPLER", + "link": 492 + }, + { + "localized_name": "sigmas", + "name": "sigmas", + "type": "SIGMAS", + "link": 493 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 494 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "LATENT", + "links": [ + 578 + ] + }, + { + "localized_name": "denoised_output", + "name": "denoised_output", + "type": "LATENT", + "links": [] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.75", + "Node name for S&R": "SamplerCustomAdvanced", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 309, + "type": "LTXVSeparateAVLatent", + "pos": [ + 5390, + 3650 + ], + "size": [ + 230, + 100 + ], + "flags": {}, + "order": 33, + "mode": 0, + "inputs": [ + { + "localized_name": "av_latent", + "name": "av_latent", + "type": "LATENT", + "link": 578 + } + ], + "outputs": [ + { + "localized_name": "video_latent", + "name": "video_latent", + "type": "LATENT", + "links": [ + 539 + ] + }, + { + "localized_name": "audio_latent", + "name": "audio_latent", + "type": "LATENT", + "links": [ + 495 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.5.1", + "Node name for S&R": "LTXVSeparateAVLatent", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 310, + "type": "CreateVideo", + "pos": [ + 6050, + 4490 + ], + "size": [ + 280, + 130 + ], + "flags": {}, + "order": 34, + "mode": 0, + "inputs": [ + { + "localized_name": "images", + "name": "images", + "type": "IMAGE", + "link": 538 + }, + { + "localized_name": "audio", + "name": "audio", + "shape": 7, + "type": "AUDIO", + "link": 534 + }, + { + "localized_name": "fps", + "name": "fps", + "type": "FLOAT", + "widget": { + "name": "fps" + }, + "link": 591 + } + ], + "outputs": [ + { + "localized_name": "VIDEO", + "name": "VIDEO", + "type": "VIDEO", + "links": [ + 536 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.5.1", + "Node name for S&R": "CreateVideo", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 24 + ] + }, + { + "id": 311, + "type": "LatentUpscaleModelLoader", + "pos": [ + 1670, + 4550 + ], + "size": [ + 400, + 110 + ], + "flags": {}, + "order": 35, + "mode": 0, + "inputs": [ + { + "localized_name": "model_name", + "name": "model_name", + "type": "COMBO", + "widget": { + "name": "model_name" + }, + "link": 607 + } + ], + "outputs": [ + { + "localized_name": "LATENT_UPSCALE_MODEL", + "name": "LATENT_UPSCALE_MODEL", + "type": "LATENT_UPSCALE_MODEL", + "links": [ + 545 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "LatentUpscaleModelLoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "ltx-2.3-spatial-upscaler-x2-1.1.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2.3/resolve/main/ltx-2.3-spatial-upscaler-x2-1.1.safetensors", + "directory": "latent_upscale_models" + } + ] + }, + "widgets_values": [ + "ltx-2.3-spatial-upscaler-x2-1.1.safetensors" + ] + }, + { + "id": 312, + "type": "PrimitiveInt", + "pos": [ + 1190, + 4470 + ], + "size": [ + 370, + 110 + ], + "flags": {}, + "order": 36, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": 597 + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 558, + 560 + ] + } + ], + "title": "Width", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.3", + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1280, + "fixed" + ] + }, + { + "id": 313, + "type": "CLIPTextEncode", + "pos": [ + 2180, + 4120 + ], + "size": [ + 600, + 170 + ], + "flags": {}, + "order": 37, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 627 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 527 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.56", + "Node name for S&R": "CLIPTextEncode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "pc game, console game, video game, cartoon, childish, ugly" + ], + "color": "#323", + "bgcolor": "#535" + }, + { + "id": 314, + "type": "CFGGuider", + "pos": [ + 3160, + 3810 + ], + "size": [ + 280, + 160 + ], + "flags": {}, + "order": 38, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 541 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 518 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 519 + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "GUIDER", + "name": "GUIDER", + "type": "GUIDER", + "links": [ + 484 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.64", + "Node name for S&R": "CFGGuider", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1 + ] + }, + { + "id": 315, + "type": "VAEDecodeTiled", + "pos": [ + 5750, + 3610 + ], + "size": [ + 280, + 200 + ], + "flags": {}, + "order": 39, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 539 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 553 + }, + { + "localized_name": "tile_size", + "name": "tile_size", + "type": "INT", + "widget": { + "name": "tile_size" + }, + "link": null + }, + { + "localized_name": "overlap", + "name": "overlap", + "type": "INT", + "widget": { + "name": "overlap" + }, + "link": null + }, + { + "localized_name": "temporal_size", + "name": "temporal_size", + "type": "INT", + "widget": { + "name": "temporal_size" + }, + "link": null + }, + { + "localized_name": "temporal_overlap", + "name": "temporal_overlap", + "type": "INT", + "widget": { + "name": "temporal_overlap" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 538 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.14.1", + "Node name for S&R": "VAEDecodeTiled", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 768, + 64, + 4096, + 4 + ] + }, + { + "id": 316, + "type": "CheckpointLoaderSimple", + "pos": [ + 1660, + 3660 + ], + "size": [ + 430, + 160 + ], + "flags": {}, + "order": 40, + "mode": 0, + "inputs": [ + { + "localized_name": "ckpt_name", + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": 601 + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 520 + ] + }, + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [] + }, + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "links": [ + 556, + 557 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.56", + "Node name for S&R": "CheckpointLoaderSimple", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "ltx-2.3-22b-dev-fp8.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2.3-fp8/resolve/main/ltx-2.3-22b-dev-fp8.safetensors", + "directory": "checkpoints" + } + ] + }, + "widgets_values": [ + "ltx-2.3-22b-dev-fp8.safetensors" + ] + }, + { + "id": 317, + "type": "LTXAVTextEncoderLoader", + "pos": [ + 1660, + 4280 + ], + "size": [ + 430, + 170 + ], + "flags": {}, + "order": 41, + "mode": 0, + "showAdvanced": false, + "inputs": [ + { + "localized_name": "text_encoder", + "name": "text_encoder", + "type": "COMBO", + "widget": { + "name": "text_encoder" + }, + "link": 606 + }, + { + "localized_name": "ckpt_name", + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": 605 + }, + { + "localized_name": "device", + "name": "device", + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [ + 615, + 627 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "LTXAVTextEncoderLoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "ltx-2.3-22b-dev-fp8.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2.3-fp8/resolve/main/ltx-2.3-22b-dev-fp8.safetensors", + "directory": "checkpoints" + }, + { + "name": "gemma_3_12B_it_fp4_mixed.safetensors", + "url": "https://huggingface.co/Comfy-Org/ltx-2/resolve/main/split_files/text_encoders/gemma_3_12B_it_fp4_mixed.safetensors", + "directory": "text_encoders" + } + ] + }, + "widgets_values": [ + "gemma_3_12B_it_fp4_mixed.safetensors", + "ltx-2.3-22b-dev-fp8.safetensors", + "default" + ] + }, + { + "id": 318, + "type": "LTXVConcatAVLatent", + "pos": [ + 3860, + 4830 + ], + "size": [ + 240, + 100 + ], + "flags": {}, + "order": 42, + "mode": 0, + "inputs": [ + { + "localized_name": "video_latent", + "name": "video_latent", + "type": "LATENT", + "link": 497 + }, + { + "localized_name": "audio_latent", + "name": "audio_latent", + "type": "LATENT", + "link": 498 + } + ], + "outputs": [ + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "links": [ + 487 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "LTXVConcatAVLatent", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 319, + "type": "PrimitiveStringMultiline", + "pos": [ + 1190, + 3680 + ], + "size": [ + 370, + 350 + ], + "flags": {}, + "order": 43, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "STRING", + "widget": { + "name": "value" + }, + "link": 595 + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 625 + ] + } + ], + "title": "Prompt", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.3", + "Node name for S&R": "PrimitiveStringMultiline", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "" + ] + }, + { + "id": 323, + "type": "ComfyMathExpression", + "pos": [ + 1210, + 5040 + ], + "size": [ + 360, + 210 + ], + "flags": { + "collapsed": true + }, + "order": 44, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 628 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": 629 + }, + { + "label": "c", + "localized_name": "values.c", + "name": "values.c", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 630, + 631 + ] + } + ], + "title": "Math Expression (length)", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ComfyMathExpression" + }, + "widgets_values": [ + "a * b + 1" + ] + } + ], + "groups": [ + { + "id": 1, + "title": "Model", + "bounding": [ + 1630, + 3550, + 480, + 1140 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 2, + "title": "Generate Low Resolution", + "bounding": [ + 3130, + 3550, + 1000, + 1140 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 3, + "title": "Prompt", + "bounding": [ + 2140, + 3550, + 960, + 1140 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 6, + "title": "Generate High Resolution", + "bounding": [ + 4670, + 3550, + 990, + 1130 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 7, + "title": "Lantent Upscale", + "bounding": [ + 4160, + 3550, + 480, + 1130 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 19, + "title": "Video Settings", + "bounding": [ + 1150, + 3550, + 460, + 1610 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 20, + "title": "Image Preprocess", + "bounding": [ + 1630, + 4720, + 830, + 440 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 21, + "title": "Empty Latent", + "bounding": [ + 2820, + 4720, + 1310, + 450 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 22, + "title": "Number conversion", + "bounding": [ + 2480, + 4720, + 310, + 440 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 512, + "origin_id": 288, + "origin_slot": 0, + "target_id": 278, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 513, + "origin_id": 307, + "origin_slot": 1, + "target_id": 278, + "target_slot": 1, + "type": "LATENT" + }, + { + "id": 478, + "origin_id": 285, + "origin_slot": 0, + "target_id": 282, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 479, + "origin_id": 284, + "origin_slot": 0, + "target_id": 282, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 480, + "origin_id": 284, + "origin_slot": 1, + "target_id": 282, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 541, + "origin_id": 285, + "origin_slot": 0, + "target_id": 314, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 518, + "origin_id": 304, + "origin_slot": 0, + "target_id": 314, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 519, + "origin_id": 304, + "origin_slot": 1, + "target_id": 314, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 483, + "origin_id": 277, + "origin_slot": 0, + "target_id": 283, + "target_slot": 0, + "type": "NOISE" + }, + { + "id": 484, + "origin_id": 314, + "origin_slot": 0, + "target_id": 283, + "target_slot": 1, + "type": "GUIDER" + }, + { + "id": 485, + "origin_id": 291, + "origin_slot": 0, + "target_id": 283, + "target_slot": 2, + "type": "SAMPLER" + }, + { + "id": 544, + "origin_id": 306, + "origin_slot": 0, + "target_id": 283, + "target_slot": 3, + "type": "SIGMAS" + }, + { + "id": 487, + "origin_id": 318, + "origin_slot": 0, + "target_id": 283, + "target_slot": 4, + "type": "LATENT" + }, + { + "id": 475, + "origin_id": 304, + "origin_slot": 0, + "target_id": 284, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 476, + "origin_id": 304, + "origin_slot": 1, + "target_id": 284, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 477, + "origin_id": 307, + "origin_slot": 0, + "target_id": 284, + "target_slot": 2, + "type": "LATENT" + }, + { + "id": 520, + "origin_id": 316, + "origin_slot": 0, + "target_id": 285, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 523, + "origin_id": 290, + "origin_slot": 0, + "target_id": 286, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 547, + "origin_id": 307, + "origin_slot": 0, + "target_id": 287, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 545, + "origin_id": 311, + "origin_slot": 0, + "target_id": 287, + "target_slot": 1, + "type": "LATENT_UPSCALE_MODEL" + }, + { + "id": 554, + "origin_id": 293, + "origin_slot": 0, + "target_id": 287, + "target_slot": 2, + "type": "VAE" + }, + { + "id": 552, + "origin_id": 293, + "origin_slot": 0, + "target_id": 288, + "target_slot": 0, + "type": "VAE" + }, + { + "id": 515, + "origin_id": 289, + "origin_slot": 0, + "target_id": 288, + "target_slot": 1, + "type": "IMAGE" + }, + { + "id": 548, + "origin_id": 287, + "origin_slot": 0, + "target_id": 288, + "target_slot": 2, + "type": "LATENT" + }, + { + "id": 543, + "origin_id": 302, + "origin_slot": 0, + "target_id": 288, + "target_slot": 4, + "type": "BOOLEAN" + }, + { + "id": 505, + "origin_id": 286, + "origin_slot": 0, + "target_id": 289, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 558, + "origin_id": 312, + "origin_slot": 0, + "target_id": 290, + "target_slot": 2, + "type": "INT" + }, + { + "id": 559, + "origin_id": 299, + "origin_slot": 0, + "target_id": 290, + "target_slot": 3, + "type": "INT" + }, + { + "id": 560, + "origin_id": 312, + "origin_slot": 0, + "target_id": 292, + "target_slot": 0, + "type": "INT" + }, + { + "id": 557, + "origin_id": 316, + "origin_slot": 2, + "target_id": 293, + "target_slot": 0, + "type": "VAE" + }, + { + "id": 562, + "origin_id": 299, + "origin_slot": 0, + "target_id": 294, + "target_slot": 0, + "type": "INT" + }, + { + "id": 561, + "origin_id": 292, + "origin_slot": 1, + "target_id": 295, + "target_slot": 0, + "type": "INT" + }, + { + "id": 563, + "origin_id": 294, + "origin_slot": 1, + "target_id": 295, + "target_slot": 1, + "type": "INT" + }, + { + "id": 556, + "origin_id": 316, + "origin_slot": 2, + "target_id": 296, + "target_slot": 0, + "type": "VAE" + }, + { + "id": 510, + "origin_id": 289, + "origin_slot": 0, + "target_id": 296, + "target_slot": 1, + "type": "IMAGE" + }, + { + "id": 511, + "origin_id": 295, + "origin_slot": 0, + "target_id": 296, + "target_slot": 2, + "type": "LATENT" + }, + { + "id": 542, + "origin_id": 302, + "origin_slot": 0, + "target_id": 296, + "target_slot": 4, + "type": "BOOLEAN" + }, + { + "id": 495, + "origin_id": 309, + "origin_slot": 1, + "target_id": 297, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 496, + "origin_id": 279, + "origin_slot": 0, + "target_id": 297, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 564, + "origin_id": 300, + "origin_slot": 0, + "target_id": 298, + "target_slot": 0, + "type": "INT" + }, + { + "id": 526, + "origin_id": 303, + "origin_slot": 0, + "target_id": 304, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 527, + "origin_id": 313, + "origin_slot": 0, + "target_id": 304, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 566, + "origin_id": 298, + "origin_slot": 0, + "target_id": 304, + "target_slot": 2, + "type": "FLOAT" + }, + { + "id": 497, + "origin_id": 296, + "origin_slot": 0, + "target_id": 318, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 498, + "origin_id": 305, + "origin_slot": 0, + "target_id": 318, + "target_slot": 1, + "type": "LATENT" + }, + { + "id": 481, + "origin_id": 279, + "origin_slot": 0, + "target_id": 305, + "target_slot": 0, + "type": "VAE" + }, + { + "id": 565, + "origin_id": 298, + "origin_slot": 1, + "target_id": 305, + "target_slot": 2, + "type": "INT" + }, + { + "id": 488, + "origin_id": 283, + "origin_slot": 0, + "target_id": 307, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 490, + "origin_id": 276, + "origin_slot": 0, + "target_id": 308, + "target_slot": 0, + "type": "NOISE" + }, + { + "id": 491, + "origin_id": 282, + "origin_slot": 0, + "target_id": 308, + "target_slot": 1, + "type": "GUIDER" + }, + { + "id": 492, + "origin_id": 280, + "origin_slot": 0, + "target_id": 308, + "target_slot": 2, + "type": "SAMPLER" + }, + { + "id": 493, + "origin_id": 281, + "origin_slot": 0, + "target_id": 308, + "target_slot": 3, + "type": "SIGMAS" + }, + { + "id": 494, + "origin_id": 278, + "origin_slot": 0, + "target_id": 308, + "target_slot": 4, + "type": "LATENT" + }, + { + "id": 578, + "origin_id": 308, + "origin_slot": 0, + "target_id": 309, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 539, + "origin_id": 309, + "origin_slot": 0, + "target_id": 315, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 553, + "origin_id": 293, + "origin_slot": 0, + "target_id": 315, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 538, + "origin_id": 315, + "origin_slot": 0, + "target_id": 310, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 534, + "origin_id": 297, + "origin_slot": 0, + "target_id": 310, + "target_slot": 1, + "type": "AUDIO" + }, + { + "id": 591, + "origin_id": 298, + "origin_slot": 0, + "target_id": 310, + "target_slot": 2, + "type": "FLOAT" + }, + { + "id": 535, + "origin_id": -10, + "origin_slot": 0, + "target_id": 290, + "target_slot": 0, + "type": "IMAGE,MASK" + }, + { + "id": 536, + "origin_id": 310, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "VIDEO" + }, + { + "id": 595, + "origin_id": -10, + "origin_slot": 1, + "target_id": 319, + "target_slot": 0, + "type": "STRING" + }, + { + "id": 597, + "origin_id": -10, + "origin_slot": 2, + "target_id": 312, + "target_slot": 0, + "type": "INT" + }, + { + "id": 598, + "origin_id": -10, + "origin_slot": 3, + "target_id": 299, + "target_slot": 0, + "type": "INT" + }, + { + "id": 599, + "origin_id": -10, + "origin_slot": 4, + "target_id": 301, + "target_slot": 0, + "type": "INT" + }, + { + "id": 601, + "origin_id": -10, + "origin_slot": 5, + "target_id": 316, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 602, + "origin_id": -10, + "origin_slot": 6, + "target_id": 285, + "target_slot": 1, + "type": "COMBO" + }, + { + "id": 604, + "origin_id": -10, + "origin_slot": 5, + "target_id": 279, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 605, + "origin_id": -10, + "origin_slot": 5, + "target_id": 317, + "target_slot": 1, + "type": "COMBO" + }, + { + "id": 606, + "origin_id": -10, + "origin_slot": 7, + "target_id": 317, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 607, + "origin_id": -10, + "origin_slot": 8, + "target_id": 311, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 615, + "origin_id": 317, + "origin_slot": 0, + "target_id": 303, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 624, + "origin_id": -10, + "origin_slot": 9, + "target_id": 300, + "target_slot": 0, + "type": "INT" + }, + { + "id": 625, + "origin_id": 319, + "origin_slot": 0, + "target_id": 303, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 627, + "origin_id": 317, + "origin_slot": 0, + "target_id": 313, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 628, + "origin_id": 301, + "origin_slot": 0, + "target_id": 323, + "target_slot": 0, + "type": "INT" + }, + { + "id": 629, + "origin_id": 300, + "origin_slot": 0, + "target_id": 323, + "target_slot": 1, + "type": "INT" + }, + { + "id": 630, + "origin_id": 323, + "origin_slot": 1, + "target_id": 305, + "target_slot": 1, + "type": "INT" + }, + { + "id": 631, + "origin_id": 323, + "origin_slot": 1, + "target_id": 295, + "target_slot": 2, + "type": "INT" + } + ], + "extra": { + "workflowRendererVersion": "Vue-corrected" + }, + "category": "Video generation and editing/Image to video", + "description": "Generates video from a single input image using LTX-2.3." + } + ] + }, + "extra": { + "ue_links": [] + } +} \ No newline at end of file diff --git a/blueprints/Image to Video (Wan 2.2).json b/blueprints/Image to Video (Wan 2.2).json index a8dafd3c9..a24adcfb6 100644 --- a/blueprints/Image to Video (Wan 2.2).json +++ b/blueprints/Image to Video (Wan 2.2).json @@ -206,7 +206,7 @@ }, "revision": 0, "config": {}, - "name": "local-Image to Video (Wan 2.2)", + "name": "Image to Video (Wan 2.2)", "inputNode": { "id": -10, "bounding": [ @@ -2027,7 +2027,8 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Video generation and editing/Image to video" + "category": "Video generation and editing/Image to video", + "description": "Image-to-video with Wan 2.2 using a start image plus text prompt to extend motion from the still frame." } ] }, diff --git a/blueprints/Pose to Image (Z-Image-Turbo).json b/blueprints/Pose to Image (Z-Image-Turbo).json index a55410ba4..5c2749efe 100644 --- a/blueprints/Pose to Image (Z-Image-Turbo).json +++ b/blueprints/Pose to Image (Z-Image-Turbo).json @@ -134,7 +134,7 @@ }, "revision": 0, "config": {}, - "name": "local-Pose to Image (Z-Image-Turbo)", + "name": "Pose to Image (Z-Image-Turbo)", "inputNode": { "id": -10, "bounding": [ @@ -1298,7 +1298,8 @@ "VHS_MetadataImage": true, "VHS_KeepIntermediate": true }, - "category": "Image generation and editing/Pose to image" + "category": "Image generation and editing/Pose to image", + "description": "Generates an image from pose keypoints using Z-Image-Turbo with text conditioning." } ] }, @@ -1319,4 +1320,4 @@ } }, "version": 0.4 -} +} \ No newline at end of file diff --git a/blueprints/Pose to Video (LTX 2.0).json b/blueprints/Pose to Video (LTX 2.0).json index ae369941c..1ce49351a 100644 --- a/blueprints/Pose to Video (LTX 2.0).json +++ b/blueprints/Pose to Video (LTX 2.0).json @@ -1,28 +1,26 @@ { - "id": "01cd475b-52df-43bf-aafa-484a5976d2d2", "revision": 0, - "last_node_id": 160, - "last_link_id": 410, + "last_node_id": 143, + "last_link_id": 0, "nodes": [ { - "id": 1, - "type": "f0e58a6b-7246-4103-9fec-73b423634b1f", + "id": 143, + "type": "68857357-cbc2-4c3a-a786-c3a58d43f9b1", "pos": [ - 210, - 3830 + 290, + 3960 ], "size": [ - 420, + 400, 500 ], "flags": { "collapsed": false }, - "order": 0, + "order": 13, "mode": 0, "inputs": [ { - "label": "prompt", "name": "text", "type": "STRING", "widget": { @@ -31,33 +29,32 @@ "link": null }, { - "label": "first_frame_strength", - "name": "strength", - "type": "FLOAT", - "widget": { - "name": "strength" - }, - "link": null - }, - { - "label": "disable_first_frame", - "name": "bypass", - "type": "BOOLEAN", - "widget": { - "name": "bypass" - }, - "link": null - }, - { - "label": "first frame", + "label": "control_images", "name": "image", "type": "IMAGE", "link": null }, { - "label": "control image", - "name": "input", - "type": "IMAGE,MASK", + "label": "first_frame", + "name": "image_1", + "type": "IMAGE", + "link": null + }, + { + "label": "image_strength", + "name": "strength_1", + "type": "FLOAT", + "widget": { + "name": "strength_1" + }, + "link": null + }, + { + "name": "noise_seed", + "type": "INT", + "widget": { + "name": "noise_seed" + }, "link": null }, { @@ -69,6 +66,7 @@ "link": null }, { + "label": "control_lora", "name": "lora_name", "type": "COMBO", "widget": { @@ -77,7 +75,15 @@ "link": null }, { - "label": "distll_lora", + "name": "text_encoder", + "type": "COMBO", + "widget": { + "name": "text_encoder" + }, + "link": null + }, + { + "label": "distill_lora", "name": "lora_name_1", "type": "COMBO", "widget": { @@ -93,30 +99,6 @@ "name": "model_name" }, "link": null - }, - { - "name": "resize_type.width", - "type": "INT", - "widget": { - "name": "resize_type.width" - }, - "link": null - }, - { - "name": "resize_type.height", - "type": "INT", - "widget": { - "name": "resize_type.height" - }, - "link": null - }, - { - "name": "length", - "type": "INT", - "widget": { - "name": "length" - }, - "link": null } ], "outputs": [ @@ -130,56 +112,49 @@ "properties": { "proxyWidgets": [ [ - "-1", + "124", "text" ], [ - "-1", - "resize_type.width" - ], - [ - "-1", - "resize_type.height" - ], - [ - "-1", - "length" - ], - [ - "-1", + "149", "strength" ], - [ - "-1", - "bypass" - ], [ "126", "noise_seed" ], [ - "126", - "control_after_generate" - ], - [ - "-1", + "103", "ckpt_name" ], [ - "-1", + "134", "lora_name" ], [ - "-1", - "model_name" + "97", + "text_encoder" ], [ - "-1", - "lora_name_1" + "105", + "lora_name" + ], + [ + "100", + "model_name" ] ], "cnr_id": "comfy-core", "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": { + "lora_name": true, + "strength": true, + "bypass": true + }, + "version": "7.7", + "input_ue_unconnectable": {} + }, "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -188,52 +163,40 @@ "secondTabOffset": 80, "secondTabWidth": 65 }, - "widgets_values": [ - "", - 1280, - 720, - 97, - 1, - false, - null, - null, - "ltx-2-19b-dev-fp8.safetensors", - "ltx-2-19b-ic-lora-pose-control.safetensors", - "ltx-2-spatial-upscaler-x2-1.0.safetensors", - "ltx-2-19b-distilled-lora-384.safetensors" - ] + "widgets_values": [], + "title": "Pose to Video (LTX 2.0)" } ], "links": [], - "groups": [], + "version": 0.4, "definitions": { "subgraphs": [ { - "id": "f0e58a6b-7246-4103-9fec-73b423634b1f", + "id": "68857357-cbc2-4c3a-a786-c3a58d43f9b1", "version": 1, "state": { - "lastGroupId": 11, - "lastNodeId": 160, - "lastLinkId": 410, + "lastGroupId": 14, + "lastNodeId": 701, + "lastLinkId": 1774, "lastRerouteId": 0 }, "revision": 0, "config": {}, - "name": "local-Pose to Video (LTX 2.0)", + "name": "Pose to Video (LTX 2.0)", "inputNode": { "id": -10, "bounding": [ - -2220, - 4180, - 153.3203125, - 280 + -2050, + 4100, + 127.029296875, + 240 ] }, "outputNode": { "id": -20, "bounding": [ - 1750.2777777777776, - 4091.1111111111113, + 1750, + 4090, 120, 60 ] @@ -246,154 +209,128 @@ "linkIds": [ 345 ], - "label": "prompt", "pos": [ - -2086.6796875, + -1942.970703125, + 4120 + ] + }, + { + "id": "35a07084-3ecf-482a-a330-b40278770ca3", + "name": "image", + "type": "IMAGE", + "linkIds": [ + 348, + 380 + ], + "label": "control_images", + "pos": [ + -1942.970703125, + 4140 + ] + }, + { + "id": "bea20802-d654-4287-a8ef-0f834314bcf9", + "name": "image_1", + "type": "IMAGE", + "linkIds": [ + 364, + 379 + ], + "label": "first_frame", + "pos": [ + -1942.970703125, + 4160 + ] + }, + { + "id": "b9b4151d-df88-40c0-a2bd-6e35b94557fe", + "name": "strength_1", + "type": "FLOAT", + "linkIds": [ + 1758, + 1759 + ], + "label": "image_strength", + "pos": [ + -1942.970703125, + 4180 + ] + }, + { + "id": "b51f6a12-9152-4526-b115-443cfd23003f", + "name": "noise_seed", + "type": "INT", + "linkIds": [ + 1767 + ], + "pos": [ + -1942.970703125, 4200 ] }, { - "id": "59430efe-1090-4e36-8afe-b21ce7f4268b", - "name": "strength", - "type": "FLOAT", + "id": "47248f12-f174-4e35-854c-fa5eebea2903", + "name": "ckpt_name", + "type": "COMBO", "linkIds": [ - 370, - 371 + 1768, + 1770, + 1771 ], - "label": "first_frame_strength", "pos": [ - -2086.6796875, + -1942.970703125, 4220 ] }, { - "id": "6145a9b9-68ed-4956-89f7-7a5ebdd5c99e", - "name": "bypass", - "type": "BOOLEAN", + "id": "6feb34cf-7972-4d3a-91fc-11070a84dc5f", + "name": "lora_name", + "type": "COMBO", "linkIds": [ - 363, - 368 + 1769 ], - "label": "disable_first_frame", + "label": "control_lora", "pos": [ - -2086.6796875, + -1942.970703125, 4240 ] }, { - "id": "f7aa8c12-bdba-4bbd-84cf-b49cfc32a1dd", - "name": "image", - "type": "IMAGE", + "id": "6b423a3e-6c0e-445d-93c0-2cc3945400d1", + "name": "text_encoder", + "type": "COMBO", "linkIds": [ - 398, - 399 + 1772 ], - "label": "first frame", "pos": [ - -2086.6796875, + -1942.970703125, 4260 ] }, { - "id": "da40a4c0-cd19-46c6-8eb3-62d0026fbe85", - "name": "input", - "type": "IMAGE,MASK", + "id": "ffd38c52-cc57-4e68-b140-94e7b03499b1", + "name": "lora_name_1", + "type": "COMBO", "linkIds": [ - 400 + 1773 ], - "label": "control image", + "label": "distill_lora", "pos": [ - -2086.6796875, + -1942.970703125, 4280 ] }, { - "id": "8005344b-99d6-4829-a619-c4e8ef640eb9", - "name": "ckpt_name", - "type": "COMBO", - "linkIds": [ - 401, - 402, - 403 - ], - "pos": [ - -2086.6796875, - 4300 - ] - }, - { - "id": "25e7c4e8-850c-4f37-bc14-e3f4b5f228c0", - "name": "lora_name", - "type": "COMBO", - "linkIds": [ - 404, - 405 - ], - "pos": [ - -2086.6796875, - 4320 - ] - }, - { - "id": "f16a18dd-947e-400a-8889-02cf998f760a", - "name": "lora_name_1", - "type": "COMBO", - "linkIds": [ - 406 - ], - "label": "distll_lora", - "pos": [ - -2086.6796875, - 4340 - ] - }, - { - "id": "1abf156c-4c85-4ee5-8671-62df3177d835", + "id": "6d8b9605-acf0-4dd7-8d45-f824c2fd5895", "name": "model_name", "type": "COMBO", "linkIds": [ - 407 + 1774 ], "label": "upscale_model", "pos": [ - -2086.6796875, - 4360 - ] - }, - { - "id": "203402cf-4253-4daf-bf78-5def9496e0af", - "name": "resize_type.width", - "type": "INT", - "linkIds": [ - 408 - ], - "pos": [ - -2086.6796875, - 4380 - ] - }, - { - "id": "e6d8ac4a-34d4-46c6-bcb2-4e66a696438c", - "name": "resize_type.height", - "type": "INT", - "linkIds": [ - 409 - ], - "pos": [ - -2086.6796875, - 4400 - ] - }, - { - "id": "6aa6cf2c-bc4f-4f8b-be62-aa15793375dc", - "name": "length", - "type": "INT", - "linkIds": [ - 410 - ], - "pos": [ - -2086.6796875, - 4420 + -1942.970703125, + 4300 ] } ], @@ -407,8 +344,8 @@ ], "localized_name": "VIDEO", "pos": [ - 1770.2777777777776, - 4111.111111111111 + 1770, + 4110 ] } ], @@ -418,15 +355,15 @@ "id": 93, "type": "CFGGuider", "pos": [ - -697.721823660531, - 3671.1105325465196 + -690, + 3710 ], "size": [ - 269.97395833333337, - 98 + 270, + 160 ], "flags": {}, - "order": 16, + "order": 7, "mode": 0, "inputs": [ { @@ -470,6 +407,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.3.64", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "CFGGuider", "enableTabs": false, "tabWidth": 65, @@ -487,12 +429,12 @@ "id": 94, "type": "KSamplerSelect", "pos": [ - -697.721823660531, - 3841.1107362825187 + -690, + 3940 ], "size": [ - 269.97395833333337, - 58 + 270, + 110 ], "flags": {}, "order": 0, @@ -521,6 +463,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.3.56", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "KSamplerSelect", "enableTabs": false, "tabWidth": 65, @@ -538,12 +485,12 @@ "id": 99, "type": "ManualSigmas", "pos": [ - 410.27824286284044, - 3851.110970278795 + 450, + 3910 ], "size": [ - 269.97395833333337, - 58 + 270, + 110 ], "flags": {}, "order": 1, @@ -572,6 +519,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "ManualSigmas", "enableTabs": false, "tabWidth": 65, @@ -589,15 +541,15 @@ "id": 100, "type": "LatentUpscaleModelLoader", "pos": [ - -69.72208571196083, - 3701.1104657166875 + -70, + 3790 ], "size": [ - 389.97395833333337, - 58 + 390, + 110 ], "flags": {}, - "order": 2, + "order": 11, "mode": 0, "inputs": [ { @@ -607,7 +559,7 @@ "widget": { "name": "model_name" }, - "link": 407 + "link": 1774 } ], "outputs": [ @@ -623,21 +575,26 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "LatentUpscaleModelLoader", - "models": [ - { - "name": "ltx-2-spatial-upscaler-x2-1.0.safetensors", - "url": "https://huggingface.co/Lightricks/LTX-2/resolve/main/ltx-2-spatial-upscaler-x2-1.0.safetensors", - "directory": "latent_upscale_models" - } - ], "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, "hasSecondTab": false, "secondTabText": "Send Back", "secondTabOffset": 80, - "secondTabWidth": 65 + "secondTabWidth": 65, + "models": [ + { + "name": "ltx-2-spatial-upscaler-x2-1.0.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2/resolve/main/ltx-2-spatial-upscaler-x2-1.0.safetensors", + "directory": "latent_upscale_models" + } + ] }, "widgets_values": [ "ltx-2-spatial-upscaler-x2-1.0.safetensors" @@ -647,15 +604,15 @@ "id": 101, "type": "LTXVConcatAVLatent", "pos": [ - 410.27824286284044, - 4101.110949206838 + 450, + 4220 ], "size": [ - 269.97395833333337, - 46 + 270, + 120 ], "flags": {}, - "order": 18, + "order": 12, "mode": 0, "inputs": [ { @@ -684,6 +641,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "LTXVConcatAVLatent", "enableTabs": false, "tabWidth": 65, @@ -692,22 +654,21 @@ "secondTabText": "Send Back", "secondTabOffset": 80, "secondTabWidth": 65 - }, - "widgets_values": [] + } }, { "id": 108, "type": "CFGGuider", "pos": [ - 410.27824286284044, - 3701.1104657166875 + 450, + 3720 ], "size": [ - 269.97395833333337, - 98 + 270, + 160 ], "flags": {}, - "order": 22, + "order": 18, "mode": 0, "inputs": [ { @@ -751,6 +712,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "CFGGuider", "enableTabs": false, "tabWidth": 65, @@ -764,19 +730,101 @@ 1 ] }, + { + "id": 111, + "type": "LTXVEmptyLatentAudio", + "pos": [ + -1100, + 4940 + ], + "size": [ + 270, + 170 + ], + "flags": {}, + "order": 20, + "mode": 0, + "inputs": [ + { + "localized_name": "audio_vae", + "name": "audio_vae", + "type": "VAE", + "link": 285 + }, + { + "localized_name": "frames_number", + "name": "frames_number", + "type": "INT", + "widget": { + "name": "frames_number" + }, + "link": 329 + }, + { + "localized_name": "frame_rate", + "name": "frame_rate", + "type": "INT", + "widget": { + "name": "frame_rate" + }, + "link": 354 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "Latent", + "name": "Latent", + "type": "LATENT", + "links": [ + 300 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.68", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "LTXVEmptyLatentAudio", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 97, + 25, + 1 + ] + }, { "id": 123, "type": "SamplerCustomAdvanced", "pos": [ - -387.72197839215096, - 3521.1103425011374 + -380, + 3530 ], "size": [ - 213.09895833333334, - 106 + 230, + 170 ], "flags": {}, - "order": 31, + "order": 29, "mode": 0, "inputs": [ { @@ -829,6 +877,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.3.60", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "SamplerCustomAdvanced", "enableTabs": false, "tabWidth": 65, @@ -837,22 +890,21 @@ "secondTabText": "Send Back", "secondTabOffset": 80, "secondTabWidth": 65 - }, - "widgets_values": [] + } }, { "id": 114, "type": "LTXVConditioning", "pos": [ - -1133.7215420073496, - 4141.110347554622 + -1130, + 4140 ], "size": [ - 269.97395833333337, - 78 + 270, + 130 ], "flags": {}, - "order": 27, + "order": 23, "mode": 0, "inputs": [ { @@ -898,6 +950,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.3.56", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "LTXVConditioning", "enableTabs": false, "tabWidth": 65, @@ -915,15 +972,15 @@ "id": 119, "type": "CLIPTextEncode", "pos": [ - -1163.7218246405453, - 3881.1109034489627 + -1160, + 3880 ], "size": [ 400, - 88 + 200 ], "flags": {}, - "order": 12, + "order": 27, "mode": 0, "inputs": [ { @@ -955,6 +1012,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.3.56", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "CLIPTextEncode", "enableTabs": false, "tabWidth": 65, @@ -974,15 +1036,15 @@ "id": 116, "type": "LTXVConcatAVLatent", "pos": [ - -519.7217122979332, - 4701.110031965835 + -520, + 4830 ], "size": [ - 187.5, - 46 + 230, + 100 ], "flags": {}, - "order": 29, + "order": 25, "mode": 0, "inputs": [ { @@ -1012,6 +1074,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "LTXVConcatAVLatent", "enableTabs": false, "tabWidth": 65, @@ -1020,22 +1087,21 @@ "secondTabText": "Send Back", "secondTabOffset": 80, "secondTabWidth": 65 - }, - "widgets_values": [] + } }, { "id": 122, "type": "LTXVSeparateAVLatent", "pos": [ - -393.72183921949465, - 3801.1107787938904 + -380, + 3810 ], "size": [ - 239.97395833333334, - 46 + 240, + 100 ], "flags": {}, - "order": 30, + "order": 28, "mode": 0, "inputs": [ { @@ -1066,6 +1132,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "LTXVSeparateAVLatent", "enableTabs": false, "tabWidth": 65, @@ -1074,22 +1145,21 @@ "secondTabText": "Send Back", "secondTabOffset": 80, "secondTabWidth": 65 - }, - "widgets_values": [] + } }, { "id": 124, "type": "CLIPTextEncode", "pos": [ - -1174.7214530029996, - 3515.1112854387566 + -1170, + 3510 ], "size": [ - 409.97395833333337, - 88 + 410, + 320 ], "flags": {}, - "order": 32, + "order": 30, "mode": 0, "inputs": [ { @@ -1121,6 +1191,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.3.56", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "CLIPTextEncode", "enableTabs": false, "tabWidth": 65, @@ -1140,15 +1215,15 @@ "id": 98, "type": "KSamplerSelect", "pos": [ - 410.27824286284044, - 3981.1101681370833 + 450, + 4070 ], "size": [ - 269.97395833333337, - 58 + 270, + 110 ], "flags": {}, - "order": 3, + "order": 2, "mode": 0, "inputs": [ { @@ -1174,6 +1249,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.3.75", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "KSamplerSelect", "enableTabs": false, "tabWidth": 65, @@ -1191,12 +1271,12 @@ "id": 105, "type": "LoraLoaderModelOnly", "pos": [ - -69.72208571196083, - 3571.110499039739 + -70, + 3570 ], "size": [ - 389.97395833333337, - 82 + 390, + 140 ], "flags": {}, "order": 15, @@ -1215,7 +1295,7 @@ "widget": { "name": "lora_name" }, - "link": 406 + "link": 1773 }, { "localized_name": "strength_model", @@ -1240,21 +1320,26 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.3.75", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "LoraLoaderModelOnly", - "models": [ - { - "name": "ltx-2-19b-distilled-lora-384.safetensors", - "url": "https://huggingface.co/Lightricks/LTX-2/resolve/main/ltx-2-19b-distilled-lora-384.safetensors", - "directory": "loras" - } - ], "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, "hasSecondTab": false, "secondTabText": "Send Back", "secondTabOffset": 80, - "secondTabWidth": 65 + "secondTabWidth": 65, + "models": [ + { + "name": "ltx-2-19b-distilled-lora-384.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2/resolve/main/ltx-2-19b-distilled-lora-384.safetensors", + "directory": "loras" + } + ] }, "widgets_values": [ "ltx-2-19b-distilled-lora-384.safetensors", @@ -1265,15 +1350,15 @@ "id": 95, "type": "LTXVScheduler", "pos": [ - -699.7218704597861, - 3981.1101681370833 + -690, + 4130 ], "size": [ - 269.97395833333337, - 154 + 270, + 170 ], "flags": {}, - "order": 17, + "order": 8, "mode": 0, "inputs": [ { @@ -1342,6 +1427,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.3.56", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "LTXVScheduler", "enableTabs": false, "tabWidth": 65, @@ -1363,15 +1453,15 @@ "id": 126, "type": "RandomNoise", "pos": [ - -697.721823660531, - 3521.1103425011374 + -690, + 3520 ], "size": [ - 269.97395833333337, - 82 + 270, + 110 ], "flags": {}, - "order": 4, + "order": 31, "mode": 0, "inputs": [ { @@ -1381,7 +1471,7 @@ "widget": { "name": "noise_seed" }, - "link": null + "link": 1767 } ], "outputs": [ @@ -1397,6 +1487,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.3.56", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "RandomNoise", "enableTabs": false, "tabWidth": 65, @@ -1408,22 +1503,22 @@ }, "widgets_values": [ 0, - "randomize" + "fixed" ] }, { "id": 107, "type": "SamplerCustomAdvanced", "pos": [ - 710.2782734905775, - 3571.110499039739 + 730, + 3570 ], "size": [ - 212.36979166666669, - 106 + 230, + 170 ], "flags": {}, - "order": 21, + "order": 17, "mode": 0, "inputs": [ { @@ -1476,6 +1571,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.3.75", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "SamplerCustomAdvanced", "enableTabs": false, "tabWidth": 65, @@ -1484,22 +1584,21 @@ "secondTabText": "Send Back", "secondTabOffset": 80, "secondTabWidth": 65 - }, - "widgets_values": [] + } }, { - "id": 143, + "id": 187, "type": "RandomNoise", "pos": [ - 410.27824286284044, - 3571.110499039739 + 450, + 3570 ], "size": [ - 269.97395833333337, - 82 + 270, + 110 ], "flags": {}, - "order": 5, + "order": 3, "mode": 0, "inputs": [ { @@ -1525,6 +1624,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.3.56", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "RandomNoise", "enableTabs": false, "tabWidth": 65, @@ -1543,12 +1647,12 @@ "id": 139, "type": "LTXVAudioVAEDecode", "pos": [ - 1130.2783163694094, - 3841.1107362825187 + 1130, + 3840 ], "size": [ - 239.97395833333334, - 46 + 240, + 100 ], "flags": {}, "order": 35, @@ -1565,7 +1669,7 @@ "localized_name": "audio_vae", "name": "audio_vae", "type": "VAE", - "link": 383 + "link": 340 } ], "outputs": [ @@ -1581,6 +1685,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "LTXVAudioVAEDecode", "enableTabs": false, "tabWidth": 65, @@ -1589,22 +1698,21 @@ "secondTabText": "Send Back", "secondTabOffset": 80, "secondTabWidth": 65 - }, - "widgets_values": [] + } }, { "id": 106, "type": "CreateVideo", "pos": [ - 1420.2783925712918, - 3761.1104019496292 + 1420, + 3760 ], "size": [ - 269.97395833333337, - 78 + 270, + 130 ], "flags": {}, - "order": 20, + "order": 16, "mode": 0, "inputs": [ { @@ -1643,6 +1751,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "CreateVideo", "enableTabs": false, "tabWidth": 65, @@ -1660,15 +1773,15 @@ "id": 134, "type": "LoraLoaderModelOnly", "pos": [ - -1649.721454901846, - 3761.1104019496292 + -1650, + 3750 ], "size": [ - 419.97395833333337, - 82 + 420, + 140 ], "flags": {}, - "order": 13, + "order": 33, "mode": 0, "inputs": [ { @@ -1684,7 +1797,7 @@ "widget": { "name": "lora_name" }, - "link": 404 + "link": 1769 }, { "localized_name": "strength_model", @@ -1710,21 +1823,26 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "LoraLoaderModelOnly", - "models": [ - { - "name": "ltx-2-19b-ic-lora-pose-control.safetensors", - "url": "https://huggingface.co/Lightricks/LTX-2-19b-IC-LoRA-Pose-Control/resolve/main/ltx-2-19b-ic-lora-pose-control.safetensors", - "directory": "loras" - } - ], "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, "hasSecondTab": false, "secondTabText": "Send Back", "secondTabOffset": 80, - "secondTabWidth": 65 + "secondTabWidth": 65, + "models": [ + { + "name": "ltx-2-19b-ic-lora-pose-control.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2-19b-IC-LoRA-Pose-Control/resolve/main/ltx-2-19b-ic-lora-pose-control.safetensors", + "directory": "loras" + } + ] }, "widgets_values": [ "ltx-2-19b-ic-lora-pose-control.safetensors", @@ -1737,12 +1855,12 @@ "id": 138, "type": "LTXVSeparateAVLatent", "pos": [ - 730.2784619127078, - 3731.1109580277 + 740, + 3810 ], "size": [ - 193.2916015625, - 46 + 230, + 100 ], "flags": {}, "order": 34, @@ -1777,6 +1895,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "LTXVSeparateAVLatent", "enableTabs": false, "tabWidth": 65, @@ -1785,22 +1908,21 @@ "secondTabText": "Send Back", "secondTabOffset": 80, "secondTabWidth": 65 - }, - "widgets_values": [] + } }, { - "id": 144, + "id": 188, "type": "VAEDecodeTiled", "pos": [ - 1120.2783619435547, - 3641.110599376351 + 1120, + 3640 ], "size": [ - 269.97395833333337, + 270, 150 ], "flags": {}, - "order": 36, + "order": 38, "mode": 0, "inputs": [ { @@ -1865,6 +1987,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "VAEDecodeTiled", "enableTabs": false, "tabWidth": 65, @@ -1885,15 +2012,15 @@ "id": 113, "type": "VAEDecode", "pos": [ - 1130.2783163694094, - 3531.1113453160738 + 1130, + 3530 ], "size": [ - 239.97395833333334, - 46 + 240, + 100 ], "flags": {}, - "order": 26, + "order": 22, "mode": 0, "inputs": [ { @@ -1920,6 +2047,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.3.75", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "VAEDecode", "enableTabs": false, "tabWidth": 65, @@ -1928,22 +2060,21 @@ "secondTabText": "Send Back", "secondTabOffset": 80, "secondTabWidth": 65 - }, - "widgets_values": [] + } }, { "id": 145, "type": "PrimitiveInt", "pos": [ - -1600, - 4940 + -1610, + 4800 ], "size": [ - 269.97395833333337, - 82 + 270, + 110 ], "flags": {}, - "order": 6, + "order": 4, "mode": 0, "inputs": [ { @@ -1969,6 +2100,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "PrimitiveInt", "enableTabs": false, "tabWidth": 65, @@ -1987,15 +2123,15 @@ "id": 148, "type": "PrimitiveFloat", "pos": [ - -1600, - 5070 + -1610, + 4930 ], "size": [ - 269.97395833333337, - 58 + 270, + 110 ], "flags": {}, - "order": 7, + "order": 5, "mode": 0, "inputs": [ { @@ -2022,6 +2158,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "PrimitiveFloat", "enableTabs": false, "tabWidth": 65, @@ -2035,19 +2176,105 @@ 24 ] }, + { + "id": 115, + "type": "EmptyLTXVLatentVideo", + "pos": [ + -1100, + 4740 + ], + "size": [ + 270, + 200 + ], + "flags": {}, + "order": 24, + "mode": 0, + "inputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 296 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 297 + }, + { + "localized_name": "length", + "name": "length", + "type": "INT", + "widget": { + "name": "length" + }, + "link": 330 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "links": [ + 360 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.60", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "EmptyLTXVLatentVideo", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 768, + 512, + 97, + 1 + ] + }, { "id": 118, "type": "Reroute", "pos": [ - -229.7217758812614, - 4211.111007032079 + -350, + 3980 ], "size": [ - 75, - 26 + 230, + 40 ], "flags": {}, - "order": 14, + "order": 26, "mode": 0, "inputs": [ { @@ -2069,22 +2296,29 @@ ], "properties": { "showOutputText": false, - "horizontal": false + "horizontal": false, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } } }, { - "id": 151, + "id": 189, "type": "LTXVImgToVideoInplace", "pos": [ - -19.72161465663438, - 4071.1107364662485 + 180, + 4040 ], "size": [ - 269.97395833333337, - 122 + 260, + 190 ], - "flags": {}, - "order": 38, + "flags": { + "collapsed": false + }, + "order": 39, "mode": 0, "inputs": [ { @@ -2097,7 +2331,7 @@ "localized_name": "image", "name": "image", "type": "IMAGE", - "link": 398 + "link": 379 }, { "localized_name": "latent", @@ -2112,7 +2346,7 @@ "widget": { "name": "strength" }, - "link": 371 + "link": 1759 }, { "localized_name": "bypass", @@ -2137,6 +2371,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "LTXVImgToVideoInplace", "enableTabs": false, "tabWidth": 65, @@ -2155,15 +2394,15 @@ "id": 104, "type": "LTXVCropGuides", "pos": [ - -9.721939801202097, - 3841.1107362825187 + -90, + 4210 ], "size": [ - 239.97395833333334, - 66 + 240, + 120 ], "flags": {}, - "order": 19, + "order": 14, "mode": 0, "inputs": [ { @@ -2215,6 +2454,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.3.68", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "LTXVCropGuides", "enableTabs": false, "tabWidth": 65, @@ -2223,22 +2467,21 @@ "secondTabText": "Send Back", "secondTabOffset": 80, "secondTabWidth": 65 - }, - "widgets_values": [] + } }, { "id": 112, "type": "LTXVLatentUpsampler", "pos": [ - -9.721939801202097, - 3961.111517352274 + -90, + 4030 ], "size": [ - 259.97395833333337, - 66 + 260, + 120 ], "flags": {}, - "order": 25, + "order": 21, "mode": 0, "inputs": [ { @@ -2274,6 +2517,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "LTXVLatentUpsampler", "enableTabs": false, "tabWidth": 65, @@ -2282,22 +2530,117 @@ "secondTabText": "Send Back", "secondTabOffset": 80, "secondTabWidth": 65 + } + }, + { + "id": 154, + "type": "MarkdownNote", + "pos": [ + -1640, + 5050 + ], + "size": [ + 350, + 170 + ], + "flags": { + "collapsed": false }, - "widgets_values": [] + "order": 6, + "mode": 0, + "inputs": [], + "outputs": [], + "title": "Frame Rate Note", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "Please make sure the frame rate value is the same in both boxes" + ], + "color": "#432", + "bgcolor": "#653" + }, + { + "id": 96, + "type": "LTXVAudioVAELoader", + "pos": [ + -1650, + 3970 + ], + "size": [ + 420, + 110 + ], + "flags": {}, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "ckpt_name", + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": 1770 + } + ], + "outputs": [ + { + "localized_name": "Audio VAE", + "name": "Audio VAE", + "type": "VAE", + "links": [ + 285, + 340 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.68", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "LTXVAudioVAELoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "ltx-2-19b-dev-fp8.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2/resolve/main/ltx-2-19b-dev-fp8.safetensors", + "directory": "checkpoints" + } + ] + }, + "widgets_values": [ + "ltx-2-19b-dev-fp8.safetensors" + ] }, { "id": 97, "type": "LTXAVTextEncoderLoader", "pos": [ - -1649.721454901846, - 4041.1110828665023 + -1650, + 4160 ], "size": [ - 419.97395833333337, - 106 + 420, + 150 ], "flags": {}, - "order": 8, + "order": 10, "mode": 0, "inputs": [ { @@ -2307,7 +2650,7 @@ "widget": { "name": "text_encoder" }, - "link": 405 + "link": 1772 }, { "localized_name": "ckpt_name", @@ -2316,7 +2659,7 @@ "widget": { "name": "ckpt_name" }, - "link": 403 + "link": 1771 }, { "localized_name": "device", @@ -2342,7 +2685,19 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "LTXAVTextEncoderLoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, "models": [ { "name": "ltx-2-19b-dev-fp8.safetensors", @@ -2354,17 +2709,10 @@ "url": "https://huggingface.co/Comfy-Org/ltx-2/resolve/main/split_files/text_encoders/gemma_3_12B_it_fp4_mixed.safetensors", "directory": "text_encoders" } - ], - "enableTabs": false, - "tabWidth": 65, - "tabXOffset": 10, - "hasSecondTab": false, - "secondTabText": "Send Back", - "secondTabOffset": 80, - "secondTabWidth": 65 + ] }, "widgets_values": [ - "ltx-2-19b-ic-lora-pose-control.safetensors", + "gemma_3_12B_it_fp4_mixed.safetensors", "ltx-2-19b-dev-fp8.safetensors", "default" ] @@ -2373,15 +2721,15 @@ "id": 103, "type": "CheckpointLoaderSimple", "pos": [ - -1649.721454901846, - 3591.1104777840524 + -1650, + 3520 ], "size": [ - 419.97395833333337, - 98 + 420, + 160 ], "flags": {}, - "order": 9, + "order": 13, "mode": 0, "inputs": [ { @@ -2391,7 +2739,7 @@ "widget": { "name": "ckpt_name" }, - "link": 401 + "link": 1768 } ], "outputs": [ @@ -2424,137 +2772,89 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.3.56", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "CheckpointLoaderSimple", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, "models": [ { "name": "ltx-2-19b-dev-fp8.safetensors", "url": "https://huggingface.co/Lightricks/LTX-2/resolve/main/ltx-2-19b-dev-fp8.safetensors", "directory": "checkpoints" } - ], - "enableTabs": false, - "tabWidth": 65, - "tabXOffset": 10, - "hasSecondTab": false, - "secondTabText": "Send Back", - "secondTabOffset": 80, - "secondTabWidth": 65 + ] }, "widgets_values": [ "ltx-2-19b-dev-fp8.safetensors" ] }, { - "id": 156, - "type": "LTXVAudioVAELoader", + "id": 110, + "type": "GetImageSize", "pos": [ - -1636.9543279290153, - 3911.095334870057 + -1610, + 4630 ], "size": [ - 399.0494791666667, - 58 + 260, + 120 ], "flags": {}, - "order": 10, + "order": 19, "mode": 0, "inputs": [ - { - "localized_name": "ckpt_name", - "name": "ckpt_name", - "type": "COMBO", - "widget": { - "name": "ckpt_name" - }, - "link": 402 - } - ], - "outputs": [ - { - "localized_name": "Audio VAE", - "name": "Audio VAE", - "type": "VAE", - "links": [ - 382, - 383 - ] - } - ], - "properties": { - "cnr_id": "comfy-core", - "ver": "0.11.0", - "Node name for S&R": "LTXVAudioVAELoader" - }, - "widgets_values": [ - "ltx-2-19b-dev-fp8.safetensors" - ] - }, - { - "id": 149, - "type": "LTXVImgToVideoInplace", - "pos": [ - -1089.7215608128167, - 4401.110560478942 - ], - "size": [ - 269.97395833333337, - 122 - ], - "flags": {}, - "order": 37, - "mode": 0, - "inputs": [ - { - "localized_name": "vae", - "name": "vae", - "type": "VAE", - "link": 359 - }, { "localized_name": "image", "name": "image", "type": "IMAGE", - "link": 399 - }, - { - "localized_name": "latent", - "name": "latent", - "type": "LATENT", - "link": 360 - }, - { - "localized_name": "strength", - "name": "strength", - "type": "FLOAT", - "widget": { - "name": "strength" - }, - "link": 370 - }, - { - "localized_name": "bypass", - "name": "bypass", - "type": "BOOLEAN", - "widget": { - "name": "bypass" - }, - "link": 363 + "link": 381 } ], "outputs": [ { - "localized_name": "latent", - "name": "latent", - "type": "LATENT", + "localized_name": "width", + "name": "width", + "type": "INT", "links": [ - 357 + 296 + ] + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "links": [ + 297 + ] + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "links": [ + 329, + 330 ] } ], "properties": { "cnr_id": "comfy-core", "ver": "0.7.0", - "Node name for S&R": "LTXVImgToVideoInplace", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "GetImageSize", "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -2562,25 +2862,21 @@ "secondTabText": "Send Back", "secondTabOffset": 80, "secondTabWidth": 65 - }, - "widgets_values": [ - 1, - false - ] + } }, { "id": 132, "type": "LTXVAddGuide", "pos": [ - -599.7217670603999, - 4421.110609115862 + -600, + 4550 ], "size": [ - 269.97395833333337, - 162 + 270, + 240 ], "flags": {}, - "order": 33, + "order": 32, "mode": 0, "inputs": [ { @@ -2611,7 +2907,7 @@ "localized_name": "image", "name": "image", "type": "IMAGE", - "link": 395 + "link": 348 }, { "localized_name": "frame_idx", @@ -2663,6 +2959,11 @@ "properties": { "cnr_id": "comfy-core", "ver": "0.3.75", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, "Node name for S&R": "LTXVAddGuide", "enableTabs": false, "tabWidth": 65, @@ -2678,114 +2979,76 @@ ] }, { - "id": 154, - "type": "MarkdownNote", + "id": 149, + "type": "LTXVImgToVideoInplace", "pos": [ - -1630, - 5190 + -1090, + 4530 ], "size": [ - 350, - 88 - ], - "flags": { - "collapsed": false - }, - "order": 11, - "mode": 0, - "inputs": [], - "outputs": [], - "title": "Frame Rate Note", - "properties": {}, - "widgets_values": [ - "Please make sure the frame rate value is the same in both boxes" - ], - "color": "#432", - "bgcolor": "#653" - }, - { - "id": 159, - "type": "ResizeImageMaskNode", - "pos": [ - -1610, - 4580 - ], - "size": [ - 284.375, - 154 + 270, + 180 ], "flags": {}, - "order": 39, + "order": 36, "mode": 0, "inputs": [ { - "localized_name": "input", - "name": "input", - "type": "IMAGE,MASK", - "link": 400 + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 359 }, { - "localized_name": "resize_type", - "name": "resize_type", - "type": "COMFY_DYNAMICCOMBO_V3", + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 364 + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "link": 360 + }, + { + "localized_name": "strength", + "name": "strength", + "type": "FLOAT", "widget": { - "name": "resize_type" + "name": "strength" }, - "link": null + "link": 1758 }, { - "localized_name": "width", - "name": "resize_type.width", - "type": "INT", + "localized_name": "bypass", + "name": "bypass", + "type": "BOOLEAN", "widget": { - "name": "resize_type.width" - }, - "link": 408 - }, - { - "localized_name": "height", - "name": "resize_type.height", - "type": "INT", - "widget": { - "name": "resize_type.height" - }, - "link": 409 - }, - { - "localized_name": "crop", - "name": "resize_type.crop", - "type": "COMBO", - "widget": { - "name": "resize_type.crop" - }, - "link": null - }, - { - "localized_name": "scale_method", - "name": "scale_method", - "type": "COMBO", - "widget": { - "name": "scale_method" + "name": "bypass" }, "link": null } ], "outputs": [ { - "localized_name": "resized", - "name": "resized", - "type": "IMAGE,MASK", + "localized_name": "latent", + "name": "latent", + "type": "LATENT", "links": [ - 391, - 392, - 395 + 357 ] } ], "properties": { "cnr_id": "comfy-core", "ver": "0.7.0", - "Node name for S&R": "ResizeImageMaskNode", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "LTXVImgToVideoInplace", "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -2795,139 +3058,69 @@ "secondTabWidth": 65 }, "widgets_values": [ - "scale dimensions", - 1280, - 720, - "center", - "lanczos" + 1, + false ] }, { - "id": 110, - "type": "GetImageSize", + "id": 155, + "type": "ImageScaleBy", "pos": [ - -1600, - 4780 + -1620, + 4440 ], "size": [ - 259.97395833333337, - 66 + 280, + 140 ], "flags": {}, - "order": 23, + "order": 37, "mode": 0, "inputs": [ { "localized_name": "image", "name": "image", "type": "IMAGE", - "link": 391 - } - ], - "outputs": [ - { - "localized_name": "width", - "name": "width", - "type": "INT", - "links": [ - 296 - ] + "link": 380 }, { - "localized_name": "height", - "name": "height", - "type": "INT", - "links": [ - 297 - ] - }, - { - "localized_name": "batch_size", - "name": "batch_size", - "type": "INT", - "links": [] - } - ], - "properties": { - "cnr_id": "comfy-core", - "ver": "0.7.0", - "Node name for S&R": "GetImageSize", - "enableTabs": false, - "tabWidth": 65, - "tabXOffset": 10, - "hasSecondTab": false, - "secondTabText": "Send Back", - "secondTabOffset": 80, - "secondTabWidth": 65 - }, - "widgets_values": [] - }, - { - "id": 115, - "type": "EmptyLTXVLatentVideo", - "pos": [ - -1099.721794809093, - 4611.11072170357 - ], - "size": [ - 269.97395833333337, - 130 - ], - "flags": {}, - "order": 28, - "mode": 0, - "inputs": [ - { - "localized_name": "width", - "name": "width", - "type": "INT", + "localized_name": "upscale_method", + "name": "upscale_method", + "type": "COMBO", "widget": { - "name": "width" + "name": "upscale_method" }, - "link": 296 + "link": null }, { - "localized_name": "height", - "name": "height", - "type": "INT", + "localized_name": "scale_by", + "name": "scale_by", + "type": "FLOAT", "widget": { - "name": "height" - }, - "link": 297 - }, - { - "localized_name": "length", - "name": "length", - "type": "INT", - "widget": { - "name": "length" - }, - "link": 410 - }, - { - "localized_name": "batch_size", - "name": "batch_size", - "type": "INT", - "widget": { - "name": "batch_size" + "name": "scale_by" }, "link": null } ], "outputs": [ { - "localized_name": "LATENT", - "name": "LATENT", - "type": "LATENT", + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", "links": [ - 360 + 381 ] } ], "properties": { "cnr_id": "comfy-core", - "ver": "0.3.60", - "Node name for S&R": "EmptyLTXVLatentVideo", + "ver": "0.5.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ImageScaleBy", "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -2937,87 +3130,8 @@ "secondTabWidth": 65 }, "widgets_values": [ - 768, - 512, - 97, - 1 - ] - }, - { - "id": 111, - "type": "LTXVEmptyLatentAudio", - "pos": [ - -1099.721794809093, - 4811.110229576288 - ], - "size": [ - 269.97395833333337, - 106 - ], - "flags": {}, - "order": 24, - "mode": 0, - "inputs": [ - { - "localized_name": "audio_vae", - "name": "audio_vae", - "type": "VAE", - "link": 382 - }, - { - "localized_name": "frames_number", - "name": "frames_number", - "type": "INT", - "widget": { - "name": "frames_number" - }, - "link": null - }, - { - "localized_name": "frame_rate", - "name": "frame_rate", - "type": "INT", - "widget": { - "name": "frame_rate" - }, - "link": 354 - }, - { - "localized_name": "batch_size", - "name": "batch_size", - "type": "INT", - "widget": { - "name": "batch_size" - }, - "link": null - } - ], - "outputs": [ - { - "localized_name": "Latent", - "name": "Latent", - "type": "LATENT", - "links": [ - 300 - ] - } - ], - "properties": { - "cnr_id": "comfy-core", - "ver": "0.3.68", - "Node name for S&R": "LTXVEmptyLatentAudio", - "enableTabs": false, - "tabWidth": 65, - "tabXOffset": 10, - "hasSecondTab": false, - "secondTabText": "Send Back", - "secondTabOffset": 80, - "secondTabWidth": 65 - }, - "widgets_values": [ - 97, - 25, - 1 + "lanczos", + 0.5 ] } ], @@ -3028,8 +3142,8 @@ "bounding": [ -1660, 3440, - 440, - 820 + 450, + 940 ], "color": "#3f789e", "font_size": 24, @@ -3041,8 +3155,8 @@ "bounding": [ -700, 3440, - 570, - 820 + 580, + 940 ], "color": "#3f789e", "font_size": 24, @@ -3054,8 +3168,8 @@ "bounding": [ -1180, 3440, - 440, - 820 + 450, + 940 ], "color": "#3f789e", "font_size": 24, @@ -3066,7 +3180,7 @@ "title": "Latent", "bounding": [ -1180, - 4290, + 4420, 1050, 680 ], @@ -3080,8 +3194,8 @@ "bounding": [ -100, 3440, - 1090, - 820 + 1110, + 940 ], "color": "#3f789e", "font_size": 24, @@ -3091,10 +3205,10 @@ "id": 6, "title": "Sampler", "bounding": [ - 350, + 410, 3480, - 620, - 750 + 590, + 880 ], "color": "#3f789e", "font_size": 24, @@ -3106,8 +3220,8 @@ "bounding": [ -90, 3480, - 430, - 310 + 450, + 480 ], "color": "#3f789e", "font_size": 24, @@ -3117,8 +3231,8 @@ "id": 11, "title": "Frame rate", "bounding": [ - -1610, - 4860, + -1620, + 4730, 290, 271.6 ], @@ -3184,6 +3298,22 @@ "target_slot": 2, "type": "CONDITIONING" }, + { + "id": 285, + "origin_id": 96, + "origin_slot": 0, + "target_id": 111, + "target_slot": 0, + "type": "VAE" + }, + { + "id": 329, + "origin_id": 110, + "origin_slot": 2, + "target_id": 111, + "target_slot": 1, + "type": "INT" + }, { "id": 260, "origin_id": 126, @@ -3240,6 +3370,14 @@ "target_slot": 1, "type": "INT" }, + { + "id": 330, + "origin_id": 110, + "origin_slot": 2, + "target_id": 115, + "target_slot": 2, + "type": "INT" + }, { "id": 325, "origin_id": 103, @@ -3360,6 +3498,14 @@ "target_slot": 0, "type": "LATENT" }, + { + "id": 340, + "origin_id": 96, + "origin_slot": 0, + "target_id": 139, + "target_slot": 1, + "type": "VAE" + }, { "id": 337, "origin_id": 138, @@ -3490,23 +3636,31 @@ }, { "id": 347, - "origin_id": 143, + "origin_id": 187, "origin_slot": 0, "target_id": 107, "target_slot": 0, "type": "NOISE" }, + { + "id": 348, + "origin_id": -10, + "origin_slot": 1, + "target_id": 132, + "target_slot": 4, + "type": "IMAGE" + }, { "id": 351, "origin_id": 138, "origin_slot": 0, - "target_id": 144, + "target_id": 188, "target_slot": 0, "type": "LATENT" }, { "id": 352, - "origin_id": 144, + "origin_id": 188, "origin_slot": 0, "target_id": 106, "target_slot": 0, @@ -3516,7 +3670,7 @@ "id": 353, "origin_id": 103, "origin_slot": 2, - "target_id": 144, + "target_id": 188, "target_slot": 1, "type": "VAE" }, @@ -3569,16 +3723,16 @@ "type": "LATENT" }, { - "id": 363, + "id": 364, "origin_id": -10, "origin_slot": 2, "target_id": 149, - "target_slot": 4, - "type": "BOOLEAN" + "target_slot": 1, + "type": "IMAGE" }, { "id": 365, - "origin_id": 151, + "origin_id": 189, "origin_slot": 0, "target_id": 101, "target_slot": 0, @@ -3588,7 +3742,7 @@ "id": 366, "origin_id": 112, "origin_slot": 0, - "target_id": 151, + "target_id": 189, "target_slot": 2, "type": "LATENT" }, @@ -3596,92 +3750,68 @@ "id": 367, "origin_id": 118, "origin_slot": 0, - "target_id": 151, + "target_id": 189, "target_slot": 0, "type": "VAE" }, { "id": 368, "origin_id": -10, - "origin_slot": 2, - "target_id": 151, + "origin_slot": 4, + "target_id": 189, "target_slot": 4, "type": "BOOLEAN" }, { - "id": 370, + "id": 379, "origin_id": -10, - "origin_slot": 1, - "target_id": 149, - "target_slot": 3, - "type": "FLOAT" - }, - { - "id": 371, - "origin_id": -10, - "origin_slot": 1, - "target_id": 151, - "target_slot": 3, - "type": "FLOAT" - }, - { - "id": 382, - "origin_id": 156, - "origin_slot": 0, - "target_id": 111, - "target_slot": 0, - "type": "VAE" - }, - { - "id": 383, - "origin_id": 156, - "origin_slot": 0, - "target_id": 139, + "origin_slot": 2, + "target_id": 189, "target_slot": 1, - "type": "VAE" + "type": "IMAGE" }, { - "id": 391, - "origin_id": 159, + "id": 380, + "origin_id": -10, + "origin_slot": 1, + "target_id": 155, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 381, + "origin_id": 155, "origin_slot": 0, "target_id": 110, "target_slot": 0, "type": "IMAGE" }, { - "id": 395, - "origin_id": 159, - "origin_slot": 0, - "target_id": 132, - "target_slot": 4, - "type": "IMAGE" - }, - { - "id": 398, - "origin_id": -10, - "origin_slot": 3, - "target_id": 151, - "target_slot": 1, - "type": "IMAGE" - }, - { - "id": 399, + "id": 1758, "origin_id": -10, "origin_slot": 3, "target_id": 149, - "target_slot": 1, - "type": "IMAGE" + "target_slot": 3, + "type": "FLOAT" }, { - "id": 400, + "id": 1759, + "origin_id": -10, + "origin_slot": 3, + "target_id": 189, + "target_slot": 3, + "type": "FLOAT" + }, + { + "id": 1767, "origin_id": -10, "origin_slot": 4, - "target_id": 159, + "target_id": 126, "target_slot": 0, - "type": "IMAGE,MASK" + "type": "INT" }, { - "id": 401, + "id": 1768, "origin_id": -10, "origin_slot": 5, "target_id": 103, @@ -3689,23 +3819,7 @@ "type": "COMBO" }, { - "id": 402, - "origin_id": -10, - "origin_slot": 5, - "target_id": 156, - "target_slot": 0, - "type": "COMBO" - }, - { - "id": 403, - "origin_id": -10, - "origin_slot": 5, - "target_id": 97, - "target_slot": 1, - "type": "COMBO" - }, - { - "id": 404, + "id": 1769, "origin_id": -10, "origin_slot": 6, "target_id": 134, @@ -3713,76 +3827,55 @@ "type": "COMBO" }, { - "id": 405, + "id": 1770, "origin_id": -10, - "origin_slot": 6, + "origin_slot": 5, + "target_id": 96, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 1771, + "origin_id": -10, + "origin_slot": 5, + "target_id": 97, + "target_slot": 1, + "type": "COMBO" + }, + { + "id": 1772, + "origin_id": -10, + "origin_slot": 7, "target_id": 97, "target_slot": 0, "type": "COMBO" }, { - "id": 406, + "id": 1773, "origin_id": -10, - "origin_slot": 7, + "origin_slot": 8, "target_id": 105, "target_slot": 1, "type": "COMBO" }, { - "id": 407, + "id": 1774, "origin_id": -10, - "origin_slot": 8, + "origin_slot": 9, "target_id": 100, "target_slot": 0, "type": "COMBO" - }, - { - "id": 408, - "origin_id": -10, - "origin_slot": 9, - "target_id": 159, - "target_slot": 2, - "type": "INT" - }, - { - "id": 409, - "origin_id": -10, - "origin_slot": 10, - "target_id": 159, - "target_slot": 3, - "type": "INT" - }, - { - "id": 410, - "origin_id": -10, - "origin_slot": 11, - "target_id": 115, - "target_slot": 2, - "type": "INT" } ], "extra": { "workflowRendererVersion": "LG" }, - "category": "Video generation and editing/Pose to video" + "category": "Video generation and editing/Pose to video", + "description": "Generates video from pose reference frames using LTX-2, with optional synchronized audio." } ] }, - "config": {}, "extra": { - "ds": { - "scale": 1.3889423076923078, - "offset": [ - 217.0560747663551, - -3703.3333333333335 - ] - }, - "frontendVersion": "1.37.10", - "workflowRendererVersion": "LG", - "VHS_latentpreview": false, - "VHS_latentpreviewrate": 0, - "VHS_MetadataImage": true, - "VHS_KeepIntermediate": true - }, - "version": 0.4 -} + "ue_links": [] + } +} \ No newline at end of file diff --git a/blueprints/Prompt Enhance.json b/blueprints/Prompt Enhance.json index 5e57548ff..e260b1203 100644 --- a/blueprints/Prompt Enhance.json +++ b/blueprints/Prompt Enhance.json @@ -270,9 +270,10 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Text generation/Prompt enhance" + "category": "Text generation/Prompt enhance", + "description": "Expands short text prompts into detailed descriptions using a text generation model for better generation quality." } ] }, "extra": {} -} +} \ No newline at end of file diff --git a/blueprints/Remove Background (BiRefNet).json b/blueprints/Remove Background (BiRefNet).json new file mode 100644 index 000000000..732a4adc4 --- /dev/null +++ b/blueprints/Remove Background (BiRefNet).json @@ -0,0 +1,397 @@ +{ + "revision": 0, + "last_node_id": 19, + "last_link_id": 0, + "nodes": [ + { + "id": 19, + "type": "5b40ca21-ba1a-41d5-b403-4d2d7acdc195", + "pos": [ + -6411.330578108367, + 1940.2638932730042 + ], + "size": [ + 349.609375, + 145.9375 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": null + }, + { + "name": "bg_removal_name", + "type": "COMBO", + "widget": { + "name": "bg_removal_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [] + }, + { + "name": "mask", + "type": "MASK", + "links": [] + } + ], + "properties": { + "proxyWidgets": [ + [ + "14", + "bg_removal_name" + ] + ] + }, + "widgets_values": [], + "title": "Remove Background (BiRefNet)" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "5b40ca21-ba1a-41d5-b403-4d2d7acdc195", + "version": 1, + "state": { + "lastGroupId": 0, + "lastNodeId": 21, + "lastLinkId": 16, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Remove Background (BiRefNet)", + "description": "Removes or replaces image backgrounds using BiRefNet segmentation and alpha compositing.", + "inputNode": { + "id": -10, + "bounding": [ + -6728.534070722246, + 1475.2619799128663, + 150.9140625, + 88 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + -6169.049695722246, + 1475.2619799128663, + 128, + 88 + ] + }, + "inputs": [ + { + "id": "7bc321cd-df31-4c39-aaf7-7f0d01326189", + "name": "image", + "type": "IMAGE", + "linkIds": [ + 5, + 7 + ], + "localized_name": "image", + "pos": [ + -6601.620008222246, + 1499.2619799128663 + ] + }, + { + "id": "e89d2cd8-daa3-4e29-8a69-851db85072cb", + "name": "bg_removal_name", + "type": "COMBO", + "linkIds": [ + 12 + ], + "pos": [ + -6601.620008222246, + 1519.2619799128663 + ] + } + ], + "outputs": [ + { + "id": "16e7863c-4c38-46c2-aa74-e82991fbfe8d", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 8 + ], + "localized_name": "IMAGE", + "pos": [ + -6145.049695722246, + 1499.2619799128663 + ] + }, + { + "id": "f7240c19-5b80-406e-a8e2-9b12440ee2d6", + "name": "mask", + "type": "MASK", + "linkIds": [ + 11 + ], + "pos": [ + -6145.049695722246, + 1519.2619799128663 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 13, + "type": "RemoveBackground", + "pos": [ + -6536.764823982709, + 1444.9963409012412 + ], + "size": [ + 302.25, + 72 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 5 + }, + { + "localized_name": "bg_removal_model", + "name": "bg_removal_model", + "type": "BACKGROUND_REMOVAL", + "link": 3 + } + ], + "outputs": [ + { + "localized_name": "mask", + "name": "mask", + "type": "MASK", + "links": [ + 4, + 11 + ] + } + ], + "properties": { + "Node name for S&R": "RemoveBackground" + } + }, + { + "id": 14, + "type": "LoadBackgroundRemovalModel", + "pos": [ + -6540.534070722246, + 1302.223464635445 + ], + "size": [ + 311.484375, + 85.515625 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "bg_removal_name", + "name": "bg_removal_name", + "type": "COMBO", + "widget": { + "name": "bg_removal_name" + }, + "link": 12 + } + ], + "outputs": [ + { + "localized_name": "bg_model", + "name": "bg_model", + "type": "BACKGROUND_REMOVAL", + "links": [ + 3 + ] + } + ], + "properties": { + "Node name for S&R": "LoadBackgroundRemovalModel", + "models": [ + { + "name": "birefnet.safetensors", + "url": "https://huggingface.co/Comfy-Org/BiRefNet/resolve/main/background_removal/birefnet.safetensors", + "directory": "background_removal" + } + ] + }, + "widgets_values": [ + "birefnet.safetensors" + ] + }, + { + "id": 15, + "type": "InvertMask", + "pos": [ + -6532.446160529669, + 1571.1111286839914 + ], + "size": [ + 285.984375, + 48 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "mask", + "name": "mask", + "type": "MASK", + "link": 4 + } + ], + "outputs": [ + { + "localized_name": "MASK", + "name": "MASK", + "type": "MASK", + "links": [ + 6 + ] + } + ], + "properties": { + "Node name for S&R": "InvertMask" + } + }, + { + "id": 16, + "type": "JoinImageWithAlpha", + "pos": [ + -6527.4370171636665, + 1674.3004951902876 + ], + "size": [ + 284.96875, + 72 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 7 + }, + { + "localized_name": "alpha", + "name": "alpha", + "type": "MASK", + "link": 6 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 8 + ] + } + ], + "properties": { + "Node name for S&R": "JoinImageWithAlpha" + } + } + ], + "groups": [], + "links": [ + { + "id": 3, + "origin_id": 14, + "origin_slot": 0, + "target_id": 13, + "target_slot": 1, + "type": "BACKGROUND_REMOVAL" + }, + { + "id": 4, + "origin_id": 13, + "origin_slot": 0, + "target_id": 15, + "target_slot": 0, + "type": "MASK" + }, + { + "id": 6, + "origin_id": 15, + "origin_slot": 0, + "target_id": 16, + "target_slot": 1, + "type": "MASK" + }, + { + "id": 5, + "origin_id": -10, + "origin_slot": 0, + "target_id": 13, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 7, + "origin_id": -10, + "origin_slot": 0, + "target_id": 16, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 8, + "origin_id": 16, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 11, + "origin_id": 13, + "origin_slot": 0, + "target_id": -20, + "target_slot": 1, + "type": "MASK" + }, + { + "id": 12, + "origin_id": -10, + "origin_slot": 1, + "target_id": 14, + "target_slot": 0, + "type": "COMBO" + } + ], + "extra": {}, + "category": "Image generation and editing/Background Removal" + } + ] + }, + "extra": {} +} \ No newline at end of file diff --git a/blueprints/Sharpen.json b/blueprints/Sharpen.json index bb79f61fc..3c4099c6b 100644 --- a/blueprints/Sharpen.json +++ b/blueprints/Sharpen.json @@ -267,7 +267,7 @@ "Node name for S&R": "GLSLShader" }, "widgets_values": [ - "#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\nuniform vec2 u_resolution;\nuniform float u_float0; // strength [0.0 – 2.0] typical: 0.3–1.0\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\n\nvoid main() {\n vec2 texel = 1.0 / u_resolution;\n \n // Sample center and neighbors\n vec4 center = texture(u_image0, v_texCoord);\n vec4 top = texture(u_image0, v_texCoord + vec2( 0.0, -texel.y));\n vec4 bottom = texture(u_image0, v_texCoord + vec2( 0.0, texel.y));\n vec4 left = texture(u_image0, v_texCoord + vec2(-texel.x, 0.0));\n vec4 right = texture(u_image0, v_texCoord + vec2( texel.x, 0.0));\n \n // Edge enhancement (Laplacian)\n vec4 edges = center * 4.0 - top - bottom - left - right;\n \n // Add edges back scaled by strength\n vec4 sharpened = center + edges * u_float0;\n \n fragColor0 = vec4(clamp(sharpened.rgb, 0.0, 1.0), center.a);\n}", + "#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\nuniform float u_float0; // strength [0.0 – 2.0] typical: 0.3–1.0\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\n\nvoid main() {\n vec2 texel = 1.0 / vec2(textureSize(u_image0, 0));\n \n // Sample center and neighbors\n vec4 center = texture(u_image0, v_texCoord);\n vec4 top = texture(u_image0, v_texCoord + vec2( 0.0, -texel.y));\n vec4 bottom = texture(u_image0, v_texCoord + vec2( 0.0, texel.y));\n vec4 left = texture(u_image0, v_texCoord + vec2(-texel.x, 0.0));\n vec4 right = texture(u_image0, v_texCoord + vec2( texel.x, 0.0));\n \n // Edge enhancement (Laplacian)\n vec4 edges = center * 4.0 - top - bottom - left - right;\n \n // Add edges back scaled by strength\n vec4 sharpened = center + edges * u_float0;\n \n fragColor0 = vec4(clamp(sharpened.rgb, 0.0, 1.0), center.a);\n}", "from_input" ] } @@ -302,8 +302,9 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image Tools/Sharpen" + "category": "Image Tools/Sharpen", + "description": "Sharpens image details using a GPU fragment shader for enhanced clarity." } ] } -} +} \ No newline at end of file diff --git a/blueprints/Text to Audio (ACE-Step 1.5).json b/blueprints/Text to Audio (ACE-Step 1.5).json index 206cf16be..5b8b8626f 100644 --- a/blueprints/Text to Audio (ACE-Step 1.5).json +++ b/blueprints/Text to Audio (ACE-Step 1.5).json @@ -222,7 +222,7 @@ }, "revision": 0, "config": {}, - "name": "local-Text to Audio (ACE-Step 1.5)", + "name": "Text to Audio (ACE-Step 1.5)", "inputNode": { "id": -10, "bounding": [ @@ -1502,7 +1502,8 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Audio/Music generation" + "category": "Audio/Music generation", + "description": "Generates audio/music from text prompts using ACE-Step 1.5, a diffusion-based audio generation model." } ] }, @@ -1518,4 +1519,4 @@ } }, "version": 0.4 -} +} \ No newline at end of file diff --git a/blueprints/Text to Image (Ernie Image Turbo).json b/blueprints/Text to Image (Ernie Image Turbo).json new file mode 100644 index 000000000..4ecdd1883 --- /dev/null +++ b/blueprints/Text to Image (Ernie Image Turbo).json @@ -0,0 +1,2112 @@ +{ + "revision": 0, + "last_node_id": 88, + "last_link_id": 0, + "nodes": [ + { + "id": 88, + "type": "2a4f0815-c4d2-4e8b-9bdf-991a8403889d", + "pos": [ + -120, + 240 + ], + "size": [ + 400, + 540 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "label": "prompt", + "name": "value", + "type": "STRING", + "widget": { + "name": "value" + }, + "link": null + }, + { + "label": "prompt_enhancement", + "name": "value_1", + "type": "BOOLEAN", + "widget": { + "name": "value_1" + }, + "link": null + }, + { + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": null + }, + { + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": null + }, + { + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": null + }, + { + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": null + }, + { + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": null + }, + { + "label": "prompt_enhancer", + "name": "clip_name_1", + "type": "COMBO", + "widget": { + "name": "clip_name_1" + }, + "link": null + }, + { + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [] + } + ], + "properties": { + "proxyWidgets": [ + [ + "94", + "value" + ], + [ + "96", + "value" + ], + [ + "71", + "width" + ], + [ + "71", + "height" + ], + [ + "70", + "seed" + ], + [ + "66", + "unet_name" + ], + [ + "62", + "clip_name" + ], + [ + "98", + "clip_name" + ], + [ + "63", + "vae_name" + ] + ], + "cnr_id": "comfy-core", + "ver": "0.18.1", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": { + "value": true, + "value_1": true + }, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [], + "title": "Text to Image (Ernie Image Turbo)" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "2a4f0815-c4d2-4e8b-9bdf-991a8403889d", + "version": 1, + "state": { + "lastGroupId": 7, + "lastNodeId": 103, + "lastLinkId": 134, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Text to Image (Ernie Image Turbo)", + "inputNode": { + "id": -10, + "bounding": [ + -1350, + 370, + 163.50390625, + 220 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 1110, + 260, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "74a4609c-67df-4ae9-ab96-9ff4e3a1c3b1", + "name": "value", + "type": "STRING", + "linkIds": [ + 128 + ], + "label": "prompt", + "pos": [ + -1206.49609375, + 390 + ] + }, + { + "id": "996f1854-7ae3-450e-821c-a9b5b7c310f9", + "name": "value_1", + "type": "BOOLEAN", + "linkIds": [ + 127 + ], + "label": "prompt_enhancement", + "pos": [ + -1206.49609375, + 410 + ] + }, + { + "id": "71e9c6e8-4285-4543-b1d3-81520088f6a4", + "name": "width", + "type": "INT", + "linkIds": [ + 104, + 129 + ], + "pos": [ + -1206.49609375, + 430 + ] + }, + { + "id": "bdb6cd97-67d9-440c-8c4c-9b7a7540edd0", + "name": "height", + "type": "INT", + "linkIds": [ + 105, + 130 + ], + "pos": [ + -1206.49609375, + 450 + ] + }, + { + "id": "18abb56c-30bf-4de5-83c1-c12376e8d14e", + "name": "seed", + "type": "INT", + "linkIds": [ + 108 + ], + "pos": [ + -1206.49609375, + 470 + ] + }, + { + "id": "e5cd06f9-64ed-4778-97ba-b165f7a79c4e", + "name": "unet_name", + "type": "COMBO", + "linkIds": [ + 109 + ], + "pos": [ + -1206.49609375, + 490 + ] + }, + { + "id": "06480e4c-4043-489b-ae68-1cf2b4246260", + "name": "clip_name", + "type": "COMBO", + "linkIds": [ + 110 + ], + "pos": [ + -1206.49609375, + 510 + ] + }, + { + "id": "8d65d01b-16b2-420d-8b7b-42077c2e4976", + "name": "clip_name_1", + "type": "COMBO", + "linkIds": [ + 132 + ], + "label": "prompt_enhancer", + "pos": [ + -1206.49609375, + 530 + ] + }, + { + "id": "697f2fdb-0fd9-4008-a895-0f9ce9e8fd88", + "name": "vae_name", + "type": "COMBO", + "linkIds": [ + 133 + ], + "pos": [ + -1206.49609375, + 550 + ] + } + ], + "outputs": [ + { + "id": "21d5fbe0-9f91-4d93-8ea8-5bbf2cd5b698", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 84 + ], + "localized_name": "IMAGE", + "pos": [ + 1130, + 280 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 71, + "type": "EmptyFlux2LatentImage", + "pos": [ + -470, + 1050 + ], + "size": [ + 270, + 170 + ], + "flags": {}, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 104 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 105 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "links": [ + 80 + ] + } + ], + "properties": { + "Node name for S&R": "EmptyFlux2LatentImage", + "cnr_id": "comfy-core", + "ver": "0.18.1", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + 1024, + 1024, + 1 + ] + }, + { + "id": 66, + "type": "UNETLoader", + "pos": [ + -470, + 320 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "unet_name", + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": 109 + }, + { + "localized_name": "weight_dtype", + "name": "weight_dtype", + "type": "COMBO", + "widget": { + "name": "weight_dtype" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 85 + ] + } + ], + "properties": { + "Node name for S&R": "UNETLoader", + "cnr_id": "comfy-core", + "ver": "0.3.73", + "models": [ + { + "name": "ernie-image-turbo.safetensors", + "url": "https://huggingface.co/Comfy-Org/ERNIE-Image/resolve/main/diffusion_models/ernie-image-turbo.safetensors", + "directory": "diffusion_models" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "ernie-image-turbo.safetensors", + "default" + ] + }, + { + "id": 65, + "type": "VAEDecode", + "pos": [ + 710, + 280 + ], + "size": [ + 230, + 100 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 73 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 74 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "slot_index": 0, + "links": [ + 84 + ] + } + ], + "properties": { + "Node name for S&R": "VAEDecode", + "cnr_id": "comfy-core", + "ver": "0.3.64", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + } + }, + { + "id": 70, + "type": "KSampler", + "pos": [ + 350, + 280 + ], + "size": [ + 320, + 350 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 85 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 76 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 113 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 80 + }, + { + "localized_name": "seed", + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": 108 + }, + { + "localized_name": "steps", + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": null + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": null + }, + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + }, + { + "localized_name": "scheduler", + "name": "scheduler", + "type": "COMBO", + "widget": { + "name": "scheduler" + }, + "link": null + }, + { + "localized_name": "denoise", + "name": "denoise", + "type": "FLOAT", + "widget": { + "name": "denoise" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 73 + ] + } + ], + "properties": { + "Node name for S&R": "KSampler", + "cnr_id": "comfy-core", + "ver": "0.3.64", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + 423299999918804, + "randomize", + 8, + 1, + "euler", + "simple", + 1 + ] + }, + { + "id": 67, + "type": "CLIPTextEncode", + "pos": [ + -140, + 320 + ], + "size": [ + 410, + 370 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 79 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 131 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 76, + 112 + ] + } + ], + "properties": { + "Node name for S&R": "CLIPTextEncode", + "cnr_id": "comfy-core", + "ver": "0.3.73", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "" + ], + "color": "#232", + "bgcolor": "#353" + }, + { + "id": 62, + "type": "CLIPLoader", + "pos": [ + -470, + 530 + ], + "size": [ + 270, + 150 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "clip_name", + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": 110 + }, + { + "localized_name": "type", + "name": "type", + "type": "COMBO", + "widget": { + "name": "type" + }, + "link": null + }, + { + "localized_name": "device", + "name": "device", + "shape": 7, + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [ + 79 + ] + } + ], + "properties": { + "Node name for S&R": "CLIPLoader", + "cnr_id": "comfy-core", + "ver": "0.3.73", + "models": [ + { + "name": "ministral-3-3b.safetensors", + "url": "https://huggingface.co/Comfy-Org/ERNIE-Image/resolve/main/text_encoders/ministral-3-3b.safetensors", + "directory": "text_encoders" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "ministral-3-3b.safetensors", + "flux2", + "default" + ] + }, + { + "id": 63, + "type": "VAELoader", + "pos": [ + -470, + 780 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "vae_name", + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": 133 + } + ], + "outputs": [ + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "links": [ + 74 + ] + } + ], + "properties": { + "Node name for S&R": "VAELoader", + "cnr_id": "comfy-core", + "ver": "0.3.73", + "models": [ + { + "name": "flux2-vae.safetensors", + "url": "https://huggingface.co/Comfy-Org/ERNIE-Image/resolve/main/vae/flux2-vae.safetensors", + "directory": "vae" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "flux2-vae.safetensors" + ] + }, + { + "id": 91, + "type": "ConditioningZeroOut", + "pos": [ + 30, + 760 + ], + "size": [ + 230, + 80 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "conditioning", + "name": "conditioning", + "type": "CONDITIONING", + "link": 112 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 113 + ] + } + ], + "properties": { + "Node name for S&R": "ConditioningZeroOut", + "cnr_id": "comfy-core", + "ver": "0.18.1", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + } + }, + { + "id": 93, + "type": "StringReplace", + "pos": [ + -500, + -650 + ], + "size": [ + 430, + 450 + ], + "flags": {}, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "string", + "name": "string", + "type": "STRING", + "widget": { + "name": "string" + }, + "link": null + }, + { + "localized_name": "find", + "name": "find", + "type": "STRING", + "widget": { + "name": "find" + }, + "link": null + }, + { + "localized_name": "replace", + "name": "replace", + "type": "STRING", + "widget": { + "name": "replace" + }, + "link": 115 + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 121 + ] + } + ], + "properties": { + "Node name for S&R": "StringReplace", + "cnr_id": "comfy-core", + "ver": "0.18.1", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "[SYSTEM_PROMPT]你是一个专业的文生图 Prompt 增强助手。你将收到用户的简短图片描述及目标生成分辨率，请据此扩写为一段内容丰富、细节充分的视觉描述，以帮助文生图模型生成高质量的图片。仅输出增强后的描述，不要包含任何解释或前缀。[/SYSTEM_PROMPT][INST]{\"prompt\": \"{prompt}\", \"width\": {width}, \"height\": {height}}[/INST]", + "{prompt}", + "" + ] + }, + { + "id": 94, + "type": "PrimitiveStringMultiline", + "pos": [ + -950, + -660 + ], + "size": [ + 400, + 200 + ], + "flags": {}, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "STRING", + "widget": { + "name": "value" + }, + "link": 128 + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 115, + 118 + ] + } + ], + "title": "String (Multiline - Prompt)", + "properties": { + "Node name for S&R": "PrimitiveStringMultiline", + "cnr_id": "comfy-core", + "ver": "0.18.1", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "" + ] + }, + { + "id": 95, + "type": "TextGenerate", + "pos": [ + 530, + -660 + ], + "size": [ + 400, + 380 + ], + "flags": {}, + "order": 10, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 116 + }, + { + "localized_name": "image", + "name": "image", + "shape": 7, + "type": "IMAGE", + "link": null + }, + { + "localized_name": "prompt", + "name": "prompt", + "type": "STRING", + "widget": { + "name": "prompt" + }, + "link": 117 + }, + { + "localized_name": "max_length", + "name": "max_length", + "type": "INT", + "widget": { + "name": "max_length" + }, + "link": null + }, + { + "localized_name": "sampling_mode", + "name": "sampling_mode", + "type": "COMFY_DYNAMICCOMBO_V3", + "widget": { + "name": "sampling_mode" + }, + "link": null + }, + { + "localized_name": "temperature", + "name": "sampling_mode.temperature", + "type": "FLOAT", + "widget": { + "name": "sampling_mode.temperature" + }, + "link": null + }, + { + "localized_name": "top_k", + "name": "sampling_mode.top_k", + "type": "INT", + "widget": { + "name": "sampling_mode.top_k" + }, + "link": null + }, + { + "localized_name": "top_p", + "name": "sampling_mode.top_p", + "type": "FLOAT", + "widget": { + "name": "sampling_mode.top_p" + }, + "link": null + }, + { + "localized_name": "min_p", + "name": "sampling_mode.min_p", + "type": "FLOAT", + "widget": { + "name": "sampling_mode.min_p" + }, + "link": null + }, + { + "localized_name": "repetition_penalty", + "name": "sampling_mode.repetition_penalty", + "type": "FLOAT", + "widget": { + "name": "sampling_mode.repetition_penalty" + }, + "link": null + }, + { + "localized_name": "seed", + "name": "sampling_mode.seed", + "type": "INT", + "widget": { + "name": "sampling_mode.seed" + }, + "link": null + }, + { + "localized_name": "sampling_mode.presence_penalty", + "name": "sampling_mode.presence_penalty", + "shape": 7, + "type": "FLOAT", + "widget": { + "name": "sampling_mode.presence_penalty" + }, + "link": null + }, + { + "localized_name": "thinking", + "name": "thinking", + "shape": 7, + "type": "BOOLEAN", + "widget": { + "name": "thinking" + }, + "link": null + }, + { + "localized_name": "use_default_template", + "name": "use_default_template", + "shape": 7, + "type": "BOOLEAN", + "widget": { + "name": "use_default_template" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "generated_text", + "name": "generated_text", + "type": "STRING", + "links": [ + 119 + ] + } + ], + "properties": { + "Node name for S&R": "TextGenerate", + "cnr_id": "comfy-core", + "ver": "0.18.1", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "", + 2048, + "on", + 0.6, + 64, + 0.8, + 0.05, + 1.05, + 0, + 0, + false, + true + ] + }, + { + "id": 96, + "type": "PrimitiveBoolean", + "pos": [ + -490, + 60 + ], + "size": [ + 270, + 100 + ], + "flags": {}, + "order": 11, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "BOOLEAN", + "widget": { + "name": "value" + }, + "link": 127 + } + ], + "outputs": [ + { + "localized_name": "BOOLEAN", + "name": "BOOLEAN", + "type": "BOOLEAN", + "links": [ + 120 + ] + } + ], + "title": "Enable prompt enhancement?", + "properties": { + "Node name for S&R": "PrimitiveBoolean", + "cnr_id": "comfy-core", + "ver": "0.18.1", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + true + ] + }, + { + "id": 97, + "type": "ComfySwitchNode", + "pos": [ + 550, + -10 + ], + "size": [ + 270, + 130 + ], + "flags": {}, + "order": 12, + "mode": 0, + "inputs": [ + { + "localized_name": "on_false", + "name": "on_false", + "type": "*", + "link": 118 + }, + { + "localized_name": "on_true", + "name": "on_true", + "type": "*", + "link": 119 + }, + { + "localized_name": "switch", + "name": "switch", + "type": "BOOLEAN", + "widget": { + "name": "switch" + }, + "link": 120 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "*", + "links": [ + 131, + 134 + ] + } + ], + "properties": { + "Node name for S&R": "ComfySwitchNode", + "cnr_id": "comfy-core", + "ver": "0.18.1", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + false + ] + }, + { + "id": 98, + "type": "CLIPLoader", + "pos": [ + -490, + -150 + ], + "size": [ + 510, + 150 + ], + "flags": {}, + "order": 13, + "mode": 0, + "inputs": [ + { + "localized_name": "clip_name", + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": 132 + }, + { + "localized_name": "type", + "name": "type", + "type": "COMBO", + "widget": { + "name": "type" + }, + "link": null + }, + { + "localized_name": "device", + "name": "device", + "shape": 7, + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [ + 116 + ] + } + ], + "title": "Load CLIP (PE)", + "properties": { + "Node name for S&R": "CLIPLoader", + "cnr_id": "comfy-core", + "ver": "0.19.0", + "models": [ + { + "name": "ernie-image-prompt-enhancer.safetensors", + "url": "https://huggingface.co/Comfy-Org/ERNIE-Image/resolve/main/text_encoders/ernie-image-prompt-enhancer.safetensors", + "directory": "text_encoders" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "ernie-image-prompt-enhancer.safetensors", + "flux2", + "default" + ] + }, + { + "id": 99, + "type": "PreviewAny", + "pos": [ + -950, + -410 + ], + "size": [ + 400, + 180 + ], + "flags": {}, + "order": 14, + "mode": 0, + "inputs": [ + { + "localized_name": "source", + "name": "source", + "type": "*", + "link": 129 + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 122 + ] + } + ], + "title": "Preview as Text (Int to String)", + "properties": { + "Node name for S&R": "PreviewAny", + "cnr_id": "comfy-core", + "ver": "0.19.0", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + null, + null, + null + ] + }, + { + "id": 100, + "type": "PreviewAny", + "pos": [ + -950, + -190 + ], + "size": [ + 400, + 180 + ], + "flags": {}, + "order": 15, + "mode": 0, + "inputs": [ + { + "localized_name": "source", + "name": "source", + "type": "*", + "link": 130 + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 124 + ] + } + ], + "title": "Preview as Text (Int to String)", + "properties": { + "Node name for S&R": "PreviewAny", + "cnr_id": "comfy-core", + "ver": "0.19.0", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + null, + null, + null + ] + }, + { + "id": 101, + "type": "StringReplace", + "pos": [ + -30, + -650 + ], + "size": [ + 230, + 450 + ], + "flags": {}, + "order": 16, + "mode": 0, + "inputs": [ + { + "localized_name": "string", + "name": "string", + "type": "STRING", + "widget": { + "name": "string" + }, + "link": 121 + }, + { + "localized_name": "find", + "name": "find", + "type": "STRING", + "widget": { + "name": "find" + }, + "link": null + }, + { + "localized_name": "replace", + "name": "replace", + "type": "STRING", + "widget": { + "name": "replace" + }, + "link": 122 + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 123 + ] + } + ], + "properties": { + "Node name for S&R": "StringReplace", + "cnr_id": "comfy-core", + "ver": "0.18.1", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "", + "{width}", + "" + ] + }, + { + "id": 102, + "type": "StringReplace", + "pos": [ + 220, + -650 + ], + "size": [ + 250, + 450 + ], + "flags": {}, + "order": 17, + "mode": 0, + "inputs": [ + { + "localized_name": "string", + "name": "string", + "type": "STRING", + "widget": { + "name": "string" + }, + "link": 123 + }, + { + "localized_name": "find", + "name": "find", + "type": "STRING", + "widget": { + "name": "find" + }, + "link": null + }, + { + "localized_name": "replace", + "name": "replace", + "type": "STRING", + "widget": { + "name": "replace" + }, + "link": 124 + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 117 + ] + } + ], + "properties": { + "Node name for S&R": "StringReplace", + "cnr_id": "comfy-core", + "ver": "0.18.1", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "", + "{height}", + "" + ] + }, + { + "id": 103, + "type": "PreviewAny", + "pos": [ + 970, + -660 + ], + "size": [ + 570, + 790 + ], + "flags": {}, + "order": 18, + "mode": 0, + "inputs": [ + { + "localized_name": "source", + "name": "source", + "type": "*", + "link": 134 + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [] + } + ], + "title": "Preview as Text (Int to String)", + "properties": { + "Node name for S&R": "PreviewAny", + "cnr_id": "comfy-core", + "ver": "0.19.0", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + null, + null, + null + ] + } + ], + "groups": [ + { + "id": 6, + "title": "Text to Image", + "bounding": [ + -510, + 200, + 1450, + 1060 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 2, + "title": "Image Size", + "bounding": [ + -490, + 950, + 300, + 290 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 3, + "title": "Prompt", + "bounding": [ + -160, + 250, + 470, + 670 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 4, + "title": "Model", + "bounding": [ + -490, + 250, + 300, + 670 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 7, + "title": "Prompt Enhancement", + "bounding": [ + -510, + -720, + 1450, + 890 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 73, + "origin_id": 70, + "origin_slot": 0, + "target_id": 65, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 74, + "origin_id": 63, + "origin_slot": 0, + "target_id": 65, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 85, + "origin_id": 66, + "origin_slot": 0, + "target_id": 70, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 76, + "origin_id": 67, + "origin_slot": 0, + "target_id": 70, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 80, + "origin_id": 71, + "origin_slot": 0, + "target_id": 70, + "target_slot": 3, + "type": "LATENT" + }, + { + "id": 79, + "origin_id": 62, + "origin_slot": 0, + "target_id": 67, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 84, + "origin_id": 65, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 104, + "origin_id": -10, + "origin_slot": 2, + "target_id": 71, + "target_slot": 0, + "type": "INT" + }, + { + "id": 105, + "origin_id": -10, + "origin_slot": 3, + "target_id": 71, + "target_slot": 1, + "type": "INT" + }, + { + "id": 108, + "origin_id": -10, + "origin_slot": 4, + "target_id": 70, + "target_slot": 4, + "type": "INT" + }, + { + "id": 109, + "origin_id": -10, + "origin_slot": 5, + "target_id": 66, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 110, + "origin_id": -10, + "origin_slot": 6, + "target_id": 62, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 112, + "origin_id": 67, + "origin_slot": 0, + "target_id": 91, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 113, + "origin_id": 91, + "origin_slot": 0, + "target_id": 70, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 115, + "origin_id": 94, + "origin_slot": 0, + "target_id": 93, + "target_slot": 2, + "type": "STRING" + }, + { + "id": 116, + "origin_id": 98, + "origin_slot": 0, + "target_id": 95, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 117, + "origin_id": 102, + "origin_slot": 0, + "target_id": 95, + "target_slot": 2, + "type": "STRING" + }, + { + "id": 118, + "origin_id": 94, + "origin_slot": 0, + "target_id": 97, + "target_slot": 0, + "type": "STRING" + }, + { + "id": 119, + "origin_id": 95, + "origin_slot": 0, + "target_id": 97, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 120, + "origin_id": 96, + "origin_slot": 0, + "target_id": 97, + "target_slot": 2, + "type": "BOOLEAN" + }, + { + "id": 121, + "origin_id": 93, + "origin_slot": 0, + "target_id": 101, + "target_slot": 0, + "type": "STRING" + }, + { + "id": 122, + "origin_id": 99, + "origin_slot": 0, + "target_id": 101, + "target_slot": 2, + "type": "STRING" + }, + { + "id": 123, + "origin_id": 101, + "origin_slot": 0, + "target_id": 102, + "target_slot": 0, + "type": "STRING" + }, + { + "id": 124, + "origin_id": 100, + "origin_slot": 0, + "target_id": 102, + "target_slot": 2, + "type": "STRING" + }, + { + "id": 127, + "origin_id": -10, + "origin_slot": 1, + "target_id": 96, + "target_slot": 0, + "type": "BOOLEAN" + }, + { + "id": 128, + "origin_id": -10, + "origin_slot": 0, + "target_id": 94, + "target_slot": 0, + "type": "STRING" + }, + { + "id": 129, + "origin_id": -10, + "origin_slot": 2, + "target_id": 99, + "target_slot": 0, + "type": "*" + }, + { + "id": 130, + "origin_id": -10, + "origin_slot": 3, + "target_id": 100, + "target_slot": 0, + "type": "*" + }, + { + "id": 131, + "origin_id": 97, + "origin_slot": 0, + "target_id": 67, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 132, + "origin_id": -10, + "origin_slot": 7, + "target_id": 98, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 133, + "origin_id": -10, + "origin_slot": 8, + "target_id": 63, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 134, + "origin_id": 97, + "origin_slot": 0, + "target_id": 103, + "target_slot": 0, + "type": "STRING" + } + ], + "extra": {}, + "category": "Image generation and editing/Text to image", + "description": "Faster ERNIE Image Turbo variant (~8B DiT, distilled for fewer sampling steps): same strengths in Chinese/English on-image text and layout-heavy graphics as the base ERNIE Image lineup, with bundled encoders and VAE." + } + ] + }, + "extra": { + "ue_links": [] + } +} diff --git a/blueprints/Text to Image (Ernie Image).json b/blueprints/Text to Image (Ernie Image).json new file mode 100644 index 000000000..2bab20d69 --- /dev/null +++ b/blueprints/Text to Image (Ernie Image).json @@ -0,0 +1,2190 @@ +{ + "revision": 0, + "last_node_id": 88, + "last_link_id": 0, + "nodes": [ + { + "id": 88, + "type": "03921aea-a70e-44b4-bc77-f6bda10f2120", + "pos": [ + -120, + 240 + ], + "size": [ + 400, + 540 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "label": "prompt", + "name": "value", + "type": "STRING", + "widget": { + "name": "value" + }, + "link": null + }, + { + "label": "prompt_enhancement", + "name": "value_1", + "type": "BOOLEAN", + "widget": { + "name": "value_1" + }, + "link": null + }, + { + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": null + }, + { + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": null + }, + { + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": null + }, + { + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": null + }, + { + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": null + }, + { + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": null + }, + { + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": null + }, + { + "label": "prompt_enhancer", + "name": "clip_name_1", + "type": "COMBO", + "widget": { + "name": "clip_name_1" + }, + "link": null + }, + { + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [] + } + ], + "properties": { + "proxyWidgets": [ + [ + "78", + "value" + ], + [ + "76", + "value" + ], + [ + "71", + "width" + ], + [ + "71", + "height" + ], + [ + "70", + "steps" + ], + [ + "70", + "cfg" + ], + [ + "70", + "seed" + ], + [ + "66", + "unet_name" + ], + [ + "62", + "clip_name" + ], + [ + "91", + "clip_name" + ], + [ + "63", + "vae_name" + ] + ], + "cnr_id": "comfy-core", + "ver": "0.18.1", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": { + "value": true, + "value_1": true + }, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [], + "title": "Text to Image (Ernie Image)" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "03921aea-a70e-44b4-bc77-f6bda10f2120", + "version": 1, + "state": { + "lastGroupId": 6, + "lastNodeId": 99, + "lastLinkId": 124, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Text to Image (Ernie Image)", + "inputNode": { + "id": -10, + "bounding": [ + -1350, + 370, + 163.50390625, + 260 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 1110, + 260, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "504de359-52a4-49aa-b6be-23c1cdb0cbde", + "name": "value", + "type": "STRING", + "linkIds": [ + 102 + ], + "label": "prompt", + "pos": [ + -1206.49609375, + 390 + ] + }, + { + "id": "29f699c6-9263-41f6-b37d-69b9fc3913dd", + "name": "value_1", + "type": "BOOLEAN", + "linkIds": [ + 103 + ], + "label": "prompt_enhancement", + "pos": [ + -1206.49609375, + 410 + ] + }, + { + "id": "968e6213-d1e9-4268-8f47-1d6b9a39a43e", + "name": "width", + "type": "INT", + "linkIds": [ + 104, + 113 + ], + "pos": [ + -1206.49609375, + 430 + ] + }, + { + "id": "181c49ef-740d-4385-aa11-79718951ccb9", + "name": "height", + "type": "INT", + "linkIds": [ + 105, + 114 + ], + "pos": [ + -1206.49609375, + 450 + ] + }, + { + "id": "1e85f808-66a1-41df-be52-334142b35419", + "name": "steps", + "type": "INT", + "linkIds": [ + 106 + ], + "pos": [ + -1206.49609375, + 470 + ] + }, + { + "id": "2806addf-a252-4aa3-a5b7-397ab36dccec", + "name": "cfg", + "type": "FLOAT", + "linkIds": [ + 107 + ], + "pos": [ + -1206.49609375, + 490 + ] + }, + { + "id": "5d036a66-5dc0-4d7c-b9a9-349e454738aa", + "name": "seed", + "type": "INT", + "linkIds": [ + 108 + ], + "pos": [ + -1206.49609375, + 510 + ] + }, + { + "id": "360f9a40-aac5-4e9c-bc98-9d55a4a58be2", + "name": "unet_name", + "type": "COMBO", + "linkIds": [ + 109 + ], + "pos": [ + -1206.49609375, + 530 + ] + }, + { + "id": "886301c7-6e88-4cec-96fa-8ae20e8340c5", + "name": "clip_name", + "type": "COMBO", + "linkIds": [ + 110 + ], + "pos": [ + -1206.49609375, + 550 + ] + }, + { + "id": "1d73a545-6d01-462f-bc61-966d4b918ff2", + "name": "clip_name_1", + "type": "COMBO", + "linkIds": [ + 120 + ], + "label": "prompt_enhancer", + "pos": [ + -1206.49609375, + 570 + ] + }, + { + "id": "8c61dc8c-e260-4b36-b73e-d36f90a0bbe3", + "name": "vae_name", + "type": "COMBO", + "linkIds": [ + 121 + ], + "pos": [ + -1206.49609375, + 590 + ] + } + ], + "outputs": [ + { + "id": "f4cb34c8-4090-4281-b428-7338a339d274", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 84 + ], + "localized_name": "IMAGE", + "pos": [ + 1130, + 280 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 71, + "type": "EmptyFlux2LatentImage", + "pos": [ + -460, + 1040 + ], + "size": [ + 270, + 170 + ], + "flags": {}, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 104 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 105 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "links": [ + 80 + ] + } + ], + "properties": { + "Node name for S&R": "EmptyFlux2LatentImage", + "cnr_id": "comfy-core", + "ver": "0.18.1", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + 1024, + 1024, + 1 + ] + }, + { + "id": 66, + "type": "UNETLoader", + "pos": [ + -470, + 320 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "unet_name", + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": 109 + }, + { + "localized_name": "weight_dtype", + "name": "weight_dtype", + "type": "COMBO", + "widget": { + "name": "weight_dtype" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 85 + ] + } + ], + "properties": { + "Node name for S&R": "UNETLoader", + "cnr_id": "comfy-core", + "ver": "0.3.73", + "models": [ + { + "name": "ernie-image.safetensors", + "url": "https://huggingface.co/Comfy-Org/ERNIE-Image/resolve/main/diffusion_models/ernie-image.safetensors", + "directory": "diffusion_models" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "ernie-image.safetensors", + "default" + ] + }, + { + "id": 65, + "type": "VAEDecode", + "pos": [ + 710, + 280 + ], + "size": [ + 230, + 100 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 73 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 74 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "slot_index": 0, + "links": [ + 84 + ] + } + ], + "properties": { + "Node name for S&R": "VAEDecode", + "cnr_id": "comfy-core", + "ver": "0.3.64", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + } + }, + { + "id": 70, + "type": "KSampler", + "pos": [ + 350, + 280 + ], + "size": [ + 320, + 350 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 85 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 76 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 83 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 80 + }, + { + "localized_name": "seed", + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": 108 + }, + { + "localized_name": "steps", + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": 106 + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": 107 + }, + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + }, + { + "localized_name": "scheduler", + "name": "scheduler", + "type": "COMBO", + "widget": { + "name": "scheduler" + }, + "link": null + }, + { + "localized_name": "denoise", + "name": "denoise", + "type": "FLOAT", + "widget": { + "name": "denoise" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 73 + ] + } + ], + "properties": { + "Node name for S&R": "KSampler", + "cnr_id": "comfy-core", + "ver": "0.3.64", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + 182596410725960, + "randomize", + 20, + 4, + "euler", + "simple", + 1 + ] + }, + { + "id": 67, + "type": "CLIPTextEncode", + "pos": [ + -140, + 320 + ], + "size": [ + 410, + 370 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 79 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 100 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 76 + ] + } + ], + "properties": { + "Node name for S&R": "CLIPTextEncode", + "cnr_id": "comfy-core", + "ver": "0.3.73", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "" + ], + "color": "#232", + "bgcolor": "#353" + }, + { + "id": 72, + "type": "CLIPTextEncode", + "pos": [ + -130, + 770 + ], + "size": [ + 390, + 140 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 82 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 83 + ] + } + ], + "properties": { + "Node name for S&R": "CLIPTextEncode", + "cnr_id": "comfy-core", + "ver": "0.3.73", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "" + ], + "color": "#223", + "bgcolor": "#335" + }, + { + "id": 83, + "type": "StringReplace", + "pos": [ + -500, + -640 + ], + "size": [ + 430, + 450 + ], + "flags": {}, + "order": 12, + "mode": 0, + "inputs": [ + { + "localized_name": "string", + "name": "string", + "type": "STRING", + "widget": { + "name": "string" + }, + "link": null + }, + { + "localized_name": "find", + "name": "find", + "type": "STRING", + "widget": { + "name": "find" + }, + "link": null + }, + { + "localized_name": "replace", + "name": "replace", + "type": "STRING", + "widget": { + "name": "replace" + }, + "link": 92 + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 115 + ] + } + ], + "properties": { + "Node name for S&R": "StringReplace", + "cnr_id": "comfy-core", + "ver": "0.18.1", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "[SYSTEM_PROMPT]你是一个专业的文生图 Prompt 增强助手。你将收到用户的简短图片描述及目标生成分辨率，请据此扩写为一段内容丰富、细节充分的视觉描述，以帮助文生图模型生成高质量的图片。仅输出增强后的描述，不要包含任何解释或前缀。[/SYSTEM_PROMPT][INST]{\"prompt\": \"{prompt}\", \"width\": {width}, \"height\": {height}}[/INST]", + "{prompt}", + "" + ] + }, + { + "id": 78, + "type": "PrimitiveStringMultiline", + "pos": [ + -950, + -650 + ], + "size": [ + 400, + 200 + ], + "flags": {}, + "order": 11, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "STRING", + "widget": { + "name": "value" + }, + "link": 102 + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 87, + 92 + ] + } + ], + "title": "String (Multiline - Prompt)", + "properties": { + "Node name for S&R": "PrimitiveStringMultiline", + "cnr_id": "comfy-core", + "ver": "0.18.1", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "" + ] + }, + { + "id": 74, + "type": "TextGenerate", + "pos": [ + 530, + -650 + ], + "size": [ + 400, + 380 + ], + "flags": {}, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 112 + }, + { + "localized_name": "image", + "name": "image", + "shape": 7, + "type": "IMAGE", + "link": null + }, + { + "localized_name": "prompt", + "name": "prompt", + "type": "STRING", + "widget": { + "name": "prompt" + }, + "link": 119 + }, + { + "localized_name": "max_length", + "name": "max_length", + "type": "INT", + "widget": { + "name": "max_length" + }, + "link": null + }, + { + "localized_name": "sampling_mode", + "name": "sampling_mode", + "type": "COMFY_DYNAMICCOMBO_V3", + "widget": { + "name": "sampling_mode" + }, + "link": null + }, + { + "localized_name": "temperature", + "name": "sampling_mode.temperature", + "type": "FLOAT", + "widget": { + "name": "sampling_mode.temperature" + }, + "link": null + }, + { + "localized_name": "top_k", + "name": "sampling_mode.top_k", + "type": "INT", + "widget": { + "name": "sampling_mode.top_k" + }, + "link": null + }, + { + "localized_name": "top_p", + "name": "sampling_mode.top_p", + "type": "FLOAT", + "widget": { + "name": "sampling_mode.top_p" + }, + "link": null + }, + { + "localized_name": "min_p", + "name": "sampling_mode.min_p", + "type": "FLOAT", + "widget": { + "name": "sampling_mode.min_p" + }, + "link": null + }, + { + "localized_name": "repetition_penalty", + "name": "sampling_mode.repetition_penalty", + "type": "FLOAT", + "widget": { + "name": "sampling_mode.repetition_penalty" + }, + "link": null + }, + { + "localized_name": "seed", + "name": "sampling_mode.seed", + "type": "INT", + "widget": { + "name": "sampling_mode.seed" + }, + "link": null + }, + { + "localized_name": "sampling_mode.presence_penalty", + "name": "sampling_mode.presence_penalty", + "shape": 7, + "type": "FLOAT", + "widget": { + "name": "sampling_mode.presence_penalty" + }, + "link": null + }, + { + "localized_name": "thinking", + "name": "thinking", + "shape": 7, + "type": "BOOLEAN", + "widget": { + "name": "thinking" + }, + "link": null + }, + { + "localized_name": "use_default_template", + "name": "use_default_template", + "shape": 7, + "type": "BOOLEAN", + "widget": { + "name": "use_default_template" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "generated_text", + "name": "generated_text", + "type": "STRING", + "links": [ + 89 + ] + } + ], + "properties": { + "Node name for S&R": "TextGenerate", + "cnr_id": "comfy-core", + "ver": "0.18.1", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "", + 2048, + "on", + 0.6, + 64, + 0.8, + 0.05, + 1.05, + 0, + 0, + false, + true + ] + }, + { + "id": 76, + "type": "PrimitiveBoolean", + "pos": [ + -500, + 60 + ], + "size": [ + 270, + 100 + ], + "flags": {}, + "order": 10, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "BOOLEAN", + "widget": { + "name": "value" + }, + "link": 103 + } + ], + "outputs": [ + { + "localized_name": "BOOLEAN", + "name": "BOOLEAN", + "type": "BOOLEAN", + "links": [ + 88 + ] + } + ], + "title": "Enable prompt enhancement?", + "properties": { + "Node name for S&R": "PrimitiveBoolean", + "cnr_id": "comfy-core", + "ver": "0.18.1", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + true + ] + }, + { + "id": 75, + "type": "ComfySwitchNode", + "pos": [ + 530, + 20 + ], + "size": [ + 270, + 130 + ], + "flags": {}, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "on_false", + "name": "on_false", + "type": "*", + "link": 87 + }, + { + "localized_name": "on_true", + "name": "on_true", + "type": "*", + "link": 89 + }, + { + "localized_name": "switch", + "name": "switch", + "type": "BOOLEAN", + "widget": { + "name": "switch" + }, + "link": 88 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "*", + "links": [ + 100, + 124 + ] + } + ], + "properties": { + "Node name for S&R": "ComfySwitchNode", + "cnr_id": "comfy-core", + "ver": "0.18.1", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + false + ] + }, + { + "id": 62, + "type": "CLIPLoader", + "pos": [ + -460, + 520 + ], + "size": [ + 270, + 150 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "clip_name", + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": 110 + }, + { + "localized_name": "type", + "name": "type", + "type": "COMBO", + "widget": { + "name": "type" + }, + "link": null + }, + { + "localized_name": "device", + "name": "device", + "shape": 7, + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [ + 79, + 82 + ] + } + ], + "properties": { + "Node name for S&R": "CLIPLoader", + "cnr_id": "comfy-core", + "ver": "0.3.73", + "models": [ + { + "name": "ministral-3-3b.safetensors", + "url": "https://huggingface.co/Comfy-Org/ERNIE-Image/resolve/main/text_encoders/ministral-3-3b.safetensors", + "directory": "text_encoders" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "ministral-3-3b.safetensors", + "flux2", + "default" + ] + }, + { + "id": 63, + "type": "VAELoader", + "pos": [ + -460, + 770 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "vae_name", + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": 121 + } + ], + "outputs": [ + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "links": [ + 74 + ] + } + ], + "properties": { + "Node name for S&R": "VAELoader", + "cnr_id": "comfy-core", + "ver": "0.3.73", + "models": [ + { + "name": "flux2-vae.safetensors", + "url": "https://huggingface.co/Comfy-Org/ERNIE-Image/resolve/main/vae/flux2-vae.safetensors", + "directory": "vae" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "flux2-vae.safetensors" + ] + }, + { + "id": 91, + "type": "CLIPLoader", + "pos": [ + -500, + -150 + ], + "size": [ + 510, + 150 + ], + "flags": {}, + "order": 13, + "mode": 0, + "inputs": [ + { + "localized_name": "clip_name", + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": 120 + }, + { + "localized_name": "type", + "name": "type", + "type": "COMBO", + "widget": { + "name": "type" + }, + "link": null + }, + { + "localized_name": "device", + "name": "device", + "shape": 7, + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [ + 112 + ] + } + ], + "title": "Load CLIP (PE)", + "properties": { + "Node name for S&R": "CLIPLoader", + "cnr_id": "comfy-core", + "ver": "0.19.0", + "models": [ + { + "name": "ernie-image-prompt-enhancer.safetensors", + "url": "https://huggingface.co/Comfy-Org/ERNIE-Image/resolve/main/text_encoders/ernie-image-prompt-enhancer.safetensors", + "directory": "text_encoders" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "ernie-image-prompt-enhancer.safetensors", + "flux2", + "default" + ] + }, + { + "id": 92, + "type": "PreviewAny", + "pos": [ + -950, + -400 + ], + "size": [ + 400, + 180 + ], + "flags": {}, + "order": 14, + "mode": 0, + "inputs": [ + { + "localized_name": "source", + "name": "source", + "type": "*", + "link": 113 + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 116 + ] + } + ], + "title": "Preview as Text (Int to String)", + "properties": { + "Node name for S&R": "PreviewAny", + "cnr_id": "comfy-core", + "ver": "0.19.0", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + null, + null, + null + ] + }, + { + "id": 93, + "type": "PreviewAny", + "pos": [ + -950, + -180 + ], + "size": [ + 400, + 180 + ], + "flags": {}, + "order": 15, + "mode": 0, + "inputs": [ + { + "localized_name": "source", + "name": "source", + "type": "*", + "link": 114 + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 118 + ] + } + ], + "title": "Preview as Text (Int to String)", + "properties": { + "Node name for S&R": "PreviewAny", + "cnr_id": "comfy-core", + "ver": "0.19.0", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + null, + null, + null + ] + }, + { + "id": 94, + "type": "StringReplace", + "pos": [ + -30, + -640 + ], + "size": [ + 230, + 450 + ], + "flags": {}, + "order": 16, + "mode": 0, + "inputs": [ + { + "localized_name": "string", + "name": "string", + "type": "STRING", + "widget": { + "name": "string" + }, + "link": 115 + }, + { + "localized_name": "find", + "name": "find", + "type": "STRING", + "widget": { + "name": "find" + }, + "link": null + }, + { + "localized_name": "replace", + "name": "replace", + "type": "STRING", + "widget": { + "name": "replace" + }, + "link": 116 + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 117 + ] + } + ], + "properties": { + "Node name for S&R": "StringReplace", + "cnr_id": "comfy-core", + "ver": "0.18.1", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "", + "{width}", + "" + ] + }, + { + "id": 95, + "type": "StringReplace", + "pos": [ + 220, + -640 + ], + "size": [ + 250, + 450 + ], + "flags": {}, + "order": 17, + "mode": 0, + "inputs": [ + { + "localized_name": "string", + "name": "string", + "type": "STRING", + "widget": { + "name": "string" + }, + "link": 117 + }, + { + "localized_name": "find", + "name": "find", + "type": "STRING", + "widget": { + "name": "find" + }, + "link": null + }, + { + "localized_name": "replace", + "name": "replace", + "type": "STRING", + "widget": { + "name": "replace" + }, + "link": 118 + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 119 + ] + } + ], + "properties": { + "Node name for S&R": "StringReplace", + "cnr_id": "comfy-core", + "ver": "0.18.1", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "", + "{height}", + "" + ] + }, + { + "id": 97, + "type": "PreviewAny", + "pos": [ + 970, + -650 + ], + "size": [ + 570, + 790 + ], + "flags": {}, + "order": 18, + "mode": 0, + "inputs": [ + { + "localized_name": "source", + "name": "source", + "type": "*", + "link": 124 + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [] + } + ], + "title": "Preview as Text (Int to String)", + "properties": { + "Node name for S&R": "PreviewAny", + "cnr_id": "comfy-core", + "ver": "0.19.0", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + null, + null, + null + ] + } + ], + "groups": [ + { + "id": 6, + "title": "Text to Image", + "bounding": [ + -510, + 200, + 1450, + 1060 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 2, + "title": "Image Size", + "bounding": [ + -480, + 940, + 310, + 290 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 3, + "title": "Prompt", + "bounding": [ + -160, + 250, + 470, + 670 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 4, + "title": "Model", + "bounding": [ + -490, + 250, + 320, + 670 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 5, + "title": "Prompt Enhancement", + "bounding": [ + -510, + -720, + 1450, + 890 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 73, + "origin_id": 70, + "origin_slot": 0, + "target_id": 65, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 74, + "origin_id": 63, + "origin_slot": 0, + "target_id": 65, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 85, + "origin_id": 66, + "origin_slot": 0, + "target_id": 70, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 76, + "origin_id": 67, + "origin_slot": 0, + "target_id": 70, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 83, + "origin_id": 72, + "origin_slot": 0, + "target_id": 70, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 80, + "origin_id": 71, + "origin_slot": 0, + "target_id": 70, + "target_slot": 3, + "type": "LATENT" + }, + { + "id": 79, + "origin_id": 62, + "origin_slot": 0, + "target_id": 67, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 100, + "origin_id": 75, + "origin_slot": 0, + "target_id": 67, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 82, + "origin_id": 62, + "origin_slot": 0, + "target_id": 72, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 92, + "origin_id": 78, + "origin_slot": 0, + "target_id": 83, + "target_slot": 2, + "type": "STRING" + }, + { + "id": 87, + "origin_id": 78, + "origin_slot": 0, + "target_id": 75, + "target_slot": 0, + "type": "STRING" + }, + { + "id": 89, + "origin_id": 74, + "origin_slot": 0, + "target_id": 75, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 88, + "origin_id": 76, + "origin_slot": 0, + "target_id": 75, + "target_slot": 2, + "type": "BOOLEAN" + }, + { + "id": 84, + "origin_id": 65, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 102, + "origin_id": -10, + "origin_slot": 0, + "target_id": 78, + "target_slot": 0, + "type": "STRING" + }, + { + "id": 103, + "origin_id": -10, + "origin_slot": 1, + "target_id": 76, + "target_slot": 0, + "type": "BOOLEAN" + }, + { + "id": 104, + "origin_id": -10, + "origin_slot": 2, + "target_id": 71, + "target_slot": 0, + "type": "INT" + }, + { + "id": 105, + "origin_id": -10, + "origin_slot": 3, + "target_id": 71, + "target_slot": 1, + "type": "INT" + }, + { + "id": 106, + "origin_id": -10, + "origin_slot": 4, + "target_id": 70, + "target_slot": 5, + "type": "INT" + }, + { + "id": 107, + "origin_id": -10, + "origin_slot": 5, + "target_id": 70, + "target_slot": 6, + "type": "FLOAT" + }, + { + "id": 108, + "origin_id": -10, + "origin_slot": 6, + "target_id": 70, + "target_slot": 4, + "type": "INT" + }, + { + "id": 109, + "origin_id": -10, + "origin_slot": 7, + "target_id": 66, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 110, + "origin_id": -10, + "origin_slot": 8, + "target_id": 62, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 112, + "origin_id": 91, + "origin_slot": 0, + "target_id": 74, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 113, + "origin_id": -10, + "origin_slot": 2, + "target_id": 92, + "target_slot": 0, + "type": "*" + }, + { + "id": 114, + "origin_id": -10, + "origin_slot": 3, + "target_id": 93, + "target_slot": 0, + "type": "*" + }, + { + "id": 115, + "origin_id": 83, + "origin_slot": 0, + "target_id": 94, + "target_slot": 0, + "type": "STRING" + }, + { + "id": 116, + "origin_id": 92, + "origin_slot": 0, + "target_id": 94, + "target_slot": 2, + "type": "STRING" + }, + { + "id": 117, + "origin_id": 94, + "origin_slot": 0, + "target_id": 95, + "target_slot": 0, + "type": "STRING" + }, + { + "id": 118, + "origin_id": 93, + "origin_slot": 0, + "target_id": 95, + "target_slot": 2, + "type": "STRING" + }, + { + "id": 119, + "origin_id": 95, + "origin_slot": 0, + "target_id": 74, + "target_slot": 2, + "type": "STRING" + }, + { + "id": 120, + "origin_id": -10, + "origin_slot": 9, + "target_id": 91, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 121, + "origin_id": -10, + "origin_slot": 10, + "target_id": 63, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 124, + "origin_id": 75, + "origin_slot": 0, + "target_id": 97, + "target_slot": 0, + "type": "STRING" + } + ], + "extra": {}, + "category": "Image generation and editing/Text to image", + "description": "Generates images from text prompts using Baidu’s open ERNIE Image (~8B DiT): bilingual in-image typography and layouts (posters, infographics, multi-panel compositions) alongside general scenes, with bundled encoders and VAE." + } + ] + }, + "extra": { + "ue_links": [] + } +} diff --git a/blueprints/Text to Image (Flux.1 Dev).json b/blueprints/Text to Image (Flux.1 Dev).json new file mode 100644 index 000000000..6d8446e81 --- /dev/null +++ b/blueprints/Text to Image (Flux.1 Dev).json @@ -0,0 +1,1047 @@ +{ + "revision": 0, + "last_node_id": 193, + "last_link_id": 0, + "nodes": [ + { + "id": 193, + "type": "1fd98b34-59ef-4d8d-afbf-58bdd7a1cd35", + "pos": [ + -1210, + -1770 + ], + "size": [ + 400, + 380 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "label": "prompt", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + }, + { + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": null + }, + { + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": null + }, + { + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": null + }, + { + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": null + }, + { + "name": "clip_name1", + "type": "COMBO", + "widget": { + "name": "clip_name1" + }, + "link": null + }, + { + "name": "clip_name2", + "type": "COMBO", + "widget": { + "name": "clip_name2" + }, + "link": null + }, + { + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [] + } + ], + "properties": { + "proxyWidgets": [ + [ + "45", + "text" + ], + [ + "27", + "width" + ], + [ + "27", + "height" + ], + [ + "31", + "seed" + ], + [ + "38", + "unet_name" + ], + [ + "40", + "clip_name1" + ], + [ + "40", + "clip_name2" + ], + [ + "39", + "vae_name" + ], + [ + "31", + "control_after_generate" + ] + ], + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.18.1" + }, + "widgets_values": [], + "title": "Text to Image (Flux.1 Dev)" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "1fd98b34-59ef-4d8d-afbf-58bdd7a1cd35", + "version": 1, + "state": { + "lastGroupId": 8, + "lastNodeId": 193, + "lastLinkId": 388, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Text to Image (Flux.1 Dev)", + "inputNode": { + "id": -10, + "bounding": [ + -1090, + 411, + 120, + 200 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 540, + 100, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "669e384e-5e26-4291-9bac-e1d1f04b4a16", + "name": "text", + "type": "STRING", + "linkIds": [ + 68 + ], + "label": "prompt", + "pos": [ + -990, + 431 + ] + }, + { + "id": "5a5c0b01-5836-4ca6-a24f-68c0a4fb9802", + "name": "width", + "type": "INT", + "linkIds": [ + 69 + ], + "pos": [ + -990, + 451 + ] + }, + { + "id": "5e01104a-ed7f-457b-aaee-934e8ecc088d", + "name": "height", + "type": "INT", + "linkIds": [ + 70 + ], + "pos": [ + -990, + 471 + ] + }, + { + "id": "ea5ea317-a484-4605-8138-8628a4b8e502", + "name": "seed", + "type": "INT", + "linkIds": [ + 382 + ], + "pos": [ + -990, + 491 + ] + }, + { + "id": "ea2332f5-bd49-4e2e-8c7a-95817dc56ed6", + "name": "unet_name", + "type": "COMBO", + "linkIds": [ + 385 + ], + "pos": [ + -990, + 511 + ] + }, + { + "id": "4fca3f43-c05f-4337-bf84-2afe67e43739", + "name": "clip_name1", + "type": "COMBO", + "linkIds": [ + 386 + ], + "pos": [ + -990, + 531 + ] + }, + { + "id": "357a679f-1370-4cd5-9269-0d5ae1986b49", + "name": "clip_name2", + "type": "COMBO", + "linkIds": [ + 387 + ], + "pos": [ + -990, + 551 + ] + }, + { + "id": "924ffec5-81f8-4585-8761-5a80d5d775bc", + "name": "vae_name", + "type": "COMBO", + "linkIds": [ + 388 + ], + "pos": [ + -990, + 571 + ] + } + ], + "outputs": [ + { + "id": "2185cb4d-8689-4cf8-b345-75319fb46a8e", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 9 + ], + "localized_name": "IMAGE", + "pos": [ + 560, + 120 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 39, + "type": "VAELoader", + "pos": [ + -800, + 670 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "vae_name", + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": 388 + } + ], + "outputs": [ + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "links": [ + 58 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.40", + "Node name for S&R": "VAELoader", + "models": [ + { + "name": "ae.safetensors", + "url": "https://huggingface.co/Comfy-Org/Lumina_Image_2.0_Repackaged/resolve/main/split_files/vae/ae.safetensors", + "directory": "vae" + } + ] + }, + "widgets_values": [ + "ae.safetensors" + ] + }, + { + "id": 38, + "type": "UNETLoader", + "pos": [ + -800, + 160 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "unet_name", + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": 385 + }, + { + "localized_name": "weight_dtype", + "name": "weight_dtype", + "type": "COMBO", + "widget": { + "name": "weight_dtype" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 61 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.40", + "Node name for S&R": "UNETLoader", + "models": [ + { + "name": "flux1-dev.safetensors", + "url": "https://huggingface.co/Comfy-Org/flux1-dev/resolve/main/flux1-dev.safetensors", + "directory": "diffusion_models" + } + ] + }, + "widgets_values": [ + "flux1-dev.safetensors", + "default" + ] + }, + { + "id": 40, + "type": "DualCLIPLoader", + "pos": [ + -800, + 380 + ], + "size": [ + 270, + 180 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "clip_name1", + "name": "clip_name1", + "type": "COMBO", + "widget": { + "name": "clip_name1" + }, + "link": 386 + }, + { + "localized_name": "clip_name2", + "name": "clip_name2", + "type": "COMBO", + "widget": { + "name": "clip_name2" + }, + "link": 387 + }, + { + "localized_name": "type", + "name": "type", + "type": "COMBO", + "widget": { + "name": "type" + }, + "link": null + }, + { + "localized_name": "device", + "name": "device", + "shape": 7, + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [ + 64 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.40", + "Node name for S&R": "DualCLIPLoader", + "models": [ + { + "name": "clip_l.safetensors", + "url": "https://huggingface.co/comfyanonymous/flux_text_encoders/resolve/main/clip_l.safetensors", + "directory": "text_encoders" + }, + { + "name": "t5xxl_fp16.safetensors", + "url": "https://huggingface.co/comfyanonymous/flux_text_encoders/resolve/main/t5xxl_fp16.safetensors", + "directory": "text_encoders" + } + ] + }, + "widgets_values": [ + "clip_l.safetensors", + "t5xxl_fp16.safetensors", + "flux", + "default" + ] + }, + { + "id": 27, + "type": "EmptySD3LatentImage", + "pos": [ + -420, + 640 + ], + "size": [ + 270, + 170 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 69 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 70 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 51 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.40", + "Node name for S&R": "EmptySD3LatentImage" + }, + "widgets_values": [ + 1024, + 1024, + 1 + ] + }, + { + "id": 45, + "type": "CLIPTextEncode", + "pos": [ + -460, + 150 + ], + "size": [ + 330, + 220 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 64 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 68 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 65, + 66 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "CLIPTextEncode" + }, + "widgets_values": [ + "" + ] + }, + { + "id": 31, + "type": "KSampler", + "pos": [ + -50, + 260 + ], + "size": [ + 320, + 350 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 61 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 65 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 63 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 51 + }, + { + "localized_name": "seed", + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": 382 + }, + { + "localized_name": "steps", + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": null + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": null + }, + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + }, + { + "localized_name": "scheduler", + "name": "scheduler", + "type": "COMBO", + "widget": { + "name": "scheduler" + }, + "link": null + }, + { + "localized_name": "denoise", + "name": "denoise", + "type": "FLOAT", + "widget": { + "name": "denoise" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 52 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.40", + "Node name for S&R": "KSampler" + }, + "widgets_values": [ + 0, + "randomize", + 20, + 1, + "euler", + "simple", + 1 + ] + }, + { + "id": 8, + "type": "VAEDecode", + "pos": [ + 20, + 120 + ], + "size": [ + 230, + 100 + ], + "flags": { + "collapsed": false + }, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 52 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 58 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "slot_index": 0, + "links": [ + 9 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.40", + "Node name for S&R": "VAEDecode" + } + }, + { + "id": 42, + "type": "ConditioningZeroOut", + "pos": [ + -350, + 420 + ], + "size": [ + 230, + 80 + ], + "flags": { + "collapsed": false + }, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "conditioning", + "name": "conditioning", + "type": "CONDITIONING", + "link": 66 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 63 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.40", + "Node name for S&R": "ConditioningZeroOut" + } + } + ], + "groups": [ + { + "id": 1, + "title": "Model", + "bounding": [ + -820, + 70, + 320, + 750 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 2, + "title": "Image Size", + "bounding": [ + -470, + 570, + 380, + 250 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 3, + "title": "Prompt", + "bounding": [ + -470, + 70, + 380, + 470 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 52, + "origin_id": 31, + "origin_slot": 0, + "target_id": 8, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 58, + "origin_id": 39, + "origin_slot": 0, + "target_id": 8, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 61, + "origin_id": 38, + "origin_slot": 0, + "target_id": 31, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 63, + "origin_id": 42, + "origin_slot": 0, + "target_id": 31, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 51, + "origin_id": 27, + "origin_slot": 0, + "target_id": 31, + "target_slot": 3, + "type": "LATENT" + }, + { + "id": 9, + "origin_id": 8, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 64, + "origin_id": 40, + "origin_slot": 0, + "target_id": 45, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 65, + "origin_id": 45, + "origin_slot": 0, + "target_id": 31, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 66, + "origin_id": 45, + "origin_slot": 0, + "target_id": 42, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 68, + "origin_id": -10, + "origin_slot": 0, + "target_id": 45, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 69, + "origin_id": -10, + "origin_slot": 1, + "target_id": 27, + "target_slot": 0, + "type": "INT" + }, + { + "id": 70, + "origin_id": -10, + "origin_slot": 2, + "target_id": 27, + "target_slot": 1, + "type": "INT" + }, + { + "id": 382, + "origin_id": -10, + "origin_slot": 3, + "target_id": 31, + "target_slot": 4, + "type": "INT" + }, + { + "id": 385, + "origin_id": -10, + "origin_slot": 4, + "target_id": 38, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 386, + "origin_id": -10, + "origin_slot": 5, + "target_id": 40, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 387, + "origin_id": -10, + "origin_slot": 6, + "target_id": 40, + "target_slot": 1, + "type": "COMBO" + }, + { + "id": 388, + "origin_id": -10, + "origin_slot": 7, + "target_id": 39, + "target_slot": 0, + "type": "COMBO" + } + ], + "extra": { + "workflowRendererVersion": "LG" + }, + "category": "Image generation and editing/Text to image", + "description": "Generates images from prompts using FLUX.1 [dev]: a 12B rectified-flow MMDiT with dual CLIP plus T5-XXL text encoders and guidance-distilled sampling for sharp prompt following versus classic DDPM diffusion." + } + ] + }, + "extra": { + "ds": { + "scale": 0.7513148009015777, + "offset": [ + 1726.1426909346173, + 146.66925047394233 + ] + }, + "ue_links": [] + } +} diff --git a/blueprints/Text to Image (Flux.1 Krea Dev).json b/blueprints/Text to Image (Flux.1 Krea Dev).json new file mode 100644 index 000000000..0d7fa03c4 --- /dev/null +++ b/blueprints/Text to Image (Flux.1 Krea Dev).json @@ -0,0 +1,1041 @@ +{ + "revision": 0, + "last_node_id": 196, + "last_link_id": 0, + "nodes": [ + { + "id": 196, + "type": "aa0a207e-bf0e-477c-a87f-f58fcf5f7749", + "pos": [ + 1010, + 130 + ], + "size": [ + 410, + 460 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + }, + { + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": null + }, + { + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": null + }, + { + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": null + }, + { + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": null + }, + { + "name": "clip_name1", + "type": "COMBO", + "widget": { + "name": "clip_name1" + }, + "link": null + }, + { + "name": "clip_name2", + "type": "COMBO", + "widget": { + "name": "clip_name2" + }, + "link": null + }, + { + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [] + } + ], + "properties": { + "proxyWidgets": [ + [ + "195", + "text" + ], + [ + "27", + "width" + ], + [ + "27", + "height" + ], + [ + "31", + "seed" + ], + [ + "38", + "unet_name" + ], + [ + "40", + "clip_name1" + ], + [ + "40", + "clip_name2" + ], + [ + "39", + "vae_name" + ] + ], + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.18.1" + }, + "widgets_values": [], + "title": "Text to Image (Flux.1 Krea Dev)" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "aa0a207e-bf0e-477c-a87f-f58fcf5f7749", + "version": 1, + "state": { + "lastGroupId": 8, + "lastNodeId": 196, + "lastLinkId": 395, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Text to Image (Flux.1 Krea Dev)", + "inputNode": { + "id": -10, + "bounding": [ + -1050, + 426, + 120, + 200 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 620, + 140, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "c2515318-6e10-4ad9-9466-e6aa855bc849", + "name": "text", + "type": "STRING", + "linkIds": [ + 71 + ], + "pos": [ + -950, + 446 + ] + }, + { + "id": "09f20672-c8a3-4180-823a-5a6af0113e4f", + "name": "width", + "type": "INT", + "linkIds": [ + 72 + ], + "pos": [ + -950, + 466 + ] + }, + { + "id": "7f54c952-896e-4356-bfb2-970e1c8f2eb7", + "name": "height", + "type": "INT", + "linkIds": [ + 73 + ], + "pos": [ + -950, + 486 + ] + }, + { + "id": "e2dc1c86-2fb4-4b80-b560-f30560af1897", + "name": "seed", + "type": "INT", + "linkIds": [ + 391 + ], + "pos": [ + -950, + 506 + ] + }, + { + "id": "34b172e7-85b2-444a-9a4d-1221f272c46e", + "name": "unet_name", + "type": "COMBO", + "linkIds": [ + 392 + ], + "pos": [ + -950, + 526 + ] + }, + { + "id": "073b7440-d943-4a2f-a3a1-fbdb8fcda9f9", + "name": "clip_name1", + "type": "COMBO", + "linkIds": [ + 393 + ], + "pos": [ + -950, + 546 + ] + }, + { + "id": "55c1286a-4aca-41fc-b967-ae3d3fa7bc85", + "name": "clip_name2", + "type": "COMBO", + "linkIds": [ + 394 + ], + "pos": [ + -950, + 566 + ] + }, + { + "id": "2241e4fc-9219-4be7-bf6d-3493b579ab5a", + "name": "vae_name", + "type": "COMBO", + "linkIds": [ + 395 + ], + "pos": [ + -950, + 586 + ] + } + ], + "outputs": [ + { + "id": "5310184a-f0a2-405f-9917-dd2a352a4fac", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 9 + ], + "localized_name": "IMAGE", + "pos": [ + 640, + 160 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 40, + "type": "DualCLIPLoader", + "pos": [ + -780, + 360 + ], + "size": [ + 270, + 180 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "clip_name1", + "name": "clip_name1", + "type": "COMBO", + "widget": { + "name": "clip_name1" + }, + "link": 393 + }, + { + "localized_name": "clip_name2", + "name": "clip_name2", + "type": "COMBO", + "widget": { + "name": "clip_name2" + }, + "link": 394 + }, + { + "localized_name": "type", + "name": "type", + "type": "COMBO", + "widget": { + "name": "type" + }, + "link": null + }, + { + "localized_name": "device", + "name": "device", + "shape": 7, + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [ + 64 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.40", + "Node name for S&R": "DualCLIPLoader", + "models": [ + { + "name": "clip_l.safetensors", + "url": "https://huggingface.co/comfyanonymous/flux_text_encoders/resolve/main/clip_l.safetensors", + "directory": "text_encoders" + }, + { + "name": "t5xxl_fp16.safetensors", + "url": "https://huggingface.co/comfyanonymous/flux_text_encoders/resolve/main/t5xxl_fp16.safetensors", + "directory": "text_encoders" + } + ] + }, + "widgets_values": [ + "clip_l.safetensors", + "t5xxl_fp16.safetensors", + "flux", + "default" + ] + }, + { + "id": 39, + "type": "VAELoader", + "pos": [ + -770, + 630 + ], + "size": [ + 240, + 110 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "vae_name", + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": 395 + } + ], + "outputs": [ + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "links": [ + 58 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.40", + "Node name for S&R": "VAELoader", + "models": [ + { + "name": "ae.safetensors", + "url": "https://huggingface.co/Comfy-Org/Lumina_Image_2.0_Repackaged/resolve/main/split_files/vae/ae.safetensors", + "directory": "vae" + } + ] + }, + "widgets_values": [ + "ae.safetensors" + ] + }, + { + "id": 38, + "type": "UNETLoader", + "pos": [ + -780, + 170 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "unet_name", + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": 392 + }, + { + "localized_name": "weight_dtype", + "name": "weight_dtype", + "type": "COMBO", + "widget": { + "name": "weight_dtype" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 61 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.40", + "Node name for S&R": "UNETLoader", + "models": [ + { + "name": "flux1-krea-dev_fp8_scaled.safetensors", + "url": "https://huggingface.co/Comfy-Org/FLUX.1-Krea-dev_ComfyUI/resolve/main/split_files/diffusion_models/flux1-krea-dev_fp8_scaled.safetensors", + "directory": "diffusion_models" + } + ] + }, + "widgets_values": [ + "flux1-krea-dev_fp8_scaled.safetensors", + "default" + ] + }, + { + "id": 195, + "type": "CLIPTextEncode", + "pos": [ + -440, + 180 + ], + "size": [ + 330, + 210 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 64 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 71 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 65, + 66 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.47", + "Node name for S&R": "CLIPTextEncode" + }, + "widgets_values": [ + "" + ] + }, + { + "id": 27, + "type": "EmptySD3LatentImage", + "pos": [ + -390, + 650 + ], + "size": [ + 270, + 170 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 72 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 73 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 51 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.40", + "Node name for S&R": "EmptySD3LatentImage" + }, + "widgets_values": [ + 1024, + 1024, + 1 + ] + }, + { + "id": 31, + "type": "KSampler", + "pos": [ + 0, + 130 + ], + "size": [ + 320, + 350 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 61 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 65 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 63 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 51 + }, + { + "localized_name": "seed", + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": 391 + }, + { + "localized_name": "steps", + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": null + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": null + }, + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + }, + { + "localized_name": "scheduler", + "name": "scheduler", + "type": "COMBO", + "widget": { + "name": "scheduler" + }, + "link": null + }, + { + "localized_name": "denoise", + "name": "denoise", + "type": "FLOAT", + "widget": { + "name": "denoise" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 52 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.40", + "Node name for S&R": "KSampler" + }, + "widgets_values": [ + 0, + "randomize", + 20, + 1, + "euler", + "simple", + 1 + ] + }, + { + "id": 8, + "type": "VAEDecode", + "pos": [ + 340, + 140 + ], + "size": [ + 230, + 100 + ], + "flags": { + "collapsed": false + }, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 52 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 58 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "slot_index": 0, + "links": [ + 9 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.40", + "Node name for S&R": "VAEDecode" + } + }, + { + "id": 42, + "type": "ConditioningZeroOut", + "pos": [ + -340, + 430 + ], + "size": [ + 230, + 80 + ], + "flags": { + "collapsed": false + }, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "conditioning", + "name": "conditioning", + "type": "CONDITIONING", + "link": 66 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 63 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.40", + "Node name for S&R": "ConditioningZeroOut" + } + } + ], + "groups": [ + { + "id": 1, + "title": "Model", + "bounding": [ + -800, + 90, + 310, + 750 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 2, + "title": "Image Size", + "bounding": [ + -460, + 560, + 400, + 280 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 3, + "title": "Prompt", + "bounding": [ + -460, + 90, + 400, + 440 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 66, + "origin_id": 195, + "origin_slot": 0, + "target_id": 42, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 52, + "origin_id": 31, + "origin_slot": 0, + "target_id": 8, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 58, + "origin_id": 39, + "origin_slot": 0, + "target_id": 8, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 61, + "origin_id": 38, + "origin_slot": 0, + "target_id": 31, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 65, + "origin_id": 195, + "origin_slot": 0, + "target_id": 31, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 63, + "origin_id": 42, + "origin_slot": 0, + "target_id": 31, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 51, + "origin_id": 27, + "origin_slot": 0, + "target_id": 31, + "target_slot": 3, + "type": "LATENT" + }, + { + "id": 64, + "origin_id": 40, + "origin_slot": 0, + "target_id": 195, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 9, + "origin_id": 8, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 71, + "origin_id": -10, + "origin_slot": 0, + "target_id": 195, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 72, + "origin_id": -10, + "origin_slot": 1, + "target_id": 27, + "target_slot": 0, + "type": "INT" + }, + { + "id": 73, + "origin_id": -10, + "origin_slot": 2, + "target_id": 27, + "target_slot": 1, + "type": "INT" + }, + { + "id": 391, + "origin_id": -10, + "origin_slot": 3, + "target_id": 31, + "target_slot": 4, + "type": "INT" + }, + { + "id": 392, + "origin_id": -10, + "origin_slot": 4, + "target_id": 38, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 393, + "origin_id": -10, + "origin_slot": 5, + "target_id": 40, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 394, + "origin_id": -10, + "origin_slot": 6, + "target_id": 40, + "target_slot": 1, + "type": "COMBO" + }, + { + "id": 395, + "origin_id": -10, + "origin_slot": 7, + "target_id": 39, + "target_slot": 0, + "type": "COMBO" + } + ], + "extra": { + "workflowRendererVersion": "LG" + }, + "category": "Image generation and editing/Text to image", + "description": "FLUX.1 Krea [dev] (Black Forest Labs × Krea): open-weight 12B rectified-flow text-to-image drop-in alongside FLUX.1 [dev], tuned away from overcooked saturation toward more natural diversity in people, realism, and style while keeping ecosystem compatibility." + } + ] + }, + "extra": { + "ds": { + "scale": 0.735584459955559, + "offset": [ + 1936.5815687336737, + 303.78330847702625 + ] + }, + "ue_links": [] + } +} diff --git a/blueprints/Text to Image (Flux.2 Dev).json b/blueprints/Text to Image (Flux.2 Dev).json new file mode 100644 index 000000000..d5ca3077d --- /dev/null +++ b/blueprints/Text to Image (Flux.2 Dev).json @@ -0,0 +1,1870 @@ +{ + "revision": 0, + "last_node_id": 123, + "last_link_id": 0, + "nodes": [ + { + "id": 123, + "type": "85066daf-feda-4c7b-bbc3-d4797e8ccf0f", + "pos": [ + -800, + 640 + ], + "size": [ + 400, + 0 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "label": "prompt", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + }, + { + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": null + }, + { + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": null + }, + { + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": null + }, + { + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": null + }, + { + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": null + }, + { + "label": "turbo_lora", + "name": "lora_name", + "type": "COMBO", + "widget": { + "name": "lora_name" + }, + "link": null + }, + { + "label": "enable_turbo_mode", + "name": "value", + "type": "BOOLEAN", + "widget": { + "name": "value" + }, + "link": null + }, + { + "name": "noise_seed", + "type": "INT", + "widget": { + "name": "noise_seed" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [] + } + ], + "properties": { + "proxyWidgets": [ + [ + "115", + "text" + ], + [ + "113", + "width" + ], + [ + "113", + "height" + ], + [ + "122", + "unet_name" + ], + [ + "111", + "clip_name" + ], + [ + "108", + "vae_name" + ], + [ + "116", + "lora_name" + ], + [ + "121", + "value" + ], + [ + "114", + "noise_seed" + ], + [ + "114", + "control_after_generate" + ] + ], + "cnr_id": "comfy-core", + "ver": "0.15.1", + "ue_properties": { + "widget_ue_connectable": { + "value": true, + "lora_name": true + }, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [], + "title": "Text to Image (Flux.2 Dev)" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "85066daf-feda-4c7b-bbc3-d4797e8ccf0f", + "version": 1, + "state": { + "lastGroupId": 6, + "lastNodeId": 123, + "lastLinkId": 232, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Text to Image (Flux.2 Dev)", + "inputNode": { + "id": -10, + "bounding": [ + -1500, + 250, + 151.744140625, + 220 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 1560, + -20, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "1f4f1091-3f97-41d8-8ed8-e8b02260cf3c", + "name": "text", + "type": "STRING", + "linkIds": [ + 206 + ], + "label": "prompt", + "pos": [ + -1368.255859375, + 270 + ] + }, + { + "id": "b9b59411-4f5f-4482-8f78-369e6d50e71c", + "name": "width", + "type": "INT", + "linkIds": [ + 222, + 231 + ], + "pos": [ + -1368.255859375, + 290 + ] + }, + { + "id": "c6de9a28-3bf6-40d0-be16-f75ec517a766", + "name": "height", + "type": "INT", + "linkIds": [ + 223, + 232 + ], + "pos": [ + -1368.255859375, + 310 + ] + }, + { + "id": "8f1b1c75-e47c-45f5-af57-74abcfe8967c", + "name": "unet_name", + "type": "COMBO", + "linkIds": [ + 225 + ], + "pos": [ + -1368.255859375, + 330 + ] + }, + { + "id": "6ac27631-1bf0-4161-9670-a662f6180b94", + "name": "clip_name", + "type": "COMBO", + "linkIds": [ + 226 + ], + "pos": [ + -1368.255859375, + 350 + ] + }, + { + "id": "932e6cbe-f716-4905-ae54-d2b3543497bd", + "name": "vae_name", + "type": "COMBO", + "linkIds": [ + 227 + ], + "pos": [ + -1368.255859375, + 370 + ] + }, + { + "id": "37400048-5e7b-427b-8b79-ea35841d5306", + "name": "lora_name", + "type": "COMBO", + "linkIds": [ + 228 + ], + "label": "turbo_lora", + "pos": [ + -1368.255859375, + 390 + ] + }, + { + "id": "333212d0-f027-476f-8b97-a921e20e340a", + "name": "value", + "type": "BOOLEAN", + "linkIds": [ + 229 + ], + "label": "enable_turbo_mode", + "pos": [ + -1368.255859375, + 410 + ] + }, + { + "id": "e7e73fad-ce6e-48d5-b719-e2abed685185", + "name": "noise_seed", + "type": "INT", + "linkIds": [ + 230 + ], + "pos": [ + -1368.255859375, + 430 + ] + } + ], + "outputs": [ + { + "id": "ed3c0a0f-a39f-453e-907f-8249c8e3335d", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 9 + ], + "localized_name": "IMAGE", + "pos": [ + 1580, + 0 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 105, + "type": "BasicGuider", + "pos": [ + 570, + 170 + ], + "size": [ + 230, + 100 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 210 + }, + { + "localized_name": "conditioning", + "name": "conditioning", + "type": "CONDITIONING", + "link": 165 + } + ], + "outputs": [ + { + "localized_name": "GUIDER", + "name": "GUIDER", + "type": "GUIDER", + "slot_index": 0, + "links": [ + 30 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "BasicGuider", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 106, + "type": "FluxGuidance", + "pos": [ + -510, + 470 + ], + "size": [ + 320, + 110 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "conditioning", + "name": "conditioning", + "type": "CONDITIONING", + "link": 41 + }, + { + "localized_name": "guidance", + "name": "guidance", + "type": "FLOAT", + "widget": { + "name": "guidance" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "slot_index": 0, + "links": [ + 165 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "FluxGuidance", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 4 + ], + "color": "#233", + "bgcolor": "#355" + }, + { + "id": 107, + "type": "KSamplerSelect", + "pos": [ + 570, + 350 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "SAMPLER", + "name": "SAMPLER", + "type": "SAMPLER", + "links": [ + 19 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "KSamplerSelect", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "euler" + ] + }, + { + "id": 108, + "type": "VAELoader", + "pos": [ + -1000, + 460 + ], + "size": [ + 300, + 110 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "vae_name", + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": 227 + } + ], + "outputs": [ + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "slot_index": 0, + "links": [ + 159 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "VAELoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "full_encoder_small_decoder.safetensors", + "url": "https://huggingface.co/black-forest-labs/FLUX.2-small-decoder/resolve/main/full_encoder_small_decoder.safetensors", + "directory": "vae" + } + ] + }, + "widgets_values": [ + "full_encoder_small_decoder.safetensors" + ] + }, + { + "id": 109, + "type": "SamplerCustomAdvanced", + "pos": [ + 860, + -20 + ], + "size": [ + 280, + 330 + ], + "flags": {}, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "noise", + "name": "noise", + "type": "NOISE", + "link": 37 + }, + { + "localized_name": "guider", + "name": "guider", + "type": "GUIDER", + "link": 30 + }, + { + "localized_name": "sampler", + "name": "sampler", + "type": "SAMPLER", + "link": 19 + }, + { + "localized_name": "sigmas", + "name": "sigmas", + "type": "SIGMAS", + "link": 132 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 161 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "LATENT", + "slot_index": 0, + "links": [ + 24 + ] + }, + { + "localized_name": "denoised_output", + "name": "denoised_output", + "type": "LATENT", + "links": null + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "SamplerCustomAdvanced", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 110, + "type": "VAEDecode", + "pos": [ + 1220, + -20 + ], + "size": [ + 230, + 100 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 24 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 159 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "slot_index": 0, + "links": [ + 9 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "VAEDecode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 111, + "type": "CLIPLoader", + "pos": [ + -1000, + 200 + ], + "size": [ + 300, + 150 + ], + "flags": {}, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "clip_name", + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": 226 + }, + { + "localized_name": "type", + "name": "type", + "type": "COMBO", + "widget": { + "name": "type" + }, + "link": null + }, + { + "localized_name": "device", + "name": "device", + "shape": 7, + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [ + 117 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "CLIPLoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "mistral_3_small_flux2_bf16.safetensors", + "url": "https://huggingface.co/Comfy-Org/flux2-dev/resolve/main/split_files/text_encoders/mistral_3_small_flux2_bf16.safetensors", + "directory": "text_encoders" + } + ] + }, + "widgets_values": [ + "mistral_3_small_flux2_bf16.safetensors", + "flux2", + "default" + ] + }, + { + "id": 112, + "type": "Flux2Scheduler", + "pos": [ + 570, + 550 + ], + "size": [ + 230, + 170 + ], + "flags": {}, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "steps", + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": 213 + }, + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 231 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 232 + } + ], + "outputs": [ + { + "localized_name": "SIGMAS", + "name": "SIGMAS", + "type": "SIGMAS", + "links": [ + 132 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "Flux2Scheduler", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 20, + 1024, + 1024 + ] + }, + { + "id": 113, + "type": "EmptyFlux2LatentImage", + "pos": [ + -980, + 660 + ], + "size": [ + 270, + 170 + ], + "flags": {}, + "order": 10, + "mode": 0, + "inputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 222 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 223 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "links": [ + 161 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "EmptyFlux2LatentImage", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1024, + 1024, + 1 + ] + }, + { + "id": 114, + "type": "RandomNoise", + "pos": [ + 570, + -20 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 11, + "mode": 0, + "inputs": [ + { + "localized_name": "noise_seed", + "name": "noise_seed", + "type": "INT", + "widget": { + "name": "noise_seed" + }, + "link": 230 + } + ], + "outputs": [ + { + "localized_name": "NOISE", + "name": "NOISE", + "type": "NOISE", + "links": [ + 37 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "RandomNoise", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1027111520328378, + "randomize" + ] + }, + { + "id": 115, + "type": "CLIPTextEncode", + "pos": [ + -630, + -40 + ], + "size": [ + 440, + 450 + ], + "flags": {}, + "order": 12, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 117 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 206 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "slot_index": 0, + "links": [ + 41 + ] + } + ], + "title": "CLIP Text Encode (Positive Prompt)", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "CLIPTextEncode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "" + ], + "color": "#232", + "bgcolor": "#353" + }, + { + "id": 116, + "type": "LoraLoaderModelOnly", + "pos": [ + -150, + 220 + ], + "size": [ + 300, + 140 + ], + "flags": {}, + "order": 13, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 221 + }, + { + "localized_name": "lora_name", + "name": "lora_name", + "type": "COMBO", + "widget": { + "name": "lora_name" + }, + "link": 228 + }, + { + "localized_name": "strength_model", + "name": "strength_model", + "type": "FLOAT", + "widget": { + "name": "strength_model" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 209 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.7.0", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "LoraLoaderModelOnly", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "Flux_2-Turbo-LoRA_comfyui.safetensors", + "url": "https://huggingface.co/ByteZSzn/Flux.2-Turbo-ComfyUI/resolve/main/Flux_2-Turbo-LoRA_comfyui.safetensors", + "directory": "loras" + } + ] + }, + "widgets_values": [ + "Flux_2-Turbo-LoRA_comfyui.safetensors", + 1 + ] + }, + { + "id": 117, + "type": "ComfySwitchNode", + "pos": [ + 220, + -30 + ], + "size": [ + 270, + 130 + ], + "flags": {}, + "order": 14, + "mode": 0, + "inputs": [ + { + "localized_name": "on_false", + "name": "on_false", + "type": "*", + "link": 208 + }, + { + "localized_name": "on_true", + "name": "on_true", + "type": "*", + "link": 209 + }, + { + "localized_name": "switch", + "name": "switch", + "type": "BOOLEAN", + "widget": { + "name": "switch" + }, + "link": 215 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "*", + "links": [ + 210 + ] + } + ], + "title": "Switch(model)", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.15.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ComfySwitchNode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + false + ] + }, + { + "id": 118, + "type": "PrimitiveInt", + "pos": [ + -140, + -30 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 211 + ] + } + ], + "title": "Steps", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.15.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 20, + "fixed" + ] + }, + { + "id": 119, + "type": "PrimitiveInt", + "pos": [ + -150, + 460 + ], + "size": [ + 300, + 110 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 212 + ] + } + ], + "title": "Steps", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.15.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 8, + "fixed" + ] + }, + { + "id": 120, + "type": "ComfySwitchNode", + "pos": [ + 220, + 260 + ], + "size": [ + 270, + 130 + ], + "flags": {}, + "order": 15, + "mode": 0, + "inputs": [ + { + "localized_name": "on_false", + "name": "on_false", + "type": "*", + "link": 211 + }, + { + "localized_name": "on_true", + "name": "on_true", + "type": "*", + "link": 212 + }, + { + "localized_name": "switch", + "name": "switch", + "type": "BOOLEAN", + "widget": { + "name": "switch" + }, + "link": 214 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "*", + "links": [ + 213 + ] + } + ], + "title": "Switch(steps)", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.15.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ComfySwitchNode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + false + ] + }, + { + "id": 121, + "type": "PrimitiveBoolean", + "pos": [ + -110, + 690 + ], + "size": [ + 270, + 100 + ], + "flags": {}, + "order": 16, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "BOOLEAN", + "widget": { + "name": "value" + }, + "link": 229 + } + ], + "outputs": [ + { + "localized_name": "BOOLEAN", + "name": "BOOLEAN", + "type": "BOOLEAN", + "links": [ + 214, + 215 + ] + } + ], + "title": "Enable Turbo LoRA", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.15.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "PrimitiveBoolean", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + false + ] + }, + { + "id": 122, + "type": "UNETLoader", + "pos": [ + -1000, + -30 + ], + "size": [ + 300, + 110 + ], + "flags": {}, + "order": 17, + "mode": 0, + "inputs": [ + { + "localized_name": "unet_name", + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": 225 + }, + { + "localized_name": "weight_dtype", + "name": "weight_dtype", + "type": "COMBO", + "widget": { + "name": "weight_dtype" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "slot_index": 0, + "links": [ + 208, + 221 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.71", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "UNETLoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "flux2_dev_fp8mixed.safetensors", + "url": "https://huggingface.co/Comfy-Org/flux2-dev/resolve/main/split_files/diffusion_models/flux2_dev_fp8mixed.safetensors", + "directory": "diffusion_models" + } + ] + }, + "widgets_values": [ + "flux2_dev_fp8mixed.safetensors", + "default" + ] + } + ], + "groups": [ + { + "id": 1, + "title": "Step 1 - Upload models", + "bounding": [ + -1040, + -110, + 380, + 710 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 2, + "title": "Custom sampler", + "bounding": [ + 540, + -110, + 640, + 870 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 4, + "title": "Step2 - Prompt", + "bounding": [ + -640, + -110, + 460, + 710 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 5, + "title": "Original", + "bounding": [ + -160, + -110, + 320, + 230 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 6, + "title": "8 Steps LoRA", + "bounding": [ + -160, + 140, + 320, + 460 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 165, + "origin_id": 106, + "origin_slot": 0, + "target_id": 105, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 41, + "origin_id": 115, + "origin_slot": 0, + "target_id": 106, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 37, + "origin_id": 114, + "origin_slot": 0, + "target_id": 109, + "target_slot": 0, + "type": "NOISE" + }, + { + "id": 30, + "origin_id": 105, + "origin_slot": 0, + "target_id": 109, + "target_slot": 1, + "type": "GUIDER" + }, + { + "id": 19, + "origin_id": 107, + "origin_slot": 0, + "target_id": 109, + "target_slot": 2, + "type": "SAMPLER" + }, + { + "id": 132, + "origin_id": 112, + "origin_slot": 0, + "target_id": 109, + "target_slot": 3, + "type": "SIGMAS" + }, + { + "id": 161, + "origin_id": 113, + "origin_slot": 0, + "target_id": 109, + "target_slot": 4, + "type": "LATENT" + }, + { + "id": 117, + "origin_id": 111, + "origin_slot": 0, + "target_id": 115, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 24, + "origin_id": 109, + "origin_slot": 0, + "target_id": 110, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 159, + "origin_id": 108, + "origin_slot": 0, + "target_id": 110, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 9, + "origin_id": 110, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 206, + "origin_id": -10, + "origin_slot": 0, + "target_id": 115, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 208, + "origin_id": 122, + "origin_slot": 0, + "target_id": 117, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 209, + "origin_id": 116, + "origin_slot": 0, + "target_id": 117, + "target_slot": 1, + "type": "MODEL" + }, + { + "id": 210, + "origin_id": 117, + "origin_slot": 0, + "target_id": 105, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 211, + "origin_id": 118, + "origin_slot": 0, + "target_id": 120, + "target_slot": 0, + "type": "INT" + }, + { + "id": 212, + "origin_id": 119, + "origin_slot": 0, + "target_id": 120, + "target_slot": 1, + "type": "INT" + }, + { + "id": 213, + "origin_id": 120, + "origin_slot": 0, + "target_id": 112, + "target_slot": 0, + "type": "INT" + }, + { + "id": 214, + "origin_id": 121, + "origin_slot": 0, + "target_id": 120, + "target_slot": 2, + "type": "BOOLEAN" + }, + { + "id": 215, + "origin_id": 121, + "origin_slot": 0, + "target_id": 117, + "target_slot": 2, + "type": "BOOLEAN" + }, + { + "id": 221, + "origin_id": 122, + "origin_slot": 0, + "target_id": 116, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 222, + "origin_id": -10, + "origin_slot": 1, + "target_id": 113, + "target_slot": 0, + "type": "INT" + }, + { + "id": 223, + "origin_id": -10, + "origin_slot": 2, + "target_id": 113, + "target_slot": 1, + "type": "INT" + }, + { + "id": 225, + "origin_id": -10, + "origin_slot": 3, + "target_id": 122, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 226, + "origin_id": -10, + "origin_slot": 4, + "target_id": 111, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 227, + "origin_id": -10, + "origin_slot": 5, + "target_id": 108, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 228, + "origin_id": -10, + "origin_slot": 6, + "target_id": 116, + "target_slot": 1, + "type": "COMBO" + }, + { + "id": 229, + "origin_id": -10, + "origin_slot": 7, + "target_id": 121, + "target_slot": 0, + "type": "BOOLEAN" + }, + { + "id": 230, + "origin_id": -10, + "origin_slot": 8, + "target_id": 114, + "target_slot": 0, + "type": "INT" + }, + { + "id": 231, + "origin_id": -10, + "origin_slot": 1, + "target_id": 112, + "target_slot": 1, + "type": "INT" + }, + { + "id": 232, + "origin_id": -10, + "origin_slot": 2, + "target_id": 112, + "target_slot": 2, + "type": "INT" + } + ], + "extra": { + "workflowRendererVersion": "LG" + }, + "category": "Image generation and editing/Text to image", + "description": "Generates images from prompts using FLUX.2 [dev]: a newer 32B rectified-flow stack with distilled guidance plus a stronger long-context multimodal encoder for complex scenes, sharper typography/UI text, anatomy, lighting, and high-resolution latent decoding." + } + ] + }, + "extra": { + "ue_links": [] + } +} diff --git a/blueprints/Text to Image (NetaYume Lumina).json b/blueprints/Text to Image (NetaYume Lumina).json new file mode 100644 index 000000000..9e11b7a86 --- /dev/null +++ b/blueprints/Text to Image (NetaYume Lumina).json @@ -0,0 +1,1470 @@ +{ + "revision": 0, + "last_node_id": 219, + "last_link_id": 0, + "nodes": [ + { + "id": 219, + "type": "fc9485c9-2acd-482e-94f1-b5fa702f2536", + "pos": [ + -1900, + 2330 + ], + "size": [ + 400, + 540 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "name": "value", + "type": "STRING", + "widget": { + "name": "value" + }, + "link": null + }, + { + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": null + }, + { + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": null + }, + { + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": null + }, + { + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [] + } + ], + "properties": { + "proxyWidgets": [ + [ + "62", + "value" + ], + [ + "53", + "width" + ], + [ + "53", + "height" + ], + [ + "55", + "seed" + ], + [ + "56", + "ckpt_name" + ], + [ + "55", + "control_after_generate" + ] + ], + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + } + }, + "widgets_values": [], + "title": "Text to Image (NetaYume Lumina)" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "fc9485c9-2acd-482e-94f1-b5fa702f2536", + "version": 1, + "state": { + "lastGroupId": 8, + "lastNodeId": 219, + "lastLinkId": 395, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Text to Image (NetaYume Lumina)", + "inputNode": { + "id": -10, + "bounding": [ + -600, + 90, + 120, + 140 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 1740.333330193419, + 286.3333328495138, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "b80a1e0c-e8a6-4c4f-8eb1-825cb7e4fdcf", + "name": "value", + "type": "STRING", + "linkIds": [ + 36 + ], + "pos": [ + -500, + 110 + ] + }, + { + "id": "6583bb32-7cff-4921-a771-1f0dcdf779e6", + "name": "width", + "type": "INT", + "linkIds": [ + 39 + ], + "pos": [ + -500, + 130 + ] + }, + { + "id": "c486937a-46c0-431b-8775-057897843cbd", + "name": "height", + "type": "INT", + "linkIds": [ + 40 + ], + "pos": [ + -500, + 150 + ] + }, + { + "id": "9c85c0cc-c906-405a-a4d9-43b93c47cb53", + "name": "seed", + "type": "INT", + "linkIds": [ + 42 + ], + "pos": [ + -500, + 170 + ] + }, + { + "id": "f7e288ec-fa1f-4a1d-b721-6b605de9cb51", + "name": "ckpt_name", + "type": "COMBO", + "linkIds": [ + 43 + ], + "pos": [ + -500, + 190 + ] + } + ], + "outputs": [ + { + "id": "ea4b872b-a294-4cbf-99a9-70e55c0f8b3e", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 16 + ], + "localized_name": "IMAGE", + "pos": [ + 1760.333330193419, + 306.3333328495138 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 53, + "type": "EmptySD3LatentImage", + "pos": [ + -220, + 370 + ], + "size": [ + 320, + 170 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 39 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 40 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 17 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.64", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "EmptySD3LatentImage" + }, + "widgets_values": [ + 1024, + 1024, + 1 + ] + }, + { + "id": 54, + "type": "ModelSamplingAuraFlow", + "pos": [ + 650, + 40 + ], + "size": [ + 310, + 110 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 12 + }, + { + "localized_name": "shift", + "name": "shift", + "type": "FLOAT", + "widget": { + "name": "shift" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "slot_index": 0, + "links": [ + 13 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.64", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ModelSamplingAuraFlow" + }, + "widgets_values": [ + 4 + ] + }, + { + "id": 55, + "type": "KSampler", + "pos": [ + 650, + 200 + ], + "size": [ + 320, + 350 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 13 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 32 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 23 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 17 + }, + { + "localized_name": "seed", + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": 42 + }, + { + "localized_name": "steps", + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": null + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": null + }, + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + }, + { + "localized_name": "scheduler", + "name": "scheduler", + "type": "COMBO", + "widget": { + "name": "scheduler" + }, + "link": null + }, + { + "localized_name": "denoise", + "name": "denoise", + "type": "FLOAT", + "widget": { + "name": "denoise" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 14 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.64", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "KSampler" + }, + "widgets_values": [ + 0, + "randomize", + 30, + 4, + "res_multistep", + "simple", + 1 + ] + }, + { + "id": 56, + "type": "CheckpointLoaderSimple", + "pos": [ + -220, + 70 + ], + "size": [ + 320, + 160 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "ckpt_name", + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": 43 + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "slot_index": 0, + "links": [ + 12 + ] + }, + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "slot_index": 1, + "links": [ + 22, + 35 + ] + }, + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "slot_index": 2, + "links": [ + 8 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.64", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "CheckpointLoaderSimple", + "models": [ + { + "name": "NetaYumev35_pretrained_all_in_one.safetensors", + "url": "https://huggingface.co/duongve/NetaYume-Lumina-Image-2.0/resolve/main/NetaYumev35_pretrained_all_in_one.safetensors", + "directory": "checkpoints" + } + ] + }, + "widgets_values": [ + "NetaYumev35_pretrained_all_in_one.safetensors" + ] + }, + { + "id": 57, + "type": "a07fdf06-1bda-4dac-bdbd-63ee8ebca1c9", + "pos": [ + 180, + 360 + ], + "size": [ + 400, + 140 + ], + "flags": {}, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 22 + }, + { + "name": "value", + "type": "STRING", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 23 + ] + } + ], + "properties": { + "proxyWidgets": [ + [ + "218", + "value" + ] + ], + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + } + }, + "widgets_values": [], + "color": "#223", + "bgcolor": "#335" + }, + { + "id": 217, + "type": "VAEDecode", + "pos": [ + 1040, + 210 + ], + "size": [ + 230, + 100 + ], + "flags": { + "collapsed": false + }, + "order": 10, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 14 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 8 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "slot_index": 0, + "links": [ + 16 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.64", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "VAEDecode" + } + }, + { + "id": 59, + "type": "MarkdownNote", + "pos": [ + 640, + -390 + ], + "size": [ + 370, + 280 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [], + "outputs": [], + "title": "Note: Prompt", + "properties": {}, + "widgets_values": [ + "Check the prompt book [here](https://nieta-art.feishu.cn/wiki/RY3GwpT59icIQlkWXEfcCqIMnQd)\n\nYou should keep the prefix part fixed until the **Prompt Start** tag\n\n@whatever in the prompt is for artist tags, such as @comfyanonymous\n\nYou can find more artist tags [here](https://gumgum10.github.io/gumgum.github.io/)\n" + ], + "color": "#222", + "bgcolor": "#000" + }, + { + "id": 60, + "type": "StringConcatenate", + "pos": [ + 170, + -370 + ], + "size": [ + 400, + 250 + ], + "flags": { + "collapsed": true + }, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "string_a", + "name": "string_a", + "type": "STRING", + "widget": { + "name": "string_a" + }, + "link": 30 + }, + { + "localized_name": "string_b", + "name": "string_b", + "type": "STRING", + "widget": { + "name": "string_b" + }, + "link": 31 + }, + { + "localized_name": "delimiter", + "name": "delimiter", + "type": "STRING", + "widget": { + "name": "delimiter" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 34 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.70", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "StringConcatenate" + }, + "widgets_values": [ + "", + "", + "" + ] + }, + { + "id": 61, + "type": "CLIPTextEncode", + "pos": [ + 170, + 60 + ], + "size": [ + 430, + 190 + ], + "flags": {}, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 35 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 34 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "slot_index": 0, + "links": [ + 32 + ] + } + ], + "title": "CLIP Text Encode (Positive Prompt)", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.64", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "CLIPTextEncode" + }, + "widgets_values": [ + "" + ], + "color": "#232", + "bgcolor": "#353" + }, + { + "id": 62, + "type": "PrimitiveStringMultiline", + "pos": [ + -240, + -210 + ], + "size": [ + 370, + 140 + ], + "flags": {}, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "STRING", + "widget": { + "name": "value" + }, + "link": 36 + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 31 + ] + } + ], + "title": "Prompt", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.70", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "PrimitiveStringMultiline" + }, + "widgets_values": [ + "" + ] + }, + { + "id": 63, + "type": "PrimitiveStringMultiline", + "pos": [ + -240, + -390 + ], + "size": [ + 370, + 140 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "STRING", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 30 + ] + } + ], + "title": "System prompt", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.70", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "PrimitiveStringMultiline" + }, + "widgets_values": [ + "You are an assistant designed to generate high quality anime images based on textual prompts. " + ] + } + ], + "groups": [ + { + "id": 1, + "title": "Model", + "bounding": [ + -250, + -30, + 370, + 280 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 2, + "title": "Image Size", + "bounding": [ + -250, + 280, + 370, + 290 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 3, + "title": "Prompt", + "bounding": [ + 150, + -30, + 460, + 600 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 4, + "title": "Prompt Builder", + "bounding": [ + -250, + -460, + 840, + 400 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 12, + "origin_id": 56, + "origin_slot": 0, + "target_id": 54, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 13, + "origin_id": 54, + "origin_slot": 0, + "target_id": 55, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 23, + "origin_id": 57, + "origin_slot": 0, + "target_id": 55, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 17, + "origin_id": 53, + "origin_slot": 0, + "target_id": 55, + "target_slot": 3, + "type": "LATENT" + }, + { + "id": 14, + "origin_id": 55, + "origin_slot": 0, + "target_id": 217, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 8, + "origin_id": 56, + "origin_slot": 2, + "target_id": 217, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 22, + "origin_id": 56, + "origin_slot": 1, + "target_id": 57, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 16, + "origin_id": 217, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 30, + "origin_id": 63, + "origin_slot": 0, + "target_id": 60, + "target_slot": 0, + "type": "STRING" + }, + { + "id": 31, + "origin_id": 62, + "origin_slot": 0, + "target_id": 60, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 32, + "origin_id": 61, + "origin_slot": 0, + "target_id": 55, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 34, + "origin_id": 60, + "origin_slot": 0, + "target_id": 61, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 35, + "origin_id": 56, + "origin_slot": 1, + "target_id": 61, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 36, + "origin_id": -10, + "origin_slot": 0, + "target_id": 62, + "target_slot": 0, + "type": "STRING" + }, + { + "id": 39, + "origin_id": -10, + "origin_slot": 1, + "target_id": 53, + "target_slot": 0, + "type": "INT" + }, + { + "id": 40, + "origin_id": -10, + "origin_slot": 2, + "target_id": 53, + "target_slot": 1, + "type": "INT" + }, + { + "id": 42, + "origin_id": -10, + "origin_slot": 3, + "target_id": 55, + "target_slot": 4, + "type": "INT" + }, + { + "id": 43, + "origin_id": -10, + "origin_slot": 4, + "target_id": 56, + "target_slot": 0, + "type": "COMBO" + } + ], + "extra": { + "workflowRendererVersion": "LG" + }, + "category": "Image generation and editing/Text to image", + "description": "Generates images from text prompts using NetaYume Lumina, fine-tuned from Neta Lumina for anime-style and illustration generation." + }, + { + "id": "a07fdf06-1bda-4dac-bdbd-63ee8ebca1c9", + "version": 1, + "state": { + "lastGroupId": 8, + "lastNodeId": 219, + "lastLinkId": 395, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "CLIP Text Encode (Negative Prompt)", + "inputNode": { + "id": -10, + "bounding": [ + -150, + 675, + 120, + 80 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 905.2780151367188, + 675, + 128.6640625, + 60 + ] + }, + "inputs": [ + { + "id": "47264a97-6fc9-454d-920f-b8a43fee0489", + "name": "clip", + "type": "CLIP", + "linkIds": [ + 5 + ], + "localized_name": "clip", + "pos": [ + -50, + 695 + ] + }, + { + "id": "7cdb7919-1dad-4bd2-928d-c543c3fd712e", + "name": "value", + "type": "STRING", + "linkIds": [ + 22 + ], + "pos": [ + -50, + 715 + ] + } + ], + "outputs": [ + { + "id": "c3f17ad9-6954-4333-bf8e-e1cf886c351b", + "name": "CONDITIONING", + "type": "CONDITIONING", + "linkIds": [ + 6 + ], + "localized_name": "CONDITIONING", + "pos": [ + 925.2780151367188, + 695 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 64, + "type": "StringConcatenate", + "pos": [ + 420, + 720 + ], + "size": [ + 400, + 200 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "string_a", + "name": "string_a", + "type": "STRING", + "widget": { + "name": "string_a" + }, + "link": 19 + }, + { + "localized_name": "string_b", + "name": "string_b", + "type": "STRING", + "widget": { + "name": "string_b" + }, + "link": 20 + }, + { + "localized_name": "delimiter", + "name": "delimiter", + "type": "STRING", + "widget": { + "name": "delimiter" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 21 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.70", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "StringConcatenate" + }, + "widgets_values": [ + "", + "", + "" + ] + }, + { + "id": 65, + "type": "PrimitiveStringMultiline", + "pos": [ + 30, + 720 + ], + "size": [ + 370, + 130 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "STRING", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 19 + ] + } + ], + "title": "System prompt", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.70", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "PrimitiveStringMultiline" + }, + "widgets_values": [ + "You are an assistant designed to generate low-quality images based on textual prompts " + ] + }, + { + "id": 218, + "type": "PrimitiveStringMultiline", + "pos": [ + 30, + 900 + ], + "size": [ + 370, + 130 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "STRING", + "widget": { + "name": "value" + }, + "link": 22 + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 20 + ] + } + ], + "title": "System prompt", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.70", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "PrimitiveStringMultiline" + }, + "widgets_values": [ + "blurry, worst quality, low quality, jpeg artifacts, signature, watermark, username, error, deformed hands, bad anatomy, extra limbs, poorly drawn hands, poorly drawn face, mutation, deformed, extra eyes, extra arms, extra legs, malformed limbs, fused fingers, too many fingers, long neck, cross-eyed, bad proportions, missing arms, missing legs, extra digit, fewer digits, cropped" + ] + }, + { + "id": 67, + "type": "CLIPTextEncode", + "pos": [ + 420, + 410 + ], + "size": [ + 430, + 190 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 5 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 21 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "slot_index": 0, + "links": [ + 6 + ] + } + ], + "title": "CLIP Text Encode (Negative Prompt)", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.64", + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "CLIPTextEncode" + }, + "widgets_values": [ + "You are an assistant designed to generate low-quality images based on textual prompts blurry, worst quality, low quality, jpeg artifacts, signature, watermark, username, error, deformed hands, bad anatomy, extra limbs, poorly drawn hands, poorly drawn face, mutation, deformed, extra eyes, extra arms, extra legs, malformed limbs, fused fingers, too many fingers, long neck, cross-eyed, bad proportions, missing arms, missing legs, extra digit, fewer digits, cropped" + ], + "color": "#223", + "bgcolor": "#335" + } + ], + "groups": [], + "links": [ + { + "id": 19, + "origin_id": 65, + "origin_slot": 0, + "target_id": 64, + "target_slot": 0, + "type": "STRING" + }, + { + "id": 20, + "origin_id": 218, + "origin_slot": 0, + "target_id": 64, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 21, + "origin_id": 64, + "origin_slot": 0, + "target_id": 67, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 5, + "origin_id": -10, + "origin_slot": 0, + "target_id": 67, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 6, + "origin_id": 67, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 22, + "origin_id": -10, + "origin_slot": 1, + "target_id": 218, + "target_slot": 0, + "type": "STRING" + } + ], + "extra": { + "workflowRendererVersion": "LG" + }, + "description": "Encodes a negative text prompt via CLIP for classifier-free guidance in anime-style generation (NetaYume Lumina)." + } + ] + }, + "extra": { + "ue_links": [] + } +} diff --git a/blueprints/Text to Image (Qwen-Image 2512).json b/blueprints/Text to Image (Qwen-Image 2512).json new file mode 100644 index 000000000..09612be8b --- /dev/null +++ b/blueprints/Text to Image (Qwen-Image 2512).json @@ -0,0 +1,1952 @@ +{ + "revision": 0, + "last_node_id": 263, + "last_link_id": 0, + "nodes": [ + { + "id": 263, + "type": "fd6ee5f8-a0a9-487a-8b44-8cb65957532a", + "pos": [ + 750, + 760 + ], + "size": [ + 400, + 0 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + }, + { + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": null + }, + { + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": null + }, + { + "label": "enable_turbo_mode", + "name": "value", + "type": "BOOLEAN", + "widget": { + "name": "value" + }, + "link": null + }, + { + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": null + }, + { + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": null + }, + { + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": null + }, + { + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": null + }, + { + "name": "lora_name", + "type": "COMBO", + "widget": { + "name": "lora_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [] + } + ], + "properties": { + "proxyWidgets": [ + [ + "249", + "text" + ], + [ + "252", + "width" + ], + [ + "252", + "height" + ], + [ + "256", + "value" + ], + [ + "253", + "seed" + ], + [ + "248", + "unet_name" + ], + [ + "245", + "clip_name" + ], + [ + "246", + "vae_name" + ], + [ + "259", + "lora_name" + ] + ], + "ue_properties": { + "widget_ue_connectable": { + "value": true + }, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.4", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [], + "title": "Text to Image (Qwen-Image 2512)" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "fd6ee5f8-a0a9-487a-8b44-8cb65957532a", + "version": 1, + "state": { + "lastGroupId": 7, + "lastNodeId": 263, + "lastLinkId": 375, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Text to Image (Qwen-Image 2512)", + "inputNode": { + "id": -10, + "bounding": [ + -1080, + 1480, + 151.744140625, + 220 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 1550, + 1460, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "74d26021-a723-4a90-8e33-5d805a7e5deb", + "name": "text", + "type": "STRING", + "linkIds": [ + 360 + ], + "pos": [ + -948.255859375, + 1500 + ] + }, + { + "id": "b55f69e6-c7cb-4641-9e1f-2cb1c1942ed0", + "name": "width", + "type": "INT", + "linkIds": [ + 361 + ], + "pos": [ + -948.255859375, + 1520 + ] + }, + { + "id": "3e80284d-aba3-43cd-ab7b-ac2a619ef18c", + "name": "height", + "type": "INT", + "linkIds": [ + 362 + ], + "pos": [ + -948.255859375, + 1540 + ] + }, + { + "id": "de06e137-6cec-4cb3-a6bb-737022310a7b", + "name": "value", + "type": "BOOLEAN", + "linkIds": [ + 370 + ], + "label": "enable_turbo_mode", + "pos": [ + -948.255859375, + 1560 + ] + }, + { + "id": "9e500dee-a5b9-481b-ac46-64bab4bd3530", + "name": "seed", + "type": "INT", + "linkIds": [ + 371 + ], + "pos": [ + -948.255859375, + 1580 + ] + }, + { + "id": "33422b12-24e5-41c6-96fc-f9a8dadd5d94", + "name": "unet_name", + "type": "COMBO", + "linkIds": [ + 372 + ], + "pos": [ + -948.255859375, + 1600 + ] + }, + { + "id": "5cf753e4-236e-468e-9a06-6b8e238badc8", + "name": "clip_name", + "type": "COMBO", + "linkIds": [ + 373 + ], + "pos": [ + -948.255859375, + 1620 + ] + }, + { + "id": "790e775c-a639-4e5f-9007-e2ee6764dc5e", + "name": "vae_name", + "type": "COMBO", + "linkIds": [ + 374 + ], + "pos": [ + -948.255859375, + 1640 + ] + }, + { + "id": "3ebed521-3fe9-4922-ae26-2483e03d9305", + "name": "lora_name", + "type": "COMBO", + "linkIds": [ + 375 + ], + "pos": [ + -948.255859375, + 1660 + ] + } + ], + "outputs": [ + { + "id": "7db1f9e2-40ee-4f9f-bb24-a0db7b96d45e", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 333 + ], + "localized_name": "IMAGE", + "pos": [ + 1570, + 1480 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 245, + "type": "CLIPLoader", + "pos": [ + -590, + 1370 + ], + "size": [ + 280, + 150 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "clip_name", + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": 373 + }, + { + "localized_name": "type", + "name": "type", + "type": "COMBO", + "widget": { + "name": "type" + }, + "link": null + }, + { + "localized_name": "device", + "name": "device", + "shape": 7, + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "slot_index": 0, + "links": [ + 314, + 315 + ] + } + ], + "properties": { + "ue_properties": { + "version": "7.7", + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.48", + "Node name for S&R": "CLIPLoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "qwen_2.5_vl_7b_fp8_scaled.safetensors", + "url": "https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/resolve/main/split_files/text_encoders/qwen_2.5_vl_7b_fp8_scaled.safetensors", + "directory": "text_encoders" + }, + { + "name": "qwen_2.5_vl_7b_fp8_scaled.safetensors", + "url": "https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/resolve/main/split_files/text_encoders/qwen_2.5_vl_7b_fp8_scaled.safetensors", + "directory": "text_encoders" + } + ] + }, + "widgets_values": [ + "qwen_2.5_vl_7b_fp8_scaled.safetensors", + "qwen_image", + "default" + ] + }, + { + "id": 246, + "type": "VAELoader", + "pos": [ + -580, + 1620 + ], + "size": [ + 280, + 110 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "vae_name", + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": 374 + } + ], + "outputs": [ + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "slot_index": 0, + "links": [ + 323 + ] + } + ], + "properties": { + "ue_properties": { + "version": "7.7", + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.48", + "Node name for S&R": "VAELoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "qwen_image_vae.safetensors", + "url": "https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/resolve/main/split_files/vae/qwen_image_vae.safetensors", + "directory": "vae" + }, + { + "name": "qwen_image_vae.safetensors", + "url": "https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/resolve/main/split_files/vae/qwen_image_vae.safetensors", + "directory": "vae" + } + ] + }, + "widgets_values": [ + "qwen_image_vae.safetensors" + ] + }, + { + "id": 247, + "type": "ModelSamplingAuraFlow", + "pos": [ + 1040, + 1110 + ], + "size": [ + 250, + 110 + ], + "flags": {}, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 367 + }, + { + "localized_name": "shift", + "name": "shift", + "type": "FLOAT", + "widget": { + "name": "shift" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 316 + ] + } + ], + "properties": { + "ue_properties": { + "version": "7.7", + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.48", + "Node name for S&R": "ModelSamplingAuraFlow", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 3.1000000000000005 + ] + }, + { + "id": 248, + "type": "UNETLoader", + "pos": [ + -590, + 1140 + ], + "size": [ + 280, + 130 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "unet_name", + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": 372 + }, + { + "localized_name": "weight_dtype", + "name": "weight_dtype", + "type": "COMBO", + "widget": { + "name": "weight_dtype" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "slot_index": 0, + "links": [ + 312, + 324 + ] + } + ], + "properties": { + "ue_properties": { + "version": "7.7", + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.48", + "Node name for S&R": "UNETLoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "qwen_image_2512_fp8_e4m3fn.safetensors", + "url": "https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/resolve/main/split_files/diffusion_models/qwen_image_2512_fp8_e4m3fn.safetensors", + "directory": "diffusion_models" + }, + { + "name": "qwen_image_2512_fp8_e4m3fn.safetensors", + "url": "https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/resolve/main/split_files/diffusion_models/qwen_image_2512_fp8_e4m3fn.safetensors", + "directory": "diffusion_models" + } + ] + }, + "widgets_values": [ + "qwen_image_2512_fp8_e4m3fn.safetensors", + "default" + ] + }, + { + "id": 249, + "type": "CLIPTextEncode", + "pos": [ + -200, + 1140 + ], + "size": [ + 360, + 420 + ], + "flags": {}, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 314 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 360 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "slot_index": 0, + "links": [ + 317 + ] + } + ], + "title": "CLIP Text Encode (Positive Prompt)", + "properties": { + "ue_properties": { + "version": "7.7", + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.48", + "Node name for S&R": "CLIPTextEncode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "" + ], + "color": "#232", + "bgcolor": "#353" + }, + { + "id": 250, + "type": "CLIPTextEncode", + "pos": [ + -200, + 1610 + ], + "size": [ + 370, + 170 + ], + "flags": {}, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 315 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "slot_index": 0, + "links": [ + 318 + ] + } + ], + "title": "CLIP Text Encode (Negative Prompt)", + "properties": { + "ue_properties": { + "version": "7.7", + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.48", + "Node name for S&R": "CLIPTextEncode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "低分辨率，低画质，肢体畸形，手指畸形，画面过饱和，蜡像感，人脸无细节，过度光滑，画面具有AI感。构图混乱。文字模糊，扭曲" + ], + "color": "#322", + "bgcolor": "#533" + }, + { + "id": 251, + "type": "VAEDecode", + "pos": [ + 1320, + 1120 + ], + "size": [ + 230, + 100 + ], + "flags": { + "collapsed": false + }, + "order": 10, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 322 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 323 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "slot_index": 0, + "links": [ + 333 + ] + } + ], + "properties": { + "ue_properties": { + "version": "7.7", + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.48", + "Node name for S&R": "VAEDecode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 252, + "type": "EmptySD3LatentImage", + "pos": [ + -550, + 1930 + ], + "size": [ + 230, + 170 + ], + "flags": {}, + "order": 11, + "mode": 0, + "inputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 361 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 362 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "links": [ + 319 + ] + } + ], + "properties": { + "ue_properties": { + "version": "7.7", + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.48", + "Node name for S&R": "EmptySD3LatentImage", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1328, + 1328, + 1 + ] + }, + { + "id": 253, + "type": "KSampler", + "pos": [ + 1040, + 1250 + ], + "size": [ + 250, + 350 + ], + "flags": {}, + "order": 12, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 316 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 317 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 318 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 319 + }, + { + "localized_name": "seed", + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": 371 + }, + { + "localized_name": "steps", + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": 368 + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": 369 + }, + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + }, + { + "localized_name": "scheduler", + "name": "scheduler", + "type": "COMBO", + "widget": { + "name": "scheduler" + }, + "link": null + }, + { + "localized_name": "denoise", + "name": "denoise", + "type": "FLOAT", + "widget": { + "name": "denoise" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 322 + ] + } + ], + "properties": { + "ue_properties": { + "version": "7.7", + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.48", + "Node name for S&R": "KSampler", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 464857551335368, + "randomize", + 50, + 4, + "euler", + "simple", + 1 + ] + }, + { + "id": 254, + "type": "PrimitiveInt", + "pos": [ + 300, + 1150 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 355 + ] + } + ], + "title": "Int (Steps)", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.12.3", + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 50, + "fixed" + ] + }, + { + "id": 255, + "type": "PrimitiveFloat", + "pos": [ + 300, + 1290 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "FLOAT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": [ + 357 + ] + } + ], + "title": "Float (CFG)", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.12.3", + "Node name for S&R": "PrimitiveFloat", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 4 + ] + }, + { + "id": 256, + "type": "PrimitiveBoolean", + "pos": [ + 300, + 2060 + ], + "size": [ + 230, + 100 + ], + "flags": {}, + "order": 13, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "BOOLEAN", + "widget": { + "name": "value" + }, + "link": 370 + } + ], + "outputs": [ + { + "localized_name": "BOOLEAN", + "name": "BOOLEAN", + "type": "BOOLEAN", + "links": [ + 326, + 358, + 359 + ] + } + ], + "title": "Enable 4 Steps LoRA?", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.12.3", + "Node name for S&R": "PrimitiveBoolean", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + false + ] + }, + { + "id": 257, + "type": "PrimitiveInt", + "pos": [ + 290, + 1540 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 347, + 354 + ] + } + ], + "title": "Int (Steps)", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.12.3", + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 4, + "fixed" + ] + }, + { + "id": 258, + "type": "PrimitiveFloat", + "pos": [ + 290, + 1670 + ], + "size": [ + 230, + 110 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "FLOAT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": [ + 356 + ] + } + ], + "title": "Float (CFG)", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.12.3", + "Node name for S&R": "PrimitiveFloat", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1 + ] + }, + { + "id": 259, + "type": "LoraLoaderModelOnly", + "pos": [ + 240, + 1820 + ], + "size": [ + 330, + 140 + ], + "flags": {}, + "order": 14, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 312 + }, + { + "localized_name": "lora_name", + "name": "lora_name", + "type": "COMBO", + "widget": { + "name": "lora_name" + }, + "link": 375 + }, + { + "localized_name": "strength_model", + "name": "strength_model", + "type": "FLOAT", + "widget": { + "name": "strength_model" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 325 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.49", + "Node name for S&R": "LoraLoaderModelOnly", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "Qwen-Image-2512-Lightning-4steps-V1.0-fp32.safetensors", + "url": "https://huggingface.co/lightx2v/Qwen-Image-2512-Lightning/resolve/main/Qwen-Image-2512-Lightning-4steps-V1.0-fp32.safetensors", + "directory": "loras" + } + ] + }, + "widgets_values": [ + "Qwen-Image-2512-Lightning-4steps-V1.0-fp32.safetensors", + 1 + ] + }, + { + "id": 260, + "type": "ComfySwitchNode", + "pos": [ + 710, + 1170 + ], + "size": [ + 230, + 130 + ], + "flags": {}, + "order": 15, + "mode": 0, + "inputs": [ + { + "localized_name": "on_false", + "name": "on_false", + "type": "*", + "link": 324 + }, + { + "localized_name": "on_true", + "name": "on_true", + "type": "*", + "link": 325 + }, + { + "localized_name": "switch", + "name": "switch", + "type": "BOOLEAN", + "widget": { + "name": "switch" + }, + "link": 326 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "*", + "links": [ + 367 + ] + } + ], + "title": "Switch (model)", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.12.3", + "Node name for S&R": "ComfySwitchNode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + false + ] + }, + { + "id": 261, + "type": "ComfySwitchNode", + "pos": [ + 710, + 1420 + ], + "size": [ + 230, + 130 + ], + "flags": {}, + "order": 16, + "mode": 0, + "inputs": [ + { + "localized_name": "on_false", + "name": "on_false", + "type": "*", + "link": 355 + }, + { + "localized_name": "on_true", + "name": "on_true", + "type": "*", + "link": 354 + }, + { + "localized_name": "switch", + "name": "switch", + "type": "BOOLEAN", + "widget": { + "name": "switch" + }, + "link": 359 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "*", + "links": [ + 368 + ] + } + ], + "title": "Switch (steps)", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.12.3", + "Node name for S&R": "ComfySwitchNode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + false + ] + }, + { + "id": 262, + "type": "ComfySwitchNode", + "pos": [ + 710, + 1660 + ], + "size": [ + 230, + 130 + ], + "flags": {}, + "order": 17, + "mode": 0, + "inputs": [ + { + "localized_name": "on_false", + "name": "on_false", + "type": "*", + "link": 357 + }, + { + "localized_name": "on_true", + "name": "on_true", + "type": "*", + "link": 356 + }, + { + "localized_name": "switch", + "name": "switch", + "type": "BOOLEAN", + "widget": { + "name": "switch" + }, + "link": 358 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "*", + "links": [ + 369 + ] + } + ], + "title": "Switch (cfg)", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.12.3", + "Node name for S&R": "ComfySwitchNode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + false + ] + } + ], + "groups": [ + { + "id": 1, + "title": "Model", + "bounding": [ + -640, + 1060, + 390, + 740 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 2, + "title": "Image size", + "bounding": [ + -630, + 1830, + 380, + 290 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 3, + "title": "Prompt", + "bounding": [ + -220, + 1060, + 400, + 740 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 5, + "title": "4-steps LoRA", + "bounding": [ + 210, + 1460, + 410, + 550 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 6, + "title": "Original Settings", + "bounding": [ + 210, + 1060, + 420, + 370 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 7, + "title": "Swtich", + "bounding": [ + 660, + 1060, + 320, + 750 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 312, + "origin_id": 248, + "origin_slot": 0, + "target_id": 259, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 314, + "origin_id": 245, + "origin_slot": 0, + "target_id": 249, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 315, + "origin_id": 245, + "origin_slot": 0, + "target_id": 250, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 322, + "origin_id": 253, + "origin_slot": 0, + "target_id": 251, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 323, + "origin_id": 246, + "origin_slot": 0, + "target_id": 251, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 316, + "origin_id": 247, + "origin_slot": 0, + "target_id": 253, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 317, + "origin_id": 249, + "origin_slot": 0, + "target_id": 253, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 318, + "origin_id": 250, + "origin_slot": 0, + "target_id": 253, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 319, + "origin_id": 252, + "origin_slot": 0, + "target_id": 253, + "target_slot": 3, + "type": "LATENT" + }, + { + "id": 324, + "origin_id": 248, + "origin_slot": 0, + "target_id": 260, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 325, + "origin_id": 259, + "origin_slot": 0, + "target_id": 260, + "target_slot": 1, + "type": "MODEL" + }, + { + "id": 326, + "origin_id": 256, + "origin_slot": 0, + "target_id": 260, + "target_slot": 2, + "type": "BOOLEAN" + }, + { + "id": 333, + "origin_id": 251, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 347, + "origin_id": 257, + "origin_slot": 0, + "target_id": 253, + "target_slot": 4, + "type": "INT" + }, + { + "id": 354, + "origin_id": 257, + "origin_slot": 0, + "target_id": 261, + "target_slot": 1, + "type": "INT" + }, + { + "id": 355, + "origin_id": 254, + "origin_slot": 0, + "target_id": 261, + "target_slot": 0, + "type": "INT" + }, + { + "id": 356, + "origin_id": 258, + "origin_slot": 0, + "target_id": 262, + "target_slot": 1, + "type": "FLOAT" + }, + { + "id": 357, + "origin_id": 255, + "origin_slot": 0, + "target_id": 262, + "target_slot": 0, + "type": "FLOAT" + }, + { + "id": 358, + "origin_id": 256, + "origin_slot": 0, + "target_id": 262, + "target_slot": 2, + "type": "BOOLEAN" + }, + { + "id": 359, + "origin_id": 256, + "origin_slot": 0, + "target_id": 261, + "target_slot": 2, + "type": "BOOLEAN" + }, + { + "id": 360, + "origin_id": -10, + "origin_slot": 0, + "target_id": 249, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 361, + "origin_id": -10, + "origin_slot": 1, + "target_id": 252, + "target_slot": 0, + "type": "INT" + }, + { + "id": 362, + "origin_id": -10, + "origin_slot": 2, + "target_id": 252, + "target_slot": 1, + "type": "INT" + }, + { + "id": 367, + "origin_id": 260, + "origin_slot": 0, + "target_id": 247, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 368, + "origin_id": 261, + "origin_slot": 0, + "target_id": 253, + "target_slot": 5, + "type": "INT" + }, + { + "id": 369, + "origin_id": 262, + "origin_slot": 0, + "target_id": 253, + "target_slot": 6, + "type": "FLOAT" + }, + { + "id": 370, + "origin_id": -10, + "origin_slot": 3, + "target_id": 256, + "target_slot": 0, + "type": "BOOLEAN" + }, + { + "id": 371, + "origin_id": -10, + "origin_slot": 4, + "target_id": 253, + "target_slot": 4, + "type": "INT" + }, + { + "id": 372, + "origin_id": -10, + "origin_slot": 5, + "target_id": 248, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 373, + "origin_id": -10, + "origin_slot": 6, + "target_id": 245, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 374, + "origin_id": -10, + "origin_slot": 7, + "target_id": 246, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 375, + "origin_id": -10, + "origin_slot": 8, + "target_id": 259, + "target_slot": 1, + "type": "COMBO" + } + ], + "extra": { + "workflowRendererVersion": "Vue-corrected" + }, + "category": "Image generation and editing/Text to image", + "description": "Generates images from text prompts using Qwen-Image-2512, with enhanced human realism and finer natural detail over the base version." + } + ] + }, + "extra": { + "ue_links": [] + } +} \ No newline at end of file diff --git a/blueprints/Text to Image (Qwen-Image).json b/blueprints/Text to Image (Qwen-Image).json new file mode 100644 index 000000000..e78d5a962 --- /dev/null +++ b/blueprints/Text to Image (Qwen-Image).json @@ -0,0 +1,1882 @@ +{ + "revision": 0, + "last_node_id": 76, + "last_link_id": 0, + "nodes": [ + { + "id": 76, + "type": "e5cfe5ba-2ae0-4bc4-869f-ab2228cb44d3", + "pos": [ + 30, + 10 + ], + "size": [ + 470, + 660 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "label": "prompt", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + }, + { + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": null + }, + { + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": null + }, + { + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": null + }, + { + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": null + }, + { + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": null + }, + { + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": null + }, + { + "label": "lightning_lora", + "name": "lora_name", + "type": "COMBO", + "widget": { + "name": "lora_name" + }, + "link": null + }, + { + "label": "enable_turbo_mode", + "name": "value", + "type": "BOOLEAN", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [] + } + ], + "properties": { + "proxyWidgets": [ + [ + "6", + "text" + ], + [ + "58", + "width" + ], + [ + "58", + "height" + ], + [ + "3", + "seed" + ], + [ + "37", + "unet_name" + ], + [ + "38", + "clip_name" + ], + [ + "39", + "vae_name" + ], + [ + "73", + "lora_name" + ], + [ + "86", + "value" + ], + [ + "3", + "control_after_generate" + ] + ], + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": { + "text": true, + "lora_name": true, + "value": true + }, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [], + "title": "Text to Image (Qwen-Image)" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "e5cfe5ba-2ae0-4bc4-869f-ab2228cb44d3", + "version": 1, + "state": { + "lastGroupId": 5, + "lastNodeId": 87, + "lastLinkId": 153, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Text to Image (Qwen-Image)", + "inputNode": { + "id": -10, + "bounding": [ + -810, + 290, + 151.744140625, + 220 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 2580, + 340, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "846fd1a5-9f4a-4e83-af40-27cafe99e5c6", + "name": "text", + "type": "STRING", + "linkIds": [ + 132 + ], + "label": "prompt", + "pos": [ + -678.255859375, + 310 + ] + }, + { + "id": "e941d29f-bb7f-4001-a956-90a9b29ae9f9", + "name": "width", + "type": "INT", + "linkIds": [ + 134 + ], + "pos": [ + -678.255859375, + 330 + ] + }, + { + "id": "df798f50-87ba-481b-b847-ca8b7c7efff3", + "name": "height", + "type": "INT", + "linkIds": [ + 135 + ], + "pos": [ + -678.255859375, + 350 + ] + }, + { + "id": "3fcf7667-f697-43ee-bdee-0d3fed39e777", + "name": "seed", + "type": "INT", + "linkIds": [ + 136 + ], + "pos": [ + -678.255859375, + 370 + ] + }, + { + "id": "e8d70f26-d9f5-4633-a39e-0bf6cf93d566", + "name": "unet_name", + "type": "COMBO", + "linkIds": [ + 137 + ], + "pos": [ + -678.255859375, + 390 + ] + }, + { + "id": "8c9b537a-c6c9-4365-96ad-dbbb82d917e0", + "name": "clip_name", + "type": "COMBO", + "linkIds": [ + 138 + ], + "pos": [ + -678.255859375, + 410 + ] + }, + { + "id": "7cc2f92b-6e2f-4e4e-a316-b61f58ed1442", + "name": "vae_name", + "type": "COMBO", + "linkIds": [ + 139 + ], + "pos": [ + -678.255859375, + 430 + ] + }, + { + "id": "3cb1ba7c-583c-4f92-afc1-71463161e2a4", + "name": "lora_name", + "type": "COMBO", + "linkIds": [ + 140 + ], + "label": "lightning_lora", + "pos": [ + -678.255859375, + 450 + ] + }, + { + "id": "4278102d-766c-4c6b-af2e-0fb9f26bbb27", + "name": "value", + "type": "BOOLEAN", + "linkIds": [ + 153 + ], + "label": "enable_turbo_mode", + "pos": [ + -678.255859375, + 470 + ] + } + ], + "outputs": [ + { + "id": "2af20250-dc7a-4643-bc84-0a180d9ca62b", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 110 + ], + "localized_name": "IMAGE", + "pos": [ + 2600, + 360 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 39, + "type": "VAELoader", + "pos": [ + -260, + 510 + ], + "size": [ + 330, + 110 + ], + "flags": {}, + "order": 11, + "mode": 0, + "inputs": [ + { + "localized_name": "vae_name", + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": 139 + } + ], + "outputs": [ + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "slot_index": 0, + "links": [ + 76 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.48", + "ue_properties": { + "version": "7.7", + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "VAELoader", + "models": [ + { + "name": "qwen_image_vae.safetensors", + "url": "https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/resolve/main/split_files/vae/qwen_image_vae.safetensors", + "directory": "vae" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "qwen_image_vae.safetensors" + ] + }, + { + "id": 38, + "type": "CLIPLoader", + "pos": [ + -260, + 280 + ], + "size": [ + 330, + 150 + ], + "flags": {}, + "order": 10, + "mode": 0, + "inputs": [ + { + "localized_name": "clip_name", + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": 138 + }, + { + "localized_name": "type", + "name": "type", + "type": "COMBO", + "widget": { + "name": "type" + }, + "link": null + }, + { + "localized_name": "device", + "name": "device", + "shape": 7, + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "slot_index": 0, + "links": [ + 74, + 75 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.48", + "ue_properties": { + "version": "7.7", + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "CLIPLoader", + "models": [ + { + "name": "qwen_2.5_vl_7b_fp8_scaled.safetensors", + "url": "https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/resolve/main/split_files/text_encoders/qwen_2.5_vl_7b_fp8_scaled.safetensors", + "directory": "text_encoders" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "qwen_2.5_vl_7b_fp8_scaled.safetensors", + "qwen_image", + "default" + ] + }, + { + "id": 58, + "type": "EmptySD3LatentImage", + "pos": [ + -240, + 810 + ], + "size": [ + 270, + 170 + ], + "flags": {}, + "order": 12, + "mode": 0, + "inputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 134 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 135 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "links": [ + 107 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.48", + "ue_properties": { + "version": "7.7", + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "EmptySD3LatentImage", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1328, + 1328, + 1 + ] + }, + { + "id": 66, + "type": "ModelSamplingAuraFlow", + "pos": [ + 1780, + 180 + ], + "size": [ + 300, + 110 + ], + "flags": {}, + "order": 13, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 147 + }, + { + "localized_name": "shift", + "name": "shift", + "type": "FLOAT", + "widget": { + "name": "shift" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 125 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.48", + "ue_properties": { + "version": "7.7", + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ModelSamplingAuraFlow", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 3.1000000000000005 + ] + }, + { + "id": 37, + "type": "UNETLoader", + "pos": [ + -260, + 80 + ], + "size": [ + 330, + 110 + ], + "flags": {}, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "unet_name", + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": 137 + }, + { + "localized_name": "weight_dtype", + "name": "weight_dtype", + "type": "COMBO", + "widget": { + "name": "weight_dtype" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "slot_index": 0, + "links": [ + 129, + 142 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.48", + "ue_properties": { + "version": "7.7", + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "UNETLoader", + "models": [ + { + "name": "qwen_image_fp8_e4m3fn.safetensors", + "url": "https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/resolve/main/split_files/diffusion_models/qwen_image_fp8_e4m3fn.safetensors", + "directory": "diffusion_models" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "qwen_image_fp8_e4m3fn.safetensors", + "default" + ] + }, + { + "id": 6, + "type": "CLIPTextEncode", + "pos": [ + 120, + 60 + ], + "size": [ + 440, + 340 + ], + "flags": {}, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 74 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 132 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "slot_index": 0, + "links": [ + 46 + ] + } + ], + "title": "CLIP Text Encode (Positive Prompt)", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.48", + "ue_properties": { + "version": "7.7", + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "CLIPTextEncode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "" + ], + "color": "#232", + "bgcolor": "#353" + }, + { + "id": 7, + "type": "CLIPTextEncode", + "pos": [ + 130, + 480 + ], + "size": [ + 430, + 180 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 75 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "slot_index": 0, + "links": [ + 52 + ] + } + ], + "title": "CLIP Text Encode (Negative Prompt)", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.48", + "ue_properties": { + "version": "7.7", + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "CLIPTextEncode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "" + ], + "color": "#322", + "bgcolor": "#533" + }, + { + "id": 8, + "type": "VAEDecode", + "pos": [ + 2190, + 350 + ], + "size": [ + 230, + 100 + ], + "flags": { + "collapsed": false + }, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 128 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 76 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "slot_index": 0, + "links": [ + 110 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.48", + "ue_properties": { + "version": "7.7", + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "VAEDecode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 73, + "type": "LoraLoaderModelOnly", + "pos": [ + 670, + 500 + ], + "size": [ + 400, + 140 + ], + "flags": {}, + "order": 14, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 129 + }, + { + "localized_name": "lora_name", + "name": "lora_name", + "type": "COMBO", + "widget": { + "name": "lora_name" + }, + "link": 140 + }, + { + "localized_name": "strength_model", + "name": "strength_model", + "type": "FLOAT", + "widget": { + "name": "strength_model" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 141 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.49", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "LoraLoaderModelOnly", + "models": [ + { + "name": "Qwen-Image-Lightning-8steps-V1.0.safetensors", + "url": "https://huggingface.co/lightx2v/Qwen-Image-Lightning/resolve/main/Qwen-Image-Lightning-8steps-V1.0.safetensors", + "directory": "loras" + } + ] + }, + "widgets_values": [ + "Qwen-Image-Lightning-8steps-V1.0.safetensors", + 1 + ] + }, + { + "id": 3, + "type": "KSampler", + "pos": [ + 1780, + 330 + ], + "size": [ + 300, + 480 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 125 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 46 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 52 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 107 + }, + { + "localized_name": "seed", + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": 136 + }, + { + "localized_name": "steps", + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": 148 + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": 149 + }, + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + }, + { + "localized_name": "scheduler", + "name": "scheduler", + "type": "COMBO", + "widget": { + "name": "scheduler" + }, + "link": null + }, + { + "localized_name": "denoise", + "name": "denoise", + "type": "FLOAT", + "widget": { + "name": "denoise" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 128 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.48", + "ue_properties": { + "version": "7.7", + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "Node name for S&R": "KSampler", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 50347169638278, + "randomize", + 8, + 1, + "euler", + "simple", + 1 + ] + }, + { + "id": 78, + "type": "ComfySwitchNode", + "pos": [ + 1320, + 180 + ], + "size": [ + 270, + 130 + ], + "flags": {}, + "order": 15, + "mode": 0, + "inputs": [ + { + "localized_name": "on_false", + "name": "on_false", + "type": "*", + "link": 142 + }, + { + "localized_name": "on_true", + "name": "on_true", + "type": "*", + "link": 141 + }, + { + "localized_name": "switch", + "name": "switch", + "type": "BOOLEAN", + "widget": { + "name": "switch" + }, + "link": 150 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "*", + "links": [ + 147 + ] + } + ], + "title": "Switch (Model)", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ComfySwitchNode" + }, + "widgets_values": [ + false + ] + }, + { + "id": 79, + "type": "PrimitiveInt", + "pos": [ + 680, + 710 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 143 + ] + } + ], + "title": "Steps", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "PrimitiveInt" + }, + "widgets_values": [ + 8, + "fixed" + ] + }, + { + "id": 81, + "type": "PrimitiveFloat", + "pos": [ + 680, + 870 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "FLOAT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": [ + 144 + ] + } + ], + "title": "CFG", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "PrimitiveFloat" + }, + "widgets_values": [ + 1 + ] + }, + { + "id": 82, + "type": "ComfySwitchNode", + "pos": [ + 1320, + 400 + ], + "size": [ + 270, + 130 + ], + "flags": {}, + "order": 16, + "mode": 0, + "inputs": [ + { + "localized_name": "on_false", + "name": "on_false", + "type": "*", + "link": 146 + }, + { + "localized_name": "on_true", + "name": "on_true", + "type": "*", + "link": 143 + }, + { + "localized_name": "switch", + "name": "switch", + "type": "BOOLEAN", + "widget": { + "name": "switch" + }, + "link": 151 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "*", + "links": [ + 148 + ] + } + ], + "title": "Switch (Steps)", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ComfySwitchNode" + }, + "widgets_values": [ + false + ] + }, + { + "id": 83, + "type": "ComfySwitchNode", + "pos": [ + 1320, + 600 + ], + "size": [ + 270, + 130 + ], + "flags": {}, + "order": 17, + "mode": 0, + "inputs": [ + { + "localized_name": "on_false", + "name": "on_false", + "type": "*", + "link": 145 + }, + { + "localized_name": "on_true", + "name": "on_true", + "type": "*", + "link": 144 + }, + { + "localized_name": "switch", + "name": "switch", + "type": "BOOLEAN", + "widget": { + "name": "switch" + }, + "link": 152 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "*", + "links": [ + 149 + ] + } + ], + "title": "Switch (CFG)", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ComfySwitchNode" + }, + "widgets_values": [ + false + ] + }, + { + "id": 84, + "type": "PrimitiveInt", + "pos": [ + 680, + 60 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 146 + ] + } + ], + "title": "Steps", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "PrimitiveInt" + }, + "widgets_values": [ + 20, + "fixed" + ] + }, + { + "id": 85, + "type": "PrimitiveFloat", + "pos": [ + 680, + 230 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "FLOAT", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": [ + 145 + ] + } + ], + "title": "CFG", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "PrimitiveFloat" + }, + "widgets_values": [ + 4 + ] + }, + { + "id": 86, + "type": "PrimitiveBoolean", + "pos": [ + 710, + 1070 + ], + "size": [ + 270, + 100 + ], + "flags": {}, + "order": 18, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "BOOLEAN", + "widget": { + "name": "value" + }, + "link": 153 + } + ], + "outputs": [ + { + "localized_name": "BOOLEAN", + "name": "BOOLEAN", + "type": "BOOLEAN", + "links": [ + 150, + 151, + 152 + ] + } + ], + "title": "Enable Lightning LoRA", + "properties": { + "cnr_id": "comfy-core", + "ver": "0.18.1", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "PrimitiveBoolean" + }, + "widgets_values": [ + false + ] + }, + { + "id": 87, + "type": "MarkdownNote", + "pos": [ + 620, + -160 + ], + "size": [ + 500, + 120 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [], + "outputs": [], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + }, + "widgets_values": [ + "Try 50 steps, if you want original the [qwen image](https://huggingface.co/Qwen/Qwen-Image)'s setting, but it will takes longer" + ], + "color": "#222", + "bgcolor": "#000" + } + ], + "groups": [ + { + "id": 1, + "title": "Step1 - Load models", + "bounding": [ + -280, + -20, + 360, + 700 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 2, + "title": "Step2 - Image size", + "bounding": [ + -280, + 710, + 360, + 300 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 3, + "title": "Step3 - Prompt", + "bounding": [ + 110, + -20, + 470, + 700 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 4, + "title": "Lightx2v 8steps LoRA", + "bounding": [ + 610, + 390, + 520, + 620 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 5, + "title": "Original Settings", + "bounding": [ + 610, + -20, + 520, + 380 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 74, + "origin_id": 38, + "origin_slot": 0, + "target_id": 6, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 75, + "origin_id": 38, + "origin_slot": 0, + "target_id": 7, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 129, + "origin_id": 37, + "origin_slot": 0, + "target_id": 73, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 128, + "origin_id": 3, + "origin_slot": 0, + "target_id": 8, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 76, + "origin_id": 39, + "origin_slot": 0, + "target_id": 8, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 125, + "origin_id": 66, + "origin_slot": 0, + "target_id": 3, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 46, + "origin_id": 6, + "origin_slot": 0, + "target_id": 3, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 52, + "origin_id": 7, + "origin_slot": 0, + "target_id": 3, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 107, + "origin_id": 58, + "origin_slot": 0, + "target_id": 3, + "target_slot": 3, + "type": "LATENT" + }, + { + "id": 110, + "origin_id": 8, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 132, + "origin_id": -10, + "origin_slot": 0, + "target_id": 6, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 134, + "origin_id": -10, + "origin_slot": 1, + "target_id": 58, + "target_slot": 0, + "type": "INT" + }, + { + "id": 135, + "origin_id": -10, + "origin_slot": 2, + "target_id": 58, + "target_slot": 1, + "type": "INT" + }, + { + "id": 136, + "origin_id": -10, + "origin_slot": 3, + "target_id": 3, + "target_slot": 4, + "type": "INT" + }, + { + "id": 137, + "origin_id": -10, + "origin_slot": 4, + "target_id": 37, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 138, + "origin_id": -10, + "origin_slot": 5, + "target_id": 38, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 139, + "origin_id": -10, + "origin_slot": 6, + "target_id": 39, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 140, + "origin_id": -10, + "origin_slot": 7, + "target_id": 73, + "target_slot": 1, + "type": "COMBO" + }, + { + "id": 141, + "origin_id": 73, + "origin_slot": 0, + "target_id": 78, + "target_slot": 1, + "type": "MODEL" + }, + { + "id": 142, + "origin_id": 37, + "origin_slot": 0, + "target_id": 78, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 143, + "origin_id": 79, + "origin_slot": 0, + "target_id": 82, + "target_slot": 1, + "type": "INT" + }, + { + "id": 144, + "origin_id": 81, + "origin_slot": 0, + "target_id": 83, + "target_slot": 1, + "type": "FLOAT" + }, + { + "id": 145, + "origin_id": 85, + "origin_slot": 0, + "target_id": 83, + "target_slot": 0, + "type": "FLOAT" + }, + { + "id": 146, + "origin_id": 84, + "origin_slot": 0, + "target_id": 82, + "target_slot": 0, + "type": "INT" + }, + { + "id": 147, + "origin_id": 78, + "origin_slot": 0, + "target_id": 66, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 148, + "origin_id": 82, + "origin_slot": 0, + "target_id": 3, + "target_slot": 5, + "type": "INT" + }, + { + "id": 149, + "origin_id": 83, + "origin_slot": 0, + "target_id": 3, + "target_slot": 6, + "type": "FLOAT" + }, + { + "id": 150, + "origin_id": 86, + "origin_slot": 0, + "target_id": 78, + "target_slot": 2, + "type": "BOOLEAN" + }, + { + "id": 151, + "origin_id": 86, + "origin_slot": 0, + "target_id": 82, + "target_slot": 2, + "type": "BOOLEAN" + }, + { + "id": 152, + "origin_id": 86, + "origin_slot": 0, + "target_id": 83, + "target_slot": 2, + "type": "BOOLEAN" + }, + { + "id": 153, + "origin_id": -10, + "origin_slot": 8, + "target_id": 86, + "target_slot": 0, + "type": "BOOLEAN" + } + ], + "extra": { + "workflowRendererVersion": "LG" + }, + "category": "Image generation and editing/Text to image", + "description": "Generates images from text prompts using Qwen-Image, Alibaba's 20B MMDiT model with excellent multilingual text rendering." + } + ] + }, + "extra": {} +} \ No newline at end of file diff --git a/blueprints/Text to Image (Z-Image-Base).json b/blueprints/Text to Image (Z-Image-Base).json new file mode 100644 index 000000000..169263712 --- /dev/null +++ b/blueprints/Text to Image (Z-Image-Base).json @@ -0,0 +1,1184 @@ +{ + "revision": 0, + "last_node_id": 126, + "last_link_id": 0, + "nodes": [ + { + "id": 126, + "type": "8a2bb267-5858-4aaf-bdcd-61002711af19", + "pos": [ + -2280, + 2850 + ], + "size": [ + 410, + 560 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "label": "prompt", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + }, + { + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": null + }, + { + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": null + }, + { + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": null + }, + { + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": null + }, + { + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": null + }, + { + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": null + }, + { + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": null + }, + { + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [] + } + ], + "properties": { + "proxyWidgets": [ + [ + "67", + "text" + ], + [ + "68", + "width" + ], + [ + "68", + "height" + ], + [ + "69", + "steps" + ], + [ + "69", + "cfg" + ], + [ + "69", + "seed" + ], + [ + "66", + "unet_name" + ], + [ + "62", + "clip_name" + ], + [ + "63", + "vae_name" + ], + [ + "69", + "control_after_generate" + ] + ], + "cnr_id": "comfy-core", + "ver": "0.13.0", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [], + "title": "Text to Image (Z-Image-Base)" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "8a2bb267-5858-4aaf-bdcd-61002711af19", + "version": 1, + "state": { + "lastGroupId": 16, + "lastNodeId": 126, + "lastLinkId": 229, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Text to Image (Z-Image-Base)", + "description": "Generates images from text prompts using Z-Image base weights with Qwen3 text encoder and bundled VAE.", + "inputNode": { + "id": -10, + "bounding": [ + -220, + 40, + 120, + 220 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 1840, + -150, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "af36fee5-4f8b-4a8e-bfa8-cb8fe7006cc3", + "name": "text", + "type": "STRING", + "linkIds": [ + 108 + ], + "label": "prompt", + "pos": [ + -120, + 60 + ] + }, + { + "id": "357f0059-e8e6-41f6-a290-c53b0a60c0ed", + "name": "width", + "type": "INT", + "linkIds": [ + 114 + ], + "pos": [ + -120, + 80 + ] + }, + { + "id": "4a442743-a9c2-4aa5-9efd-05d43f3322d3", + "name": "height", + "type": "INT", + "linkIds": [ + 115 + ], + "pos": [ + -120, + 100 + ] + }, + { + "id": "a0fc336b-d349-418e-8415-318653f7b6b3", + "name": "steps", + "type": "INT", + "linkIds": [ + 116 + ], + "pos": [ + -120, + 120 + ] + }, + { + "id": "2f253ace-1e1a-415f-9b95-a10430bd5749", + "name": "cfg", + "type": "FLOAT", + "linkIds": [ + 117 + ], + "pos": [ + -120, + 140 + ] + }, + { + "id": "18a6ad37-23aa-4bf7-a0cd-1d6ca6e2a128", + "name": "seed", + "type": "INT", + "linkIds": [ + 118 + ], + "pos": [ + -120, + 160 + ] + }, + { + "id": "d1fc4937-8505-4ec6-9fc4-a33ef7b45eee", + "name": "unet_name", + "type": "COMBO", + "linkIds": [ + 119 + ], + "pos": [ + -120, + 180 + ] + }, + { + "id": "db45dd49-d990-4ceb-a849-f96341874cdd", + "name": "clip_name", + "type": "COMBO", + "linkIds": [ + 120 + ], + "pos": [ + -120, + 200 + ] + }, + { + "id": "37b8eac6-9b1b-452b-81f3-0ba9e34a576a", + "name": "vae_name", + "type": "COMBO", + "linkIds": [ + 121 + ], + "pos": [ + -120, + 220 + ] + } + ], + "outputs": [ + { + "id": "f2bea309-bfe7-4ccb-9ffe-9475bf1da2ae", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 79 + ], + "localized_name": "IMAGE", + "pos": [ + 1860, + -130 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 67, + "type": "CLIPTextEncode", + "pos": [ + 600, + -90 + ], + "size": [ + 410, + 320 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 78 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 108 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 75 + ] + } + ], + "title": "CLIP Text Encode (Positive Prompt)", + "properties": { + "Node name for S&R": "CLIPTextEncode", + "cnr_id": "comfy-core", + "ver": "0.3.73", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "" + ], + "color": "#232", + "bgcolor": "#353" + }, + { + "id": 68, + "type": "EmptySD3LatentImage", + "pos": [ + 240, + 620 + ], + "size": [ + 260, + 170 + ], + "flags": {}, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 114 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 115 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 77 + ] + } + ], + "properties": { + "Node name for S&R": "EmptySD3LatentImage", + "cnr_id": "comfy-core", + "ver": "0.3.64", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1024, + 1024, + 1 + ] + }, + { + "id": 63, + "type": "VAELoader", + "pos": [ + 230, + 340 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "vae_name", + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": 121 + } + ], + "outputs": [ + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "links": [ + 73 + ] + } + ], + "properties": { + "Node name for S&R": "VAELoader", + "cnr_id": "comfy-core", + "ver": "0.3.73", + "models": [ + { + "name": "ae.safetensors", + "url": "https://huggingface.co/Comfy-Org/z_image_turbo/resolve/main/split_files/vae/ae.safetensors", + "directory": "vae" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "ae.safetensors" + ] + }, + { + "id": 62, + "type": "CLIPLoader", + "pos": [ + 230, + 110 + ], + "size": [ + 270, + 150 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "clip_name", + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": 120 + }, + { + "localized_name": "type", + "name": "type", + "type": "COMBO", + "widget": { + "name": "type" + }, + "link": null + }, + { + "localized_name": "device", + "name": "device", + "shape": 7, + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [ + 78, + 82 + ] + } + ], + "properties": { + "Node name for S&R": "CLIPLoader", + "cnr_id": "comfy-core", + "ver": "0.3.73", + "models": [ + { + "name": "qwen_3_4b.safetensors", + "url": "https://huggingface.co/Comfy-Org/z_image_turbo/resolve/main/split_files/text_encoders/qwen_3_4b.safetensors", + "directory": "text_encoders" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "qwen_3_4b.safetensors", + "lumina2", + "default" + ] + }, + { + "id": 65, + "type": "VAEDecode", + "pos": [ + 1450, + -150 + ], + "size": [ + 230, + 100 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 72 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 73 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "slot_index": 0, + "links": [ + 79 + ] + } + ], + "properties": { + "Node name for S&R": "VAEDecode", + "cnr_id": "comfy-core", + "ver": "0.3.64", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 70, + "type": "ModelSamplingAuraFlow", + "pos": [ + 1100, + -150 + ], + "size": [ + 310, + 110 + ], + "flags": {}, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 109 + }, + { + "localized_name": "shift", + "name": "shift", + "type": "FLOAT", + "widget": { + "name": "shift" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "slot_index": 0, + "links": [ + 74 + ] + } + ], + "properties": { + "Node name for S&R": "ModelSamplingAuraFlow", + "cnr_id": "comfy-core", + "ver": "0.3.64", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 3 + ] + }, + { + "id": 66, + "type": "UNETLoader", + "pos": [ + 230, + -90 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "unet_name", + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": 119 + }, + { + "localized_name": "weight_dtype", + "name": "weight_dtype", + "type": "COMBO", + "widget": { + "name": "weight_dtype" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 109 + ] + } + ], + "properties": { + "Node name for S&R": "UNETLoader", + "cnr_id": "comfy-core", + "ver": "0.3.73", + "models": [ + { + "name": "z_image_bf16.safetensors", + "url": "https://huggingface.co/Comfy-Org/z_image/resolve/main/split_files/diffusion_models/z_image_bf16.safetensors", + "directory": "diffusion_models" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "z_image_bf16.safetensors", + "default" + ] + }, + { + "id": 71, + "type": "CLIPTextEncode", + "pos": [ + 600, + 310 + ], + "size": [ + 390, + 140 + ], + "flags": {}, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 82 + }, + { + "label": "prompt", + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 83 + ] + } + ], + "title": "CLIP Text Encode (Negative Prompt)", + "properties": { + "Node name for S&R": "CLIPTextEncode", + "cnr_id": "comfy-core", + "ver": "0.3.73", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "" + ], + "color": "#323", + "bgcolor": "#535" + }, + { + "id": 69, + "type": "KSampler", + "pos": [ + 1100, + 10 + ], + "size": [ + 310, + 440 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 74 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 75 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 83 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 77 + }, + { + "localized_name": "seed", + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": 118 + }, + { + "localized_name": "steps", + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": 116 + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": 117 + }, + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + }, + { + "localized_name": "scheduler", + "name": "scheduler", + "type": "COMBO", + "widget": { + "name": "scheduler" + }, + "link": null + }, + { + "localized_name": "denoise", + "name": "denoise", + "type": "FLOAT", + "widget": { + "name": "denoise" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 72 + ] + } + ], + "properties": { + "Node name for S&R": "KSampler", + "cnr_id": "comfy-core", + "ver": "0.3.64", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 0, + "randomize", + 25, + 4, + "res_multistep", + "simple", + 1 + ] + }, + { + "id": 87, + "type": "MarkdownNote", + "pos": [ + 1110, + -360 + ], + "size": [ + 300, + 120 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [], + "outputs": [], + "properties": {}, + "widgets_values": [ + "- Steps: 30～50\n- cfg: 3～5" + ], + "color": "#222", + "bgcolor": "#000", + "title": "Original Settings" + } + ], + "groups": [ + { + "id": 2, + "title": "Step2 - Image size", + "bounding": [ + 200, + 530, + 330, + 287.9999544955691 + ], + "color": "#3f789e", + "flags": {} + }, + { + "id": 3, + "title": "Step3 - Prompt", + "bounding": [ + 570, + -200, + 470, + 700 + ], + "color": "#3f789e", + "flags": {} + }, + { + "id": 4, + "title": "Step1 - Load models", + "bounding": [ + 200, + -200, + 330, + 700 + ], + "color": "#3f789e", + "flags": {} + } + ], + "links": [ + { + "id": 78, + "origin_id": 62, + "origin_slot": 0, + "target_id": 67, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 74, + "origin_id": 70, + "origin_slot": 0, + "target_id": 69, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 75, + "origin_id": 67, + "origin_slot": 0, + "target_id": 69, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 83, + "origin_id": 71, + "origin_slot": 0, + "target_id": 69, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 77, + "origin_id": 68, + "origin_slot": 0, + "target_id": 69, + "target_slot": 3, + "type": "LATENT" + }, + { + "id": 82, + "origin_id": 62, + "origin_slot": 0, + "target_id": 71, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 72, + "origin_id": 69, + "origin_slot": 0, + "target_id": 65, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 73, + "origin_id": 63, + "origin_slot": 0, + "target_id": 65, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 79, + "origin_id": 65, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 108, + "origin_id": -10, + "origin_slot": 0, + "target_id": 67, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 109, + "origin_id": 66, + "origin_slot": 0, + "target_id": 70, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 114, + "origin_id": -10, + "origin_slot": 1, + "target_id": 68, + "target_slot": 0, + "type": "INT" + }, + { + "id": 115, + "origin_id": -10, + "origin_slot": 2, + "target_id": 68, + "target_slot": 1, + "type": "INT" + }, + { + "id": 116, + "origin_id": -10, + "origin_slot": 3, + "target_id": 69, + "target_slot": 5, + "type": "INT" + }, + { + "id": 117, + "origin_id": -10, + "origin_slot": 4, + "target_id": 69, + "target_slot": 6, + "type": "FLOAT" + }, + { + "id": 118, + "origin_id": -10, + "origin_slot": 5, + "target_id": 69, + "target_slot": 4, + "type": "INT" + }, + { + "id": 119, + "origin_id": -10, + "origin_slot": 6, + "target_id": 66, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 120, + "origin_id": -10, + "origin_slot": 7, + "target_id": 62, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 121, + "origin_id": -10, + "origin_slot": 8, + "target_id": 63, + "target_slot": 0, + "type": "COMBO" + } + ], + "extra": { + "workflowRendererVersion": "LG" + }, + "category": "Image generation and editing/Text to image" + } + ] + }, + "extra": {} +} \ No newline at end of file diff --git a/blueprints/Text to Image (Z-Image-Turbo).json b/blueprints/Text to Image (Z-Image-Turbo).json index 6aa80e327..2501486fa 100644 --- a/blueprints/Text to Image (Z-Image-Turbo).json +++ b/blueprints/Text to Image (Z-Image-Turbo).json @@ -1,22 +1,21 @@ { - "id": "1c3eaa76-5cfa-4dc7-8571-97a570324e01", "revision": 0, - "last_node_id": 34, - "last_link_id": 40, + "last_node_id": 57, + "last_link_id": 0, "nodes": [ { - "id": 5, - "type": "dfe9eb32-97c0-43a5-90d5-4fd37768d91b", + "id": 57, + "type": "f2fdebf6-dfaf-43b6-9eb2-7f70613cfdc1", "pos": [ - -2.5766491043910378e-05, - 1229.999928629805 + 130, + 200 ], "size": [ 400, 470 ], "flags": {}, - "order": 0, + "order": 1, "mode": 0, "inputs": [ { @@ -44,6 +43,22 @@ }, "link": null }, + { + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": null + }, + { + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": null + }, { "name": "unet_name", "type": "COMBO", @@ -80,15 +95,15 @@ "properties": { "proxyWidgets": [ [ - "-1", + "27", "text" ], [ - "-1", + "13", "width" ], [ - "-1", + "13", "height" ], [ @@ -97,19 +112,23 @@ ], [ "3", - "control_after_generate" + "steps" ], [ - "-1", + "28", "unet_name" ], [ - "-1", + "30", "clip_name" ], [ - "-1", + "29", "vae_name" + ], + [ + "3", + "control_after_generate" ] ], "cnr_id": "comfy-core", @@ -122,48 +141,40 @@ "secondTabOffset": 80, "secondTabWidth": 65 }, - "widgets_values": [ - "", - 1024, - 1024, - null, - null, - "z_image_turbo_bf16.safetensors", - "qwen_3_4b.safetensors", - "ae.safetensors" - ] + "widgets_values": [], + "title": "Text to Image (Z-Image-Turbo)" } ], "links": [], - "groups": [], + "version": 0.4, "definitions": { "subgraphs": [ { - "id": "dfe9eb32-97c0-43a5-90d5-4fd37768d91b", + "id": "f2fdebf6-dfaf-43b6-9eb2-7f70613cfdc1", "version": 1, "state": { "lastGroupId": 4, - "lastNodeId": 34, - "lastLinkId": 40, + "lastNodeId": 61, + "lastLinkId": 75, "lastRerouteId": 0 }, "revision": 0, "config": {}, - "name": "local-Text to Image (Z-Image-Turbo)", + "name": "Text to Image (Z-Image-Turbo)", "inputNode": { "id": -10, "bounding": [ - -80, - 425, + -560, + 480, 120, - 160 + 200 ] }, "outputNode": { "id": -20, "bounding": [ - 1490, - 415, + 1670, + 320, 120, 60 ] @@ -178,8 +189,8 @@ ], "label": "prompt", "pos": [ - 20, - 445 + -460, + 500 ] }, { @@ -190,8 +201,8 @@ 35 ], "pos": [ - 20, - 465 + -460, + 520 ] }, { @@ -202,44 +213,68 @@ 36 ], "pos": [ - 20, - 485 + -460, + 540 ] }, { - "id": "23087d15-8412-4fbd-b71e-9b6d7ef76de1", + "id": "f77677f7-6bf6-4c19-a71f-c4a553d5981e", + "name": "seed", + "type": "INT", + "linkIds": [ + 71 + ], + "pos": [ + -460, + 560 + ] + }, + { + "id": "ef9a9fb1-5983-4bc9-a60b-cf5aec48bff1", + "name": "steps", + "type": "INT", + "linkIds": [ + 72 + ], + "pos": [ + -460, + 580 + ] + }, + { + "id": "a20a1b30-785f-4a04-bb6d-3d61adab9764", "name": "unet_name", "type": "COMBO", "linkIds": [ - 38 + 73 ], "pos": [ - 20, - 505 + -460, + 600 ] }, { - "id": "0677f5c3-2a3f-43d4-98ac-a4c56d5efdc0", + "id": "4af8fc2b-4655-4086-8240-45f8cb38c6f6", "name": "clip_name", "type": "COMBO", "linkIds": [ - 39 + 74 ], "pos": [ - 20, - 525 + -460, + 620 ] }, { - "id": "c85c0445-2641-48b1-bbca-95057edf2fcf", + "id": "4d518693-2807-439c-9cb6-cffd23ccba2c", "name": "vae_name", "type": "COMBO", "linkIds": [ - 40 + 75 ], "pos": [ - 20, - 545 + -460, + 640 ] } ], @@ -253,8 +288,8 @@ ], "localized_name": "IMAGE", "pos": [ - 1510, - 435 + 1690, + 340 ] } ], @@ -264,15 +299,15 @@ "id": 30, "type": "CLIPLoader", "pos": [ - 109.99997264844609, - 329.99999029608756 + 30, + 420 ], "size": [ - 269.9869791666667, - 106 + 270, + 150 ], "flags": {}, - "order": 0, + "order": 7, "mode": 0, "inputs": [ { @@ -282,7 +317,7 @@ "widget": { "name": "clip_name" }, - "link": 39 + "link": 74 }, { "localized_name": "type", @@ -315,9 +350,9 @@ } ], "properties": { + "Node name for S&R": "CLIPLoader", "cnr_id": "comfy-core", "ver": "0.3.73", - "Node name for S&R": "CLIPLoader", "models": [ { "name": "qwen_3_4b.safetensors", @@ -343,15 +378,15 @@ "id": 29, "type": "VAELoader", "pos": [ - 109.99997264844609, - 479.9999847172637 + 30, + 650 ], "size": [ - 269.9869791666667, - 58 + 270, + 110 ], "flags": {}, - "order": 1, + "order": 6, "mode": 0, "inputs": [ { @@ -361,7 +396,7 @@ "widget": { "name": "vae_name" }, - "link": 40 + "link": 75 } ], "outputs": [ @@ -375,9 +410,9 @@ } ], "properties": { + "Node name for S&R": "VAELoader", "cnr_id": "comfy-core", "ver": "0.3.73", - "Node name for S&R": "VAELoader", "models": [ { "name": "ae.safetensors", @@ -401,12 +436,12 @@ "id": 33, "type": "ConditioningZeroOut", "pos": [ - 639.9999103333332, - 620.0000271257795 + 630, + 960 ], "size": [ - 204.134765625, - 26 + 230, + 80 ], "flags": {}, "order": 8, @@ -430,9 +465,9 @@ } ], "properties": { + "Node name for S&R": "ConditioningZeroOut", "cnr_id": "comfy-core", "ver": "0.3.73", - "Node name for S&R": "ConditioningZeroOut", "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -440,22 +475,21 @@ "secondTabText": "Send Back", "secondTabOffset": 80, "secondTabWidth": 65 - }, - "widgets_values": [] + } }, { "id": 8, "type": "VAEDecode", "pos": [ - 1219.9999088104782, - 160.00009184959066 + 1320, + 230 ], "size": [ - 209.98697916666669, - 46 + 230, + 100 ], "flags": {}, - "order": 5, + "order": 1, "mode": 0, "inputs": [ { @@ -483,9 +517,9 @@ } ], "properties": { + "Node name for S&R": "VAEDecode", "cnr_id": "comfy-core", "ver": "0.3.64", - "Node name for S&R": "VAEDecode", "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -493,22 +527,21 @@ "secondTabText": "Send Back", "secondTabOffset": 80, "secondTabWidth": 65 - }, - "widgets_values": [] + } }, { "id": 28, "type": "UNETLoader", "pos": [ - 109.99997264844609, - 200.0000502647102 + 30, + 230 ], "size": [ - 269.9869791666667, - 82 + 270, + 110 ], "flags": {}, - "order": 2, + "order": 5, "mode": 0, "inputs": [ { @@ -518,7 +551,7 @@ "widget": { "name": "unet_name" }, - "link": 38 + "link": 73 }, { "localized_name": "weight_dtype", @@ -541,9 +574,9 @@ } ], "properties": { + "Node name for S&R": "UNETLoader", "cnr_id": "comfy-core", "ver": "0.3.73", - "Node name for S&R": "UNETLoader", "models": [ { "name": "z_image_turbo_bf16.safetensors", @@ -568,15 +601,15 @@ "id": 27, "type": "CLIPTextEncode", "pos": [ - 429.99997828947767, - 200.0000502647102 + 400, + 230 ], "size": [ - 409.9869791666667, - 319.9869791666667 + 450, + 650 ], "flags": {}, - "order": 7, + "order": 4, "mode": 0, "inputs": [ { @@ -607,9 +640,9 @@ } ], "properties": { + "Node name for S&R": "CLIPTextEncode", "cnr_id": "comfy-core", "ver": "0.3.73", - "Node name for S&R": "CLIPTextEncode", "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -626,15 +659,15 @@ "id": 13, "type": "EmptySD3LatentImage", "pos": [ - 109.99997264844609, - 629.9999791384399 + 40, + 890 ], "size": [ - 259.9869791666667, - 106 + 260, + 170 ], "flags": {}, - "order": 6, + "order": 3, "mode": 0, "inputs": [ { @@ -677,9 +710,9 @@ } ], "properties": { + "Node name for S&R": "EmptySD3LatentImage", "cnr_id": "comfy-core", "ver": "0.3.64", - "Node name for S&R": "EmptySD3LatentImage", "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -694,19 +727,77 @@ 1 ] }, + { + "id": 11, + "type": "ModelSamplingAuraFlow", + "pos": [ + 950, + 230 + ], + "size": [ + 310, + 110 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 26 + }, + { + "localized_name": "shift", + "name": "shift", + "type": "FLOAT", + "widget": { + "name": "shift" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "slot_index": 0, + "links": [ + 13 + ] + } + ], + "properties": { + "Node name for S&R": "ModelSamplingAuraFlow", + "cnr_id": "comfy-core", + "ver": "0.3.64", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 3 + ] + }, { "id": 3, "type": "KSampler", "pos": [ - 879.9999615530063, - 269.9999774911694 + 950, + 400 ], "size": [ - 314.9869791666667, - 262 + 320, + 350 ], "flags": {}, - "order": 4, + "order": 0, "mode": 0, "inputs": [ { @@ -740,7 +831,7 @@ "widget": { "name": "seed" }, - "link": null + "link": 71 }, { "localized_name": "steps", @@ -749,7 +840,7 @@ "widget": { "name": "steps" }, - "link": null + "link": 72 }, { "localized_name": "cfg", @@ -800,9 +891,9 @@ } ], "properties": { + "Node name for S&R": "KSampler", "cnr_id": "comfy-core", "ver": "0.3.64", - "Node name for S&R": "KSampler", "enableTabs": false, "tabWidth": 65, "tabXOffset": 10, @@ -814,81 +905,23 @@ "widgets_values": [ 0, "randomize", - 4, + 8, 1, "res_multistep", "simple", 1 ] - }, - { - "id": 11, - "type": "ModelSamplingAuraFlow", - "pos": [ - 879.9999615530063, - 160.00009184959066 - ], - "size": [ - 309.9869791666667, - 58 - ], - "flags": {}, - "order": 3, - "mode": 0, - "inputs": [ - { - "localized_name": "model", - "name": "model", - "type": "MODEL", - "link": 26 - }, - { - "localized_name": "shift", - "name": "shift", - "type": "FLOAT", - "widget": { - "name": "shift" - }, - "link": null - } - ], - "outputs": [ - { - "localized_name": "MODEL", - "name": "MODEL", - "type": "MODEL", - "slot_index": 0, - "links": [ - 13 - ] - } - ], - "properties": { - "cnr_id": "comfy-core", - "ver": "0.3.64", - "Node name for S&R": "ModelSamplingAuraFlow", - "enableTabs": false, - "tabWidth": 65, - "tabXOffset": 10, - "hasSecondTab": false, - "secondTabText": "Send Back", - "secondTabOffset": 80, - "secondTabWidth": 65 - }, - "widgets_values": [ - 3 - ] } ], "groups": [ { "id": 2, - "title": "Image size", + "title": "Step2 - Image size", "bounding": [ - 100, - 560, - 290, - 200 + 10, + 820, + 320, + 280 ], "color": "#3f789e", "font_size": 24, @@ -896,12 +929,12 @@ }, { "id": 3, - "title": "Prompt", + "title": "Step3 - Prompt", "bounding": [ - 410, + 360, 130, - 450, - 540 + 530, + 970 ], "color": "#3f789e", "font_size": 24, @@ -909,12 +942,12 @@ }, { "id": 4, - "title": "Models", + "title": "Step1 - Load models", "bounding": [ - 100, + 0, 130, - 290, - 413.6 + 330, + 660 ], "color": "#3f789e", "font_size": 24, @@ -1027,25 +1060,41 @@ "type": "INT" }, { - "id": 38, + "id": 71, "origin_id": -10, "origin_slot": 3, + "target_id": 3, + "target_slot": 4, + "type": "INT" + }, + { + "id": 72, + "origin_id": -10, + "origin_slot": 4, + "target_id": 3, + "target_slot": 5, + "type": "INT" + }, + { + "id": 73, + "origin_id": -10, + "origin_slot": 5, "target_id": 28, "target_slot": 0, "type": "COMBO" }, { - "id": 39, + "id": 74, "origin_id": -10, - "origin_slot": 4, + "origin_slot": 6, "target_id": 30, "target_slot": 0, "type": "COMBO" }, { - "id": 40, + "id": 75, "origin_id": -10, - "origin_slot": 5, + "origin_slot": 7, "target_id": 29, "target_slot": 0, "type": "COMBO" @@ -1054,25 +1103,10 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image generation and editing/Text to image" + "category": "Image generation and editing/Text to image", + "description": "Generates images from text prompts using Z-Image-Turbo, Alibaba's distilled 6B DiT model." } ] }, - "config": {}, - "extra": { - "frontendVersion": "1.37.10", - "workflowRendererVersion": "LG", - "VHS_latentpreview": false, - "VHS_latentpreviewrate": 0, - "VHS_MetadataImage": true, - "VHS_KeepIntermediate": true, - "ds": { - "scale": 0.8401370345180755, - "offset": [ - 940.0587067393087, - -830.7121087564725 - ] - } - }, - "version": 0.4 -} + "extra": {} +} \ No newline at end of file diff --git a/blueprints/Text to Image.json b/blueprints/Text to Image.json new file mode 100644 index 000000000..ffe3682ff --- /dev/null +++ b/blueprints/Text to Image.json @@ -0,0 +1,1132 @@ +{ + "revision": 0, + "last_node_id": 71, + "last_link_id": 0, + "nodes": [ + { + "id": 71, + "type": "2d5985c9-deef-41ae-9c34-6353d3d7d1ef", + "pos": [ + 90, + 800 + ], + "size": [ + 400, + 80 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "label": "prompt", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + }, + { + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": null + }, + { + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": null + }, + { + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": null + }, + { + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": null + }, + { + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": null + }, + { + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [] + } + ], + "title": "Text to Image", + "properties": { + "proxyWidgets": [ + [ + "67", + "text" + ], + [ + "68", + "width" + ], + [ + "68", + "height" + ], + [ + "66", + "unet_name" + ], + [ + "62", + "clip_name" + ], + [ + "63", + "vae_name" + ], + [ + "70", + "steps" + ], + [ + "70", + "control_after_generate" + ] + ], + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": { + "text": true + }, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [] + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "2d5985c9-deef-41ae-9c34-6353d3d7d1ef", + "version": 1, + "state": { + "lastGroupId": 4, + "lastNodeId": 71, + "lastLinkId": 70, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Text to Image", + "inputNode": { + "id": -10, + "bounding": [ + -80, + 425, + 120, + 180 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 1490, + 415, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "fb178669-e742-4a53-8a69-7df59834dfd8", + "name": "text", + "type": "STRING", + "linkIds": [ + 34 + ], + "label": "prompt", + "pos": [ + 20, + 445 + ] + }, + { + "id": "dd780b3c-23e9-46ff-8469-156008f42e5a", + "name": "width", + "type": "INT", + "linkIds": [ + 35 + ], + "pos": [ + 20, + 465 + ] + }, + { + "id": "7b08d546-6bb0-4ef9-82e9-ffae5e1ee6bc", + "name": "height", + "type": "INT", + "linkIds": [ + 36 + ], + "pos": [ + 20, + 485 + ] + }, + { + "id": "8ed4eb73-a2bf-4766-8bf4-c5890b560596", + "name": "unet_name", + "type": "COMBO", + "linkIds": [ + 38 + ], + "pos": [ + 20, + 505 + ] + }, + { + "id": "f362d639-d412-4b5d-8490-1e9995dc5f82", + "name": "clip_name", + "type": "COMBO", + "linkIds": [ + 39 + ], + "pos": [ + 20, + 525 + ] + }, + { + "id": "ee25ac16-de63-4b74-bbbb-5b29fdc1efcf", + "name": "vae_name", + "type": "COMBO", + "linkIds": [ + 40 + ], + "pos": [ + 20, + 545 + ] + }, + { + "id": "51cbcd61-9218-4bcb-89ac-ecdfb1ef8892", + "name": "steps", + "type": "INT", + "linkIds": [ + 70 + ], + "pos": [ + 20, + 565 + ] + } + ], + "outputs": [ + { + "id": "1fa72a21-ce00-4952-814e-1f2ffbe87d1d", + "name": "IMAGE", + "type": "IMAGE", + "linkIds": [ + 16 + ], + "localized_name": "IMAGE", + "pos": [ + 1510, + 435 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 62, + "type": "CLIPLoader", + "pos": [ + 110, + 330 + ], + "size": [ + 270, + 110 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "clip_name", + "name": "clip_name", + "type": "COMBO", + "widget": { + "name": "clip_name" + }, + "link": 39 + }, + { + "localized_name": "type", + "name": "type", + "type": "COMBO", + "widget": { + "name": "type" + }, + "link": null + }, + { + "localized_name": "device", + "name": "device", + "shape": 7, + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [ + 28 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "CLIPLoader", + "models": [ + { + "name": "qwen_3_4b.safetensors", + "url": "https://huggingface.co/Comfy-Org/z_image_turbo/resolve/main/split_files/text_encoders/qwen_3_4b.safetensors", + "directory": "text_encoders" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "qwen_3_4b.safetensors", + "lumina2", + "default" + ] + }, + { + "id": 63, + "type": "VAELoader", + "pos": [ + 110, + 480 + ], + "size": [ + 270, + 60 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "vae_name", + "name": "vae_name", + "type": "COMBO", + "widget": { + "name": "vae_name" + }, + "link": 40 + } + ], + "outputs": [ + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "links": [ + 27 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "VAELoader", + "models": [ + { + "name": "ae.safetensors", + "url": "https://huggingface.co/Comfy-Org/z_image_turbo/resolve/main/split_files/vae/ae.safetensors", + "directory": "vae" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "ae.safetensors" + ] + }, + { + "id": 64, + "type": "ConditioningZeroOut", + "pos": [ + 640, + 620 + ], + "size": [ + 210, + 30 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "conditioning", + "name": "conditioning", + "type": "CONDITIONING", + "link": 32 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 33 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ConditioningZeroOut", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 65, + "type": "VAEDecode", + "pos": [ + 1220, + 160 + ], + "size": [ + 210, + 50 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 14 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 27 + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "slot_index": 0, + "links": [ + 16 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.64", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "VAEDecode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 66, + "type": "UNETLoader", + "pos": [ + 110, + 200 + ], + "size": [ + 270, + 90 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "unet_name", + "name": "unet_name", + "type": "COMBO", + "widget": { + "name": "unet_name" + }, + "link": 38 + }, + { + "localized_name": "weight_dtype", + "name": "weight_dtype", + "type": "COMBO", + "widget": { + "name": "weight_dtype" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 26 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "UNETLoader", + "models": [ + { + "name": "z_image_turbo_bf16.safetensors", + "url": "https://huggingface.co/Comfy-Org/z_image_turbo/resolve/main/split_files/diffusion_models/z_image_turbo_bf16.safetensors", + "directory": "diffusion_models" + } + ], + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "z_image_turbo_bf16.safetensors", + "default" + ] + }, + { + "id": 67, + "type": "CLIPTextEncode", + "pos": [ + 430, + 200 + ], + "size": [ + 410, + 370 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 28 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 34 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 30, + 32 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.73", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "CLIPTextEncode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "" + ] + }, + { + "id": 68, + "type": "EmptySD3LatentImage", + "pos": [ + 110, + 630 + ], + "size": [ + 260, + 110 + ], + "flags": {}, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 35 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 36 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 17 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.64", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "EmptySD3LatentImage", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1024, + 1024, + 1 + ] + }, + { + "id": 69, + "type": "ModelSamplingAuraFlow", + "pos": [ + 880, + 160 + ], + "size": [ + 310, + 60 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 26 + }, + { + "localized_name": "shift", + "name": "shift", + "type": "FLOAT", + "widget": { + "name": "shift" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "slot_index": 0, + "links": [ + 13 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.64", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "ModelSamplingAuraFlow", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 3 + ] + }, + { + "id": 70, + "type": "KSampler", + "pos": [ + 880, + 270 + ], + "size": [ + 320, + 270 + ], + "flags": {}, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 13 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 30 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 33 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 17 + }, + { + "localized_name": "seed", + "name": "seed", + "type": "INT", + "widget": { + "name": "seed" + }, + "link": null + }, + { + "localized_name": "steps", + "name": "steps", + "type": "INT", + "widget": { + "name": "steps" + }, + "link": 70 + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": null + }, + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + }, + { + "localized_name": "scheduler", + "name": "scheduler", + "type": "COMBO", + "widget": { + "name": "scheduler" + }, + "link": null + }, + { + "localized_name": "denoise", + "name": "denoise", + "type": "FLOAT", + "widget": { + "name": "denoise" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "slot_index": 0, + "links": [ + 14 + ] + } + ], + "properties": { + "cnr_id": "comfy-core", + "ver": "0.3.64", + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "Node name for S&R": "KSampler", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 0, + "randomize", + 8, + 1, + "res_multistep", + "simple", + 1 + ] + } + ], + "groups": [ + { + "id": 2, + "title": "Step2 - Image size", + "bounding": [ + 100, + 560, + 290, + 200 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 3, + "title": "Step3 - Prompt", + "bounding": [ + 410, + 130, + 450, + 540 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 4, + "title": "Step1 - Load models", + "bounding": [ + 100, + 130, + 290, + 413.6 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 32, + "origin_id": 67, + "origin_slot": 0, + "target_id": 64, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 26, + "origin_id": 66, + "origin_slot": 0, + "target_id": 69, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 14, + "origin_id": 70, + "origin_slot": 0, + "target_id": 65, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 27, + "origin_id": 63, + "origin_slot": 0, + "target_id": 65, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 13, + "origin_id": 69, + "origin_slot": 0, + "target_id": 70, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 30, + "origin_id": 67, + "origin_slot": 0, + "target_id": 70, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 33, + "origin_id": 64, + "origin_slot": 0, + "target_id": 70, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 17, + "origin_id": 68, + "origin_slot": 0, + "target_id": 70, + "target_slot": 3, + "type": "LATENT" + }, + { + "id": 28, + "origin_id": 62, + "origin_slot": 0, + "target_id": 67, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 16, + "origin_id": 65, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 34, + "origin_id": -10, + "origin_slot": 0, + "target_id": 67, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 35, + "origin_id": -10, + "origin_slot": 1, + "target_id": 68, + "target_slot": 0, + "type": "INT" + }, + { + "id": 36, + "origin_id": -10, + "origin_slot": 2, + "target_id": 68, + "target_slot": 1, + "type": "INT" + }, + { + "id": 38, + "origin_id": -10, + "origin_slot": 3, + "target_id": 66, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 39, + "origin_id": -10, + "origin_slot": 4, + "target_id": 62, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 40, + "origin_id": -10, + "origin_slot": 5, + "target_id": 63, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 70, + "origin_id": -10, + "origin_slot": 6, + "target_id": 70, + "target_slot": 5, + "type": "INT" + } + ], + "extra": { + "workflowRendererVersion": "LG" + }, + "category": "Image generation and editing/Text to image", + "description": "Generates images from text prompts using Z-Image-Turbo defaults with Qwen3 text encoder and VAE." + } + ] + }, + "extra": {} +} diff --git a/blueprints/Text to Video (LTX-2.3).json b/blueprints/Text to Video (LTX-2.3).json new file mode 100644 index 000000000..f44a216dd --- /dev/null +++ b/blueprints/Text to Video (LTX-2.3).json @@ -0,0 +1,4297 @@ +{ + "revision": 0, + "last_node_id": 324, + "last_link_id": 0, + "nodes": [ + { + "id": 324, + "type": "871cf29d-2726-43a5-b61e-01fa939d699d", + "pos": [ + -300, + 4290 + ], + "size": [ + 400, + 170 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "name": "value", + "type": "STRING", + "widget": { + "name": "value" + }, + "link": null + }, + { + "label": "width", + "name": "value_2", + "type": "INT", + "widget": { + "name": "value_2" + }, + "link": null + }, + { + "label": "height", + "name": "value_3", + "type": "INT", + "widget": { + "name": "value_3" + }, + "link": null + }, + { + "label": "duration", + "name": "value_4", + "type": "INT", + "widget": { + "name": "value_4" + }, + "link": null + }, + { + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": null + }, + { + "label": "distilled_lora", + "name": "lora_name", + "type": "COMBO", + "widget": { + "name": "lora_name" + }, + "link": null + }, + { + "name": "text_encoder", + "type": "COMBO", + "widget": { + "name": "text_encoder" + }, + "link": null + }, + { + "label": "latent_upscale_model", + "name": "model_name", + "type": "COMBO", + "widget": { + "name": "model_name" + }, + "link": null + }, + { + "label": "fps", + "name": "value_1", + "type": "INT", + "widget": { + "name": "value_1" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "VIDEO", + "name": "VIDEO", + "type": "VIDEO", + "links": [] + } + ], + "properties": { + "proxyWidgets": [ + [ + "320", + "value" + ], + [ + "314", + "value" + ], + [ + "301", + "value" + ], + [ + "303", + "value" + ], + [ + "318", + "ckpt_name" + ], + [ + "287", + "lora_name" + ], + [ + "319", + "text_encoder" + ], + [ + "313", + "model_name" + ], + [ + "302", + "value" + ], + [ + "279", + "noise_seed" + ], + [ + "279", + "control_after_generate" + ] + ], + "ue_properties": { + "widget_ue_connectable": { + "value_1": true, + "value_2": true, + "value_3": true, + "value_4": true, + "lora_name": true, + "model_name": true + }, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.3", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [], + "title": "Text to Video (LTX-2.3)" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "871cf29d-2726-43a5-b61e-01fa939d699d", + "version": 1, + "state": { + "lastGroupId": 26, + "lastNodeId": 324, + "lastLinkId": 631, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Text to Video (LTX-2.3)", + "inputNode": { + "id": -10, + "bounding": [ + 720, + 4240, + 162.162109375, + 220 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + 6100, + 4160, + 120, + 60 + ] + }, + "inputs": [ + { + "id": "9494c550-4172-49c6-930e-5b508f775e77", + "name": "value", + "type": "STRING", + "linkIds": [ + 595 + ], + "pos": [ + 862.162109375, + 4260 + ] + }, + { + "id": "58dbb3f6-f924-4548-96ef-e0e34610bd4e", + "name": "value_2", + "type": "INT", + "linkIds": [ + 597 + ], + "label": "width", + "pos": [ + 862.162109375, + 4280 + ] + }, + { + "id": "6086d5b8-2586-448c-a641-dd14d76dd102", + "name": "value_3", + "type": "INT", + "linkIds": [ + 598 + ], + "label": "height", + "pos": [ + 862.162109375, + 4300 + ] + }, + { + "id": "feb8c2eb-ae48-4fa8-bc24-929552d656c3", + "name": "value_4", + "type": "INT", + "linkIds": [ + 599 + ], + "label": "duration", + "pos": [ + 862.162109375, + 4320 + ] + }, + { + "id": "d7255058-319a-4880-8f9a-7e542c8f3c3c", + "name": "ckpt_name", + "type": "COMBO", + "linkIds": [ + 601, + 604, + 605 + ], + "pos": [ + 862.162109375, + 4340 + ] + }, + { + "id": "4afce68d-8f65-4342-9d6d-ae0a7688c3e3", + "name": "lora_name", + "type": "COMBO", + "linkIds": [ + 602 + ], + "label": "distilled_lora", + "pos": [ + 862.162109375, + 4360 + ] + }, + { + "id": "ab842b4b-c977-4679-b421-424722785b57", + "name": "text_encoder", + "type": "COMBO", + "linkIds": [ + 606 + ], + "pos": [ + 862.162109375, + 4380 + ] + }, + { + "id": "9e47372d-28d9-4311-91e9-e90d03f4eb43", + "name": "model_name", + "type": "COMBO", + "linkIds": [ + 607 + ], + "label": "latent_upscale_model", + "pos": [ + 862.162109375, + 4400 + ] + }, + { + "id": "7951b137-465e-4844-b05f-88b89f0e1ba8", + "name": "value_1", + "type": "INT", + "linkIds": [ + 627 + ], + "label": "fps", + "pos": [ + 862.162109375, + 4420 + ] + } + ], + "outputs": [ + { + "id": "954ef307-c897-4eea-8b5c-5c6ce15a5357", + "name": "VIDEO", + "type": "VIDEO", + "linkIds": [ + 536 + ], + "localized_name": "VIDEO", + "pos": [ + 6120, + 4180 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 278, + "type": "RandomNoise", + "pos": [ + 4720, + 3750 + ], + "size": [ + 280, + 110 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [ + { + "localized_name": "noise_seed", + "name": "noise_seed", + "type": "INT", + "widget": { + "name": "noise_seed" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "NOISE", + "name": "NOISE", + "type": "NOISE", + "links": [ + 490 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.75", + "Node name for S&R": "RandomNoise", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 42, + "fixed" + ] + }, + { + "id": 279, + "type": "RandomNoise", + "pos": [ + 3200, + 3900 + ], + "size": [ + 280, + 110 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "noise_seed", + "name": "noise_seed", + "type": "INT", + "widget": { + "name": "noise_seed" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "NOISE", + "name": "NOISE", + "type": "NOISE", + "links": [ + 483 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.56", + "Node name for S&R": "RandomNoise", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 343011291748534, + "randomize" + ] + }, + { + "id": 280, + "type": "LTXVConcatAVLatent", + "pos": [ + 4730, + 4520 + ], + "size": [ + 280, + 100 + ], + "flags": {}, + "order": 8, + "mode": 0, + "inputs": [ + { + "localized_name": "video_latent", + "name": "video_latent", + "type": "LATENT", + "link": 512 + }, + { + "localized_name": "audio_latent", + "name": "audio_latent", + "type": "LATENT", + "link": 513 + } + ], + "outputs": [ + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "links": [ + 494 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.5.1", + "Node name for S&R": "LTXVConcatAVLatent", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 281, + "type": "LTXVAudioVAELoader", + "pos": [ + 1660, + 4140 + ], + "size": [ + 430, + 110 + ], + "flags": {}, + "order": 9, + "mode": 0, + "inputs": [ + { + "localized_name": "ckpt_name", + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": 604 + } + ], + "outputs": [ + { + "localized_name": "Audio VAE", + "name": "Audio VAE", + "type": "VAE", + "links": [ + 481, + 496 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.68", + "Node name for S&R": "LTXVAudioVAELoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "ltx-2.3-22b-dev-fp8.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2.3-fp8/resolve/main/ltx-2.3-22b-dev-fp8.safetensors", + "directory": "checkpoints" + } + ] + }, + "widgets_values": [ + "ltx-2.3-22b-dev-fp8.safetensors" + ] + }, + { + "id": 282, + "type": "KSamplerSelect", + "pos": [ + 4720, + 4160 + ], + "size": [ + 280, + 110 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "SAMPLER", + "name": "SAMPLER", + "type": "SAMPLER", + "links": [ + 492 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.75", + "Node name for S&R": "KSamplerSelect", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "euler_cfg_pp" + ] + }, + { + "id": 283, + "type": "ManualSigmas", + "pos": [ + 4720, + 4340 + ], + "size": [ + 280, + 110 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "sigmas", + "name": "sigmas", + "type": "STRING", + "widget": { + "name": "sigmas" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "SIGMAS", + "name": "SIGMAS", + "type": "SIGMAS", + "links": [ + 493 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.5.1", + "Node name for S&R": "ManualSigmas", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "0.85, 0.7250, 0.4219, 0.0" + ] + }, + { + "id": 284, + "type": "CFGGuider", + "pos": [ + 4720, + 3930 + ], + "size": [ + 280, + 160 + ], + "flags": {}, + "order": 10, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 478 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 479 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 480 + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "GUIDER", + "name": "GUIDER", + "type": "GUIDER", + "links": [ + 491 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.71", + "Node name for S&R": "CFGGuider", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1 + ] + }, + { + "id": 285, + "type": "SamplerCustomAdvanced", + "pos": [ + 3620, + 3990 + ], + "size": [ + 230, + 170 + ], + "flags": {}, + "order": 11, + "mode": 0, + "inputs": [ + { + "localized_name": "noise", + "name": "noise", + "type": "NOISE", + "link": 483 + }, + { + "localized_name": "guider", + "name": "guider", + "type": "GUIDER", + "link": 484 + }, + { + "localized_name": "sampler", + "name": "sampler", + "type": "SAMPLER", + "link": 485 + }, + { + "localized_name": "sigmas", + "name": "sigmas", + "type": "SIGMAS", + "link": 544 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 487 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "LATENT", + "links": [ + 488 + ] + }, + { + "localized_name": "denoised_output", + "name": "denoised_output", + "type": "LATENT", + "links": [] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.60", + "Node name for S&R": "SamplerCustomAdvanced", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 286, + "type": "LTXVCropGuides", + "pos": [ + 3900, + 3700 + ], + "size": [ + 250, + 120 + ], + "flags": {}, + "order": 12, + "mode": 0, + "inputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 475 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 476 + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "link": 477 + } + ], + "outputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "links": [ + 479 + ] + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "links": [ + 480 + ] + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "slot_index": 2, + "links": [] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.68", + "Node name for S&R": "LTXVCropGuides", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 287, + "type": "LoraLoaderModelOnly", + "pos": [ + 1660, + 3910 + ], + "size": [ + 430, + 140 + ], + "flags": {}, + "order": 13, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 520 + }, + { + "localized_name": "lora_name", + "name": "lora_name", + "type": "COMBO", + "widget": { + "name": "lora_name" + }, + "link": 602 + }, + { + "localized_name": "strength_model", + "name": "strength_model", + "type": "FLOAT", + "widget": { + "name": "strength_model" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 478, + 541 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.75", + "Node name for S&R": "LoraLoaderModelOnly", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "ltx-2.3-22b-distilled-lora-384.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2.3/resolve/main/ltx-2.3-22b-distilled-lora-384.safetensors", + "directory": "loras" + } + ] + }, + "widgets_values": [ + "ltx-2.3-22b-distilled-lora-384.safetensors", + 0.5 + ] + }, + { + "id": 288, + "type": "ResizeImagesByLongerEdge", + "pos": [ + 2120, + 5040 + ], + "size": [ + 310, + 110 + ], + "flags": { + "collapsed": false + }, + "order": 14, + "mode": 0, + "inputs": [ + { + "localized_name": "images", + "name": "images", + "type": "IMAGE", + "link": 523 + }, + { + "localized_name": "longer_edge", + "name": "longer_edge", + "type": "INT", + "widget": { + "name": "longer_edge" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "images", + "name": "images", + "type": "IMAGE", + "links": [ + 505 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.5.1", + "Node name for S&R": "ResizeImagesByLongerEdge", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1536 + ] + }, + { + "id": 289, + "type": "LTXVLatentUpsampler", + "pos": [ + 4270, + 3910 + ], + "size": [ + 330, + 120 + ], + "flags": {}, + "order": 15, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 547 + }, + { + "localized_name": "upscale_model", + "name": "upscale_model", + "type": "LATENT_UPSCALE_MODEL", + "link": 545 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 554 + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "links": [ + 548 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.14.1", + "Node name for S&R": "LTXVLatentUpsampler", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 290, + "type": "LTXVImgToVideoInplace", + "pos": [ + 4280, + 4150 + ], + "size": [ + 300, + 180 + ], + "flags": {}, + "order": 16, + "mode": 0, + "inputs": [ + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 552 + }, + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 515 + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "link": 548 + }, + { + "localized_name": "strength", + "name": "strength", + "type": "FLOAT", + "widget": { + "name": "strength" + }, + "link": null + }, + { + "localized_name": "bypass", + "name": "bypass", + "type": "BOOLEAN", + "widget": { + "name": "bypass" + }, + "link": 543 + } + ], + "outputs": [ + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "links": [ + 512 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "LTXVImgToVideoInplace", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1, + false + ] + }, + { + "id": 291, + "type": "LTXVPreprocess", + "pos": [ + 2130, + 5190 + ], + "size": [ + 290, + 110 + ], + "flags": {}, + "order": 17, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 505 + }, + { + "localized_name": "img_compression", + "name": "img_compression", + "type": "INT", + "widget": { + "name": "img_compression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "output_image", + "name": "output_image", + "type": "IMAGE", + "links": [ + 510, + 515 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "LTXVPreprocess", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 18 + ] + }, + { + "id": 292, + "type": "ResizeImageMaskNode", + "pos": [ + 1670, + 5040 + ], + "size": [ + 300, + 160 + ], + "flags": {}, + "order": 18, + "mode": 0, + "inputs": [ + { + "localized_name": "input", + "name": "input", + "type": "IMAGE,MASK", + "link": 626 + }, + { + "localized_name": "resize_type", + "name": "resize_type", + "type": "COMFY_DYNAMICCOMBO_V3", + "widget": { + "name": "resize_type" + }, + "link": null + }, + { + "localized_name": "width", + "name": "resize_type.width", + "type": "INT", + "widget": { + "name": "resize_type.width" + }, + "link": 558 + }, + { + "localized_name": "height", + "name": "resize_type.height", + "type": "INT", + "widget": { + "name": "resize_type.height" + }, + "link": 559 + }, + { + "localized_name": "crop", + "name": "resize_type.crop", + "type": "COMBO", + "widget": { + "name": "resize_type.crop" + }, + "link": null + }, + { + "localized_name": "scale_method", + "name": "scale_method", + "type": "COMBO", + "widget": { + "name": "scale_method" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "resized", + "name": "resized", + "type": "*", + "links": [ + 523 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "ResizeImageMaskNode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "scale dimensions", + 1920, + 1088, + "center", + "lanczos" + ] + }, + { + "id": 293, + "type": "KSamplerSelect", + "pos": [ + 3200, + 4350 + ], + "size": [ + 280, + 110 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "sampler_name", + "name": "sampler_name", + "type": "COMBO", + "widget": { + "name": "sampler_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "SAMPLER", + "name": "SAMPLER", + "type": "SAMPLER", + "links": [ + 485 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.56", + "Node name for S&R": "KSamplerSelect", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "euler_ancestral_cfg_pp" + ] + }, + { + "id": 294, + "type": "ComfyMathExpression", + "pos": [ + 2530, + 5070 + ], + "size": [ + 230, + 170 + ], + "flags": { + "collapsed": true + }, + "order": 19, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 560 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 561 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.3", + "Node name for S&R": "ComfyMathExpression", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "a/2" + ] + }, + { + "id": 295, + "type": "Reroute", + "pos": [ + 3930, + 4090 + ], + "size": [ + 80, + 30 + ], + "flags": {}, + "order": 20, + "mode": 0, + "inputs": [ + { + "name": "", + "type": "*", + "link": 557 + } + ], + "outputs": [ + { + "name": "", + "type": "VAE", + "links": [ + 552, + 553, + 554 + ] + } + ], + "properties": { + "showOutputText": false, + "horizontal": false, + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + } + } + }, + { + "id": 296, + "type": "ComfyMathExpression", + "pos": [ + 2530, + 5130 + ], + "size": [ + 230, + 170 + ], + "flags": { + "collapsed": true + }, + "order": 21, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 562 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 563 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.3", + "Node name for S&R": "ComfyMathExpression", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "a/2" + ] + }, + { + "id": 297, + "type": "EmptyLTXVLatentVideo", + "pos": [ + 2980, + 5200 + ], + "size": [ + 280, + 200 + ], + "flags": {}, + "order": 22, + "mode": 0, + "inputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "widget": { + "name": "width" + }, + "link": 561 + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "widget": { + "name": "height" + }, + "link": 563 + }, + { + "localized_name": "length", + "name": "length", + "type": "INT", + "widget": { + "name": "length" + }, + "link": 631 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "LATENT", + "name": "LATENT", + "type": "LATENT", + "links": [ + 511 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.60", + "Node name for S&R": "EmptyLTXVLatentVideo", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 768, + 512, + 97, + 1 + ] + }, + { + "id": 298, + "type": "LTXVImgToVideoInplace", + "pos": [ + 3420, + 4990 + ], + "size": [ + 280, + 180 + ], + "flags": {}, + "order": 23, + "mode": 0, + "inputs": [ + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 556 + }, + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 510 + }, + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "link": 511 + }, + { + "localized_name": "strength", + "name": "strength", + "type": "FLOAT", + "widget": { + "name": "strength" + }, + "link": null + }, + { + "localized_name": "bypass", + "name": "bypass", + "type": "BOOLEAN", + "widget": { + "name": "bypass" + }, + "link": 542 + } + ], + "outputs": [ + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "links": [ + 497 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "LTXVImgToVideoInplace", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 0.7, + false + ] + }, + { + "id": 299, + "type": "LTXVAudioVAEDecode", + "pos": [ + 5770, + 3940 + ], + "size": [ + 270, + 100 + ], + "flags": {}, + "order": 24, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 495 + }, + { + "label": "Audio VAE", + "localized_name": "audio_vae", + "name": "audio_vae", + "type": "VAE", + "link": 496 + } + ], + "outputs": [ + { + "localized_name": "Audio", + "name": "Audio", + "type": "AUDIO", + "links": [ + 534 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "LTXVAudioVAEDecode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 300, + "type": "ComfyMathExpression", + "pos": [ + 2530, + 5270 + ], + "size": [ + 230, + 170 + ], + "flags": { + "collapsed": true + }, + "order": 25, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 564 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": [ + 566, + 591 + ] + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 565 + ] + } + ], + "title": "Math Expression (fps)", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.3", + "Node name for S&R": "ComfyMathExpression", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "a" + ] + }, + { + "id": 301, + "type": "PrimitiveInt", + "pos": [ + 1160, + 4530 + ], + "size": [ + 370, + 110 + ], + "flags": {}, + "order": 26, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": 598 + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 559, + 562 + ] + } + ], + "title": "Height", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.3", + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 720, + "fixed" + ] + }, + { + "id": 302, + "type": "PrimitiveInt", + "pos": [ + 1160, + 4680 + ], + "size": [ + 370, + 110 + ], + "flags": {}, + "order": 27, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": 627 + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 564, + 629 + ] + } + ], + "title": "Frame Rate", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.3", + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 25, + "fixed" + ] + }, + { + "id": 303, + "type": "PrimitiveInt", + "pos": [ + 1160, + 4230 + ], + "size": [ + 370, + 110 + ], + "flags": {}, + "order": 28, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": 599 + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 628 + ] + } + ], + "title": "Duration", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 5, + "fixed" + ] + }, + { + "id": 304, + "type": "PrimitiveBoolean", + "pos": [ + 1170, + 4080 + ], + "size": [ + 370, + 100 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "BOOLEAN", + "widget": { + "name": "value" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "BOOLEAN", + "name": "BOOLEAN", + "type": "BOOLEAN", + "links": [ + 542, + 543 + ] + } + ], + "title": "Switch to Text to Video?", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.0", + "Node name for S&R": "PrimitiveBoolean", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + true + ] + }, + { + "id": 305, + "type": "CLIPTextEncode", + "pos": [ + 2170, + 3640 + ], + "size": [ + 550, + 740 + ], + "flags": {}, + "order": 29, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 615 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 623 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 526 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.56", + "Node name for S&R": "CLIPTextEncode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "" + ], + "color": "#232", + "bgcolor": "#353" + }, + { + "id": 306, + "type": "LTXVConditioning", + "pos": [ + 2790, + 3670 + ], + "size": [ + 280, + 130 + ], + "flags": {}, + "order": 30, + "mode": 0, + "inputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 526 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 527 + }, + { + "localized_name": "frame_rate", + "name": "frame_rate", + "type": "FLOAT", + "widget": { + "name": "frame_rate" + }, + "link": 566 + } + ], + "outputs": [ + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "links": [ + 475, + 518 + ] + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "links": [ + 476, + 519 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.56", + "Node name for S&R": "LTXVConditioning", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 24 + ] + }, + { + "id": 307, + "type": "LTXVEmptyLatentAudio", + "pos": [ + 2970, + 4970 + ], + "size": [ + 280, + 170 + ], + "flags": {}, + "order": 31, + "mode": 0, + "inputs": [ + { + "localized_name": "audio_vae", + "name": "audio_vae", + "type": "VAE", + "link": 481 + }, + { + "localized_name": "frames_number", + "name": "frames_number", + "type": "INT", + "widget": { + "name": "frames_number" + }, + "link": 630 + }, + { + "localized_name": "frame_rate", + "name": "frame_rate", + "type": "INT", + "widget": { + "name": "frame_rate" + }, + "link": 565 + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "widget": { + "name": "batch_size" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "Latent", + "name": "Latent", + "type": "LATENT", + "links": [ + 498 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.68", + "Node name for S&R": "LTXVEmptyLatentAudio", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 97, + 25, + 1 + ] + }, + { + "id": 308, + "type": "ManualSigmas", + "pos": [ + 3200, + 4550 + ], + "size": [ + 500, + 110 + ], + "flags": {}, + "order": 6, + "mode": 0, + "inputs": [ + { + "localized_name": "sigmas", + "name": "sigmas", + "type": "STRING", + "widget": { + "name": "sigmas" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "SIGMAS", + "name": "SIGMAS", + "type": "SIGMAS", + "links": [ + 544 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.14.1", + "Node name for S&R": "ManualSigmas", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "1.0, 0.99375, 0.9875, 0.98125, 0.975, 0.909375, 0.725, 0.421875, 0.0" + ] + }, + { + "id": 309, + "type": "LTXVSeparateAVLatent", + "pos": [ + 3890, + 3910 + ], + "size": [ + 250, + 100 + ], + "flags": {}, + "order": 32, + "mode": 0, + "inputs": [ + { + "localized_name": "av_latent", + "name": "av_latent", + "type": "LATENT", + "link": 488 + } + ], + "outputs": [ + { + "localized_name": "video_latent", + "name": "video_latent", + "type": "LATENT", + "links": [ + 477, + 547 + ] + }, + { + "localized_name": "audio_latent", + "name": "audio_latent", + "type": "LATENT", + "links": [ + 513 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.5.1", + "Node name for S&R": "LTXVSeparateAVLatent", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 310, + "type": "SamplerCustomAdvanced", + "pos": [ + 5070, + 3750 + ], + "size": [ + 230, + 170 + ], + "flags": {}, + "order": 33, + "mode": 0, + "inputs": [ + { + "localized_name": "noise", + "name": "noise", + "type": "NOISE", + "link": 490 + }, + { + "localized_name": "guider", + "name": "guider", + "type": "GUIDER", + "link": 491 + }, + { + "localized_name": "sampler", + "name": "sampler", + "type": "SAMPLER", + "link": 492 + }, + { + "localized_name": "sigmas", + "name": "sigmas", + "type": "SIGMAS", + "link": 493 + }, + { + "localized_name": "latent_image", + "name": "latent_image", + "type": "LATENT", + "link": 494 + } + ], + "outputs": [ + { + "localized_name": "output", + "name": "output", + "type": "LATENT", + "links": [ + 578 + ] + }, + { + "localized_name": "denoised_output", + "name": "denoised_output", + "type": "LATENT", + "links": [] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.75", + "Node name for S&R": "SamplerCustomAdvanced", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 311, + "type": "LTXVSeparateAVLatent", + "pos": [ + 5410, + 3750 + ], + "size": [ + 230, + 100 + ], + "flags": {}, + "order": 34, + "mode": 0, + "inputs": [ + { + "localized_name": "av_latent", + "name": "av_latent", + "type": "LATENT", + "link": 578 + } + ], + "outputs": [ + { + "localized_name": "video_latent", + "name": "video_latent", + "type": "LATENT", + "links": [ + 539 + ] + }, + { + "localized_name": "audio_latent", + "name": "audio_latent", + "type": "LATENT", + "links": [ + 495 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.5.1", + "Node name for S&R": "LTXVSeparateAVLatent", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 312, + "type": "CreateVideo", + "pos": [ + 5740, + 4610 + ], + "size": [ + 280, + 130 + ], + "flags": {}, + "order": 35, + "mode": 0, + "inputs": [ + { + "localized_name": "images", + "name": "images", + "type": "IMAGE", + "link": 538 + }, + { + "localized_name": "audio", + "name": "audio", + "shape": 7, + "type": "AUDIO", + "link": 534 + }, + { + "localized_name": "fps", + "name": "fps", + "type": "FLOAT", + "widget": { + "name": "fps" + }, + "link": 591 + } + ], + "outputs": [ + { + "localized_name": "VIDEO", + "name": "VIDEO", + "type": "VIDEO", + "links": [ + 536 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.5.1", + "Node name for S&R": "CreateVideo", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 24 + ] + }, + { + "id": 313, + "type": "LatentUpscaleModelLoader", + "pos": [ + 1670, + 4600 + ], + "size": [ + 400, + 110 + ], + "flags": {}, + "order": 36, + "mode": 0, + "inputs": [ + { + "localized_name": "model_name", + "name": "model_name", + "type": "COMBO", + "widget": { + "name": "model_name" + }, + "link": 607 + } + ], + "outputs": [ + { + "localized_name": "LATENT_UPSCALE_MODEL", + "name": "LATENT_UPSCALE_MODEL", + "type": "LATENT_UPSCALE_MODEL", + "links": [ + 545 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "LatentUpscaleModelLoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "ltx-2.3-spatial-upscaler-x2-1.1.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2.3/resolve/main/ltx-2.3-spatial-upscaler-x2-1.1.safetensors", + "directory": "latent_upscale_models" + } + ] + }, + "widgets_values": [ + "ltx-2.3-spatial-upscaler-x2-1.1.safetensors" + ] + }, + { + "id": 314, + "type": "PrimitiveInt", + "pos": [ + 1160, + 4380 + ], + "size": [ + 370, + 110 + ], + "flags": {}, + "order": 37, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "INT", + "widget": { + "name": "value" + }, + "link": 597 + } + ], + "outputs": [ + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 558, + 560 + ] + } + ], + "title": "Width", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.3", + "Node name for S&R": "PrimitiveInt", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1280, + "fixed" + ] + }, + { + "id": 315, + "type": "CLIPTextEncode", + "pos": [ + 2180, + 4480 + ], + "size": [ + 530, + 240 + ], + "flags": {}, + "order": 38, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 625 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 527 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.56", + "Node name for S&R": "CLIPTextEncode", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "pc game, console game, video game, cartoon, childish, ugly" + ], + "color": "#323", + "bgcolor": "#535" + }, + { + "id": 316, + "type": "CFGGuider", + "pos": [ + 3200, + 4100 + ], + "size": [ + 280, + 160 + ], + "flags": {}, + "order": 39, + "mode": 0, + "inputs": [ + { + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 541 + }, + { + "localized_name": "positive", + "name": "positive", + "type": "CONDITIONING", + "link": 518 + }, + { + "localized_name": "negative", + "name": "negative", + "type": "CONDITIONING", + "link": 519 + }, + { + "localized_name": "cfg", + "name": "cfg", + "type": "FLOAT", + "widget": { + "name": "cfg" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "GUIDER", + "name": "GUIDER", + "type": "GUIDER", + "links": [ + 484 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.64", + "Node name for S&R": "CFGGuider", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 1 + ] + }, + { + "id": 317, + "type": "VAEDecodeTiled", + "pos": [ + 5760, + 3650 + ], + "size": [ + 280, + 200 + ], + "flags": {}, + "order": 40, + "mode": 0, + "inputs": [ + { + "localized_name": "samples", + "name": "samples", + "type": "LATENT", + "link": 539 + }, + { + "localized_name": "vae", + "name": "vae", + "type": "VAE", + "link": 553 + }, + { + "localized_name": "tile_size", + "name": "tile_size", + "type": "INT", + "widget": { + "name": "tile_size" + }, + "link": null + }, + { + "localized_name": "overlap", + "name": "overlap", + "type": "INT", + "widget": { + "name": "overlap" + }, + "link": null + }, + { + "localized_name": "temporal_size", + "name": "temporal_size", + "type": "INT", + "widget": { + "name": "temporal_size" + }, + "link": null + }, + { + "localized_name": "temporal_overlap", + "name": "temporal_overlap", + "type": "INT", + "widget": { + "name": "temporal_overlap" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 538 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.14.1", + "Node name for S&R": "VAEDecodeTiled", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 768, + 64, + 4096, + 4 + ] + }, + { + "id": 318, + "type": "CheckpointLoaderSimple", + "pos": [ + 1660, + 3660 + ], + "size": [ + 430, + 160 + ], + "flags": {}, + "order": 41, + "mode": 0, + "inputs": [ + { + "localized_name": "ckpt_name", + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": 601 + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 520 + ] + }, + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [] + }, + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "links": [ + 556, + 557 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.3.56", + "Node name for S&R": "CheckpointLoaderSimple", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "ltx-2.3-22b-dev-fp8.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2.3-fp8/resolve/main/ltx-2.3-22b-dev-fp8.safetensors", + "directory": "checkpoints" + } + ] + }, + "widgets_values": [ + "ltx-2.3-22b-dev-fp8.safetensors" + ] + }, + { + "id": 319, + "type": "LTXAVTextEncoderLoader", + "pos": [ + 1660, + 4340 + ], + "size": [ + 430, + 170 + ], + "flags": {}, + "order": 42, + "mode": 0, + "inputs": [ + { + "localized_name": "text_encoder", + "name": "text_encoder", + "type": "COMBO", + "widget": { + "name": "text_encoder" + }, + "link": 606 + }, + { + "localized_name": "ckpt_name", + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": 605 + }, + { + "localized_name": "device", + "name": "device", + "type": "COMBO", + "widget": { + "name": "device" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [ + 615, + 625 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "LTXAVTextEncoderLoader", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "ltx-2.3-22b-dev-fp8.safetensors", + "url": "https://huggingface.co/Lightricks/LTX-2.3-fp8/resolve/main/ltx-2.3-22b-dev-fp8.safetensors", + "directory": "checkpoints" + }, + { + "name": "gemma_3_12B_it_fp4_mixed.safetensors", + "url": "https://huggingface.co/Comfy-Org/ltx-2/resolve/main/split_files/text_encoders/gemma_3_12B_it_fp4_mixed.safetensors", + "directory": "text_encoders" + } + ] + }, + "widgets_values": [ + "gemma_3_12B_it_fp4_mixed.safetensors", + "ltx-2.3-22b-dev-fp8.safetensors", + "default" + ] + }, + { + "id": 320, + "type": "PrimitiveStringMultiline", + "pos": [ + 1160, + 3680 + ], + "size": [ + 370, + 350 + ], + "flags": {}, + "order": 43, + "mode": 0, + "inputs": [ + { + "localized_name": "value", + "name": "value", + "type": "STRING", + "widget": { + "name": "value" + }, + "link": 595 + } + ], + "outputs": [ + { + "localized_name": "STRING", + "name": "STRING", + "type": "STRING", + "links": [ + 623 + ] + } + ], + "title": "Prompt", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.3", + "Node name for S&R": "PrimitiveStringMultiline", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "" + ] + }, + { + "id": 321, + "type": "LTXVConcatAVLatent", + "pos": [ + 3820, + 4990 + ], + "size": [ + 240, + 100 + ], + "flags": {}, + "order": 44, + "mode": 0, + "inputs": [ + { + "localized_name": "video_latent", + "name": "video_latent", + "type": "LATENT", + "link": 497 + }, + { + "localized_name": "audio_latent", + "name": "audio_latent", + "type": "LATENT", + "link": 498 + } + ], + "outputs": [ + { + "localized_name": "latent", + "name": "latent", + "type": "LATENT", + "links": [ + 487 + ] + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.7.0", + "Node name for S&R": "LTXVConcatAVLatent", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 322, + "type": "LoadImage", + "pos": [ + 1150, + 4940 + ], + "size": [ + 400, + 480 + ], + "flags": {}, + "order": 7, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "COMBO", + "widget": { + "name": "image" + }, + "link": null + }, + { + "localized_name": "choose file to upload", + "name": "upload", + "type": "IMAGEUPLOAD", + "widget": { + "name": "upload" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "IMAGE", + "name": "IMAGE", + "type": "IMAGE", + "links": [ + 626 + ] + }, + { + "localized_name": "MASK", + "name": "MASK", + "type": "MASK", + "links": null + } + ], + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "version": "7.7", + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.16.3", + "Node name for S&R": "LoadImage", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "example.png", + "image" + ] + }, + { + "id": 323, + "type": "ComfyMathExpression", + "pos": [ + 2540, + 5370 + ], + "size": [ + 260, + 190 + ], + "flags": { + "collapsed": true + }, + "order": 45, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 628 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": 629 + }, + { + "label": "c", + "localized_name": "values.c", + "name": "values.c", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 630, + 631 + ] + } + ], + "title": "Math Expression (length)", + "properties": { + "ue_properties": { + "widget_ue_connectable": {}, + "input_ue_unconnectable": {} + }, + "cnr_id": "comfy-core", + "ver": "0.18.1", + "Node name for S&R": "ComfyMathExpression" + }, + "widgets_values": [ + "a * b + 1" + ] + } + ], + "groups": [ + { + "id": 1, + "title": "Model", + "bounding": [ + 1630, + 3550, + 480, + 1270 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 2, + "title": "Generate Low Resolution", + "bounding": [ + 3150, + 3550, + 1020, + 1270 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 3, + "title": "Prompt", + "bounding": [ + 2140, + 3550, + 980, + 1270 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 6, + "title": "Generate High Resolution", + "bounding": [ + 4690, + 3550, + 960, + 1270 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 7, + "title": "Lantent Upscale", + "bounding": [ + 4200, + 3550, + 460, + 1270 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 19, + "title": "Video Settings", + "bounding": [ + 1110, + 3550, + 490, + 1270 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 20, + "title": "Image Preprocess", + "bounding": [ + 1630, + 4850, + 830, + 610 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 21, + "title": "Empty Latent", + "bounding": [ + 2830, + 4850, + 1340, + 610 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 22, + "title": "Number conversion", + "bounding": [ + 2490, + 4850, + 320, + 610 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + }, + { + "id": 26, + "title": "Image will not affect the video", + "bounding": [ + 1110, + 4850, + 490, + 610 + ], + "color": "#3f789e", + "font_size": 24, + "flags": {} + } + ], + "links": [ + { + "id": 512, + "origin_id": 290, + "origin_slot": 0, + "target_id": 280, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 513, + "origin_id": 309, + "origin_slot": 1, + "target_id": 280, + "target_slot": 1, + "type": "LATENT" + }, + { + "id": 478, + "origin_id": 287, + "origin_slot": 0, + "target_id": 284, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 479, + "origin_id": 286, + "origin_slot": 0, + "target_id": 284, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 480, + "origin_id": 286, + "origin_slot": 1, + "target_id": 284, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 541, + "origin_id": 287, + "origin_slot": 0, + "target_id": 316, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 518, + "origin_id": 306, + "origin_slot": 0, + "target_id": 316, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 519, + "origin_id": 306, + "origin_slot": 1, + "target_id": 316, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 483, + "origin_id": 279, + "origin_slot": 0, + "target_id": 285, + "target_slot": 0, + "type": "NOISE" + }, + { + "id": 484, + "origin_id": 316, + "origin_slot": 0, + "target_id": 285, + "target_slot": 1, + "type": "GUIDER" + }, + { + "id": 485, + "origin_id": 293, + "origin_slot": 0, + "target_id": 285, + "target_slot": 2, + "type": "SAMPLER" + }, + { + "id": 544, + "origin_id": 308, + "origin_slot": 0, + "target_id": 285, + "target_slot": 3, + "type": "SIGMAS" + }, + { + "id": 487, + "origin_id": 321, + "origin_slot": 0, + "target_id": 285, + "target_slot": 4, + "type": "LATENT" + }, + { + "id": 475, + "origin_id": 306, + "origin_slot": 0, + "target_id": 286, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 476, + "origin_id": 306, + "origin_slot": 1, + "target_id": 286, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 477, + "origin_id": 309, + "origin_slot": 0, + "target_id": 286, + "target_slot": 2, + "type": "LATENT" + }, + { + "id": 520, + "origin_id": 318, + "origin_slot": 0, + "target_id": 287, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 523, + "origin_id": 292, + "origin_slot": 0, + "target_id": 288, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 547, + "origin_id": 309, + "origin_slot": 0, + "target_id": 289, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 545, + "origin_id": 313, + "origin_slot": 0, + "target_id": 289, + "target_slot": 1, + "type": "LATENT_UPSCALE_MODEL" + }, + { + "id": 554, + "origin_id": 295, + "origin_slot": 0, + "target_id": 289, + "target_slot": 2, + "type": "VAE" + }, + { + "id": 552, + "origin_id": 295, + "origin_slot": 0, + "target_id": 290, + "target_slot": 0, + "type": "VAE" + }, + { + "id": 515, + "origin_id": 291, + "origin_slot": 0, + "target_id": 290, + "target_slot": 1, + "type": "IMAGE" + }, + { + "id": 548, + "origin_id": 289, + "origin_slot": 0, + "target_id": 290, + "target_slot": 2, + "type": "LATENT" + }, + { + "id": 543, + "origin_id": 304, + "origin_slot": 0, + "target_id": 290, + "target_slot": 4, + "type": "BOOLEAN" + }, + { + "id": 505, + "origin_id": 288, + "origin_slot": 0, + "target_id": 291, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 558, + "origin_id": 314, + "origin_slot": 0, + "target_id": 292, + "target_slot": 2, + "type": "INT" + }, + { + "id": 559, + "origin_id": 301, + "origin_slot": 0, + "target_id": 292, + "target_slot": 3, + "type": "INT" + }, + { + "id": 560, + "origin_id": 314, + "origin_slot": 0, + "target_id": 294, + "target_slot": 0, + "type": "INT" + }, + { + "id": 557, + "origin_id": 318, + "origin_slot": 2, + "target_id": 295, + "target_slot": 0, + "type": "VAE" + }, + { + "id": 562, + "origin_id": 301, + "origin_slot": 0, + "target_id": 296, + "target_slot": 0, + "type": "INT" + }, + { + "id": 561, + "origin_id": 294, + "origin_slot": 1, + "target_id": 297, + "target_slot": 0, + "type": "INT" + }, + { + "id": 563, + "origin_id": 296, + "origin_slot": 1, + "target_id": 297, + "target_slot": 1, + "type": "INT" + }, + { + "id": 556, + "origin_id": 318, + "origin_slot": 2, + "target_id": 298, + "target_slot": 0, + "type": "VAE" + }, + { + "id": 510, + "origin_id": 291, + "origin_slot": 0, + "target_id": 298, + "target_slot": 1, + "type": "IMAGE" + }, + { + "id": 511, + "origin_id": 297, + "origin_slot": 0, + "target_id": 298, + "target_slot": 2, + "type": "LATENT" + }, + { + "id": 542, + "origin_id": 304, + "origin_slot": 0, + "target_id": 298, + "target_slot": 4, + "type": "BOOLEAN" + }, + { + "id": 495, + "origin_id": 311, + "origin_slot": 1, + "target_id": 299, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 496, + "origin_id": 281, + "origin_slot": 0, + "target_id": 299, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 564, + "origin_id": 302, + "origin_slot": 0, + "target_id": 300, + "target_slot": 0, + "type": "INT" + }, + { + "id": 526, + "origin_id": 305, + "origin_slot": 0, + "target_id": 306, + "target_slot": 0, + "type": "CONDITIONING" + }, + { + "id": 527, + "origin_id": 315, + "origin_slot": 0, + "target_id": 306, + "target_slot": 1, + "type": "CONDITIONING" + }, + { + "id": 566, + "origin_id": 300, + "origin_slot": 0, + "target_id": 306, + "target_slot": 2, + "type": "FLOAT" + }, + { + "id": 497, + "origin_id": 298, + "origin_slot": 0, + "target_id": 321, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 498, + "origin_id": 307, + "origin_slot": 0, + "target_id": 321, + "target_slot": 1, + "type": "LATENT" + }, + { + "id": 481, + "origin_id": 281, + "origin_slot": 0, + "target_id": 307, + "target_slot": 0, + "type": "VAE" + }, + { + "id": 565, + "origin_id": 300, + "origin_slot": 1, + "target_id": 307, + "target_slot": 2, + "type": "INT" + }, + { + "id": 488, + "origin_id": 285, + "origin_slot": 0, + "target_id": 309, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 490, + "origin_id": 278, + "origin_slot": 0, + "target_id": 310, + "target_slot": 0, + "type": "NOISE" + }, + { + "id": 491, + "origin_id": 284, + "origin_slot": 0, + "target_id": 310, + "target_slot": 1, + "type": "GUIDER" + }, + { + "id": 492, + "origin_id": 282, + "origin_slot": 0, + "target_id": 310, + "target_slot": 2, + "type": "SAMPLER" + }, + { + "id": 493, + "origin_id": 283, + "origin_slot": 0, + "target_id": 310, + "target_slot": 3, + "type": "SIGMAS" + }, + { + "id": 494, + "origin_id": 280, + "origin_slot": 0, + "target_id": 310, + "target_slot": 4, + "type": "LATENT" + }, + { + "id": 578, + "origin_id": 310, + "origin_slot": 0, + "target_id": 311, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 539, + "origin_id": 311, + "origin_slot": 0, + "target_id": 317, + "target_slot": 0, + "type": "LATENT" + }, + { + "id": 553, + "origin_id": 295, + "origin_slot": 0, + "target_id": 317, + "target_slot": 1, + "type": "VAE" + }, + { + "id": 538, + "origin_id": 317, + "origin_slot": 0, + "target_id": 312, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 534, + "origin_id": 299, + "origin_slot": 0, + "target_id": 312, + "target_slot": 1, + "type": "AUDIO" + }, + { + "id": 591, + "origin_id": 300, + "origin_slot": 0, + "target_id": 312, + "target_slot": 2, + "type": "FLOAT" + }, + { + "id": 536, + "origin_id": 312, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "VIDEO" + }, + { + "id": 595, + "origin_id": -10, + "origin_slot": 0, + "target_id": 320, + "target_slot": 0, + "type": "STRING" + }, + { + "id": 597, + "origin_id": -10, + "origin_slot": 1, + "target_id": 314, + "target_slot": 0, + "type": "INT" + }, + { + "id": 598, + "origin_id": -10, + "origin_slot": 2, + "target_id": 301, + "target_slot": 0, + "type": "INT" + }, + { + "id": 599, + "origin_id": -10, + "origin_slot": 3, + "target_id": 303, + "target_slot": 0, + "type": "INT" + }, + { + "id": 601, + "origin_id": -10, + "origin_slot": 4, + "target_id": 318, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 602, + "origin_id": -10, + "origin_slot": 5, + "target_id": 287, + "target_slot": 1, + "type": "COMBO" + }, + { + "id": 604, + "origin_id": -10, + "origin_slot": 4, + "target_id": 281, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 605, + "origin_id": -10, + "origin_slot": 4, + "target_id": 319, + "target_slot": 1, + "type": "COMBO" + }, + { + "id": 606, + "origin_id": -10, + "origin_slot": 6, + "target_id": 319, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 607, + "origin_id": -10, + "origin_slot": 7, + "target_id": 313, + "target_slot": 0, + "type": "COMBO" + }, + { + "id": 615, + "origin_id": 319, + "origin_slot": 0, + "target_id": 305, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 623, + "origin_id": 320, + "origin_slot": 0, + "target_id": 305, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 625, + "origin_id": 319, + "origin_slot": 0, + "target_id": 315, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 626, + "origin_id": 322, + "origin_slot": 0, + "target_id": 292, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 627, + "origin_id": -10, + "origin_slot": 8, + "target_id": 302, + "target_slot": 0, + "type": "INT" + }, + { + "id": 628, + "origin_id": 303, + "origin_slot": 0, + "target_id": 323, + "target_slot": 0, + "type": "INT" + }, + { + "id": 629, + "origin_id": 302, + "origin_slot": 0, + "target_id": 323, + "target_slot": 1, + "type": "INT" + }, + { + "id": 630, + "origin_id": 323, + "origin_slot": 1, + "target_id": 307, + "target_slot": 1, + "type": "INT" + }, + { + "id": 631, + "origin_id": 323, + "origin_slot": 1, + "target_id": 297, + "target_slot": 2, + "type": "INT" + } + ], + "extra": { + "workflowRendererVersion": "Vue-corrected" + }, + "category": "Video generation and editing/Text to video", + "description": "Generates video from text prompts using LTX-2.3, Lightricks' video diffusion model." + } + ] + }, + "extra": { + "ue_links": [] + } +} \ No newline at end of file diff --git a/blueprints/Text to Video (Wan 2.2).json b/blueprints/Text to Video (Wan 2.2).json index 0ce485b67..a264a490d 100644 --- a/blueprints/Text to Video (Wan 2.2).json +++ b/blueprints/Text to Video (Wan 2.2).json @@ -1572,7 +1572,8 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Video generation and editing/Text to video" + "category": "Video generation and editing/Text to video", + "description": "Generates video from text prompts using Wan2.2, Alibaba's diffusion video model." } ] }, @@ -1586,4 +1587,4 @@ "VHS_KeepIntermediate": true }, "version": 0.4 -} +} \ No newline at end of file diff --git a/blueprints/Unsharp Mask.json b/blueprints/Unsharp Mask.json index b673eb703..79a4c954f 100644 --- a/blueprints/Unsharp Mask.json +++ b/blueprints/Unsharp Mask.json @@ -383,7 +383,7 @@ "Node name for S&R": "GLSLShader" }, "widgets_values": [ - "#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\nuniform vec2 u_resolution;\nuniform float u_float0; // amount [0.0 - 3.0] typical: 0.5-1.5\nuniform float u_float1; // radius [0.5 - 10.0] blur radius in pixels\nuniform float u_float2; // threshold [0.0 - 0.1] min difference to sharpen\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\n\nfloat gaussian(float x, float sigma) {\n return exp(-(x * x) / (2.0 * sigma * sigma));\n}\n\nfloat getLuminance(vec3 color) {\n return dot(color, vec3(0.2126, 0.7152, 0.0722));\n}\n\nvoid main() {\n vec2 texel = 1.0 / u_resolution;\n float radius = max(u_float1, 0.5);\n float amount = u_float0;\n float threshold = u_float2;\n\n vec4 original = texture(u_image0, v_texCoord);\n\n // Gaussian blur for the \"unsharp\" mask\n int samples = int(ceil(radius));\n float sigma = radius / 2.0;\n\n vec4 blurred = vec4(0.0);\n float totalWeight = 0.0;\n\n for (int x = -samples; x <= samples; x++) {\n for (int y = -samples; y <= samples; y++) {\n vec2 offset = vec2(float(x), float(y)) * texel;\n vec4 sample_color = texture(u_image0, v_texCoord + offset);\n\n float dist = length(vec2(float(x), float(y)));\n float weight = gaussian(dist, sigma);\n blurred += sample_color * weight;\n totalWeight += weight;\n }\n }\n blurred /= totalWeight;\n\n // Unsharp mask = original - blurred\n vec3 mask = original.rgb - blurred.rgb;\n\n // Luminance-based threshold with smooth falloff\n float lumaDelta = abs(getLuminance(original.rgb) - getLuminance(blurred.rgb));\n float thresholdScale = smoothstep(0.0, threshold, lumaDelta);\n mask *= thresholdScale;\n\n // Sharpen: original + mask * amount\n vec3 sharpened = original.rgb + mask * amount;\n\n fragColor0 = vec4(clamp(sharpened, 0.0, 1.0), original.a);\n}\n", + "#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\nuniform float u_float0; // amount [0.0 - 3.0] typical: 0.5-1.5\nuniform float u_float1; // radius [0.5 - 10.0] blur radius in pixels\nuniform float u_float2; // threshold [0.0 - 0.1] min difference to sharpen\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\n\nfloat gaussian(float x, float sigma) {\n return exp(-(x * x) / (2.0 * sigma * sigma));\n}\n\nfloat getLuminance(vec3 color) {\n return dot(color, vec3(0.2126, 0.7152, 0.0722));\n}\n\nvoid main() {\n vec2 texel = 1.0 / vec2(textureSize(u_image0, 0));\n float radius = max(u_float1, 0.5);\n float amount = u_float0;\n float threshold = u_float2;\n\n vec4 original = texture(u_image0, v_texCoord);\n\n // Gaussian blur for the \"unsharp\" mask\n int samples = int(ceil(radius));\n float sigma = radius / 2.0;\n\n vec4 blurred = vec4(0.0);\n float totalWeight = 0.0;\n\n for (int x = -samples; x <= samples; x++) {\n for (int y = -samples; y <= samples; y++) {\n vec2 offset = vec2(float(x), float(y)) * texel;\n vec4 sample_color = texture(u_image0, v_texCoord + offset);\n\n float dist = length(vec2(float(x), float(y)));\n float weight = gaussian(dist, sigma);\n blurred += sample_color * weight;\n totalWeight += weight;\n }\n }\n blurred /= totalWeight;\n\n // Unsharp mask = original - blurred\n vec3 mask = original.rgb - blurred.rgb;\n\n // Luminance-based threshold with smooth falloff\n float lumaDelta = abs(getLuminance(original.rgb) - getLuminance(blurred.rgb));\n float thresholdScale = smoothstep(0.0, threshold, lumaDelta);\n mask *= thresholdScale;\n\n // Sharpen: original + mask * amount\n vec3 sharpened = original.rgb + mask * amount;\n\n fragColor0 = vec4(clamp(sharpened, 0.0, 1.0), original.a);\n}\n", "from_input" ] } @@ -434,8 +434,9 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Image Tools/Sharpen" + "category": "Image Tools/Sharpen", + "description": "Enhances edge contrast via unsharp masking for a sharper image appearance." } ] } -} +} \ No newline at end of file diff --git a/blueprints/Video Captioning (Gemini).json b/blueprints/Video Captioning (Gemini).json index ea6dc8bee..7642b23c1 100644 --- a/blueprints/Video Captioning (Gemini).json +++ b/blueprints/Video Captioning (Gemini).json @@ -307,7 +307,8 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Text generation/Video Captioning" + "category": "Text generation/Video Captioning", + "description": "Generates descriptive captions for video input using Google's Gemini multimodal LLM." } ] } diff --git a/blueprints/Video Inpaint(Wan2.1 VACE).json b/blueprints/Video Inpaint(Wan2.1 VACE).json index f404e6773..a658be5f8 100644 --- a/blueprints/Video Inpaint(Wan2.1 VACE).json +++ b/blueprints/Video Inpaint(Wan2.1 VACE).json @@ -165,7 +165,7 @@ }, "revision": 0, "config": {}, - "name": "local-Video Inpaint(Wan2.1 VACE)", + "name": "Video Inpaint (Wan 2.1 VACE)", "inputNode": { "id": -10, "bounding": [ @@ -2368,7 +2368,8 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Video generation and editing/Inpaint video" + "category": "Video generation and editing/Inpaint video", + "description": "Inpaints masked regions in video frames using Wan 2.1 VACE." } ] }, diff --git a/blueprints/Video Segmentation (SAM3).json b/blueprints/Video Segmentation (SAM3).json new file mode 100644 index 000000000..4d9a13412 --- /dev/null +++ b/blueprints/Video Segmentation (SAM3).json @@ -0,0 +1,827 @@ +{ + "revision": 0, + "last_node_id": 130, + "last_link_id": 0, + "nodes": [ + { + "id": 130, + "type": "7937cf78-b52b-40a3-93b2-b4e2e5f98df1", + "pos": [ + -1210, + -2780 + ], + "size": [ + 300, + 370 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "name": "video", + "type": "VIDEO", + "link": null + }, + { + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": null + }, + { + "name": "bboxes", + "type": "BOUNDING_BOX", + "link": null + }, + { + "name": "positive_coords", + "type": "STRING", + "link": null + }, + { + "name": "negative_coords", + "type": "STRING", + "link": null + }, + { + "name": "threshold", + "type": "FLOAT", + "widget": { + "name": "threshold" + }, + "link": null + }, + { + "name": "refine_iterations", + "type": "INT", + "widget": { + "name": "refine_iterations" + }, + "link": null + }, + { + "name": "individual_masks", + "type": "BOOLEAN", + "widget": { + "name": "individual_masks" + }, + "link": null + }, + { + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "masks", + "name": "masks", + "type": "MASK", + "links": [] + }, + { + "localized_name": "bboxes", + "name": "bboxes", + "type": "BOUNDING_BOX", + "links": [] + }, + { + "name": "audio", + "type": "AUDIO", + "links": null + }, + { + "name": "fps", + "type": "FLOAT", + "links": null + } + ], + "properties": { + "proxyWidgets": [ + [ + "125", + "text" + ], + [ + "126", + "threshold" + ], + [ + "126", + "refine_iterations" + ], + [ + "126", + "individual_masks" + ], + [ + "127", + "ckpt_name" + ] + ], + "cnr_id": "comfy-core", + "ver": "0.19.3", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [], + "title": "Video Segmentation (SAM3)" + } + ], + "links": [], + "version": 0.4, + "definitions": { + "subgraphs": [ + { + "id": "7937cf78-b52b-40a3-93b2-b4e2e5f98df1", + "version": 1, + "state": { + "lastGroupId": 0, + "lastNodeId": 130, + "lastLinkId": 299, + "lastRerouteId": 0 + }, + "revision": 0, + "config": {}, + "name": "Video Segmentation (SAM3)", + "inputNode": { + "id": -10, + "bounding": [ + -2260, + -3450, + 136.369140625, + 220 + ] + }, + "outputNode": { + "id": -20, + "bounding": [ + -1050, + -3510, + 120, + 120 + ] + }, + "inputs": [ + { + "id": "680ffd88-32fe-48be-88d6-91ea44d5eaee", + "name": "video", + "type": "VIDEO", + "linkIds": [ + 252 + ], + "pos": [ + -2143.630859375, + -3430 + ] + }, + { + "id": "ceaf249c-32d7-4624-8bf6-e590e347ed90", + "name": "text", + "type": "STRING", + "linkIds": [ + 254 + ], + "pos": [ + -2143.630859375, + -3410 + ] + }, + { + "id": "1ffbff36-da0c-4854-8cb4-88ad31e64f99", + "name": "bboxes", + "type": "BOUNDING_BOX", + "linkIds": [ + 255 + ], + "pos": [ + -2143.630859375, + -3390 + ] + }, + { + "id": "67b7f4c7-cec0-4e00-b154-23cc1abf880e", + "name": "positive_coords", + "type": "STRING", + "linkIds": [ + 256 + ], + "pos": [ + -2143.630859375, + -3370 + ] + }, + { + "id": "b090a498-2bde-46b9-9554-18501401d687", + "name": "negative_coords", + "type": "STRING", + "linkIds": [ + 257 + ], + "pos": [ + -2143.630859375, + -3350 + ] + }, + { + "id": "1a76dfcf-ce95-46af-bba5-c42160c683dd", + "name": "threshold", + "type": "FLOAT", + "linkIds": [ + 261 + ], + "pos": [ + -2143.630859375, + -3330 + ] + }, + { + "id": "999523fa-c476-4c53-80c3-0a2f554d18ab", + "name": "refine_iterations", + "type": "INT", + "linkIds": [ + 262 + ], + "pos": [ + -2143.630859375, + -3310 + ] + }, + { + "id": "d2371011-7fe5-4a39-b0c1-df2e0bbd6ece", + "name": "individual_masks", + "type": "BOOLEAN", + "linkIds": [ + 263 + ], + "pos": [ + -2143.630859375, + -3290 + ] + }, + { + "id": "675a8b37-17db-48d1-853c-2fe5d6a74582", + "name": "ckpt_name", + "type": "COMBO", + "linkIds": [ + 273 + ], + "pos": [ + -2143.630859375, + -3270 + ] + } + ], + "outputs": [ + { + "id": "ff50da09-1e59-4a58-9b7f-be1a00aa5913", + "name": "masks", + "type": "MASK", + "linkIds": [ + 231 + ], + "localized_name": "masks", + "pos": [ + -1030, + -3490 + ] + }, + { + "id": "8f622e40-8528-4078-b7d3-147e9f872194", + "name": "bboxes", + "type": "BOUNDING_BOX", + "linkIds": [ + 232 + ], + "localized_name": "bboxes", + "pos": [ + -1030, + -3470 + ] + }, + { + "id": "6c9924ec-f0fa-4509-83ea-8f97f5889bcc", + "name": "audio", + "type": "AUDIO", + "linkIds": [ + 259 + ], + "pos": [ + -1030, + -3450 + ] + }, + { + "id": "82c1cddc-ab11-44eb-9e2f-1a5c7ea5645b", + "name": "fps", + "type": "FLOAT", + "linkIds": [ + 260 + ], + "pos": [ + -1030, + -3430 + ] + } + ], + "widgets": [], + "nodes": [ + { + "id": 125, + "type": "CLIPTextEncode", + "pos": [ + -2010, + -3040 + ], + "size": [ + 400, + 200 + ], + "flags": {}, + "order": 1, + "mode": 0, + "inputs": [ + { + "localized_name": "clip", + "name": "clip", + "type": "CLIP", + "link": 240 + }, + { + "localized_name": "text", + "name": "text", + "type": "STRING", + "widget": { + "name": "text" + }, + "link": 254 + } + ], + "outputs": [ + { + "localized_name": "CONDITIONING", + "name": "CONDITIONING", + "type": "CONDITIONING", + "links": [ + 200 + ] + } + ], + "properties": { + "Node name for S&R": "CLIPTextEncode", + "cnr_id": "comfy-core", + "ver": "0.19.3", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + "" + ] + }, + { + "id": 126, + "type": "SAM3_Detect", + "pos": [ + -1520, + -3520 + ], + "size": [ + 270, + 290 + ], + "flags": {}, + "order": 2, + "mode": 0, + "inputs": [ + { + "label": "model", + "localized_name": "model", + "name": "model", + "type": "MODEL", + "link": 237 + }, + { + "label": "image", + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 253 + }, + { + "label": "conditioning", + "localized_name": "conditioning", + "name": "conditioning", + "shape": 7, + "type": "CONDITIONING", + "link": 200 + }, + { + "label": "bboxes", + "localized_name": "bboxes", + "name": "bboxes", + "shape": 7, + "type": "BOUNDING_BOX", + "link": 255 + }, + { + "label": "positive_coords", + "localized_name": "positive_coords", + "name": "positive_coords", + "shape": 7, + "type": "STRING", + "link": 256 + }, + { + "label": "negative_coords", + "localized_name": "negative_coords", + "name": "negative_coords", + "shape": 7, + "type": "STRING", + "link": 257 + }, + { + "localized_name": "threshold", + "name": "threshold", + "type": "FLOAT", + "widget": { + "name": "threshold" + }, + "link": 261 + }, + { + "localized_name": "refine_iterations", + "name": "refine_iterations", + "type": "INT", + "widget": { + "name": "refine_iterations" + }, + "link": 262 + }, + { + "localized_name": "individual_masks", + "name": "individual_masks", + "type": "BOOLEAN", + "widget": { + "name": "individual_masks" + }, + "link": 263 + } + ], + "outputs": [ + { + "localized_name": "masks", + "name": "masks", + "type": "MASK", + "links": [ + 231 + ] + }, + { + "localized_name": "bboxes", + "name": "bboxes", + "type": "BOUNDING_BOX", + "links": [ + 232 + ] + } + ], + "properties": { + "Node name for S&R": "SAM3_Detect", + "cnr_id": "comfy-core", + "ver": "0.19.3", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + }, + "widgets_values": [ + 0.5, + 2, + false + ] + }, + { + "id": 127, + "type": "CheckpointLoaderSimple", + "pos": [ + -1970, + -3310 + ], + "size": [ + 330, + 160 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "ckpt_name", + "name": "ckpt_name", + "type": "COMBO", + "widget": { + "name": "ckpt_name" + }, + "link": 273 + } + ], + "outputs": [ + { + "localized_name": "MODEL", + "name": "MODEL", + "type": "MODEL", + "links": [ + 237 + ] + }, + { + "localized_name": "CLIP", + "name": "CLIP", + "type": "CLIP", + "links": [ + 240 + ] + }, + { + "localized_name": "VAE", + "name": "VAE", + "type": "VAE", + "links": null + } + ], + "properties": { + "Node name for S&R": "CheckpointLoaderSimple", + "cnr_id": "comfy-core", + "ver": "0.19.3", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65, + "models": [ + { + "name": "sam3.1_multiplex_fp16.safetensors", + "url": "https://huggingface.co/Comfy-Org/sam3.1/resolve/main/checkpoints/sam3.1_multiplex_fp16.safetensors", + "directory": "checkpoints" + } + ] + }, + "widgets_values": [ + "sam3.1_multiplex_fp16.safetensors" + ] + }, + { + "id": 128, + "type": "GetVideoComponents", + "pos": [ + -1910, + -3540 + ], + "size": [ + 230, + 120 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "video", + "name": "video", + "type": "VIDEO", + "link": 252 + } + ], + "outputs": [ + { + "localized_name": "images", + "name": "images", + "type": "IMAGE", + "links": [ + 253 + ] + }, + { + "localized_name": "audio", + "name": "audio", + "type": "AUDIO", + "links": [ + 259 + ] + }, + { + "localized_name": "fps", + "name": "fps", + "type": "FLOAT", + "links": [ + 260 + ] + } + ], + "properties": { + "Node name for S&R": "GetVideoComponents", + "cnr_id": "comfy-core", + "ver": "0.19.3", + "enableTabs": false, + "tabWidth": 65, + "tabXOffset": 10, + "hasSecondTab": false, + "secondTabText": "Send Back", + "secondTabOffset": 80, + "secondTabWidth": 65 + } + }, + { + "id": 129, + "type": "Note", + "pos": [ + -1980, + -2790 + ], + "size": [ + 370, + 250 + ], + "flags": {}, + "order": 0, + "mode": 0, + "inputs": [], + "outputs": [], + "title": "Note: Prompt format", + "properties": {}, + "widgets_values": [ + "Max tokens for this model is only 32, to separately prompt multiple subjects you can separate prompts with comma, and set the max amount of objects detected for each prompt with :N\n\nFor example above test prompt finds 2 cakes, one apron, 4 window panels" + ], + "color": "#432", + "bgcolor": "#653" + } + ], + "groups": [], + "links": [ + { + "id": 237, + "origin_id": 127, + "origin_slot": 0, + "target_id": 126, + "target_slot": 0, + "type": "MODEL" + }, + { + "id": 200, + "origin_id": 125, + "origin_slot": 0, + "target_id": 126, + "target_slot": 2, + "type": "CONDITIONING" + }, + { + "id": 240, + "origin_id": 127, + "origin_slot": 1, + "target_id": 125, + "target_slot": 0, + "type": "CLIP" + }, + { + "id": 231, + "origin_id": 126, + "origin_slot": 0, + "target_id": -20, + "target_slot": 0, + "type": "MASK" + }, + { + "id": 232, + "origin_id": 126, + "origin_slot": 1, + "target_id": -20, + "target_slot": 1, + "type": "BOUNDING_BOX" + }, + { + "id": 252, + "origin_id": -10, + "origin_slot": 0, + "target_id": 128, + "target_slot": 0, + "type": "VIDEO" + }, + { + "id": 253, + "origin_id": 128, + "origin_slot": 0, + "target_id": 126, + "target_slot": 1, + "type": "IMAGE" + }, + { + "id": 254, + "origin_id": -10, + "origin_slot": 1, + "target_id": 125, + "target_slot": 1, + "type": "STRING" + }, + { + "id": 255, + "origin_id": -10, + "origin_slot": 2, + "target_id": 126, + "target_slot": 3, + "type": "BOUNDING_BOX" + }, + { + "id": 256, + "origin_id": -10, + "origin_slot": 3, + "target_id": 126, + "target_slot": 4, + "type": "STRING" + }, + { + "id": 257, + "origin_id": -10, + "origin_slot": 4, + "target_id": 126, + "target_slot": 5, + "type": "STRING" + }, + { + "id": 259, + "origin_id": 128, + "origin_slot": 1, + "target_id": -20, + "target_slot": 2, + "type": "AUDIO" + }, + { + "id": 260, + "origin_id": 128, + "origin_slot": 2, + "target_id": -20, + "target_slot": 3, + "type": "FLOAT" + }, + { + "id": 261, + "origin_id": -10, + "origin_slot": 5, + "target_id": 126, + "target_slot": 6, + "type": "FLOAT" + }, + { + "id": 262, + "origin_id": -10, + "origin_slot": 6, + "target_id": 126, + "target_slot": 7, + "type": "INT" + }, + { + "id": 263, + "origin_id": -10, + "origin_slot": 7, + "target_id": 126, + "target_slot": 8, + "type": "BOOLEAN" + }, + { + "id": 273, + "origin_id": -10, + "origin_slot": 8, + "target_id": 127, + "target_slot": 0, + "type": "COMBO" + } + ], + "extra": {}, + "category": "Video Tools", + "description": "Segments video into temporally consistent masks using Meta SAM3 from text or interactive prompts." + } + ] + }, + "extra": {} +} diff --git a/blueprints/Video Stitch.json b/blueprints/Video Stitch.json index 020896d78..2ac78b328 100644 --- a/blueprints/Video Stitch.json +++ b/blueprints/Video Stitch.json @@ -1,21 +1,21 @@ { "revision": 0, - "last_node_id": 84, + "last_node_id": 85, "last_link_id": 0, "nodes": [ { - "id": 84, - "type": "8e8aa94a-647e-436d-8440-8ee4691864de", + "id": 85, + "type": "637913e7-0206-46ba-8ded-70ae3a7c2e19", "pos": [ - -6100, - 2620 + -880, + -2260 ], "size": [ 290, 160 ], "flags": {}, - "order": 0, + "order": 2, "mode": 0, "inputs": [ { @@ -76,31 +76,26 @@ "properties": { "proxyWidgets": [ [ - "-1", + "79", "direction" ], [ - "-1", + "79", "match_image_size" ], [ - "-1", + "79", "spacing_width" ], [ - "-1", + "79", "spacing_color" ] ], "cnr_id": "comfy-core", "ver": "0.13.0" }, - "widgets_values": [ - "right", - true, - 0, - "white" - ], + "widgets_values": [], "title": "Video Stitch" } ], @@ -109,12 +104,12 @@ "definitions": { "subgraphs": [ { - "id": "8e8aa94a-647e-436d-8440-8ee4691864de", + "id": "637913e7-0206-46ba-8ded-70ae3a7c2e19", "version": 1, "state": { "lastGroupId": 1, - "lastNodeId": 84, - "lastLinkId": 262, + "lastNodeId": 97, + "lastLinkId": 282, "lastRerouteId": 0 }, "revision": 0, @@ -123,8 +118,8 @@ "inputNode": { "id": -10, "bounding": [ - -6580, - 2649, + -6810, + 2580, 143.55859375, 160 ] @@ -132,8 +127,8 @@ "outputNode": { "id": -20, "bounding": [ - -5720, - 2659, + -4770, + 2600, 120, 60 ] @@ -149,8 +144,8 @@ "localized_name": "video", "label": "Before Video", "pos": [ - -6456.44140625, - 2669 + -6686.44140625, + 2600 ] }, { @@ -163,8 +158,8 @@ "localized_name": "video_1", "label": "After Video", "pos": [ - -6456.44140625, - 2689 + -6686.44140625, + 2620 ] }, { @@ -175,8 +170,8 @@ 259 ], "pos": [ - -6456.44140625, - 2709 + -6686.44140625, + 2640 ] }, { @@ -187,8 +182,8 @@ 260 ], "pos": [ - -6456.44140625, - 2729 + -6686.44140625, + 2660 ] }, { @@ -199,8 +194,8 @@ 261 ], "pos": [ - -6456.44140625, - 2749 + -6686.44140625, + 2680 ] }, { @@ -211,8 +206,8 @@ 262 ], "pos": [ - -6456.44140625, - 2769 + -6686.44140625, + 2700 ] } ], @@ -226,8 +221,8 @@ ], "localized_name": "VIDEO", "pos": [ - -5700, - 2679 + -4750, + 2620 ] } ], @@ -238,11 +233,11 @@ "type": "GetVideoComponents", "pos": [ -6390, - 2560 + 2600 ], "size": [ - 193.530859375, - 66 + 230, + 120 ], "flags": {}, "order": 1, @@ -278,9 +273,9 @@ } ], "properties": { + "Node name for S&R": "GetVideoComponents", "cnr_id": "comfy-core", - "ver": "0.13.0", - "Node name for S&R": "GetVideoComponents" + "ver": "0.13.0" } }, { @@ -291,8 +286,8 @@ 2420 ], "size": [ - 193.530859375, - 66 + 230, + 120 ], "flags": {}, "order": 0, @@ -332,21 +327,254 @@ } ], "properties": { + "Node name for S&R": "GetVideoComponents", "cnr_id": "comfy-core", - "ver": "0.13.0", - "Node name for S&R": "GetVideoComponents" + "ver": "0.13.0" } }, + { + "id": 90, + "type": "GetImageSize", + "pos": [ + -6390, + 3030 + ], + "size": [ + 230, + 120 + ], + "flags": {}, + "order": 4, + "mode": 0, + "inputs": [ + { + "localized_name": "image", + "name": "image", + "type": "IMAGE", + "link": 266 + } + ], + "outputs": [ + { + "localized_name": "width", + "name": "width", + "type": "INT", + "links": [ + 274 + ] + }, + { + "localized_name": "height", + "name": "height", + "type": "INT", + "links": [ + 276 + ] + }, + { + "localized_name": "batch_size", + "name": "batch_size", + "type": "INT", + "links": null + } + ], + "properties": { + "Node name for S&R": "GetImageSize" + } + }, + { + "id": 80, + "type": "CreateVideo", + "pos": [ + -5190, + 2420 + ], + "size": [ + 270, + 130 + ], + "flags": {}, + "order": 3, + "mode": 0, + "inputs": [ + { + "localized_name": "images", + "name": "images", + "type": "IMAGE", + "link": 282 + }, + { + "localized_name": "audio", + "name": "audio", + "shape": 7, + "type": "AUDIO", + "link": 251 + }, + { + "localized_name": "fps", + "name": "fps", + "type": "FLOAT", + "widget": { + "name": "fps" + }, + "link": 252 + } + ], + "outputs": [ + { + "localized_name": "VIDEO", + "name": "VIDEO", + "type": "VIDEO", + "links": [ + 255 + ] + } + ], + "properties": { + "Node name for S&R": "CreateVideo", + "cnr_id": "comfy-core", + "ver": "0.13.0" + }, + "widgets_values": [ + 30 + ] + }, + { + "id": 95, + "type": "ComfyMathExpression", + "pos": [ + -6040, + 3020 + ], + "size": [ + 400, + 200 + ], + "flags": {}, + "order": 5, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 274 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 279 + ] + } + ], + "properties": { + "Node name for S&R": "ComfyMathExpression" + }, + "widgets_values": [ + "a & ~1" + ] + }, + { + "id": 96, + "type": "ComfyMathExpression", + "pos": [ + -6040, + 3290 + ], + "size": [ + 400, + 200 + ], + "flags": {}, + "order": 6, + "mode": 0, + "inputs": [ + { + "label": "a", + "localized_name": "values.a", + "name": "values.a", + "type": "FLOAT,INT", + "link": 276 + }, + { + "label": "b", + "localized_name": "values.b", + "name": "values.b", + "shape": 7, + "type": "FLOAT,INT", + "link": null + }, + { + "localized_name": "expression", + "name": "expression", + "type": "STRING", + "widget": { + "name": "expression" + }, + "link": null + } + ], + "outputs": [ + { + "localized_name": "FLOAT", + "name": "FLOAT", + "type": "FLOAT", + "links": null + }, + { + "localized_name": "INT", + "name": "INT", + "type": "INT", + "links": [ + 280 + ] + } + ], + "properties": { + "Node name for S&R": "ComfyMathExpression" + }, + "widgets_values": [ + "a & ~1" + ] + }, { "id": 79, "type": "ImageStitch", "pos": [ -6390, - 2700 + 2780 ], "size": [ 270, - 150 + 160 ], "flags": {}, "order": 2, @@ -408,14 +636,15 @@ "name": "IMAGE", "type": "IMAGE", "links": [ - 250 + 266, + 281 ] } ], "properties": { + "Node name for S&R": "ImageStitch", "cnr_id": "comfy-core", - "ver": "0.13.0", - "Node name for S&R": "ImageStitch" + "ver": "0.13.0" }, "widgets_values": [ "right", @@ -425,60 +654,91 @@ ] }, { - "id": 80, - "type": "CreateVideo", + "id": 97, + "type": "ResizeImageMaskNode", "pos": [ - -6040, - 2610 + -5560, + 2790 ], "size": [ 270, - 78 + 160 ], "flags": {}, - "order": 3, + "order": 7, "mode": 0, "inputs": [ { - "localized_name": "images", - "name": "images", - "type": "IMAGE", - "link": 250 + "localized_name": "input", + "name": "input", + "type": "IMAGE,MASK", + "link": 281 }, { - "localized_name": "audio", - "name": "audio", - "shape": 7, - "type": "AUDIO", - "link": 251 - }, - { - "localized_name": "fps", - "name": "fps", - "type": "FLOAT", + "localized_name": "resize_type", + "name": "resize_type", + "type": "COMFY_DYNAMICCOMBO_V3", "widget": { - "name": "fps" + "name": "resize_type" }, - "link": 252 + "link": null + }, + { + "localized_name": "width", + "name": "resize_type.width", + "type": "INT", + "widget": { + "name": "resize_type.width" + }, + "link": 279 + }, + { + "localized_name": "height", + "name": "resize_type.height", + "type": "INT", + "widget": { + "name": "resize_type.height" + }, + "link": 280 + }, + { + "localized_name": "crop", + "name": "resize_type.crop", + "type": "COMBO", + "widget": { + "name": "resize_type.crop" + }, + "link": null + }, + { + "localized_name": "scale_method", + "name": "scale_method", + "type": "COMBO", + "widget": { + "name": "scale_method" + }, + "link": null } ], "outputs": [ { - "localized_name": "VIDEO", - "name": "VIDEO", - "type": "VIDEO", + "localized_name": "resized", + "name": "resized", + "type": "*", "links": [ - 255 + 282 ] } ], "properties": { - "cnr_id": "comfy-core", - "ver": "0.13.0", - "Node name for S&R": "CreateVideo" + "Node name for S&R": "ResizeImageMaskNode" }, "widgets_values": [ - 30 + "scale dimensions", + 512, + 512, + "center", + "area" ] } ], @@ -500,14 +760,6 @@ "target_slot": 1, "type": "IMAGE" }, - { - "id": 250, - "origin_id": 79, - "origin_slot": 0, - "target_id": 80, - "target_slot": 0, - "type": "IMAGE" - }, { "id": 251, "origin_id": 77, @@ -579,13 +831,71 @@ "target_id": 79, "target_slot": 5, "type": "COMBO" + }, + { + "id": 266, + "origin_id": 79, + "origin_slot": 0, + "target_id": 90, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 274, + "origin_id": 90, + "origin_slot": 0, + "target_id": 95, + "target_slot": 0, + "type": "INT" + }, + { + "id": 276, + "origin_id": 90, + "origin_slot": 1, + "target_id": 96, + "target_slot": 0, + "type": "INT" + }, + { + "id": 279, + "origin_id": 95, + "origin_slot": 1, + "target_id": 97, + "target_slot": 2, + "type": "INT" + }, + { + "id": 280, + "origin_id": 96, + "origin_slot": 1, + "target_id": 97, + "target_slot": 3, + "type": "INT" + }, + { + "id": 281, + "origin_id": 79, + "origin_slot": 0, + "target_id": 97, + "target_slot": 0, + "type": "IMAGE" + }, + { + "id": 282, + "origin_id": 97, + "origin_slot": 0, + "target_id": 80, + "target_slot": 0, + "type": "IMAGE" } ], "extra": { "workflowRendererVersion": "LG" }, - "category": "Video Tools/Stitch videos" + "category": "Video Tools/Stitch videos", + "description": "Stitches multiple video clips into a single sequential video file." } ] - } -} + }, + "extra": {} +} \ No newline at end of file diff --git a/blueprints/Video Upscale(GAN x4).json b/blueprints/Video Upscale(GAN x4).json index b61dc88d7..73476e36b 100644 --- a/blueprints/Video Upscale(GAN x4).json +++ b/blueprints/Video Upscale(GAN x4).json @@ -412,9 +412,10 @@ "extra": { "workflowRendererVersion": "LG" }, - "category": "Video generation and editing/Enhance video" + "category": "Video generation and editing/Enhance video", + "description": "Upscales video to 4× resolution using a GAN-based upscaling model." } ] }, "extra": {} -} +} \ No newline at end of file diff --git a/comfy/background_removal/birefnet.json b/comfy/background_removal/birefnet.json new file mode 100644 index 000000000..f0960af39 --- /dev/null +++ b/comfy/background_removal/birefnet.json @@ -0,0 +1,7 @@ +{ + "model_type": "birefnet", + "image_std": [1.0, 1.0, 1.0], + "image_mean": [0.0, 0.0, 0.0], + "image_size": 1024, + "resize_to_original": true +} diff --git a/comfy/background_removal/birefnet.py b/comfy/background_removal/birefnet.py new file mode 100644 index 000000000..df54b2b90 --- /dev/null +++ b/comfy/background_removal/birefnet.py @@ -0,0 +1,689 @@ +import torch +import comfy.ops +import numpy as np +import torch.nn as nn +from functools import partial +import torch.nn.functional as F +from torchvision.ops import deform_conv2d +from comfy.ldm.modules.attention import optimized_attention_for_device + +CXT = [3072, 1536, 768, 384][1:][::-1][-3:] + +class Attention(nn.Module): + def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, device=None, dtype=None, operations=None): + super().__init__() + + self.dim = dim + self.num_heads = num_heads + head_dim = dim // num_heads + self.scale = qk_scale or head_dim ** -0.5 + + self.q = operations.Linear(dim, dim, bias=qkv_bias, device=device, dtype=dtype) + self.kv = operations.Linear(dim, dim * 2, bias=qkv_bias, device=device, dtype=dtype) + self.proj = operations.Linear(dim, dim, device=device, dtype=dtype) + + def forward(self, x): + B, N, C = x.shape + optimized_attention = optimized_attention_for_device(x.device, mask=False, small_input=True) + q = self.q(x).reshape(B, N, self.num_heads, C // self.num_heads).permute(0, 2, 1, 3) + kv = self.kv(x).reshape(B, -1, 2, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4) + k, v = kv[0], kv[1] + + x = optimized_attention( + q, k, v, heads=self.num_heads, skip_output_reshape=True, skip_reshape=True + ).transpose(1, 2).reshape(B, N, C) + x = self.proj(x) + + return x + +class Mlp(nn.Module): + def __init__(self, in_features, hidden_features=None, out_features=None, device=None, dtype=None, operations=None): + super().__init__() + out_features = out_features or in_features + hidden_features = hidden_features or in_features + self.fc1 = operations.Linear(in_features, hidden_features, device=device, dtype=dtype) + self.act = nn.GELU() + self.fc2 = operations.Linear(hidden_features, out_features, device=device, dtype=dtype) + + def forward(self, x): + x = self.fc1(x) + x = self.act(x) + x = self.fc2(x) + return x + + +def window_partition(x, window_size): + B, H, W, C = x.shape + x = x.view(B, H // window_size, window_size, W // window_size, window_size, C) + windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C) + return windows + + +def window_reverse(windows, window_size, H, W): + B = int(windows.shape[0] / (H * W / window_size / window_size)) + x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1) + x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1) + return x + + +class WindowAttention(nn.Module): + def __init__(self, dim, window_size, num_heads, qkv_bias=True, qk_scale=None, device=None, dtype=None, operations=None): + + super().__init__() + self.dim = dim + self.window_size = window_size # Wh, Ww + self.num_heads = num_heads + head_dim = dim // num_heads + self.scale = qk_scale or head_dim ** -0.5 + + self.relative_position_bias_table = nn.Parameter( + torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads, device=device, dtype=dtype)) + + coords_h = torch.arange(self.window_size[0]) + coords_w = torch.arange(self.window_size[1]) + coords = torch.stack(torch.meshgrid([coords_h, coords_w], indexing='ij')) # 2, Wh, Ww + coords_flatten = torch.flatten(coords, 1) # 2, Wh*Ww + relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :] # 2, Wh*Ww, Wh*Ww + relative_coords = relative_coords.permute(1, 2, 0).contiguous() # Wh*Ww, Wh*Ww, 2 + relative_coords[:, :, 0] += self.window_size[0] - 1 + relative_coords[:, :, 1] += self.window_size[1] - 1 + relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1 + relative_position_index = relative_coords.sum(-1) # Wh*Ww, Wh*Ww + self.register_buffer("relative_position_index", relative_position_index) + + self.qkv = operations.Linear(dim, dim * 3, bias=qkv_bias, device=device, dtype=dtype) + self.proj = operations.Linear(dim, dim, device=device, dtype=dtype) + self.softmax = nn.Softmax(dim=-1) + + def forward(self, x, mask=None): + B_, N, C = x.shape + qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4) + q, k, v = qkv[0], qkv[1], qkv[2] + + q = q * self.scale + attn = (q @ k.transpose(-2, -1)) + + relative_position_bias = self.relative_position_bias_table[self.relative_position_index.long().view(-1)].view( + self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1) # Wh*Ww,Wh*Ww,nH + relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous() # nH, Wh*Ww, Wh*Ww + attn = attn + relative_position_bias.unsqueeze(0) + + if mask is not None: + nW = mask.shape[0] + attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0) + attn = attn.view(-1, self.num_heads, N, N) + attn = self.softmax(attn) + else: + attn = self.softmax(attn) + + x = (attn @ v).transpose(1, 2).reshape(B_, N, C) + x = self.proj(x) + return x + + +class SwinTransformerBlock(nn.Module): + def __init__(self, dim, num_heads, window_size=7, shift_size=0, + mlp_ratio=4., qkv_bias=True, qk_scale=None, + norm_layer=nn.LayerNorm, device=None, dtype=None, operations=None): + super().__init__() + self.dim = dim + self.num_heads = num_heads + self.window_size = window_size + self.shift_size = shift_size + self.mlp_ratio = mlp_ratio + + self.norm1 = norm_layer(dim, device=device, dtype=dtype) + self.attn = WindowAttention( + dim, window_size=(self.window_size, self.window_size), num_heads=num_heads, + qkv_bias=qkv_bias, qk_scale=qk_scale, device=device, dtype=dtype, operations=operations) + + self.norm2 = norm_layer(dim, device=device, dtype=dtype) + mlp_hidden_dim = int(dim * mlp_ratio) + self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, device=device, dtype=dtype, operations=operations) + + self.H = None + self.W = None + + def forward(self, x, mask_matrix): + B, L, C = x.shape + H, W = self.H, self.W + + shortcut = x + x = self.norm1(x) + x = x.view(B, H, W, C) + + pad_l = pad_t = 0 + pad_r = (self.window_size - W % self.window_size) % self.window_size + pad_b = (self.window_size - H % self.window_size) % self.window_size + x = F.pad(x, (0, 0, pad_l, pad_r, pad_t, pad_b)) + _, Hp, Wp, _ = x.shape + + if self.shift_size > 0: + shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2)) + attn_mask = mask_matrix + else: + shifted_x = x + attn_mask = None + + x_windows = window_partition(shifted_x, self.window_size) + x_windows = x_windows.view(-1, self.window_size * self.window_size, C) + + attn_windows = self.attn(x_windows, mask=attn_mask) + + attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C) + shifted_x = window_reverse(attn_windows, self.window_size, Hp, Wp) # B H' W' C + + if self.shift_size > 0: + x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2)) + else: + x = shifted_x + + if pad_r > 0 or pad_b > 0: + x = x[:, :H, :W, :].contiguous() + + x = x.view(B, H * W, C) + + x = shortcut + x + x = x + self.mlp(self.norm2(x)) + + return x + + +class PatchMerging(nn.Module): + def __init__(self, dim, device=None, dtype=None, operations=None): + super().__init__() + self.dim = dim + self.reduction = operations.Linear(4 * dim, 2 * dim, bias=False, device=device, dtype=dtype) + self.norm = operations.LayerNorm(4 * dim, device=device, dtype=dtype) + + def forward(self, x, H, W): + B, L, C = x.shape + x = x.view(B, H, W, C) + + # padding + pad_input = (H % 2 == 1) or (W % 2 == 1) + if pad_input: + x = F.pad(x, (0, 0, 0, W % 2, 0, H % 2)) + + x0 = x[:, 0::2, 0::2, :] # B H/2 W/2 C + x1 = x[:, 1::2, 0::2, :] # B H/2 W/2 C + x2 = x[:, 0::2, 1::2, :] # B H/2 W/2 C + x3 = x[:, 1::2, 1::2, :] # B H/2 W/2 C + x = torch.cat([x0, x1, x2, x3], -1) # B H/2 W/2 4*C + x = x.view(B, -1, 4 * C) # B H/2*W/2 4*C + + x = self.norm(x) + x = self.reduction(x) + + return x + + +class BasicLayer(nn.Module): + def __init__(self, + dim, + depth, + num_heads, + window_size=7, + mlp_ratio=4., + qkv_bias=True, + qk_scale=None, + norm_layer=nn.LayerNorm, + downsample=None, + device=None, dtype=None, operations=None): + super().__init__() + self.window_size = window_size + self.shift_size = window_size // 2 + self.depth = depth + + # build blocks + self.blocks = nn.ModuleList([ + SwinTransformerBlock( + dim=dim, + num_heads=num_heads, + window_size=window_size, + shift_size=0 if (i % 2 == 0) else window_size // 2, + mlp_ratio=mlp_ratio, + qkv_bias=qkv_bias, + qk_scale=qk_scale, + norm_layer=norm_layer, + device=device, dtype=dtype, operations=operations) + for i in range(depth)]) + + # patch merging layer + if downsample is not None: + self.downsample = downsample(dim=dim, device=device, dtype=dtype, operations=operations) + else: + self.downsample = None + + def forward(self, x, H, W): + Hp = int(np.ceil(H / self.window_size)) * self.window_size + Wp = int(np.ceil(W / self.window_size)) * self.window_size + img_mask = torch.zeros((1, Hp, Wp, 1), device=x.device) # 1 Hp Wp 1 + h_slices = (slice(0, -self.window_size), + slice(-self.window_size, -self.shift_size), + slice(-self.shift_size, None)) + w_slices = (slice(0, -self.window_size), + slice(-self.window_size, -self.shift_size), + slice(-self.shift_size, None)) + cnt = 0 + for h in h_slices: + for w in w_slices: + img_mask[:, h, w, :] = cnt + cnt += 1 + + mask_windows = window_partition(img_mask, self.window_size) + mask_windows = mask_windows.view(-1, self.window_size * self.window_size) + attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2) + attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0)) + + for blk in self.blocks: + blk.H, blk.W = H, W + x = blk(x, attn_mask) + if self.downsample is not None: + x_down = self.downsample(x, H, W) + Wh, Ww = (H + 1) // 2, (W + 1) // 2 + return x, H, W, x_down, Wh, Ww + else: + return x, H, W, x, H, W + + +class PatchEmbed(nn.Module): + def __init__(self, patch_size=4, in_channels=3, embed_dim=96, norm_layer=None, device=None, dtype=None, operations=None): + super().__init__() + patch_size = (patch_size, patch_size) + self.patch_size = patch_size + + self.in_channels = in_channels + self.embed_dim = embed_dim + + self.proj = operations.Conv2d(in_channels, embed_dim, kernel_size=patch_size, stride=patch_size, device=device, dtype=dtype) + if norm_layer is not None: + self.norm = norm_layer(embed_dim, device=device, dtype=dtype) + else: + self.norm = None + + def forward(self, x): + _, _, H, W = x.size() + if W % self.patch_size[1] != 0: + x = F.pad(x, (0, self.patch_size[1] - W % self.patch_size[1])) + if H % self.patch_size[0] != 0: + x = F.pad(x, (0, 0, 0, self.patch_size[0] - H % self.patch_size[0])) + + x = self.proj(x) # B C Wh Ww + if self.norm is not None: + Wh, Ww = x.size(2), x.size(3) + x = x.flatten(2).transpose(1, 2) + x = self.norm(x) + x = x.transpose(1, 2).view(-1, self.embed_dim, Wh, Ww) + + return x + + +class SwinTransformer(nn.Module): + def __init__(self, + pretrain_img_size=224, + patch_size=4, + in_channels=3, + embed_dim=96, + depths=[2, 2, 6, 2], + num_heads=[3, 6, 12, 24], + window_size=7, + mlp_ratio=4., + qkv_bias=True, + qk_scale=None, + patch_norm=True, + out_indices=(0, 1, 2, 3), + frozen_stages=-1, + device=None, dtype=None, operations=None): + super().__init__() + + norm_layer = partial(operations.LayerNorm, device=device, dtype=dtype) + self.pretrain_img_size = pretrain_img_size + self.num_layers = len(depths) + self.embed_dim = embed_dim + self.patch_norm = patch_norm + self.out_indices = out_indices + self.frozen_stages = frozen_stages + + self.patch_embed = PatchEmbed( + patch_size=patch_size, in_channels=in_channels, embed_dim=embed_dim, + device=device, dtype=dtype, operations=operations, + norm_layer=norm_layer if self.patch_norm else None) + + self.layers = nn.ModuleList() + for i_layer in range(self.num_layers): + layer = BasicLayer( + dim=int(embed_dim * 2 ** i_layer), + depth=depths[i_layer], + num_heads=num_heads[i_layer], + window_size=window_size, + mlp_ratio=mlp_ratio, + qkv_bias=qkv_bias, + qk_scale=qk_scale, + norm_layer=norm_layer, + downsample=PatchMerging if (i_layer < self.num_layers - 1) else None, + device=device, dtype=dtype, operations=operations) + self.layers.append(layer) + + num_features = [int(embed_dim * 2 ** i) for i in range(self.num_layers)] + self.num_features = num_features + + for i_layer in out_indices: + layer = norm_layer(num_features[i_layer]) + layer_name = f'norm{i_layer}' + self.add_module(layer_name, layer) + + + def forward(self, x): + x = self.patch_embed(x) + + Wh, Ww = x.size(2), x.size(3) + + outs = [] + x = x.flatten(2).transpose(1, 2) + for i in range(self.num_layers): + layer = self.layers[i] + x_out, H, W, x, Wh, Ww = layer(x, Wh, Ww) + + if i in self.out_indices: + norm_layer = getattr(self, f'norm{i}') + x_out = norm_layer(x_out) + + out = x_out.view(-1, H, W, self.num_features[i]).permute(0, 3, 1, 2).contiguous() + outs.append(out) + + return tuple(outs) + +class DeformableConv2d(nn.Module): + def __init__(self, + in_channels, + out_channels, + kernel_size=3, + stride=1, + padding=1, + bias=False, device=None, dtype=None, operations=None): + + super(DeformableConv2d, self).__init__() + + kernel_size = kernel_size if type(kernel_size) is tuple else (kernel_size, kernel_size) + self.stride = stride if type(stride) is tuple else (stride, stride) + self.padding = padding + + self.offset_conv = operations.Conv2d(in_channels, + 2 * kernel_size[0] * kernel_size[1], + kernel_size=kernel_size, + stride=stride, + padding=self.padding, + bias=True, device=device, dtype=dtype) + + self.modulator_conv = operations.Conv2d(in_channels, + 1 * kernel_size[0] * kernel_size[1], + kernel_size=kernel_size, + stride=stride, + padding=self.padding, + bias=True, device=device, dtype=dtype) + + self.regular_conv = operations.Conv2d(in_channels, + out_channels=out_channels, + kernel_size=kernel_size, + stride=stride, + padding=self.padding, + bias=bias, device=device, dtype=dtype) + + def forward(self, x): + offset = self.offset_conv(x) + modulator = 2. * torch.sigmoid(self.modulator_conv(x)) + weight, bias, offload_info = comfy.ops.cast_bias_weight(self.regular_conv, x, offloadable=True) + + x = deform_conv2d( + input=x, + offset=offset, + weight=weight, + bias=None, + padding=self.padding, + mask=modulator, + stride=self.stride, + ) + comfy.ops.uncast_bias_weight(self.regular_conv, weight, bias, offload_info) + return x + +class BasicDecBlk(nn.Module): + def __init__(self, in_channels=64, out_channels=64, inter_channels=64, device=None, dtype=None, operations=None): + super(BasicDecBlk, self).__init__() + inter_channels = 64 + self.conv_in = operations.Conv2d(in_channels, inter_channels, 3, 1, padding=1, device=device, dtype=dtype) + self.relu_in = nn.ReLU(inplace=True) + self.dec_att = ASPPDeformable(in_channels=inter_channels, device=device, dtype=dtype, operations=operations) + self.conv_out = operations.Conv2d(inter_channels, out_channels, 3, 1, padding=1, device=device, dtype=dtype) + self.bn_in = operations.BatchNorm2d(inter_channels, device=device, dtype=dtype) + self.bn_out = operations.BatchNorm2d(out_channels, device=device, dtype=dtype) + + def forward(self, x): + x = self.conv_in(x) + x = self.bn_in(x) + x = self.relu_in(x) + x = self.dec_att(x) + x = self.conv_out(x) + x = self.bn_out(x) + return x + + +class BasicLatBlk(nn.Module): + def __init__(self, in_channels=64, out_channels=64, device=None, dtype=None, operations=None): + super(BasicLatBlk, self).__init__() + self.conv = operations.Conv2d(in_channels, out_channels, 1, 1, 0, device=device, dtype=dtype) + + def forward(self, x): + x = self.conv(x) + return x + + +class _ASPPModuleDeformable(nn.Module): + def __init__(self, in_channels, planes, kernel_size, padding, device, dtype, operations): + super(_ASPPModuleDeformable, self).__init__() + self.atrous_conv = DeformableConv2d(in_channels, planes, kernel_size=kernel_size, + stride=1, padding=padding, bias=False, device=device, dtype=dtype, operations=operations) + self.bn = operations.BatchNorm2d(planes, device=device, dtype=dtype) + self.relu = nn.ReLU(inplace=True) + + def forward(self, x): + x = self.atrous_conv(x) + x = self.bn(x) + + return self.relu(x) + + +class ASPPDeformable(nn.Module): + def __init__(self, in_channels, out_channels=None, parallel_block_sizes=[1, 3, 7], device=None, dtype=None, operations=None): + super(ASPPDeformable, self).__init__() + self.down_scale = 1 + if out_channels is None: + out_channels = in_channels + self.in_channelster = 256 // self.down_scale + + self.aspp1 = _ASPPModuleDeformable(in_channels, self.in_channelster, 1, padding=0, device=device, dtype=dtype, operations=operations) + self.aspp_deforms = nn.ModuleList([ + _ASPPModuleDeformable(in_channels, self.in_channelster, conv_size, padding=int(conv_size//2), device=device, dtype=dtype, operations=operations) + for conv_size in parallel_block_sizes + ]) + + self.global_avg_pool = nn.Sequential(nn.AdaptiveAvgPool2d((1, 1)), + operations.Conv2d(in_channels, self.in_channelster, 1, stride=1, bias=False, device=device, dtype=dtype), + operations.BatchNorm2d(self.in_channelster, device=device, dtype=dtype), + nn.ReLU(inplace=True)) + self.conv1 = operations.Conv2d(self.in_channelster * (2 + len(self.aspp_deforms)), out_channels, 1, bias=False, device=device, dtype=dtype) + self.bn1 = operations.BatchNorm2d(out_channels, device=device, dtype=dtype) + self.relu = nn.ReLU(inplace=True) + + def forward(self, x): + x1 = self.aspp1(x) + x_aspp_deforms = [aspp_deform(x) for aspp_deform in self.aspp_deforms] + x5 = self.global_avg_pool(x) + x5 = F.interpolate(x5, size=x1.size()[2:], mode='bilinear', align_corners=True) + x = torch.cat((x1, *x_aspp_deforms, x5), dim=1) + + x = self.conv1(x) + x = self.bn1(x) + x = self.relu(x) + + return x + +class BiRefNet(nn.Module): + def __init__(self, config=None, dtype=None, device=None, operations=None): + super(BiRefNet, self).__init__() + self.bb = SwinTransformer(embed_dim=192, depths=[2, 2, 18, 2], num_heads=[6, 12, 24, 48], window_size=12, device=device, dtype=dtype, operations=operations) + + channels = [1536, 768, 384, 192] + channels = [c * 2 for c in channels] + self.cxt = channels[1:][::-1][-3:] + self.squeeze_module = nn.Sequential(*[ + BasicDecBlk(channels[0]+sum(self.cxt), channels[0], device=device, dtype=dtype, operations=operations) + for _ in range(1) + ]) + + self.decoder = Decoder(channels, device=device, dtype=dtype, operations=operations) + + def forward_enc(self, x): + x1, x2, x3, x4 = self.bb(x) + B, C, H, W = x.shape + x1_, x2_, x3_, x4_ = self.bb(F.interpolate(x, size=(H//2, W//2), mode='bilinear', align_corners=True)) + x1 = torch.cat([x1, F.interpolate(x1_, size=x1.shape[2:], mode='bilinear', align_corners=True)], dim=1) + x2 = torch.cat([x2, F.interpolate(x2_, size=x2.shape[2:], mode='bilinear', align_corners=True)], dim=1) + x3 = torch.cat([x3, F.interpolate(x3_, size=x3.shape[2:], mode='bilinear', align_corners=True)], dim=1) + x4 = torch.cat([x4, F.interpolate(x4_, size=x4.shape[2:], mode='bilinear', align_corners=True)], dim=1) + x4 = torch.cat( + ( + *[ + F.interpolate(x1, size=x4.shape[2:], mode='bilinear', align_corners=True), + F.interpolate(x2, size=x4.shape[2:], mode='bilinear', align_corners=True), + F.interpolate(x3, size=x4.shape[2:], mode='bilinear', align_corners=True), + ][-len(CXT):], + x4 + ), + dim=1 + ) + return (x1, x2, x3, x4) + + def forward_ori(self, x): + (x1, x2, x3, x4) = self.forward_enc(x) + x4 = self.squeeze_module(x4) + features = [x, x1, x2, x3, x4] + scaled_preds = self.decoder(features) + return scaled_preds + + def forward(self, pixel_values, intermediate_output=None): + scaled_preds = self.forward_ori(pixel_values) + return scaled_preds + + +class Decoder(nn.Module): + def __init__(self, channels, device, dtype, operations): + super(Decoder, self).__init__() + # factory kwargs + fk = {"device":device, "dtype":dtype, "operations":operations} + DecoderBlock = partial(BasicDecBlk, **fk) + LateralBlock = partial(BasicLatBlk, **fk) + DBlock = partial(SimpleConvs, **fk) + + self.split = True + N_dec_ipt = 64 + ic = 64 + ipt_cha_opt = 1 + self.ipt_blk5 = DBlock(2**10*3 if self.split else 3, [N_dec_ipt, channels[0]//8][ipt_cha_opt], inter_channels=ic) + self.ipt_blk4 = DBlock(2**8*3 if self.split else 3, [N_dec_ipt, channels[0]//8][ipt_cha_opt], inter_channels=ic) + self.ipt_blk3 = DBlock(2**6*3 if self.split else 3, [N_dec_ipt, channels[1]//8][ipt_cha_opt], inter_channels=ic) + self.ipt_blk2 = DBlock(2**4*3 if self.split else 3, [N_dec_ipt, channels[2]//8][ipt_cha_opt], inter_channels=ic) + self.ipt_blk1 = DBlock(2**0*3 if self.split else 3, [N_dec_ipt, channels[3]//8][ipt_cha_opt], inter_channels=ic) + + self.decoder_block4 = DecoderBlock(channels[0]+([N_dec_ipt, channels[0]//8][ipt_cha_opt]), channels[1]) + self.decoder_block3 = DecoderBlock(channels[1]+([N_dec_ipt, channels[0]//8][ipt_cha_opt]), channels[2]) + self.decoder_block2 = DecoderBlock(channels[2]+([N_dec_ipt, channels[1]//8][ipt_cha_opt]), channels[3]) + self.decoder_block1 = DecoderBlock(channels[3]+([N_dec_ipt, channels[2]//8][ipt_cha_opt]), channels[3]//2) + + fk = {"device":device, "dtype":dtype} + + self.conv_out1 = nn.Sequential(operations.Conv2d(channels[3]//2+([N_dec_ipt, channels[3]//8][ipt_cha_opt]), 1, 1, 1, 0, **fk)) + + self.lateral_block4 = LateralBlock(channels[1], channels[1]) + self.lateral_block3 = LateralBlock(channels[2], channels[2]) + self.lateral_block2 = LateralBlock(channels[3], channels[3]) + + self.conv_ms_spvn_4 = operations.Conv2d(channels[1], 1, 1, 1, 0, **fk) + self.conv_ms_spvn_3 = operations.Conv2d(channels[2], 1, 1, 1, 0, **fk) + self.conv_ms_spvn_2 = operations.Conv2d(channels[3], 1, 1, 1, 0, **fk) + + _N = 16 + + self.gdt_convs_4 = nn.Sequential(operations.Conv2d(channels[0] // 2, _N, 3, 1, 1, **fk), operations.BatchNorm2d(_N, **fk), nn.ReLU(inplace=True)) + self.gdt_convs_3 = nn.Sequential(operations.Conv2d(channels[1] // 2, _N, 3, 1, 1, **fk), operations.BatchNorm2d(_N, **fk), nn.ReLU(inplace=True)) + self.gdt_convs_2 = nn.Sequential(operations.Conv2d(channels[2] // 2, _N, 3, 1, 1, **fk), operations.BatchNorm2d(_N, **fk), nn.ReLU(inplace=True)) + + [setattr(self, f"gdt_convs_pred_{i}", nn.Sequential(operations.Conv2d(_N, 1, 1, 1, 0, **fk))) for i in range(2, 5)] + [setattr(self, f"gdt_convs_attn_{i}", nn.Sequential(operations.Conv2d(_N, 1, 1, 1, 0, **fk))) for i in range(2, 5)] + + def get_patches_batch(self, x, p): + _size_h, _size_w = p.shape[2:] + patches_batch = [] + for idx in range(x.shape[0]): + columns_x = torch.split(x[idx], split_size_or_sections=_size_w, dim=-1) + patches_x = [] + for column_x in columns_x: + patches_x += [p.unsqueeze(0) for p in torch.split(column_x, split_size_or_sections=_size_h, dim=-2)] + patch_sample = torch.cat(patches_x, dim=1) + patches_batch.append(patch_sample) + return torch.cat(patches_batch, dim=0) + + def forward(self, features): + x, x1, x2, x3, x4 = features + + patches_batch = self.get_patches_batch(x, x4) if self.split else x + x4 = torch.cat((x4, self.ipt_blk5(F.interpolate(patches_batch, size=x4.shape[2:], mode='bilinear', align_corners=True))), 1) + p4 = self.decoder_block4(x4) + p4_gdt = self.gdt_convs_4(p4) + gdt_attn_4 = self.gdt_convs_attn_4(p4_gdt).sigmoid() + p4 = p4 * gdt_attn_4 + _p4 = F.interpolate(p4, size=x3.shape[2:], mode='bilinear', align_corners=True) + _p3 = _p4 + self.lateral_block4(x3) + + patches_batch = self.get_patches_batch(x, _p3) if self.split else x + _p3 = torch.cat((_p3, self.ipt_blk4(F.interpolate(patches_batch, size=x3.shape[2:], mode='bilinear', align_corners=True))), 1) + p3 = self.decoder_block3(_p3) + + p3_gdt = self.gdt_convs_3(p3) + gdt_attn_3 = self.gdt_convs_attn_3(p3_gdt).sigmoid() + p3 = p3 * gdt_attn_3 + _p3 = F.interpolate(p3, size=x2.shape[2:], mode='bilinear', align_corners=True) + _p2 = _p3 + self.lateral_block3(x2) + + patches_batch = self.get_patches_batch(x, _p2) if self.split else x + _p2 = torch.cat((_p2, self.ipt_blk3(F.interpolate(patches_batch, size=x2.shape[2:], mode='bilinear', align_corners=True))), 1) + p2 = self.decoder_block2(_p2) + + p2_gdt = self.gdt_convs_2(p2) + gdt_attn_2 = self.gdt_convs_attn_2(p2_gdt).sigmoid() + p2 = p2 * gdt_attn_2 + + _p2 = F.interpolate(p2, size=x1.shape[2:], mode='bilinear', align_corners=True) + _p1 = _p2 + self.lateral_block2(x1) + + patches_batch = self.get_patches_batch(x, _p1) if self.split else x + _p1 = torch.cat((_p1, self.ipt_blk2(F.interpolate(patches_batch, size=x1.shape[2:], mode='bilinear', align_corners=True))), 1) + _p1 = self.decoder_block1(_p1) + _p1 = F.interpolate(_p1, size=x.shape[2:], mode='bilinear', align_corners=True) + + patches_batch = self.get_patches_batch(x, _p1) if self.split else x + _p1 = torch.cat((_p1, self.ipt_blk1(F.interpolate(patches_batch, size=x.shape[2:], mode='bilinear', align_corners=True))), 1) + p1_out = self.conv_out1(_p1) + return p1_out + + +class SimpleConvs(nn.Module): + def __init__( + self, in_channels: int, out_channels: int, inter_channels=64, device=None, dtype=None, operations=None + ) -> None: + super().__init__() + self.conv1 = operations.Conv2d(in_channels, inter_channels, 3, 1, 1, device=device, dtype=dtype) + self.conv_out = operations.Conv2d(inter_channels, out_channels, 3, 1, 1, device=device, dtype=dtype) + + def forward(self, x): + return self.conv_out(self.conv1(x)) diff --git a/comfy/bg_removal_model.py b/comfy/bg_removal_model.py new file mode 100644 index 000000000..7877afd7f --- /dev/null +++ b/comfy/bg_removal_model.py @@ -0,0 +1,78 @@ +from .utils import load_torch_file +import os +import json +import torch +import logging + +import comfy.ops +import comfy.model_patcher +import comfy.model_management +import comfy.clip_model +import comfy.background_removal.birefnet + +BG_REMOVAL_MODELS = { + "birefnet": comfy.background_removal.birefnet.BiRefNet +} + +class BackgroundRemovalModel(): + def __init__(self, json_config): + with open(json_config) as f: + config = json.load(f) + + self.image_size = config.get("image_size", 1024) + self.image_mean = config.get("image_mean", [0.0, 0.0, 0.0]) + self.image_std = config.get("image_std", [1.0, 1.0, 1.0]) + self.model_type = config.get("model_type", "birefnet") + self.config = config.copy() + model_class = BG_REMOVAL_MODELS.get(self.model_type) + + self.load_device = comfy.model_management.text_encoder_device() + offload_device = comfy.model_management.text_encoder_offload_device() + self.dtype = comfy.model_management.text_encoder_dtype(self.load_device) + self.model = model_class(config, self.dtype, offload_device, comfy.ops.manual_cast) + self.model.eval() + + self.patcher = comfy.model_patcher.CoreModelPatcher(self.model, load_device=self.load_device, offload_device=offload_device) + + def load_sd(self, sd): + return self.model.load_state_dict(sd, strict=False, assign=self.patcher.is_dynamic()) + + def get_sd(self): + return self.model.state_dict() + + def encode_image(self, image): + comfy.model_management.load_model_gpu(self.patcher) + H, W = image.shape[1], image.shape[2] + pixel_values = comfy.clip_model.clip_preprocess(image.to(self.load_device), size=self.image_size, mean=self.image_mean, std=self.image_std, crop=False) + out = self.model(pixel_values=pixel_values) + out = torch.nn.functional.interpolate(out, size=(H, W), mode="bicubic", antialias=False) + + mask = out.sigmoid().to(device=comfy.model_management.intermediate_device(), dtype=comfy.model_management.intermediate_dtype()) + if mask.ndim == 3: + mask = mask.unsqueeze(0) + if mask.shape[1] != 1: + mask = mask.movedim(-1, 1) + + return mask + + +def load_background_removal_model(sd): + if "bb.layers.1.blocks.0.attn.relative_position_index" in sd: + json_config = os.path.join(os.path.join(os.path.dirname(os.path.realpath(__file__)), "background_removal"), "birefnet.json") + else: + return None + + bg_model = BackgroundRemovalModel(json_config) + m, u = bg_model.load_sd(sd) + if len(m) > 0: + logging.warning("missing background removal: {}".format(m)) + u = set(u) + keys = list(sd.keys()) + for k in keys: + if k not in u: + sd.pop(k) + return bg_model + +def load(ckpt_path): + sd = load_torch_file(ckpt_path) + return load_background_removal_model(sd) diff --git a/comfy/cli_args.py b/comfy/cli_args.py index dbaadf723..9dadb0093 100644 --- a/comfy/cli_args.py +++ b/comfy/cli_args.py @@ -90,8 +90,8 @@ parser.add_argument("--force-channels-last", action="store_true", help="Force ch parser.add_argument("--directml", type=int, nargs="?", metavar="DIRECTML_DEVICE", const=-1, help="Use torch-directml.") parser.add_argument("--oneapi-device-selector", type=str, default=None, metavar="SELECTOR_STRING", help="Sets the oneAPI device(s) this instance will use.") -parser.add_argument("--disable-ipex-optimize", action="store_true", help="Disables ipex.optimize default when loading models with Intel's Extension for Pytorch.") parser.add_argument("--supports-fp8-compute", action="store_true", help="ComfyUI will act like if the device supports fp8 compute.") +parser.add_argument("--enable-triton-backend", action="store_true", help="ComfyUI will enable the use of Triton backend in comfy-kitchen. Is disabled at launch by default.") class LatentPreviewMethod(enum.Enum): NoPreviews = "none" @@ -238,6 +238,8 @@ database_default_path = os.path.abspath( ) parser.add_argument("--database-url", type=str, default=f"sqlite:///{database_default_path}", help="Specify the database URL, e.g. for an in-memory database you can use 'sqlite:///:memory:'.") parser.add_argument("--enable-assets", action="store_true", help="Enable the assets system (API routes, database synchronization, and background scanning).") +parser.add_argument("--feature-flag", type=str, action='append', default=[], metavar="KEY[=VALUE]", help="Set a server feature flag. Use KEY=VALUE to set an explicit value, or bare KEY to set it to true. Can be specified multiple times. Boolean values (true/false) and numbers are auto-converted. Examples: --feature-flag show_signin_button=true or --feature-flag show_signin_button") +parser.add_argument("--list-feature-flags", action="store_true", help="Print the registry of known CLI-settable feature flags as JSON and exit.") if comfy.options.args_parsing: args = parser.parse_args() diff --git a/comfy/context_windows.py b/comfy/context_windows.py index cb44ee6e8..db57537a2 100644 --- a/comfy/context_windows.py +++ b/comfy/context_windows.py @@ -63,7 +63,11 @@ class IndexListContextWindow(ContextWindowABC): dim = self.dim if dim == 0 and full.shape[dim] == 1: return full - idx = tuple([slice(None)] * dim + [self.index_list]) + indices = self.index_list + anchor_idx = getattr(self, 'causal_anchor_index', None) + if anchor_idx is not None and anchor_idx >= 0: + indices = [anchor_idx] + list(indices) + idx = tuple([slice(None)] * dim + [indices]) window = full[idx] if retain_index_list: idx = tuple([slice(None)] * dim + [retain_index_list]) @@ -113,7 +117,14 @@ def slice_cond(cond_value, window: IndexListContextWindow, x_in: torch.Tensor, d # skip leading latent positions that have no corresponding conditioning (e.g. reference frames) if temporal_offset > 0: - indices = [i - temporal_offset for i in window.index_list[temporal_offset:]] + anchor_idx = getattr(window, 'causal_anchor_index', None) + if anchor_idx is not None and anchor_idx >= 0: + # anchor occupies one of the no-cond positions, so skip one fewer from window.index_list + skip_count = temporal_offset - 1 + else: + skip_count = temporal_offset + + indices = [i - temporal_offset for i in window.index_list[skip_count:]] indices = [i for i in indices if 0 <= i] else: indices = list(window.index_list) @@ -150,7 +161,8 @@ class ContextFuseMethod: ContextResults = collections.namedtuple("ContextResults", ['window_idx', 'sub_conds_out', 'sub_conds', 'window']) class IndexListContextHandler(ContextHandlerABC): def __init__(self, context_schedule: ContextSchedule, fuse_method: ContextFuseMethod, context_length: int=1, context_overlap: int=0, context_stride: int=1, - closed_loop: bool=False, dim:int=0, freenoise: bool=False, cond_retain_index_list: list[int]=[], split_conds_to_windows: bool=False): + closed_loop: bool=False, dim:int=0, freenoise: bool=False, cond_retain_index_list: list[int]=[], split_conds_to_windows: bool=False, + causal_window_fix: bool=True): self.context_schedule = context_schedule self.fuse_method = fuse_method self.context_length = context_length @@ -162,6 +174,7 @@ class IndexListContextHandler(ContextHandlerABC): self.freenoise = freenoise self.cond_retain_index_list = [int(x.strip()) for x in cond_retain_index_list.split(",")] if cond_retain_index_list else [] self.split_conds_to_windows = split_conds_to_windows + self.causal_window_fix = causal_window_fix self.callbacks = {} @@ -318,6 +331,14 @@ class IndexListContextHandler(ContextHandlerABC): # allow processing to end between context window executions for faster Cancel comfy.model_management.throw_exception_if_processing_interrupted() + # causal_window_fix: prepend a pre-window frame that will be stripped post-forward + anchor_applied = False + if self.causal_window_fix: + anchor_idx = window.index_list[0] - 1 + if 0 <= anchor_idx < x_in.size(self.dim): + window.causal_anchor_index = anchor_idx + anchor_applied = True + for callback in comfy.patcher_extension.get_all_callbacks(IndexListCallbacks.EVALUATE_CONTEXT_WINDOWS, self.callbacks): callback(self, model, x_in, conds, timestep, model_options, window_idx, window, model_options, device, first_device) @@ -332,6 +353,12 @@ class IndexListContextHandler(ContextHandlerABC): if device is not None: for i in range(len(sub_conds_out)): sub_conds_out[i] = sub_conds_out[i].to(x_in.device) + + # strip causal_window_fix anchor if applied + if anchor_applied: + for i in range(len(sub_conds_out)): + sub_conds_out[i] = sub_conds_out[i].narrow(self.dim, 1, sub_conds_out[i].shape[self.dim] - 1) + results.append(ContextResults(window_idx, sub_conds_out, sub_conds, window)) return results diff --git a/comfy/deploy_environment.py b/comfy/deploy_environment.py new file mode 100644 index 000000000..8c99a3584 --- /dev/null +++ b/comfy/deploy_environment.py @@ -0,0 +1,34 @@ +import functools +import logging +import os + +logger = logging.getLogger(__name__) + +_DEFAULT_DEPLOY_ENV = "local-git" +_ENV_FILENAME = ".comfy_environment" + +# Resolve the ComfyUI install directory (the parent of this `comfy/` package). +# We deliberately avoid `folder_paths.base_path` here because that is overridden +# by the `--base-directory` CLI arg to a user-supplied path, whereas the +# `.comfy_environment` marker is written by launchers/installers next to the +# ComfyUI install itself. +_COMFY_INSTALL_DIR = os.path.dirname(os.path.dirname(os.path.realpath(__file__))) + + +@functools.cache +def get_deploy_environment() -> str: + env_file = os.path.join(_COMFY_INSTALL_DIR, _ENV_FILENAME) + try: + with open(env_file, encoding="utf-8") as f: + # Cap the read so a malformed or maliciously crafted file (e.g. + # a single huge line with no newline) can't blow up memory. + first_line = f.readline(128).strip() + value = "".join(c for c in first_line if 32 <= ord(c) < 127) + if value: + return value + except FileNotFoundError: + pass + except Exception as e: + logger.error("Failed to read %s: %s", env_file, e) + + return _DEFAULT_DEPLOY_ENV diff --git a/comfy/hooks.py b/comfy/hooks.py index 1a76c7ba4..5458fc3d8 100644 --- a/comfy/hooks.py +++ b/comfy/hooks.py @@ -93,7 +93,7 @@ class Hook: self.hook_scope = hook_scope '''Scope of where this hook should apply in terms of the conds used in sampling run.''' self.custom_should_register = default_should_register - '''Can be overriden with a compatible function to decide if this hook should be registered without the need to override .should_register''' + '''Can be overridden with a compatible function to decide if this hook should be registered without the need to override .should_register''' @property def strength(self): diff --git a/comfy/k_diffusion/sampling.py b/comfy/k_diffusion/sampling.py index 6978eb717..11db46d94 100644 --- a/comfy/k_diffusion/sampling.py +++ b/comfy/k_diffusion/sampling.py @@ -242,6 +242,7 @@ def sample_euler_ancestral_RF(model, x, sigmas, extra_args=None, callback=None, extra_args = {} if extra_args is None else extra_args seed = extra_args.get("seed", None) noise_sampler = default_noise_sampler(x, seed=seed) if noise_sampler is None else noise_sampler + s_noise = s_noise * getattr(model.inner_model.model_patcher.get_model_object('model_sampling'), "noise_scale", 1.0) s_in = x.new_ones([x.shape[0]]) for i in trange(len(sigmas) - 1, disable=disable): denoised = model(x, sigmas[i] * s_in, **extra_args) @@ -373,6 +374,7 @@ def sample_dpm_2_ancestral_RF(model, x, sigmas, extra_args=None, callback=None, extra_args = {} if extra_args is None else extra_args seed = extra_args.get("seed", None) noise_sampler = default_noise_sampler(x, seed=seed) if noise_sampler is None else noise_sampler + s_noise = s_noise * getattr(model.inner_model.model_patcher.get_model_object('model_sampling'), "noise_scale", 1.0) s_in = x.new_ones([x.shape[0]]) for i in trange(len(sigmas) - 1, disable=disable): denoised = model(x, sigmas[i] * s_in, **extra_args) @@ -686,6 +688,7 @@ def sample_dpmpp_2s_ancestral_RF(model, x, sigmas, extra_args=None, callback=Non extra_args = {} if extra_args is None else extra_args seed = extra_args.get("seed", None) noise_sampler = default_noise_sampler(x, seed=seed) if noise_sampler is None else noise_sampler + s_noise = s_noise * getattr(model.inner_model.model_patcher.get_model_object('model_sampling'), "noise_scale", 1.0) s_in = x.new_ones([x.shape[0]]) sigma_fn = lambda lbda: (lbda.exp() + 1) ** -1 lambda_fn = lambda sigma: ((1-sigma)/sigma).log() @@ -747,6 +750,7 @@ def sample_dpmpp_sde(model, x, sigmas, extra_args=None, callback=None, disable=N sigma_fn = partial(half_log_snr_to_sigma, model_sampling=model_sampling) lambda_fn = partial(sigma_to_half_log_snr, model_sampling=model_sampling) sigmas = offset_first_sigma_for_snr(sigmas, model_sampling) + s_noise = s_noise * getattr(model_sampling, "noise_scale", 1.0) for i in trange(len(sigmas) - 1, disable=disable): denoised = model(x, sigmas[i] * s_in, **extra_args) @@ -832,6 +836,7 @@ def sample_dpmpp_2m_sde(model, x, sigmas, extra_args=None, callback=None, disabl model_sampling = model.inner_model.model_patcher.get_model_object('model_sampling') lambda_fn = partial(sigma_to_half_log_snr, model_sampling=model_sampling) sigmas = offset_first_sigma_for_snr(sigmas, model_sampling) + s_noise = s_noise * getattr(model_sampling, "noise_scale", 1.0) old_denoised = None h, h_last = None, None @@ -889,6 +894,7 @@ def sample_dpmpp_3m_sde(model, x, sigmas, extra_args=None, callback=None, disabl model_sampling = model.inner_model.model_patcher.get_model_object('model_sampling') lambda_fn = partial(sigma_to_half_log_snr, model_sampling=model_sampling) sigmas = offset_first_sigma_for_snr(sigmas, model_sampling) + s_noise = s_noise * getattr(model_sampling, "noise_scale", 1.0) denoised_1, denoised_2 = None, None h, h_1, h_2 = None, None, None @@ -1006,23 +1012,39 @@ def sample_ddpm(model, x, sigmas, extra_args=None, callback=None, disable=None, return generic_step_sampler(model, x, sigmas, extra_args, callback, disable, noise_sampler, DDPMSampler_step) @torch.no_grad() -def sample_lcm(model, x, sigmas, extra_args=None, callback=None, disable=None, noise_sampler=None): +def sample_lcm(model, x, sigmas, extra_args=None, callback=None, disable=None, noise_sampler=None, s_noise=1.0, s_noise_end=None, noise_clip_std=0.0): + + # s_noise / s_noise_end: per-step noise multiplier, linearly interpolated across steps + # noise_clip_std: clamp injected noise to +/- N stddevs (0 disables). + extra_args = {} if extra_args is None else extra_args seed = extra_args.get("seed", None) noise_sampler = default_noise_sampler(x, seed=seed) if noise_sampler is None else noise_sampler s_in = x.new_ones([x.shape[0]]) - for i in trange(len(sigmas) - 1, disable=disable): + n_steps = max(1, len(sigmas) - 1) + model_sampling = model.inner_model.model_patcher.get_model_object('model_sampling') + + s_start = float(s_noise) + s_end = s_start if s_noise_end is None else float(s_noise_end) + for i in trange(n_steps, disable=disable): denoised = model(x, sigmas[i] * s_in, **extra_args) if callback is not None: callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigmas[i], 'denoised': denoised}) x = denoised if sigmas[i + 1] > 0: - x = model.inner_model.inner_model.model_sampling.noise_scaling(sigmas[i + 1], noise_sampler(sigmas[i], sigmas[i + 1]), x) + noise = noise_sampler(sigmas[i], sigmas[i + 1]) + if noise_clip_std > 0: + clip_val = noise_clip_std * noise.std() + noise = noise.clamp(min=-clip_val, max=clip_val) + t = (i / (n_steps - 1)) if n_steps > 1 else 0.0 + s_noise_i = s_start + (s_end - s_start) * t + if s_noise_i != 1.0: + noise = noise * s_noise_i + x = model_sampling.noise_scaling(sigmas[i + 1], noise, x) return x - @torch.no_grad() def sample_heunpp2(model, x, sigmas, extra_args=None, callback=None, disable=None, s_churn=0., s_tmin=0., s_tmax=float('inf'), s_noise=1.): # From MIT licensed: https://github.com/Carzit/sd-webui-samplers-scheduler/ @@ -1249,6 +1271,7 @@ def sample_euler_ancestral_cfg_pp(model, x, sigmas, extra_args=None, callback=No model_sampling = model.inner_model.model_patcher.get_model_object("model_sampling") lambda_fn = partial(sigma_to_half_log_snr, model_sampling=model_sampling) + s_noise = s_noise * getattr(model_sampling, "noise_scale", 1.0) uncond_denoised = None @@ -1296,6 +1319,7 @@ def sample_dpmpp_2s_ancestral_cfg_pp(model, x, sigmas, extra_args=None, callback extra_args = {} if extra_args is None else extra_args seed = extra_args.get("seed", None) noise_sampler = default_noise_sampler(x, seed=seed) if noise_sampler is None else noise_sampler + s_noise = s_noise * getattr(model.inner_model.model_patcher.get_model_object('model_sampling'), "noise_scale", 1.0) temp = [0] def post_cfg_function(args): @@ -1371,6 +1395,7 @@ def res_multistep(model, x, sigmas, extra_args=None, callback=None, disable=None extra_args = {} if extra_args is None else extra_args seed = extra_args.get("seed", None) noise_sampler = default_noise_sampler(x, seed=seed) if noise_sampler is None else noise_sampler + s_noise = s_noise * getattr(model.inner_model.model_patcher.get_model_object('model_sampling'), "noise_scale", 1.0) s_in = x.new_ones([x.shape[0]]) sigma_fn = lambda t: t.neg().exp() t_fn = lambda sigma: sigma.log().neg() @@ -1504,6 +1529,7 @@ def sample_er_sde(model, x, sigmas, extra_args=None, callback=None, disable=None extra_args = {} if extra_args is None else extra_args seed = extra_args.get("seed", None) noise_sampler = default_noise_sampler(x, seed=seed) if noise_sampler is None else noise_sampler + s_noise = s_noise * getattr(model.inner_model.model_patcher.get_model_object('model_sampling'), "noise_scale", 1.0) s_in = x.new_ones([x.shape[0]]) def default_er_sde_noise_scaler(x): @@ -1574,9 +1600,10 @@ def sample_seeds_2(model, x, sigmas, extra_args=None, callback=None, disable=Non seed = extra_args.get("seed", None) noise_sampler = default_noise_sampler(x, seed=seed) if noise_sampler is None else noise_sampler s_in = x.new_ones([x.shape[0]]) - inject_noise = eta > 0 and s_noise > 0 model_sampling = model.inner_model.model_patcher.get_model_object('model_sampling') + s_noise = s_noise * getattr(model_sampling, "noise_scale", 1.0) + inject_noise = eta > 0 and s_noise > 0 sigma_fn = partial(half_log_snr_to_sigma, model_sampling=model_sampling) lambda_fn = partial(sigma_to_half_log_snr, model_sampling=model_sampling) sigmas = offset_first_sigma_for_snr(sigmas, model_sampling) @@ -1645,9 +1672,10 @@ def sample_seeds_3(model, x, sigmas, extra_args=None, callback=None, disable=Non seed = extra_args.get("seed", None) noise_sampler = default_noise_sampler(x, seed=seed) if noise_sampler is None else noise_sampler s_in = x.new_ones([x.shape[0]]) - inject_noise = eta > 0 and s_noise > 0 model_sampling = model.inner_model.model_patcher.get_model_object('model_sampling') + s_noise = s_noise * getattr(model_sampling, "noise_scale", 1.0) + inject_noise = eta > 0 and s_noise > 0 sigma_fn = partial(half_log_snr_to_sigma, model_sampling=model_sampling) lambda_fn = partial(sigma_to_half_log_snr, model_sampling=model_sampling) sigmas = offset_first_sigma_for_snr(sigmas, model_sampling) @@ -1713,6 +1741,7 @@ def sample_sa_solver(model, x, sigmas, extra_args=None, callback=None, disable=F s_in = x.new_ones([x.shape[0]]) model_sampling = model.inner_model.model_patcher.get_model_object("model_sampling") + s_noise = s_noise * getattr(model_sampling, "noise_scale", 1.0) sigmas = offset_first_sigma_for_snr(sigmas, model_sampling) lambdas = sigma_to_half_log_snr(sigmas, model_sampling=model_sampling) @@ -1810,3 +1839,119 @@ def sample_sa_solver(model, x, sigmas, extra_args=None, callback=None, disable=F def sample_sa_solver_pece(model, x, sigmas, extra_args=None, callback=None, disable=False, tau_func=None, s_noise=1.0, noise_sampler=None, predictor_order=3, corrector_order=4, simple_order_2=False): """Stochastic Adams Solver with PECE (Predict–Evaluate–Correct–Evaluate) mode (NeurIPS 2023).""" return sample_sa_solver(model, x, sigmas, extra_args=extra_args, callback=callback, disable=disable, tau_func=tau_func, s_noise=s_noise, noise_sampler=noise_sampler, predictor_order=predictor_order, corrector_order=corrector_order, use_pece=True, simple_order_2=simple_order_2) + + +@torch.no_grad() +def sample_ar_video(model, x, sigmas, extra_args=None, callback=None, disable=None, + num_frame_per_block=1): + """ + Autoregressive video sampler: block-by-block denoising with KV cache + and flow-match re-noising for Causal Forcing / Self-Forcing models. + + Requires a Causal-WAN compatible model (diffusion_model must expose + init_kv_caches / init_crossattn_caches) and 5-D latents [B,C,T,H,W]. + + All AR-loop parameters are passed via the SamplerARVideo node, not read + from the checkpoint or transformer_options. + """ + extra_args = {} if extra_args is None else extra_args + model_options = extra_args.get("model_options", {}) + transformer_options = model_options.get("transformer_options", {}) + + if x.ndim != 5: + raise ValueError( + f"ar_video sampler requires 5-D video latents [B,C,T,H,W], got {x.ndim}-D tensor with shape {x.shape}. " + "This sampler is only compatible with autoregressive video models (e.g. Causal-WAN)." + ) + + inner_model = model.inner_model.inner_model + causal_model = inner_model.diffusion_model + + if not (hasattr(causal_model, "init_kv_caches") and hasattr(causal_model, "init_crossattn_caches")): + raise TypeError( + "ar_video sampler requires a Causal-WAN compatible model whose diffusion_model " + "exposes init_kv_caches() and init_crossattn_caches(). The loaded checkpoint " + "does not support this interface — choose a different sampler." + ) + + seed = extra_args.get("seed", 0) + + bs, c, lat_t, lat_h, lat_w = x.shape + frame_seq_len = -(-lat_h // 2) * -(-lat_w // 2) # ceiling division + num_blocks = -(-lat_t // num_frame_per_block) # ceiling division + device = x.device + model_dtype = inner_model.get_dtype() + + kv_caches = causal_model.init_kv_caches(bs, lat_t * frame_seq_len, device, model_dtype) + crossattn_caches = causal_model.init_crossattn_caches(bs, device, model_dtype) + + output = torch.zeros_like(x) + s_in = x.new_ones([x.shape[0]]) + current_start_frame = 0 + + # I2V: seed KV cache with the initial image latent before the denoising loop + initial_latent = transformer_options.get("ar_config", {}).get("initial_latent", None) + if initial_latent is not None: + initial_latent = inner_model.process_latent_in(initial_latent).to(device=device, dtype=model_dtype) + n_init = initial_latent.shape[2] + output[:, :, :n_init] = initial_latent + + ar_state = {"start_frame": 0, "kv_caches": kv_caches, "crossattn_caches": crossattn_caches} + transformer_options["ar_state"] = ar_state + zero_sigma = sigmas.new_zeros([1]) + _ = model(initial_latent, zero_sigma * s_in, **extra_args) + + current_start_frame = n_init + remaining = lat_t - n_init + num_blocks = -(-remaining // num_frame_per_block) + + num_sigma_steps = len(sigmas) - 1 + total_real_steps = num_blocks * num_sigma_steps + step_count = 0 + + try: + for block_idx in trange(num_blocks, disable=disable): + bf = min(num_frame_per_block, lat_t - current_start_frame) + fs, fe = current_start_frame, current_start_frame + bf + noisy_input = x[:, :, fs:fe] + + ar_state = { + "start_frame": current_start_frame, + "kv_caches": kv_caches, + "crossattn_caches": crossattn_caches, + } + transformer_options["ar_state"] = ar_state + + for i in range(num_sigma_steps): + denoised = model(noisy_input, sigmas[i] * s_in, **extra_args) + + if callback is not None: + scaled_i = step_count * num_sigma_steps // total_real_steps + callback({"x": noisy_input, "i": scaled_i, "sigma": sigmas[i], + "sigma_hat": sigmas[i], "denoised": denoised}) + + if sigmas[i + 1] == 0: + noisy_input = denoised + else: + sigma_next = sigmas[i + 1] + torch.manual_seed(seed + block_idx * 1000 + i) + fresh_noise = torch.randn_like(denoised) + noisy_input = (1.0 - sigma_next) * denoised + sigma_next * fresh_noise + + for cache in kv_caches: + cache["end"] -= bf * frame_seq_len + + step_count += 1 + + output[:, :, fs:fe] = noisy_input + + for cache in kv_caches: + cache["end"] -= bf * frame_seq_len + zero_sigma = sigmas.new_zeros([1]) + _ = model(noisy_input, zero_sigma * s_in, **extra_args) + + current_start_frame += bf + finally: + transformer_options.pop("ar_state", None) + + return output diff --git a/comfy/latent_formats.py b/comfy/latent_formats.py index 6a57bca1c..d527eec4a 100644 --- a/comfy/latent_formats.py +++ b/comfy/latent_formats.py @@ -9,6 +9,7 @@ class LatentFormat: latent_rgb_factors_reshape = None taesd_decoder_name = None spacial_downscale_ratio = 8 + temporal_downscale_ratio = 1 def process_in(self, latent): return latent * self.scale_factor @@ -224,6 +225,7 @@ class Flux2(LatentFormat): self.latent_rgb_factors_bias = [-0.0329, -0.0718, -0.0851] self.latent_rgb_factors_reshape = lambda t: t.reshape(t.shape[0], 32, 2, 2, t.shape[-2], t.shape[-1]).permute(0, 1, 4, 2, 5, 3).reshape(t.shape[0], 32, t.shape[-2] * 2, t.shape[-1] * 2) + self.taesd_decoder_name = "taef2_decoder" def process_in(self, latent): return latent @@ -234,6 +236,7 @@ class Flux2(LatentFormat): class Mochi(LatentFormat): latent_channels = 12 latent_dimensions = 3 + temporal_downscale_ratio = 6 def __init__(self): self.scale_factor = 1.0 @@ -277,6 +280,7 @@ class LTXV(LatentFormat): latent_channels = 128 latent_dimensions = 3 spacial_downscale_ratio = 32 + temporal_downscale_ratio = 8 def __init__(self): self.latent_rgb_factors = [ @@ -420,6 +424,7 @@ class LTXAV(LTXV): class HunyuanVideo(LatentFormat): latent_channels = 16 latent_dimensions = 3 + temporal_downscale_ratio = 4 scale_factor = 0.476986 latent_rgb_factors = [ [-0.0395, -0.0331, 0.0445], @@ -446,6 +451,7 @@ class HunyuanVideo(LatentFormat): class Cosmos1CV8x8x8(LatentFormat): latent_channels = 16 latent_dimensions = 3 + temporal_downscale_ratio = 8 latent_rgb_factors = [ [ 0.1817, 0.2284, 0.2423], @@ -471,6 +477,7 @@ class Cosmos1CV8x8x8(LatentFormat): class Wan21(LatentFormat): latent_channels = 16 latent_dimensions = 3 + temporal_downscale_ratio = 4 latent_rgb_factors = [ [-0.1299, -0.1692, 0.2932], @@ -733,6 +740,7 @@ class HunyuanVideo15(LatentFormat): latent_channels = 32 latent_dimensions = 3 spacial_downscale_ratio = 16 + temporal_downscale_ratio = 4 scale_factor = 1.03682 taesd_decoder_name = "lighttaehy1_5" @@ -783,3 +791,36 @@ class ZImagePixelSpace(ChromaRadiance): No VAE encoding/decoding — the model operates directly on RGB pixels. """ pass + + +class HiDreamO1Pixel(ChromaRadiance): + """Pixel-space latent format for HiDream-O1. + No VAE — model patches/unpatches raw RGB internally with patch_size=32. + """ + pass + +class CogVideoX(LatentFormat): + """Latent format for CogVideoX-2b (THUDM/CogVideoX-2b). + + scale_factor matches the vae/config.json scaling_factor for the 2b variant. + The 5b-class checkpoints (CogVideoX-5b, CogVideoX-1.5-5B, CogVideoX-Fun-V1.5-*) + use a different value; see CogVideoX1_5 below. + """ + latent_channels = 16 + latent_dimensions = 3 + temporal_downscale_ratio = 4 + + def __init__(self): + self.scale_factor = 1.15258426 + + +class CogVideoX1_5(CogVideoX): + """Latent format for 5b-class CogVideoX checkpoints. + + Covers THUDM/CogVideoX-5b, THUDM/CogVideoX-1.5-5B, and the CogVideoX-Fun + V1.5-5b family (including VOID inpainting). All of these have + scaling_factor=0.7 in their vae/config.json. Auto-selected in + supported_models.CogVideoX_T2V based on transformer hidden dim. + """ + def __init__(self): + self.scale_factor = 0.7 diff --git a/comfy/ldm/cogvideo/__init__.py b/comfy/ldm/cogvideo/__init__.py new file mode 100644 index 000000000..e69de29bb diff --git a/comfy/ldm/cogvideo/model.py b/comfy/ldm/cogvideo/model.py new file mode 100644 index 000000000..fb475ed53 --- /dev/null +++ b/comfy/ldm/cogvideo/model.py @@ -0,0 +1,573 @@ +# CogVideoX 3D Transformer - ported to ComfyUI native ops +# Architecture reference: diffusers CogVideoXTransformer3DModel +# Style reference: comfy/ldm/wan/model.py + +import math +import torch +import torch.nn as nn +import torch.nn.functional as F + +from comfy.ldm.modules.attention import optimized_attention +import comfy.patcher_extension +import comfy.ldm.common_dit + + +def _get_1d_rotary_pos_embed(dim, pos, theta=10000.0): + """Returns (cos, sin) each with shape [seq_len, dim]. + + Frequencies are computed at dim//2 resolution then repeat_interleaved + to full dim, matching CogVideoX's interleaved (real, imag) pair format. + """ + freqs = 1.0 / (theta ** (torch.arange(0, dim, 2, dtype=torch.float32, device=pos.device) / dim)) + angles = torch.outer(pos.float(), freqs.float()) + cos = angles.cos().repeat_interleave(2, dim=-1).float() + sin = angles.sin().repeat_interleave(2, dim=-1).float() + return (cos, sin) + + +def apply_rotary_emb(x, freqs_cos_sin): + """Apply CogVideoX rotary embedding to query or key tensor. + + x: [B, heads, seq_len, head_dim] + freqs_cos_sin: (cos, sin) each [seq_len, head_dim//2] + + Uses interleaved pair rotation (same as diffusers CogVideoX/Flux). + head_dim is reshaped to (-1, 2) pairs, rotated, then flattened back. + """ + cos, sin = freqs_cos_sin + cos = cos[None, None, :, :].to(x.device) + sin = sin[None, None, :, :].to(x.device) + + # Interleaved pairs: [B, H, S, D] -> [B, H, S, D//2, 2] -> (real, imag) + x_real, x_imag = x.reshape(*x.shape[:-1], -1, 2).unbind(-1) + x_rotated = torch.stack([-x_imag, x_real], dim=-1).flatten(3) + + return (x.float() * cos + x_rotated.float() * sin).to(x.dtype) + + +def get_timestep_embedding(timesteps, dim, flip_sin_to_cos=True, downscale_freq_shift=0, scale=1, max_period=10000): + half = dim // 2 + freqs = torch.exp(-math.log(max_period) * torch.arange(start=0, end=half, dtype=torch.float32, device=timesteps.device) / half) + args = timesteps[:, None].float() * freqs[None] * scale + embedding = torch.cat([torch.sin(args), torch.cos(args)], dim=-1) + if flip_sin_to_cos: + embedding = torch.cat([embedding[:, half:], embedding[:, :half]], dim=-1) + if dim % 2: + embedding = torch.cat([embedding, torch.zeros_like(embedding[:, :1])], dim=-1) + return embedding + + +def get_3d_sincos_pos_embed(embed_dim, spatial_size, temporal_size, spatial_interpolation_scale=1.0, temporal_interpolation_scale=1.0, device=None): + if isinstance(spatial_size, int): + spatial_size = (spatial_size, spatial_size) + + grid_w = torch.arange(spatial_size[0], dtype=torch.float32, device=device) / spatial_interpolation_scale + grid_h = torch.arange(spatial_size[1], dtype=torch.float32, device=device) / spatial_interpolation_scale + grid_t = torch.arange(temporal_size, dtype=torch.float32, device=device) / temporal_interpolation_scale + + grid_t, grid_h, grid_w = torch.meshgrid(grid_t, grid_h, grid_w, indexing="ij") + + embed_dim_spatial = 2 * (embed_dim // 3) + embed_dim_temporal = embed_dim // 3 + + pos_embed_spatial = _get_2d_sincos_pos_embed(embed_dim_spatial, grid_h, grid_w, device=device) + pos_embed_temporal = _get_1d_sincos_pos_embed(embed_dim_temporal, grid_t[:, 0, 0], device=device) + + T, H, W = grid_t.shape + pos_embed_temporal = pos_embed_temporal.unsqueeze(1).unsqueeze(1).expand(-1, H, W, -1) + pos_embed = torch.cat([pos_embed_temporal, pos_embed_spatial], dim=-1) + + return pos_embed + + +def _get_2d_sincos_pos_embed(embed_dim, grid_h, grid_w, device=None): + T, H, W = grid_h.shape + half_dim = embed_dim // 2 + pos_h = _get_1d_sincos_pos_embed(half_dim, grid_h.reshape(-1), device=device).reshape(T, H, W, half_dim) + pos_w = _get_1d_sincos_pos_embed(half_dim, grid_w.reshape(-1), device=device).reshape(T, H, W, half_dim) + return torch.cat([pos_h, pos_w], dim=-1) + + +def _get_1d_sincos_pos_embed(embed_dim, pos, device=None): + half = embed_dim // 2 + freqs = torch.exp(-math.log(10000.0) * torch.arange(start=0, end=half, dtype=torch.float32, device=device) / half) + args = pos.float().reshape(-1)[:, None] * freqs[None] + embedding = torch.cat([torch.cos(args), torch.sin(args)], dim=-1) + if embed_dim % 2: + embedding = torch.cat([embedding, torch.zeros_like(embedding[:, :1])], dim=-1) + return embedding + + + +class CogVideoXPatchEmbed(nn.Module): + def __init__(self, patch_size=2, patch_size_t=None, in_channels=16, dim=1920, + text_dim=4096, bias=True, sample_width=90, sample_height=60, + sample_frames=49, temporal_compression_ratio=4, + max_text_seq_length=226, spatial_interpolation_scale=1.875, + temporal_interpolation_scale=1.0, use_positional_embeddings=True, + use_learned_positional_embeddings=True, + device=None, dtype=None, operations=None): + super().__init__() + self.patch_size = patch_size + self.patch_size_t = patch_size_t + self.dim = dim + self.sample_height = sample_height + self.sample_width = sample_width + self.sample_frames = sample_frames + self.temporal_compression_ratio = temporal_compression_ratio + self.max_text_seq_length = max_text_seq_length + self.spatial_interpolation_scale = spatial_interpolation_scale + self.temporal_interpolation_scale = temporal_interpolation_scale + self.use_positional_embeddings = use_positional_embeddings + self.use_learned_positional_embeddings = use_learned_positional_embeddings + + if patch_size_t is None: + self.proj = operations.Conv2d(in_channels, dim, kernel_size=patch_size, stride=patch_size, bias=bias, device=device, dtype=dtype) + else: + self.proj = operations.Linear(in_channels * patch_size * patch_size * patch_size_t, dim, device=device, dtype=dtype) + + self.text_proj = operations.Linear(text_dim, dim, device=device, dtype=dtype) + + if use_positional_embeddings or use_learned_positional_embeddings: + persistent = use_learned_positional_embeddings + pos_embedding = self._get_positional_embeddings(sample_height, sample_width, sample_frames) + self.register_buffer("pos_embedding", pos_embedding, persistent=persistent) + + def _get_positional_embeddings(self, sample_height, sample_width, sample_frames, device=None): + post_patch_height = sample_height // self.patch_size + post_patch_width = sample_width // self.patch_size + post_time_compression_frames = (sample_frames - 1) // self.temporal_compression_ratio + 1 + if self.patch_size_t is not None: + post_time_compression_frames = post_time_compression_frames // self.patch_size_t + num_patches = post_patch_height * post_patch_width * post_time_compression_frames + + pos_embedding = get_3d_sincos_pos_embed( + self.dim, + (post_patch_width, post_patch_height), + post_time_compression_frames, + self.spatial_interpolation_scale, + self.temporal_interpolation_scale, + device=device, + ) + pos_embedding = pos_embedding.reshape(-1, self.dim) + joint_pos_embedding = pos_embedding.new_zeros( + 1, self.max_text_seq_length + num_patches, self.dim, requires_grad=False + ) + joint_pos_embedding.data[:, self.max_text_seq_length:].copy_(pos_embedding) + return joint_pos_embedding + + def forward(self, text_embeds, image_embeds): + input_dtype = text_embeds.dtype + text_embeds = self.text_proj(text_embeds.to(self.text_proj.weight.dtype)).to(input_dtype) + batch_size, num_frames, channels, height, width = image_embeds.shape + + proj_dtype = self.proj.weight.dtype + if self.patch_size_t is None: + image_embeds = image_embeds.reshape(-1, channels, height, width) + image_embeds = self.proj(image_embeds.to(proj_dtype)).to(input_dtype) + image_embeds = image_embeds.view(batch_size, num_frames, *image_embeds.shape[1:]) + image_embeds = image_embeds.flatten(3).transpose(2, 3) + image_embeds = image_embeds.flatten(1, 2) + else: + p = self.patch_size + p_t = self.patch_size_t + image_embeds = image_embeds.permute(0, 1, 3, 4, 2) + image_embeds = image_embeds.reshape( + batch_size, num_frames // p_t, p_t, height // p, p, width // p, p, channels + ) + image_embeds = image_embeds.permute(0, 1, 3, 5, 7, 2, 4, 6).flatten(4, 7).flatten(1, 3) + image_embeds = self.proj(image_embeds.to(proj_dtype)).to(input_dtype) + + embeds = torch.cat([text_embeds, image_embeds], dim=1).contiguous() + + if self.use_positional_embeddings or self.use_learned_positional_embeddings: + text_seq_length = text_embeds.shape[1] + num_image_patches = image_embeds.shape[1] + + if self.use_learned_positional_embeddings: + image_pos = self.pos_embedding[ + :, self.max_text_seq_length:self.max_text_seq_length + num_image_patches + ].to(device=embeds.device, dtype=embeds.dtype) + else: + image_pos = get_3d_sincos_pos_embed( + self.dim, + (width // self.patch_size, height // self.patch_size), + num_image_patches // ((height // self.patch_size) * (width // self.patch_size)), + self.spatial_interpolation_scale, + self.temporal_interpolation_scale, + device=embeds.device, + ).reshape(1, num_image_patches, self.dim).to(dtype=embeds.dtype) + + # Build joint: zeros for text + sincos for image + joint_pos = torch.zeros(1, text_seq_length + num_image_patches, self.dim, device=embeds.device, dtype=embeds.dtype) + joint_pos[:, text_seq_length:] = image_pos + embeds = embeds + joint_pos + + return embeds + + +class CogVideoXLayerNormZero(nn.Module): + def __init__(self, time_dim, dim, elementwise_affine=True, eps=1e-5, bias=True, + device=None, dtype=None, operations=None): + super().__init__() + self.silu = nn.SiLU() + self.linear = operations.Linear(time_dim, 6 * dim, bias=bias, device=device, dtype=dtype) + self.norm = operations.LayerNorm(dim, eps=eps, elementwise_affine=elementwise_affine, device=device, dtype=dtype) + + def forward(self, hidden_states, encoder_hidden_states, temb): + shift, scale, gate, enc_shift, enc_scale, enc_gate = self.linear(self.silu(temb)).chunk(6, dim=1) + hidden_states = self.norm(hidden_states) * (1 + scale)[:, None, :] + shift[:, None, :] + encoder_hidden_states = self.norm(encoder_hidden_states) * (1 + enc_scale)[:, None, :] + enc_shift[:, None, :] + return hidden_states, encoder_hidden_states, gate[:, None, :], enc_gate[:, None, :] + + +class CogVideoXAdaLayerNorm(nn.Module): + def __init__(self, time_dim, dim, elementwise_affine=True, eps=1e-5, + device=None, dtype=None, operations=None): + super().__init__() + self.silu = nn.SiLU() + self.linear = operations.Linear(time_dim, 2 * dim, device=device, dtype=dtype) + self.norm = operations.LayerNorm(dim, eps=eps, elementwise_affine=elementwise_affine, device=device, dtype=dtype) + + def forward(self, x, temb): + temb = self.linear(self.silu(temb)) + shift, scale = temb.chunk(2, dim=1) + x = self.norm(x) * (1 + scale)[:, None, :] + shift[:, None, :] + return x + + +class CogVideoXBlock(nn.Module): + def __init__(self, dim, num_heads, head_dim, time_dim, + eps=1e-5, ff_inner_dim=None, ff_bias=True, + device=None, dtype=None, operations=None): + super().__init__() + self.dim = dim + self.num_heads = num_heads + self.head_dim = head_dim + + self.norm1 = CogVideoXLayerNormZero(time_dim, dim, eps=eps, device=device, dtype=dtype, operations=operations) + + # Self-attention (joint text + latent) + self.q = operations.Linear(dim, dim, bias=True, device=device, dtype=dtype) + self.k = operations.Linear(dim, dim, bias=True, device=device, dtype=dtype) + self.v = operations.Linear(dim, dim, bias=True, device=device, dtype=dtype) + self.norm_q = operations.LayerNorm(head_dim, eps=1e-6, elementwise_affine=True, device=device, dtype=dtype) + self.norm_k = operations.LayerNorm(head_dim, eps=1e-6, elementwise_affine=True, device=device, dtype=dtype) + self.attn_out = operations.Linear(dim, dim, bias=True, device=device, dtype=dtype) + + self.norm2 = CogVideoXLayerNormZero(time_dim, dim, eps=eps, device=device, dtype=dtype, operations=operations) + + # Feed-forward (GELU approximate) + inner_dim = ff_inner_dim or dim * 4 + self.ff_proj = operations.Linear(dim, inner_dim, bias=ff_bias, device=device, dtype=dtype) + self.ff_out = operations.Linear(inner_dim, dim, bias=ff_bias, device=device, dtype=dtype) + + def forward(self, hidden_states, encoder_hidden_states, temb, image_rotary_emb=None, transformer_options=None): + if transformer_options is None: + transformer_options = {} + text_seq_length = encoder_hidden_states.size(1) + + # Norm & modulate + norm_hidden, norm_encoder, gate_msa, enc_gate_msa = self.norm1(hidden_states, encoder_hidden_states, temb) + + # Joint self-attention + qkv_input = torch.cat([norm_encoder, norm_hidden], dim=1) + b, s, _ = qkv_input.shape + n, d = self.num_heads, self.head_dim + + q = self.q(qkv_input).view(b, s, n, d) + k = self.k(qkv_input).view(b, s, n, d) + v = self.v(qkv_input) + + q = self.norm_q(q).view(b, s, n, d) + k = self.norm_k(k).view(b, s, n, d) + + # Apply rotary embeddings to image tokens only (diffusers format: [B, heads, seq, head_dim]) + if image_rotary_emb is not None: + q_img = q[:, text_seq_length:].transpose(1, 2) # [B, heads, img_seq, head_dim] + k_img = k[:, text_seq_length:].transpose(1, 2) + q_img = apply_rotary_emb(q_img, image_rotary_emb) + k_img = apply_rotary_emb(k_img, image_rotary_emb) + q = torch.cat([q[:, :text_seq_length], q_img.transpose(1, 2)], dim=1) + k = torch.cat([k[:, :text_seq_length], k_img.transpose(1, 2)], dim=1) + + attn_out = optimized_attention( + q.reshape(b, s, n * d), + k.reshape(b, s, n * d), + v, + heads=self.num_heads, + transformer_options=transformer_options, + ) + + attn_out = self.attn_out(attn_out) + + attn_encoder, attn_hidden = attn_out.split([text_seq_length, s - text_seq_length], dim=1) + + hidden_states = hidden_states + gate_msa * attn_hidden + encoder_hidden_states = encoder_hidden_states + enc_gate_msa * attn_encoder + + # Norm & modulate for FF + norm_hidden, norm_encoder, gate_ff, enc_gate_ff = self.norm2(hidden_states, encoder_hidden_states, temb) + + # Feed-forward (GELU on concatenated text + latent) + ff_input = torch.cat([norm_encoder, norm_hidden], dim=1) + ff_output = self.ff_out(F.gelu(self.ff_proj(ff_input), approximate="tanh")) + + hidden_states = hidden_states + gate_ff * ff_output[:, text_seq_length:] + encoder_hidden_states = encoder_hidden_states + enc_gate_ff * ff_output[:, :text_seq_length] + + return hidden_states, encoder_hidden_states + + +class CogVideoXTransformer3DModel(nn.Module): + def __init__(self, + num_attention_heads=30, + attention_head_dim=64, + in_channels=16, + out_channels=16, + flip_sin_to_cos=True, + freq_shift=0, + time_embed_dim=512, + ofs_embed_dim=None, + text_embed_dim=4096, + num_layers=30, + dropout=0.0, + attention_bias=True, + sample_width=90, + sample_height=60, + sample_frames=49, + patch_size=2, + patch_size_t=None, + temporal_compression_ratio=4, + max_text_seq_length=226, + spatial_interpolation_scale=1.875, + temporal_interpolation_scale=1.0, + use_rotary_positional_embeddings=False, + use_learned_positional_embeddings=False, + patch_bias=True, + image_model=None, + device=None, + dtype=None, + operations=None, + ): + super().__init__() + self.dtype = dtype + dim = num_attention_heads * attention_head_dim + self.dim = dim + self.num_attention_heads = num_attention_heads + self.attention_head_dim = attention_head_dim + self.in_channels = in_channels + self.out_channels = out_channels + self.patch_size = patch_size + self.patch_size_t = patch_size_t + self.max_text_seq_length = max_text_seq_length + self.use_rotary_positional_embeddings = use_rotary_positional_embeddings + + # 1. Patch embedding + self.patch_embed = CogVideoXPatchEmbed( + patch_size=patch_size, + patch_size_t=patch_size_t, + in_channels=in_channels, + dim=dim, + text_dim=text_embed_dim, + bias=patch_bias, + sample_width=sample_width, + sample_height=sample_height, + sample_frames=sample_frames, + temporal_compression_ratio=temporal_compression_ratio, + max_text_seq_length=max_text_seq_length, + spatial_interpolation_scale=spatial_interpolation_scale, + temporal_interpolation_scale=temporal_interpolation_scale, + use_positional_embeddings=not use_rotary_positional_embeddings, + use_learned_positional_embeddings=use_learned_positional_embeddings, + device=device, dtype=torch.float32, operations=operations, + ) + + # 2. Time embedding + self.time_proj_dim = dim + self.time_proj_flip = flip_sin_to_cos + self.time_proj_shift = freq_shift + self.time_embedding_linear_1 = operations.Linear(dim, time_embed_dim, device=device, dtype=dtype) + self.time_embedding_act = nn.SiLU() + self.time_embedding_linear_2 = operations.Linear(time_embed_dim, time_embed_dim, device=device, dtype=dtype) + + # Optional OFS embedding (CogVideoX 1.5 I2V) + self.ofs_proj_dim = ofs_embed_dim + if ofs_embed_dim: + self.ofs_embedding_linear_1 = operations.Linear(ofs_embed_dim, ofs_embed_dim, device=device, dtype=dtype) + self.ofs_embedding_act = nn.SiLU() + self.ofs_embedding_linear_2 = operations.Linear(ofs_embed_dim, ofs_embed_dim, device=device, dtype=dtype) + else: + self.ofs_embedding_linear_1 = None + + # 3. Transformer blocks + self.blocks = nn.ModuleList([ + CogVideoXBlock( + dim=dim, + num_heads=num_attention_heads, + head_dim=attention_head_dim, + time_dim=time_embed_dim, + eps=1e-5, + device=device, dtype=dtype, operations=operations, + ) + for _ in range(num_layers) + ]) + + self.norm_final = operations.LayerNorm(dim, eps=1e-5, elementwise_affine=True, device=device, dtype=dtype) + + # 4. Output + self.norm_out = CogVideoXAdaLayerNorm( + time_dim=time_embed_dim, dim=dim, eps=1e-5, + device=device, dtype=dtype, operations=operations, + ) + + if patch_size_t is None: + output_dim = patch_size * patch_size * out_channels + else: + output_dim = patch_size * patch_size * patch_size_t * out_channels + + self.proj_out = operations.Linear(dim, output_dim, device=device, dtype=dtype) + + self.spatial_interpolation_scale = spatial_interpolation_scale + self.temporal_interpolation_scale = temporal_interpolation_scale + self.temporal_compression_ratio = temporal_compression_ratio + + def forward(self, x, timestep, context, ofs=None, transformer_options=None, **kwargs): + if transformer_options is None: + transformer_options = {} + return comfy.patcher_extension.WrapperExecutor.new_class_executor( + self._forward, + self, + comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.DIFFUSION_MODEL, transformer_options) + ).execute(x, timestep, context, ofs, transformer_options, **kwargs) + + def _forward(self, x, timestep, context, ofs=None, transformer_options=None, **kwargs): + if transformer_options is None: + transformer_options = {} + # ComfyUI passes [B, C, T, H, W] + batch_size, channels, t, h, w = x.shape + + # Pad to patch size (temporal + spatial), same pattern as WAN + p_t = self.patch_size_t if self.patch_size_t is not None else 1 + x = comfy.ldm.common_dit.pad_to_patch_size(x, (p_t, self.patch_size, self.patch_size)) + + # CogVideoX expects [B, T, C, H, W] + x = x.permute(0, 2, 1, 3, 4) + batch_size, num_frames, channels, height, width = x.shape + + # Time embedding + t_emb = get_timestep_embedding(timestep, self.time_proj_dim, self.time_proj_flip, self.time_proj_shift) + t_emb = t_emb.to(dtype=x.dtype) + emb = self.time_embedding_linear_2(self.time_embedding_act(self.time_embedding_linear_1(t_emb))) + + if self.ofs_embedding_linear_1 is not None and ofs is not None: + ofs_emb = get_timestep_embedding(ofs, self.ofs_proj_dim, self.time_proj_flip, self.time_proj_shift) + ofs_emb = ofs_emb.to(dtype=x.dtype) + ofs_emb = self.ofs_embedding_linear_2(self.ofs_embedding_act(self.ofs_embedding_linear_1(ofs_emb))) + emb = emb + ofs_emb + + # Patch embedding + hidden_states = self.patch_embed(context, x) + + text_seq_length = context.shape[1] + encoder_hidden_states = hidden_states[:, :text_seq_length] + hidden_states = hidden_states[:, text_seq_length:] + + # Rotary embeddings (if used) + image_rotary_emb = None + if self.use_rotary_positional_embeddings: + post_patch_height = height // self.patch_size + post_patch_width = width // self.patch_size + if self.patch_size_t is None: + post_time = num_frames + else: + post_time = num_frames // self.patch_size_t + image_rotary_emb = self._get_rotary_emb(post_patch_height, post_patch_width, post_time, device=x.device) + + # Transformer blocks + for i, block in enumerate(self.blocks): + hidden_states, encoder_hidden_states = block( + hidden_states=hidden_states, + encoder_hidden_states=encoder_hidden_states, + temb=emb, + image_rotary_emb=image_rotary_emb, + transformer_options=transformer_options, + ) + + hidden_states = self.norm_final(hidden_states) + + # Output projection + hidden_states = self.norm_out(hidden_states, temb=emb) + hidden_states = self.proj_out(hidden_states) + + # Unpatchify + p = self.patch_size + p_t = self.patch_size_t + + if p_t is None: + output = hidden_states.reshape(batch_size, num_frames, height // p, width // p, -1, p, p) + output = output.permute(0, 1, 4, 2, 5, 3, 6).flatten(5, 6).flatten(3, 4) + else: + output = hidden_states.reshape( + batch_size, (num_frames + p_t - 1) // p_t, height // p, width // p, -1, p_t, p, p + ) + output = output.permute(0, 1, 5, 4, 2, 6, 3, 7).flatten(6, 7).flatten(4, 5).flatten(1, 2) + + # Back to ComfyUI format [B, C, T, H, W] and crop padding + output = output.permute(0, 2, 1, 3, 4)[:, :, :t, :h, :w] + return output + + def _get_rotary_emb(self, h, w, t, device): + """Compute CogVideoX 3D rotary positional embeddings. + + For CogVideoX 1.5 (patch_size_t != None): uses "slice" mode — grid positions + are integer arange computed at max_size, then sliced to actual size. + For CogVideoX 1.0 (patch_size_t == None): uses "linspace" mode with crop coords + scaled by spatial_interpolation_scale. + """ + d = self.attention_head_dim + dim_t = d // 4 + dim_h = d // 8 * 3 + dim_w = d // 8 * 3 + + if self.patch_size_t is not None: + # CogVideoX 1.5: "slice" mode — positions are simple integer indices + # Compute at max(sample_size, actual_size) then slice to actual + base_h = self.patch_embed.sample_height // self.patch_size + base_w = self.patch_embed.sample_width // self.patch_size + max_h = max(base_h, h) + max_w = max(base_w, w) + + grid_h = torch.arange(max_h, device=device, dtype=torch.float32) + grid_w = torch.arange(max_w, device=device, dtype=torch.float32) + grid_t = torch.arange(t, device=device, dtype=torch.float32) + else: + # CogVideoX 1.0: "linspace" mode with interpolation scale + grid_h = torch.linspace(0, h - 1, h, device=device, dtype=torch.float32) * self.spatial_interpolation_scale + grid_w = torch.linspace(0, w - 1, w, device=device, dtype=torch.float32) * self.spatial_interpolation_scale + grid_t = torch.arange(t, device=device, dtype=torch.float32) + + freqs_t = _get_1d_rotary_pos_embed(dim_t, grid_t) + freqs_h = _get_1d_rotary_pos_embed(dim_h, grid_h) + freqs_w = _get_1d_rotary_pos_embed(dim_w, grid_w) + + t_cos, t_sin = freqs_t + h_cos, h_sin = freqs_h + w_cos, w_sin = freqs_w + + # Slice to actual size (for "slice" mode where grids may be larger) + t_cos, t_sin = t_cos[:t], t_sin[:t] + h_cos, h_sin = h_cos[:h], h_sin[:h] + w_cos, w_sin = w_cos[:w], w_sin[:w] + + # Broadcast and concatenate into [T*H*W, head_dim] + t_cos = t_cos[:, None, None, :].expand(-1, h, w, -1) + t_sin = t_sin[:, None, None, :].expand(-1, h, w, -1) + h_cos = h_cos[None, :, None, :].expand(t, -1, w, -1) + h_sin = h_sin[None, :, None, :].expand(t, -1, w, -1) + w_cos = w_cos[None, None, :, :].expand(t, h, -1, -1) + w_sin = w_sin[None, None, :, :].expand(t, h, -1, -1) + + cos = torch.cat([t_cos, h_cos, w_cos], dim=-1).reshape(t * h * w, -1) + sin = torch.cat([t_sin, h_sin, w_sin], dim=-1).reshape(t * h * w, -1) + return (cos, sin) diff --git a/comfy/ldm/cogvideo/vae.py b/comfy/ldm/cogvideo/vae.py new file mode 100644 index 000000000..d4e6f321e --- /dev/null +++ b/comfy/ldm/cogvideo/vae.py @@ -0,0 +1,566 @@ +# CogVideoX VAE - ported to ComfyUI native ops +# Architecture reference: diffusers AutoencoderKLCogVideoX +# Style reference: comfy/ldm/wan/vae.py + +import numpy as np + +import torch +import torch.nn as nn +import torch.nn.functional as F + +import comfy.ops +ops = comfy.ops.disable_weight_init + + +class CausalConv3d(nn.Module): + """Causal 3D convolution with temporal padding. + + Uses comfy.ops.Conv3d with autopad='causal_zero' fast path: when input has + a single temporal frame and no cache, the 3D conv weight is sliced to act + as a 2D conv, avoiding computation on zero-padded temporal dimensions. + """ + def __init__(self, in_channels, out_channels, kernel_size, stride=1, dilation=1, pad_mode="constant"): + super().__init__() + if isinstance(kernel_size, int): + kernel_size = (kernel_size,) * 3 + + time_kernel, height_kernel, width_kernel = kernel_size + self.time_kernel_size = time_kernel + self.pad_mode = pad_mode + + height_pad = (height_kernel - 1) // 2 + width_pad = (width_kernel - 1) // 2 + self.time_causal_padding = (width_pad, width_pad, height_pad, height_pad, time_kernel - 1, 0) + + stride = stride if isinstance(stride, tuple) else (stride, 1, 1) + dilation = (dilation, 1, 1) + self.conv = ops.Conv3d( + in_channels, out_channels, kernel_size, + stride=stride, dilation=dilation, + padding=(0, height_pad, width_pad), + ) + + def forward(self, x, conv_cache=None): + if self.pad_mode == "replicate": + x = F.pad(x, self.time_causal_padding, mode="replicate") + conv_cache = None + else: + kernel_t = self.time_kernel_size + if kernel_t > 1: + if conv_cache is None and x.shape[2] == 1: + # Fast path: single frame, no cache. All temporal padding + # frames are copies of the input (replicate-style), so the + # 3D conv reduces to a 2D conv with summed temporal kernel. + w = comfy.ops.cast_to_input(self.conv.weight, x) + b = comfy.ops.cast_to_input(self.conv.bias, x) if self.conv.bias is not None else None + w2d = w.sum(dim=2, keepdim=True) + out = F.conv3d(x, w2d, b, + self.conv.stride, self.conv.padding, + self.conv.dilation, self.conv.groups) + return out, None + cached = [conv_cache] if conv_cache is not None else [x[:, :, :1]] * (kernel_t - 1) + x = torch.cat(cached + [x], dim=2) + conv_cache = x[:, :, -self.time_kernel_size + 1:].clone() if self.time_kernel_size > 1 else None + + out = self.conv(x) + return out, conv_cache + + +def _interpolate_zq(zq, target_size): + """Interpolate latent z to target (T, H, W), matching CogVideoX's first-frame-special handling.""" + t = target_size[0] + if t > 1 and t % 2 == 1: + z_first = F.interpolate(zq[:, :, :1], size=(1, target_size[1], target_size[2])) + z_rest = F.interpolate(zq[:, :, 1:], size=(t - 1, target_size[1], target_size[2])) + return torch.cat([z_first, z_rest], dim=2) + return F.interpolate(zq, size=target_size) + + +class SpatialNorm3D(nn.Module): + """Spatially conditioned normalization.""" + def __init__(self, f_channels, zq_channels, groups=32): + super().__init__() + self.norm_layer = ops.GroupNorm(num_channels=f_channels, num_groups=groups, eps=1e-6, affine=True) + self.conv_y = CausalConv3d(zq_channels, f_channels, kernel_size=1, stride=1) + self.conv_b = CausalConv3d(zq_channels, f_channels, kernel_size=1, stride=1) + + def forward(self, f, zq, conv_cache=None): + new_cache = {} + conv_cache = conv_cache or {} + + if zq.shape[-3:] != f.shape[-3:]: + zq = _interpolate_zq(zq, f.shape[-3:]) + + conv_y, new_cache["conv_y"] = self.conv_y(zq, conv_cache=conv_cache.get("conv_y")) + conv_b, new_cache["conv_b"] = self.conv_b(zq, conv_cache=conv_cache.get("conv_b")) + + return self.norm_layer(f) * conv_y + conv_b, new_cache + + +class ResnetBlock3D(nn.Module): + """3D ResNet block with optional spatial norm.""" + def __init__(self, in_channels, out_channels=None, temb_channels=512, groups=32, + eps=1e-6, act_fn="silu", spatial_norm_dim=None, pad_mode="first"): + super().__init__() + out_channels = out_channels or in_channels + self.in_channels = in_channels + self.out_channels = out_channels + self.spatial_norm_dim = spatial_norm_dim + + if act_fn == "silu": + self.nonlinearity = nn.SiLU() + elif act_fn == "swish": + self.nonlinearity = nn.SiLU() + else: + self.nonlinearity = nn.SiLU() + + if spatial_norm_dim is None: + self.norm1 = ops.GroupNorm(num_channels=in_channels, num_groups=groups, eps=eps) + self.norm2 = ops.GroupNorm(num_channels=out_channels, num_groups=groups, eps=eps) + else: + self.norm1 = SpatialNorm3D(in_channels, spatial_norm_dim, groups=groups) + self.norm2 = SpatialNorm3D(out_channels, spatial_norm_dim, groups=groups) + + self.conv1 = CausalConv3d(in_channels, out_channels, kernel_size=3, pad_mode=pad_mode) + + if temb_channels > 0: + self.temb_proj = ops.Linear(temb_channels, out_channels) + + self.conv2 = CausalConv3d(out_channels, out_channels, kernel_size=3, pad_mode=pad_mode) + + if in_channels != out_channels: + self.conv_shortcut = ops.Conv3d(in_channels, out_channels, kernel_size=1, stride=1, padding=0) + else: + self.conv_shortcut = None + + def forward(self, x, temb=None, zq=None, conv_cache=None): + new_cache = {} + conv_cache = conv_cache or {} + residual = x + + if zq is not None: + x, new_cache["norm1"] = self.norm1(x, zq, conv_cache=conv_cache.get("norm1")) + else: + x = self.norm1(x) + + x = self.nonlinearity(x) + x, new_cache["conv1"] = self.conv1(x, conv_cache=conv_cache.get("conv1")) + + if temb is not None and hasattr(self, "temb_proj"): + x = x + self.temb_proj(self.nonlinearity(temb))[:, :, None, None, None] + + if zq is not None: + x, new_cache["norm2"] = self.norm2(x, zq, conv_cache=conv_cache.get("norm2")) + else: + x = self.norm2(x) + + x = self.nonlinearity(x) + x, new_cache["conv2"] = self.conv2(x, conv_cache=conv_cache.get("conv2")) + + if self.conv_shortcut is not None: + residual = self.conv_shortcut(residual) + + return x + residual, new_cache + + +class Downsample3D(nn.Module): + """3D downsampling with optional temporal compression.""" + def __init__(self, in_channels, out_channels, kernel_size=3, stride=2, padding=0, compress_time=False): + super().__init__() + self.conv = ops.Conv2d(in_channels, out_channels, kernel_size=kernel_size, stride=stride, padding=padding) + self.compress_time = compress_time + + def forward(self, x): + if self.compress_time: + b, c, t, h, w = x.shape + x = x.permute(0, 3, 4, 1, 2).reshape(b * h * w, c, t) + if t % 2 == 1: + x_first, x_rest = x[..., 0], x[..., 1:] + if x_rest.shape[-1] > 0: + x_rest = F.avg_pool1d(x_rest, kernel_size=2, stride=2) + x = torch.cat([x_first[..., None], x_rest], dim=-1) + x = x.reshape(b, h, w, c, x.shape[-1]).permute(0, 3, 4, 1, 2) + else: + x = F.avg_pool1d(x, kernel_size=2, stride=2) + x = x.reshape(b, h, w, c, x.shape[-1]).permute(0, 3, 4, 1, 2) + + pad = (0, 1, 0, 1) + x = F.pad(x, pad, mode="constant", value=0) + b, c, t, h, w = x.shape + x = x.permute(0, 2, 1, 3, 4).reshape(b * t, c, h, w) + x = self.conv(x) + x = x.reshape(b, t, x.shape[1], x.shape[2], x.shape[3]).permute(0, 2, 1, 3, 4) + return x + + +class Upsample3D(nn.Module): + """3D upsampling with optional temporal decompression.""" + def __init__(self, in_channels, out_channels, kernel_size=3, stride=1, padding=1, compress_time=False): + super().__init__() + self.conv = ops.Conv2d(in_channels, out_channels, kernel_size=kernel_size, stride=stride, padding=padding) + self.compress_time = compress_time + + def forward(self, x): + if self.compress_time: + if x.shape[2] > 1 and x.shape[2] % 2 == 1: + x_first, x_rest = x[:, :, 0], x[:, :, 1:] + x_first = F.interpolate(x_first, scale_factor=2.0) + x_rest = F.interpolate(x_rest, scale_factor=2.0) + x = torch.cat([x_first[:, :, None, :, :], x_rest], dim=2) + elif x.shape[2] > 1: + x = F.interpolate(x, scale_factor=2.0) + else: + x = x.squeeze(2) + x = F.interpolate(x, scale_factor=2.0) + x = x[:, :, None, :, :] + else: + b, c, t, h, w = x.shape + x = x.permute(0, 2, 1, 3, 4).reshape(b * t, c, h, w) + x = F.interpolate(x, scale_factor=2.0) + x = x.reshape(b, t, c, *x.shape[2:]).permute(0, 2, 1, 3, 4) + + b, c, t, h, w = x.shape + x = x.permute(0, 2, 1, 3, 4).reshape(b * t, c, h, w) + x = self.conv(x) + x = x.reshape(b, t, *x.shape[1:]).permute(0, 2, 1, 3, 4) + return x + + +class DownBlock3D(nn.Module): + def __init__(self, in_channels, out_channels, temb_channels=0, num_layers=1, + eps=1e-6, act_fn="silu", groups=32, add_downsample=True, + compress_time=False, pad_mode="first"): + super().__init__() + self.resnets = nn.ModuleList([ + ResnetBlock3D( + in_channels=in_channels if i == 0 else out_channels, + out_channels=out_channels, + temb_channels=temb_channels, + groups=groups, eps=eps, act_fn=act_fn, pad_mode=pad_mode, + ) + for i in range(num_layers) + ]) + self.downsamplers = nn.ModuleList([Downsample3D(out_channels, out_channels, compress_time=compress_time)]) if add_downsample else None + + def forward(self, x, temb=None, zq=None, conv_cache=None): + new_cache = {} + conv_cache = conv_cache or {} + for i, resnet in enumerate(self.resnets): + x, new_cache[f"resnet_{i}"] = resnet(x, temb, zq, conv_cache=conv_cache.get(f"resnet_{i}")) + if self.downsamplers is not None: + for ds in self.downsamplers: + x = ds(x) + return x, new_cache + + +class MidBlock3D(nn.Module): + def __init__(self, in_channels, temb_channels=0, num_layers=1, + eps=1e-6, act_fn="silu", groups=32, spatial_norm_dim=None, pad_mode="first"): + super().__init__() + self.resnets = nn.ModuleList([ + ResnetBlock3D( + in_channels=in_channels, out_channels=in_channels, + temb_channels=temb_channels, groups=groups, eps=eps, + act_fn=act_fn, spatial_norm_dim=spatial_norm_dim, pad_mode=pad_mode, + ) + for _ in range(num_layers) + ]) + + def forward(self, x, temb=None, zq=None, conv_cache=None): + new_cache = {} + conv_cache = conv_cache or {} + for i, resnet in enumerate(self.resnets): + x, new_cache[f"resnet_{i}"] = resnet(x, temb, zq, conv_cache=conv_cache.get(f"resnet_{i}")) + return x, new_cache + + +class UpBlock3D(nn.Module): + def __init__(self, in_channels, out_channels, temb_channels=0, num_layers=1, + eps=1e-6, act_fn="silu", groups=32, spatial_norm_dim=16, + add_upsample=True, compress_time=False, pad_mode="first"): + super().__init__() + self.resnets = nn.ModuleList([ + ResnetBlock3D( + in_channels=in_channels if i == 0 else out_channels, + out_channels=out_channels, + temb_channels=temb_channels, groups=groups, eps=eps, + act_fn=act_fn, spatial_norm_dim=spatial_norm_dim, pad_mode=pad_mode, + ) + for i in range(num_layers) + ]) + self.upsamplers = nn.ModuleList([Upsample3D(out_channels, out_channels, compress_time=compress_time)]) if add_upsample else None + + def forward(self, x, temb=None, zq=None, conv_cache=None): + new_cache = {} + conv_cache = conv_cache or {} + for i, resnet in enumerate(self.resnets): + x, new_cache[f"resnet_{i}"] = resnet(x, temb, zq, conv_cache=conv_cache.get(f"resnet_{i}")) + if self.upsamplers is not None: + for us in self.upsamplers: + x = us(x) + return x, new_cache + + +class Encoder3D(nn.Module): + def __init__(self, in_channels=3, out_channels=16, + block_out_channels=(128, 256, 256, 512), + layers_per_block=3, act_fn="silu", + eps=1e-6, groups=32, pad_mode="first", + temporal_compression_ratio=4): + super().__init__() + temporal_compress_level = int(np.log2(temporal_compression_ratio)) + + self.conv_in = CausalConv3d(in_channels, block_out_channels[0], kernel_size=3, pad_mode=pad_mode) + + self.down_blocks = nn.ModuleList() + output_channel = block_out_channels[0] + for i in range(len(block_out_channels)): + input_channel = output_channel + output_channel = block_out_channels[i] + is_final = i == len(block_out_channels) - 1 + compress_time = i < temporal_compress_level + + self.down_blocks.append(DownBlock3D( + in_channels=input_channel, out_channels=output_channel, + temb_channels=0, num_layers=layers_per_block, + eps=eps, act_fn=act_fn, groups=groups, + add_downsample=not is_final, compress_time=compress_time, + )) + + self.mid_block = MidBlock3D( + in_channels=block_out_channels[-1], temb_channels=0, + num_layers=2, eps=eps, act_fn=act_fn, groups=groups, pad_mode=pad_mode, + ) + + self.norm_out = ops.GroupNorm(groups, block_out_channels[-1], eps=1e-6) + self.conv_act = nn.SiLU() + self.conv_out = CausalConv3d(block_out_channels[-1], 2 * out_channels, kernel_size=3, pad_mode=pad_mode) + + def forward(self, x, conv_cache=None): + new_cache = {} + conv_cache = conv_cache or {} + + x, new_cache["conv_in"] = self.conv_in(x, conv_cache=conv_cache.get("conv_in")) + + for i, block in enumerate(self.down_blocks): + key = f"down_block_{i}" + x, new_cache[key] = block(x, None, None, conv_cache.get(key)) + + x, new_cache["mid_block"] = self.mid_block(x, None, None, conv_cache=conv_cache.get("mid_block")) + + x = self.norm_out(x) + x = self.conv_act(x) + x, new_cache["conv_out"] = self.conv_out(x, conv_cache=conv_cache.get("conv_out")) + + return x, new_cache + + +class Decoder3D(nn.Module): + def __init__(self, in_channels=16, out_channels=3, + block_out_channels=(128, 256, 256, 512), + layers_per_block=3, act_fn="silu", + eps=1e-6, groups=32, pad_mode="first", + temporal_compression_ratio=4): + super().__init__() + reversed_channels = list(reversed(block_out_channels)) + temporal_compress_level = int(np.log2(temporal_compression_ratio)) + + self.conv_in = CausalConv3d(in_channels, reversed_channels[0], kernel_size=3, pad_mode=pad_mode) + + self.mid_block = MidBlock3D( + in_channels=reversed_channels[0], temb_channels=0, + num_layers=2, eps=eps, act_fn=act_fn, groups=groups, + spatial_norm_dim=in_channels, pad_mode=pad_mode, + ) + + self.up_blocks = nn.ModuleList() + output_channel = reversed_channels[0] + for i in range(len(block_out_channels)): + prev_channel = output_channel + output_channel = reversed_channels[i] + is_final = i == len(block_out_channels) - 1 + compress_time = i < temporal_compress_level + + self.up_blocks.append(UpBlock3D( + in_channels=prev_channel, out_channels=output_channel, + temb_channels=0, num_layers=layers_per_block + 1, + eps=eps, act_fn=act_fn, groups=groups, + spatial_norm_dim=in_channels, + add_upsample=not is_final, compress_time=compress_time, + )) + + self.norm_out = SpatialNorm3D(reversed_channels[-1], in_channels, groups=groups) + self.conv_act = nn.SiLU() + self.conv_out = CausalConv3d(reversed_channels[-1], out_channels, kernel_size=3, pad_mode=pad_mode) + + def forward(self, sample, conv_cache=None): + new_cache = {} + conv_cache = conv_cache or {} + + x, new_cache["conv_in"] = self.conv_in(sample, conv_cache=conv_cache.get("conv_in")) + + x, new_cache["mid_block"] = self.mid_block(x, None, sample, conv_cache=conv_cache.get("mid_block")) + + for i, block in enumerate(self.up_blocks): + key = f"up_block_{i}" + x, new_cache[key] = block(x, None, sample, conv_cache=conv_cache.get(key)) + + x, new_cache["norm_out"] = self.norm_out(x, sample, conv_cache=conv_cache.get("norm_out")) + x = self.conv_act(x) + x, new_cache["conv_out"] = self.conv_out(x, conv_cache=conv_cache.get("conv_out")) + + return x, new_cache + + + +class AutoencoderKLCogVideoX(nn.Module): + """CogVideoX VAE. Spatial tiling/slicing handled by ComfyUI's VAE wrapper. + + Uses rolling temporal decode: conv_in + mid_block + temporal up_blocks run + on the full (low-res) tensor, then the expensive spatial-only up_blocks + + norm_out + conv_out are processed in small temporal chunks with conv_cache + carrying causal state between chunks. This keeps peak VRAM proportional to + chunk_size rather than total frame count. + """ + + def __init__(self, + in_channels=3, out_channels=3, + block_out_channels=(128, 256, 256, 512), + latent_channels=16, layers_per_block=3, + act_fn="silu", eps=1e-6, groups=32, + temporal_compression_ratio=4, + ): + super().__init__() + self.latent_channels = latent_channels + self.temporal_compression_ratio = temporal_compression_ratio + + self.encoder = Encoder3D( + in_channels=in_channels, out_channels=latent_channels, + block_out_channels=block_out_channels, layers_per_block=layers_per_block, + act_fn=act_fn, eps=eps, groups=groups, + temporal_compression_ratio=temporal_compression_ratio, + ) + self.decoder = Decoder3D( + in_channels=latent_channels, out_channels=out_channels, + block_out_channels=block_out_channels, layers_per_block=layers_per_block, + act_fn=act_fn, eps=eps, groups=groups, + temporal_compression_ratio=temporal_compression_ratio, + ) + + self.num_latent_frames_batch_size = 2 + self.num_sample_frames_batch_size = 8 + + def encode(self, x): + t = x.shape[2] + frame_batch = self.num_sample_frames_batch_size + remainder = t % frame_batch + conv_cache = None + enc = [] + + # Process remainder frames first so only the first chunk can have an + # odd temporal dimension — where Downsample3D's first-frame-special + # handling in temporal compression is actually correct. + if remainder > 0: + chunk, conv_cache = self.encoder(x[:, :, :remainder], conv_cache=conv_cache) + enc.append(chunk.to(x.device)) + + for start in range(remainder, t, frame_batch): + chunk, conv_cache = self.encoder(x[:, :, start:start + frame_batch], conv_cache=conv_cache) + enc.append(chunk.to(x.device)) + + enc = torch.cat(enc, dim=2) + mean, _ = enc.chunk(2, dim=1) + return mean + + def decode(self, z): + return self._decode_rolling(z) + + def _decode_batched(self, z): + """Original batched decode - processes 2 latent frames through full decoder.""" + t = z.shape[2] + frame_batch = self.num_latent_frames_batch_size + num_batches = max(t // frame_batch, 1) + conv_cache = None + dec = [] + for i in range(num_batches): + remaining = t % frame_batch + start = frame_batch * i + (0 if i == 0 else remaining) + end = frame_batch * (i + 1) + remaining + chunk, conv_cache = self.decoder(z[:, :, start:end], conv_cache=conv_cache) + dec.append(chunk.cpu()) + return torch.cat(dec, dim=2).to(z.device) + + def _decode_rolling(self, z): + """Rolling decode - processes low-res layers on full tensor, then rolls + through expensive high-res layers in temporal chunks.""" + decoder = self.decoder + device = z.device + + # Determine which up_blocks have temporal upsample vs spatial-only. + # Temporal up_blocks are cheap (low res), spatial-only are expensive. + temporal_compress_level = int(np.log2(self.temporal_compression_ratio)) + split_at = temporal_compress_level # first N up_blocks do temporal upsample + + # Phase 1: conv_in + mid_block + temporal up_blocks on full tensor (low/medium res) + x, _ = decoder.conv_in(z) + x, _ = decoder.mid_block(x, None, z) + + for i in range(split_at): + x, _ = decoder.up_blocks[i](x, None, z) + + # Phase 2: remaining spatial-only up_blocks + norm_out + conv_out in temporal chunks + remaining_blocks = list(range(split_at, len(decoder.up_blocks))) + chunk_size = 4 # pixel frames per chunk through high-res layers + t_expanded = x.shape[2] + + if t_expanded <= chunk_size or len(remaining_blocks) == 0: + # Small enough to process in one go + for i in remaining_blocks: + x, _ = decoder.up_blocks[i](x, None, z) + x, _ = decoder.norm_out(x, z) + x = decoder.conv_act(x) + x, _ = decoder.conv_out(x) + return x + + # Expand z temporally once to match Phase 2's time dimension. + # z stays at latent spatial resolution so this is small (~16 MB vs ~1.3 GB + # for the old approach of pre-interpolating to every pixel resolution). + z_time_expanded = _interpolate_zq(z, (t_expanded, z.shape[3], z.shape[4])) + + # Process in temporal chunks, interpolating spatially per-chunk to avoid + # allocating full [B, C, t_expanded, H, W] tensors at each resolution. + dec_out = [] + conv_caches = {} + + for chunk_start in range(0, t_expanded, chunk_size): + chunk_end = min(chunk_start + chunk_size, t_expanded) + x_chunk = x[:, :, chunk_start:chunk_end] + z_t_chunk = z_time_expanded[:, :, chunk_start:chunk_end] + z_spatial_cache = {} + + for i in remaining_blocks: + block = decoder.up_blocks[i] + cache_key = f"up_block_{i}" + hw_key = (x_chunk.shape[3], x_chunk.shape[4]) + if hw_key not in z_spatial_cache: + if z_t_chunk.shape[3] == hw_key[0] and z_t_chunk.shape[4] == hw_key[1]: + z_spatial_cache[hw_key] = z_t_chunk + else: + z_spatial_cache[hw_key] = F.interpolate(z_t_chunk, size=(z_t_chunk.shape[2], hw_key[0], hw_key[1])) + x_chunk, new_cache = block(x_chunk, None, z_spatial_cache[hw_key], conv_cache=conv_caches.get(cache_key)) + conv_caches[cache_key] = new_cache + + hw_key = (x_chunk.shape[3], x_chunk.shape[4]) + if hw_key not in z_spatial_cache: + z_spatial_cache[hw_key] = F.interpolate(z_t_chunk, size=(z_t_chunk.shape[2], hw_key[0], hw_key[1])) + x_chunk, new_cache = decoder.norm_out(x_chunk, z_spatial_cache[hw_key], conv_cache=conv_caches.get("norm_out")) + conv_caches["norm_out"] = new_cache + x_chunk = decoder.conv_act(x_chunk) + x_chunk, new_cache = decoder.conv_out(x_chunk, conv_cache=conv_caches.get("conv_out")) + conv_caches["conv_out"] = new_cache + + dec_out.append(x_chunk.cpu()) + del z_spatial_cache + + del x, z_time_expanded + return torch.cat(dec_out, dim=2).to(device) diff --git a/comfy/ldm/ernie/model.py b/comfy/ldm/ernie/model.py new file mode 100644 index 000000000..eba661aec --- /dev/null +++ b/comfy/ldm/ernie/model.py @@ -0,0 +1,301 @@ +import math +import torch +import torch.nn as nn +import torch.nn.functional as F + +from comfy.ldm.modules.attention import optimized_attention +import comfy.model_management + +def rope(pos: torch.Tensor, dim: int, theta: int) -> torch.Tensor: + assert dim % 2 == 0 + if not comfy.model_management.supports_fp64(pos.device): + device = torch.device("cpu") + else: + device = pos.device + + scale = torch.arange(0, dim, 2, dtype=torch.float64, device=device) / dim + omega = 1.0 / (theta**scale) + out = torch.einsum("...n,d->...nd", pos.to(device), omega) + out = torch.stack([torch.cos(out), torch.sin(out)], dim=0) + return out.to(dtype=torch.float32, device=pos.device) + +def apply_rotary_emb(x_in: torch.Tensor, freqs_cis: torch.Tensor) -> torch.Tensor: + rot_dim = freqs_cis.shape[-1] + x, x_pass = x_in[..., :rot_dim], x_in[..., rot_dim:] + cos_ = freqs_cis[0] + sin_ = freqs_cis[1] + x1, x2 = x.chunk(2, dim=-1) + x_rotated = torch.cat((-x2, x1), dim=-1) + return torch.cat((x * cos_ + x_rotated * sin_, x_pass), dim=-1) + +class ErnieImageEmbedND3(nn.Module): + def __init__(self, dim: int, theta: int, axes_dim: tuple): + super().__init__() + self.dim = dim + self.theta = theta + self.axes_dim = list(axes_dim) + + def forward(self, ids: torch.Tensor) -> torch.Tensor: + emb = torch.cat([rope(ids[..., i], self.axes_dim[i], self.theta) for i in range(3)], dim=-1) + emb = emb.unsqueeze(3) # [2, B, S, 1, head_dim//2] + return torch.stack([emb, emb], dim=-1).reshape(*emb.shape[:-1], -1) # [B, S, 1, head_dim] + +class ErnieImagePatchEmbedDynamic(nn.Module): + def __init__(self, in_channels: int, embed_dim: int, patch_size: int, operations, device=None, dtype=None): + super().__init__() + self.patch_size = patch_size + self.proj = operations.Conv2d(in_channels, embed_dim, kernel_size=patch_size, stride=patch_size, bias=True, device=device, dtype=dtype) + + def forward(self, x: torch.Tensor) -> torch.Tensor: + x = self.proj(x) + batch_size, dim, height, width = x.shape + return x.reshape(batch_size, dim, height * width).transpose(1, 2).contiguous() + +class Timesteps(nn.Module): + def __init__(self, num_channels: int, flip_sin_to_cos: bool = False): + super().__init__() + self.num_channels = num_channels + self.flip_sin_to_cos = flip_sin_to_cos + + def forward(self, timesteps: torch.Tensor) -> torch.Tensor: + half_dim = self.num_channels // 2 + exponent = -math.log(10000) * torch.arange(half_dim, dtype=torch.float32, device=timesteps.device) / half_dim + emb = torch.exp(exponent) + emb = timesteps[:, None].float() * emb[None, :] + if self.flip_sin_to_cos: + emb = torch.cat([torch.cos(emb), torch.sin(emb)], dim=-1) + else: + emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=-1) + return emb + +class TimestepEmbedding(nn.Module): + def __init__(self, in_channels: int, time_embed_dim: int, operations, device=None, dtype=None): + super().__init__() + Linear = operations.Linear + self.linear_1 = Linear(in_channels, time_embed_dim, bias=True, device=device, dtype=dtype) + self.act = nn.SiLU() + self.linear_2 = Linear(time_embed_dim, time_embed_dim, bias=True, device=device, dtype=dtype) + + def forward(self, sample: torch.Tensor) -> torch.Tensor: + sample = self.linear_1(sample) + sample = self.act(sample) + sample = self.linear_2(sample) + return sample + +class ErnieImageAttention(nn.Module): + def __init__(self, query_dim: int, heads: int, dim_head: int, eps: float = 1e-6, operations=None, device=None, dtype=None): + super().__init__() + self.heads = heads + self.head_dim = dim_head + self.inner_dim = heads * dim_head + + Linear = operations.Linear + RMSNorm = operations.RMSNorm + + self.to_q = Linear(query_dim, self.inner_dim, bias=False, device=device, dtype=dtype) + self.to_k = Linear(query_dim, self.inner_dim, bias=False, device=device, dtype=dtype) + self.to_v = Linear(query_dim, self.inner_dim, bias=False, device=device, dtype=dtype) + + self.norm_q = RMSNorm(dim_head, eps=eps, elementwise_affine=True, device=device, dtype=dtype) + self.norm_k = RMSNorm(dim_head, eps=eps, elementwise_affine=True, device=device, dtype=dtype) + + self.to_out = nn.ModuleList([Linear(self.inner_dim, query_dim, bias=False, device=device, dtype=dtype)]) + + def forward(self, x: torch.Tensor, attention_mask: torch.Tensor = None, image_rotary_emb: torch.Tensor = None) -> torch.Tensor: + B, S, _ = x.shape + + q_flat = self.to_q(x) + k_flat = self.to_k(x) + v_flat = self.to_v(x) + + query = q_flat.view(B, S, self.heads, self.head_dim) + key = k_flat.view(B, S, self.heads, self.head_dim) + + query = self.norm_q(query) + key = self.norm_k(key) + + if image_rotary_emb is not None: + query = apply_rotary_emb(query, image_rotary_emb) + key = apply_rotary_emb(key, image_rotary_emb) + + q_flat = query.reshape(B, S, -1) + k_flat = key.reshape(B, S, -1) + + hidden_states = optimized_attention(q_flat, k_flat, v_flat, self.heads, mask=attention_mask) + + return self.to_out[0](hidden_states) + +class ErnieImageFeedForward(nn.Module): + def __init__(self, hidden_size: int, ffn_hidden_size: int, operations, device=None, dtype=None): + super().__init__() + Linear = operations.Linear + self.gate_proj = Linear(hidden_size, ffn_hidden_size, bias=False, device=device, dtype=dtype) + self.up_proj = Linear(hidden_size, ffn_hidden_size, bias=False, device=device, dtype=dtype) + self.linear_fc2 = Linear(ffn_hidden_size, hidden_size, bias=False, device=device, dtype=dtype) + + def forward(self, x: torch.Tensor) -> torch.Tensor: + return self.linear_fc2(self.up_proj(x) * F.gelu(self.gate_proj(x))) + +class ErnieImageSharedAdaLNBlock(nn.Module): + def __init__(self, hidden_size: int, num_heads: int, ffn_hidden_size: int, eps: float = 1e-6, operations=None, device=None, dtype=None): + super().__init__() + RMSNorm = operations.RMSNorm + + self.adaLN_sa_ln = RMSNorm(hidden_size, eps=eps, device=device, dtype=dtype) + self.self_attention = ErnieImageAttention( + query_dim=hidden_size, + dim_head=hidden_size // num_heads, + heads=num_heads, + eps=eps, + operations=operations, + device=device, + dtype=dtype + ) + self.adaLN_mlp_ln = RMSNorm(hidden_size, eps=eps, device=device, dtype=dtype) + self.mlp = ErnieImageFeedForward(hidden_size, ffn_hidden_size, operations=operations, device=device, dtype=dtype) + + def forward(self, x, rotary_pos_emb, temb, attention_mask=None): + shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = temb + + residual = x + x_norm = self.adaLN_sa_ln(x) + x_norm = x_norm * (1 + scale_msa) + shift_msa + + attn_out = self.self_attention(x_norm, attention_mask=attention_mask, image_rotary_emb=rotary_pos_emb) + x = residual + gate_msa * attn_out + + residual = x + x_norm = self.adaLN_mlp_ln(x) + x_norm = x_norm * (1 + scale_mlp) + shift_mlp + + return residual + gate_mlp * self.mlp(x_norm) + +class ErnieImageAdaLNContinuous(nn.Module): + def __init__(self, hidden_size: int, eps: float = 1e-6, operations=None, device=None, dtype=None): + super().__init__() + LayerNorm = operations.LayerNorm + Linear = operations.Linear + self.norm = LayerNorm(hidden_size, elementwise_affine=False, eps=eps, device=device, dtype=dtype) + self.linear = Linear(hidden_size, hidden_size * 2, device=device, dtype=dtype) + + def forward(self, x: torch.Tensor, conditioning: torch.Tensor) -> torch.Tensor: + scale, shift = self.linear(conditioning).chunk(2, dim=-1) + x = self.norm(x) + x = torch.addcmul(shift.unsqueeze(1), x, 1 + scale.unsqueeze(1)) + return x + +class ErnieImageModel(nn.Module): + def __init__( + self, + hidden_size: int = 4096, + num_attention_heads: int = 32, + num_layers: int = 36, + ffn_hidden_size: int = 12288, + in_channels: int = 128, + out_channels: int = 128, + patch_size: int = 1, + text_in_dim: int = 3072, + rope_theta: int = 256, + rope_axes_dim: tuple = (32, 48, 48), + eps: float = 1e-6, + qk_layernorm: bool = True, + device=None, + dtype=None, + operations=None, + **kwargs + ): + super().__init__() + self.dtype = dtype + self.hidden_size = hidden_size + self.num_heads = num_attention_heads + self.head_dim = hidden_size // num_attention_heads + self.patch_size = patch_size + self.out_channels = out_channels + + Linear = operations.Linear + + self.x_embedder = ErnieImagePatchEmbedDynamic(in_channels, hidden_size, patch_size, operations, device, dtype) + self.text_proj = Linear(text_in_dim, hidden_size, bias=False, device=device, dtype=dtype) if text_in_dim != hidden_size else None + + self.time_proj = Timesteps(hidden_size, flip_sin_to_cos=False) + self.time_embedding = TimestepEmbedding(hidden_size, hidden_size, operations, device, dtype) + + self.pos_embed = ErnieImageEmbedND3(dim=self.head_dim, theta=rope_theta, axes_dim=rope_axes_dim) + + self.adaLN_modulation = nn.Sequential( + nn.SiLU(), + Linear(hidden_size, 6 * hidden_size, device=device, dtype=dtype) + ) + + self.layers = nn.ModuleList([ + ErnieImageSharedAdaLNBlock(hidden_size, num_attention_heads, ffn_hidden_size, eps, operations, device, dtype) + for _ in range(num_layers) + ]) + + self.final_norm = ErnieImageAdaLNContinuous(hidden_size, eps, operations, device, dtype) + self.final_linear = Linear(hidden_size, patch_size * patch_size * out_channels, device=device, dtype=dtype) + + def forward(self, x, timesteps, context, **kwargs): + device, dtype = x.device, x.dtype + B, C, H, W = x.shape + p, Hp, Wp = self.patch_size, H // self.patch_size, W // self.patch_size + N_img = Hp * Wp + + img_bsh = self.x_embedder(x) + + text_bth = context + if self.text_proj is not None and text_bth.numel() > 0: + text_bth = self.text_proj(text_bth) + Tmax = text_bth.shape[1] + + hidden_states = torch.cat([img_bsh, text_bth], dim=1) + + text_ids = torch.zeros((B, Tmax, 3), device=device, dtype=torch.float32) + text_ids[:, :, 0] = torch.linspace(0, Tmax - 1, steps=Tmax, device=x.device, dtype=torch.float32) + index = float(Tmax) + + transformer_options = kwargs.get("transformer_options", {}) + rope_options = transformer_options.get("rope_options", None) + + h_len, w_len = float(Hp), float(Wp) + h_offset, w_offset = 0.0, 0.0 + + if rope_options is not None: + h_len = (h_len - 1.0) * rope_options.get("scale_y", 1.0) + 1.0 + w_len = (w_len - 1.0) * rope_options.get("scale_x", 1.0) + 1.0 + index += rope_options.get("shift_t", 0.0) + h_offset += rope_options.get("shift_y", 0.0) + w_offset += rope_options.get("shift_x", 0.0) + + image_ids = torch.zeros((Hp, Wp, 3), device=device, dtype=torch.float32) + image_ids[:, :, 0] = image_ids[:, :, 1] + index + image_ids[:, :, 1] = image_ids[:, :, 1] + torch.linspace(h_offset, h_len - 1 + h_offset, steps=Hp, device=device, dtype=torch.float32).unsqueeze(1) + image_ids[:, :, 2] = image_ids[:, :, 2] + torch.linspace(w_offset, w_len - 1 + w_offset, steps=Wp, device=device, dtype=torch.float32).unsqueeze(0) + + image_ids = image_ids.view(1, N_img, 3).expand(B, -1, -1) + + rotary_pos_emb = self.pos_embed(torch.cat([image_ids, text_ids], dim=1)).to(x.dtype) + del image_ids, text_ids + + sample = self.time_proj(timesteps).to(dtype) + c = self.time_embedding(sample) + + shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = [ + t.unsqueeze(1).contiguous() for t in self.adaLN_modulation(c).chunk(6, dim=-1) + ] + + temb = [shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp] + for layer in self.layers: + hidden_states = layer(hidden_states, rotary_pos_emb, temb) + + hidden_states = self.final_norm(hidden_states, c).type_as(hidden_states) + + patches = self.final_linear(hidden_states)[:, :N_img, :] + output = ( + patches.view(B, Hp, Wp, p, p, self.out_channels) + .permute(0, 5, 1, 3, 2, 4) + .contiguous() + .view(B, self.out_channels, H, W) + ) + + return output diff --git a/comfy/ldm/flux/math.py b/comfy/ldm/flux/math.py index 824daf5e6..6d0aed827 100644 --- a/comfy/ldm/flux/math.py +++ b/comfy/ldm/flux/math.py @@ -16,7 +16,7 @@ def attention(q: Tensor, k: Tensor, v: Tensor, pe: Tensor, mask=None, transforme def rope(pos: Tensor, dim: int, theta: int) -> Tensor: assert dim % 2 == 0 - if comfy.model_management.is_device_mps(pos.device) or comfy.model_management.is_intel_xpu() or comfy.model_management.is_directml_enabled(): + if not comfy.model_management.supports_fp64(pos.device): device = torch.device("cpu") else: device = pos.device diff --git a/comfy/ldm/hidream_o1/attention.py b/comfy/ldm/hidream_o1/attention.py new file mode 100644 index 000000000..1b68f1771 --- /dev/null +++ b/comfy/ldm/hidream_o1/attention.py @@ -0,0 +1,41 @@ +"""HiDream-O1 two-pass attention: tokens [0, ar_len) are causal, [ar_len, T) +attend full K/V. Splitting Q at the boundary avoids the (B, 1, T, T) additive +mask the general-purpose path would build (~500 MB at T~16K) and lets the +gen half hit the user's preferred backend via optimized_attention. +""" + +import torch + +import comfy.ops +from comfy.ldm.modules.attention import optimized_attention + + +def make_two_pass_attention(ar_len: int, transformer_options=None): + """Build a two-pass attention callable. AR pass uses SDPA-causal directly, gen pass routes through optimized_attention. + The AR pass goes through SDPA directand bypasses wrappers, it is only ~1% of T at typical edit sizes. + """ + + def two_pass_attention(q, k, v, heads, **kwargs): + B, H, T, D = q.shape + + if T < k.shape[2]: # KV-cache hot path: Q is shorter than K/V (cached AR prefix is in K/V only), all fresh Q positions are in the gen region, single full-attention call + out = optimized_attention(q, k, v, heads, mask=None, skip_reshape=True, skip_output_reshape=True, transformer_options=transformer_options) + elif ar_len >= T: + out = comfy.ops.scaled_dot_product_attention(q, k, v, attn_mask=None, dropout_p=0.0, is_causal=True) + elif ar_len <= 0: + out = optimized_attention(q, k, v, heads, mask=None, skip_reshape=True, skip_output_reshape=True, transformer_options=transformer_options) + else: + out_ar = comfy.ops.scaled_dot_product_attention( + q[:, :, :ar_len], k[:, :, :ar_len], v[:, :, :ar_len], + attn_mask=None, dropout_p=0.0, is_causal=True, + ) + out_gen = optimized_attention( + q[:, :, ar_len:], k, v, heads, + mask=None, skip_reshape=True, skip_output_reshape=True, + transformer_options=transformer_options, + ) + out = torch.cat([out_ar, out_gen], dim=2) + + return out.transpose(1, 2).reshape(B, T, H * D) + + return two_pass_attention diff --git a/comfy/ldm/hidream_o1/conditioning.py b/comfy/ldm/hidream_o1/conditioning.py new file mode 100644 index 000000000..7496f0035 --- /dev/null +++ b/comfy/ldm/hidream_o1/conditioning.py @@ -0,0 +1,230 @@ +"""HiDream-O1 conditioning prep — ref-image dual path + extra_conds assembly. + +Each ref image goes through two paths: a 32x32 patchified stream concatenated +to the noised target, and a Qwen3-VL ViT path producing tokens that scatter +into input_ids at <|image_pad|> positions. +""" + +from typing import List + +import torch + +import comfy.utils +from comfy.text_encoders.qwen_vl import process_qwen2vl_images + +from .utils import (PATCH_SIZE, calculate_dimensions, cond_image_size, ref_max_size, resize_tensor) + +# Qwen3-VL ViT preprocessing constants (preprocessor_config.json). +VIT_PATCH = 16 +VIT_MERGE = 2 +VIT_IMAGE_MEAN = [0.5, 0.5, 0.5] +VIT_IMAGE_STD = [0.5, 0.5, 0.5] + + +def prepare_ref_images( + ref_images: List[torch.Tensor], + target_h: int, + target_w: int, + device: torch.device, + dtype: torch.dtype, +): + """Build the dual-path tensors for K reference images at (target_h, target_w). + + Returns None for K=0, else a dict with ref_patches, ref_pixel_values, + ref_image_grid_thw, per_ref_vit_tokens, per_ref_patch_grids. + """ + K = len(ref_images) + if K == 0: + return None + max_size = ref_max_size(max(target_h, target_w), K) + cis = cond_image_size(K) + + refs_t = [img[0].clamp(0, 1).permute(2, 0, 1).unsqueeze(0).contiguous().float() for img in ref_images] + refs_t = [resize_tensor(t, max_size, PATCH_SIZE) for t in refs_t] + + # 32-patch path. + ref_patches_per = [] + per_ref_patch_grids = [] + for t in refs_t: + t_norm = (t.squeeze(0) - 0.5) / 0.5 # (3, H, W) in [-1, 1] + h_p, w_p = t_norm.shape[-2] // PATCH_SIZE, t_norm.shape[-1] // PATCH_SIZE + per_ref_patch_grids.append((h_p, w_p)) + patches = ( + t_norm.reshape(3, h_p, PATCH_SIZE, w_p, PATCH_SIZE) + .permute(1, 3, 0, 2, 4) + .reshape(h_p * w_p, 3 * PATCH_SIZE * PATCH_SIZE) + ) + ref_patches_per.append(patches) + ref_patches = torch.cat(ref_patches_per, dim=0).unsqueeze(0).to(device=device, dtype=dtype) + + # ViT path. + refs_vlm_t = [] + for t in refs_t: + _, _, h, w = t.shape + cond_w, cond_h = calculate_dimensions(cis, w / h) + cond_w = max(cond_w, VIT_PATCH * VIT_MERGE) + cond_h = max(cond_h, VIT_PATCH * VIT_MERGE) + refs_vlm_t.append(comfy.utils.common_upscale(t, cond_w, cond_h, "lanczos", "disabled")) + + pv_list, grid_list, per_ref_vit_tokens = [], [], [] + for t_v in refs_vlm_t: + pv, grid_thw = process_qwen2vl_images( + t_v.permute(0, 2, 3, 1), + min_pixels=0, max_pixels=10**12, + patch_size=VIT_PATCH, merge_size=VIT_MERGE, + image_mean=VIT_IMAGE_MEAN, image_std=VIT_IMAGE_STD, + ) + grid_thw = grid_thw[0] + pv_list.append(pv.to(device=device, dtype=dtype)) + grid_list.append(grid_thw.to(device=device)) + # Post-merge token count = number of <|image_pad|> tokens this image expands to in input_ids. + gh, gw = int(grid_thw[1].item()), int(grid_thw[2].item()) + per_ref_vit_tokens.append((gh // VIT_MERGE) * (gw // VIT_MERGE)) + + return { + "ref_patches": ref_patches, + "ref_pixel_values": torch.cat(pv_list, dim=0), + "ref_image_grid_thw": torch.stack(grid_list, dim=0), + "per_ref_vit_tokens": per_ref_vit_tokens, + "per_ref_patch_grids": per_ref_patch_grids, + } + + +def build_ref_input_ids( + text_input_ids: torch.Tensor, + per_ref_vit_tokens: List[int], + image_token_id: int, + vision_start_id: int, + vision_end_id: int, +): + """Splice [vision_start, image_pad*N, vision_end] blocks into input_ids + after the [im_start, user, \\n] prefix (matches original chat template). + """ + ids = text_input_ids[0].tolist() + inserted = [] + for n_pad in per_ref_vit_tokens: + inserted.extend([vision_start_id] + [image_token_id] * n_pad + [vision_end_id]) + new_ids = ids[:3] + inserted + ids[3:] # 3 = len([im_start, user, \n]) + return torch.tensor([new_ids], dtype=text_input_ids.dtype, device=text_input_ids.device) + + +def build_extra_conds( + text_input_ids: torch.Tensor, + noise: torch.Tensor, + ref_images: List[torch.Tensor] = None, + target_patch_size: int = 32, +): + """Assemble all conditioning tensors for HiDreamO1Transformer.forward: + input_ids (with ref-vision tokens spliced in for the edit/IP path), + position_ids (MRoPE), token_types, vinput_mask, plus the ref + dual-path tensors when refs are provided. + """ + from .utils import get_rope_index_fix_point + from comfy.text_encoders.hidream_o1 import ( + IMAGE_TOKEN_ID, VISION_START_ID, VISION_END_ID, + ) + + if text_input_ids.dim() == 1: + text_input_ids = text_input_ids.unsqueeze(0) + text_input_ids = text_input_ids.long().to(noise.device) + B = noise.shape[0] + if text_input_ids.shape[0] == 1 and B > 1: + text_input_ids = text_input_ids.expand(B, -1) + + H, W = noise.shape[-2], noise.shape[-1] + h_p, w_p = H // target_patch_size, W // target_patch_size + image_len = h_p * w_p + image_grid_thw_tgt = torch.tensor( + [[1, h_p, w_p]], dtype=torch.long, device=text_input_ids.device, + ) + + out = {} + if ref_images: + ref = prepare_ref_images(ref_images, H, W, device=noise.device, dtype=noise.dtype) + text_input_ids = build_ref_input_ids( + text_input_ids, ref["per_ref_vit_tokens"], + IMAGE_TOKEN_ID, VISION_START_ID, VISION_END_ID, + ) + new_txt_len = text_input_ids.shape[1] + + # Each ref's patchified stream gets a [vision_start, image_pad*N-1] + # block in the position-id stream after the noised target. + ref_grid_lengths = [hp * wp for (hp, wp) in ref["per_ref_patch_grids"]] + tgt_vision = torch.full((1, image_len), IMAGE_TOKEN_ID, + dtype=text_input_ids.dtype, device=text_input_ids.device) + tgt_vision[:, 0] = VISION_START_ID + ref_vision_blocks = [] + for rl in ref_grid_lengths: + blk = torch.full((1, rl), IMAGE_TOKEN_ID, + dtype=text_input_ids.dtype, device=text_input_ids.device) + blk[:, 0] = VISION_START_ID + ref_vision_blocks.append(blk) + ref_vision_cat = torch.cat([tgt_vision] + ref_vision_blocks, dim=1) + input_ids_pad = torch.cat([text_input_ids, ref_vision_cat], dim=-1) + total_ref_patches_len = sum(ref_grid_lengths) + total_len = new_txt_len + image_len + total_ref_patches_len + + # K (ViT, post-merge) + 1 (target) + K (ref-patches) image grids. + K = len(ref_images) + igthw_cond = ref["ref_image_grid_thw"].clone() + igthw_cond[:, 1] //= 2 + igthw_cond[:, 2] //= 2 + image_grid_thw_ref = torch.tensor( + [[1, hp, wp] for (hp, wp) in ref["per_ref_patch_grids"]], + dtype=torch.long, device=text_input_ids.device, + ) + igthw_all = torch.cat([ + igthw_cond.to(text_input_ids.device), + image_grid_thw_tgt, + image_grid_thw_ref, + ], dim=0) + position_ids, _ = get_rope_index_fix_point( + spatial_merge_size=1, + image_token_id=IMAGE_TOKEN_ID, + vision_start_token_id=VISION_START_ID, + input_ids=input_ids_pad, image_grid_thw=igthw_all, + attention_mask=None, + skip_vision_start_token=[0] * K + [1] + [1] * K, + fix_point=4096, + ) + + # tms + target_image + ref_patches are all gen. + tms_pos = new_txt_len - 1 + ar_len = tms_pos + token_types = torch.zeros(B, total_len, dtype=torch.long, device=noise.device) + token_types[:, tms_pos:] = 1 + vinput_mask = torch.zeros(B, total_len, dtype=torch.bool, device=noise.device) + vinput_mask[:, new_txt_len:] = True + + # Leading batch dim sidesteps CONDRegular.process_cond's repeat_to_batch_size truncation + out["ref_pixel_values"] = ref["ref_pixel_values"].unsqueeze(0) + out["ref_image_grid_thw"] = ref["ref_image_grid_thw"].unsqueeze(0) + out["ref_patches"] = ref["ref_patches"] + else: + # T2I: text + noised target only, vision_start replaces the first image token + txt_len = text_input_ids.shape[1] + total_len = txt_len + image_len + vision_tokens = torch.full((B, image_len), IMAGE_TOKEN_ID, + dtype=text_input_ids.dtype, device=text_input_ids.device) + vision_tokens[:, 0] = VISION_START_ID + input_ids_pad = torch.cat([text_input_ids, vision_tokens], dim=-1) + position_ids, _ = get_rope_index_fix_point( + spatial_merge_size=1, + image_token_id=IMAGE_TOKEN_ID, + vision_start_token_id=VISION_START_ID, + input_ids=input_ids_pad, image_grid_thw=image_grid_thw_tgt, + attention_mask=None, + skip_vision_start_token=[1], + ) + ar_len = txt_len - 1 + token_types = torch.zeros(B, total_len, dtype=torch.long, device=noise.device) + token_types[:, ar_len:] = 1 + vinput_mask = torch.zeros(B, total_len, dtype=torch.bool, device=noise.device) + vinput_mask[:, txt_len:] = True + + out["input_ids"] = text_input_ids + out["position_ids"] = position_ids[:, 0].unsqueeze(0) # Collapse position_ids batch and add a leading dim so CONDRegular's batch-resize doesn't truncate the 3-axis MRoPE dim + out["token_types"] = token_types + out["vinput_mask"] = vinput_mask + out["ar_len"] = ar_len + return out diff --git a/comfy/ldm/hidream_o1/model.py b/comfy/ldm/hidream_o1/model.py new file mode 100644 index 000000000..a223e706f --- /dev/null +++ b/comfy/ldm/hidream_o1/model.py @@ -0,0 +1,306 @@ +"""HiDream-O1-Image transformer. + +Pixel-space DiT built on Qwen3-VL: the vision tower (Qwen35VisionModel) +encodes ref images, the Qwen3-VL-8B decoder (Llama2_ with interleaved MRoPE) +processes a unified text+image sequence, and 32x32 patch embed/unembed +shims map raw RGB in and out of LLM hidden space. The Qwen3-VL deepstack +mergers go unused — their weights are dropped at load. +""" + +from dataclasses import dataclass, field +from typing import List, Optional + +import einops +import torch +import torch.nn as nn + +import comfy.patcher_extension +from comfy.ldm.modules.diffusionmodules.mmdit import TimestepEmbedder +from comfy.text_encoders.llama import Llama2_ +from comfy.text_encoders.qwen35 import Qwen35VisionModel + +from .attention import make_two_pass_attention + + +IMAGE_TOKEN_ID = 151655 # Qwen3-VL <|image_pad|> +TMS_TOKEN_ID = 151673 # HiDream-O1 <|tms_token|> +PATCH_SIZE = 32 + + +@dataclass +class HiDreamO1TextConfig: + """Qwen3-VL-8B text-decoder dims (matches public Qwen3-VL-8B-Instruct).""" + vocab_size: int = 151936 + hidden_size: int = 4096 + intermediate_size: int = 12288 + num_hidden_layers: int = 36 + num_attention_heads: int = 32 + num_key_value_heads: int = 8 + head_dim: int = 128 + max_position_embeddings: int = 128000 + rms_norm_eps: float = 1e-6 + rope_theta: float = 5000000.0 + rope_scale: Optional[float] = None + rope_dims: List[int] = field(default_factory=lambda: [24, 20, 20]) + interleaved_mrope: bool = True + transformer_type: str = "llama" + rms_norm_add: bool = False + mlp_activation: str = "silu" + qkv_bias: bool = False + q_norm: str = "gemma3" + k_norm: str = "gemma3" + final_norm: bool = True + lm_head: bool = False + stop_tokens: List[int] = field(default_factory=lambda: [151643, 151645]) + + +QWEN3VL_VISION_DEFAULTS = dict( + hidden_size=1152, + num_heads=16, + intermediate_size=4304, + depth=27, + patch_size=16, + temporal_patch_size=2, + in_channels=3, + spatial_merge_size=2, + num_position_embeddings=2304, + deepstack_visual_indexes=(8, 16, 24), + out_hidden_size=4096, # final merger projects directly into LLM hidden +) + + +class BottleneckPatchEmbed(nn.Module): + # 3072 -> 1024 -> 4096 (raw 32x32 RGB patch -> bottleneck -> LLM hidden). + def __init__(self, patch_size=32, in_chans=3, pca_dim=1024, embed_dim=4096, bias=True, device=None, dtype=None, ops=None): + super().__init__() + self.proj1 = ops.Linear(patch_size * patch_size * in_chans, pca_dim, bias=False, device=device, dtype=dtype) + self.proj2 = ops.Linear(pca_dim, embed_dim, bias=bias, device=device, dtype=dtype) + + def forward(self, x): + return self.proj2(self.proj1(x)) + + +class FinalLayer(nn.Module): + # 4096 -> 3072 (LLM hidden -> flat pixel patch). + def __init__(self, hidden_size, patch_size=32, out_channels=3, device=None, dtype=None, ops=None): + super().__init__() + self.linear = ops.Linear(hidden_size, patch_size * patch_size * out_channels, bias=True, device=device, dtype=dtype) + + def forward(self, x): + return self.linear(x) + + +class HiDreamO1Transformer(nn.Module): + """HiDream-O1 unified pixel-level transformer.""" + + def __init__(self, image_model=None, dtype=None, device=None, operations=None, + text_config_overrides=None, vision_config_overrides=None, **kwargs): + super().__init__() + self.dtype = dtype + + text_cfg = HiDreamO1TextConfig(**(text_config_overrides or {})) + vision_cfg = dict(QWEN3VL_VISION_DEFAULTS) + if vision_config_overrides: + vision_cfg.update(vision_config_overrides) + vision_cfg["out_hidden_size"] = text_cfg.hidden_size + + self.text_config = text_cfg + self.vision_config = vision_cfg + self.hidden_size = text_cfg.hidden_size + self.patch_size = PATCH_SIZE + self.in_channels = 3 + self.tms_token_id = TMS_TOKEN_ID + + self.visual = Qwen35VisionModel(vision_cfg, device=device, dtype=dtype, ops=operations) + self.language_model = Llama2_(text_cfg, device=device, dtype=dtype, ops=operations) + self.t_embedder1 = TimestepEmbedder( + text_cfg.hidden_size, device=device, dtype=dtype, operations=operations, + ) + self.x_embedder = BottleneckPatchEmbed( + patch_size=self.patch_size, in_chans=self.in_channels, + pca_dim=text_cfg.hidden_size // 4, embed_dim=text_cfg.hidden_size, + bias=True, device=device, dtype=dtype, ops=operations, + ) + self.final_layer2 = FinalLayer( + text_cfg.hidden_size, patch_size=self.patch_size, + out_channels=self.in_channels, device=device, dtype=dtype, ops=operations, + ) + + self._visual_cache = None + self._kv_cache_entries = [] + + def clear_kv_cache(self): + self._kv_cache_entries = [] + self._visual_cache = None + + def forward(self, x, timesteps, context=None, transformer_options={}, **kwargs): + return comfy.patcher_extension.WrapperExecutor.new_class_executor( + self._forward, + self, + comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.DIFFUSION_MODEL, transformer_options) + ).execute(x, timesteps, context, transformer_options, **kwargs) + + def _forward(self, x, timesteps, context=None, transformer_options={}, input_ids=None, attention_mask=None, position_ids=None, + vinput_mask=None, ar_len=None, ref_pixel_values=None, ref_image_grid_thw=None, ref_patches=None, **kwargs): + """Returns flow-match velocity (x - x_pred) / sigma""" + + if input_ids is None or position_ids is None: + raise ValueError("HiDreamO1Transformer requires input_ids and position_ids in conditioning") + + B, _, H, W = x.shape + h_p, w_p = H // self.patch_size, W // self.patch_size + tgt_image_len = h_p * w_p + + z = einops.rearrange( + x, 'B C (H p1) (W p2) -> B (H W) (C p1 p2)', + p1=self.patch_size, p2=self.patch_size, + ) + vinputs = torch.cat([z, ref_patches.to(z.dtype)], dim=1) if ref_patches is not None else z + + inputs_embeds = self.language_model.embed_tokens(input_ids).to(x.dtype) + + if ref_pixel_values is not None and ref_image_grid_thw is not None: + # ViT output is constant across sampling steps within a generation + # identity-key by the input tensor so refs don't recompute every step. + cached = self._visual_cache + if cached is not None and cached[0] is ref_pixel_values: + image_embeds = cached[1] + else: + ref_pv = ref_pixel_values.to(inputs_embeds.device) + ref_grid = ref_image_grid_thw.to(inputs_embeds.device).long() + # extra_conds wraps with a leading batch dim; refs are model-level so [0] always recovers them. + if ref_pv.dim() == 3: + ref_pv = ref_pv[0] + if ref_grid.dim() == 3: + ref_grid = ref_grid[0] + image_embeds = self.visual(ref_pv, ref_grid).to(inputs_embeds.dtype) + self._visual_cache = (ref_pixel_values, image_embeds) + # image_pad positions identical across batch (input_ids shared cond/uncond). + image_idx = (input_ids[0] == IMAGE_TOKEN_ID).nonzero(as_tuple=True)[0] + if image_idx.shape[0] != image_embeds.shape[0]: + raise ValueError( + f"Image-token count {image_idx.shape[0]} != ViT output count " + f"{image_embeds.shape[0]}; check tokenizer/processor alignment." + ) + inputs_embeds[:, image_idx] = image_embeds.unsqueeze(0).expand(B, -1, -1) + + sigma = timesteps.float() / 1000.0 + t_pixeldit = 1.0 - sigma + t_emb = self.t_embedder1(t_pixeldit * 1000, inputs_embeds.dtype) + tms_mask_3d = (input_ids == self.tms_token_id).unsqueeze(-1).expand_as(inputs_embeds) + inputs_embeds = torch.where(tms_mask_3d, t_emb.unsqueeze(1).expand_as(inputs_embeds), inputs_embeds) + + vinputs_embedded = self.x_embedder(vinputs.to(inputs_embeds.dtype)) + inputs_embeds = torch.cat([inputs_embeds, vinputs_embedded], dim=1) + + # extra_conds stores position_ids as (1, 3, T); process_cond repeats dim 0 to B. Take row 0. + freqs_cis = self.language_model.compute_freqs_cis(position_ids[0].to(x.device), x.device) + freqs_cis = tuple(t.to(x.dtype) for t in freqs_cis) + + two_pass_attn = make_two_pass_attention(ar_len, transformer_options=transformer_options) + patches_replace = transformer_options.get("patches_replace", {}) + blocks_replace = patches_replace.get("dit", {}) + transformer_options["total_blocks"] = len(self.language_model.layers) + transformer_options["block_type"] = "double" + + # Cache prefix K/V across steps. Key includes input_ids (prompt), ref_id + # (refs scatter into inputs_embeds), and position_ids (RoPE baked into cached K). + can_cache = not blocks_replace and ar_len > 0 + cache_len = ar_len if can_cache else 0 + ref_id = id(ref_pixel_values) if ref_pixel_values is not None else None + pos_ids_key = position_ids[..., :cache_len] if can_cache else position_ids + cache_entries = self._kv_cache_entries + # Drop stale entries from a previous device (model was unloaded and reloaded). + if cache_entries and cache_entries[0]["input_ids"].device != input_ids.device: + cache_entries = [] + self._kv_cache_entries = [] + kv_cache = None + if can_cache: + for entry in cache_entries: + ck = entry["input_ids"] + ep = entry["position_ids"] + if (entry["cache_len"] == cache_len + and ck.shape == input_ids.shape and torch.equal(ck, input_ids) + and entry["ref_id"] == ref_id + and ep.shape == pos_ids_key.shape and torch.equal(ep, pos_ids_key)): + kv_cache = entry + break + + if kv_cache is not None: + # Hot path: project Q/K/V only for fresh positions; past_key_value prepends cached AR K/V. + hidden_states = inputs_embeds[:, cache_len:] + sliced_freqs = tuple(t[..., cache_len:, :] for t in freqs_cis) + for i, layer in enumerate(self.language_model.layers): + transformer_options["block_index"] = i + K_i, V_i = kv_cache["kv"][i] + hidden_states, _ = layer( + x=hidden_states, attention_mask=None, freqs_cis=sliced_freqs, optimized_attention=two_pass_attn, + past_key_value=(K_i, V_i, cache_len), + ) + else: + # Cold path: run full sequence; if cacheable, snapshot K/V at AR positions. + snapshots = [] if can_cache else None + past_kv_cold = () if can_cache else None + hidden_states = inputs_embeds + for i, layer in enumerate(self.language_model.layers): + transformer_options["block_index"] = i + if ("double_block", i) in blocks_replace: + def block_wrap(args, _layer=layer): + out = {} + out["x"], _ = _layer( + x=args["x"], attention_mask=args.get("attention_mask"), + freqs_cis=args["freqs_cis"], optimized_attention=args["optimized_attention"], + past_key_value=None, + ) + return out + out = blocks_replace[("double_block", i)]( + {"x": hidden_states, "attention_mask": None, + "freqs_cis": freqs_cis, "optimized_attention": two_pass_attn, + "transformer_options": transformer_options}, + {"original_block": block_wrap}, + ) + hidden_states = out["x"] + else: + hidden_states, present_kv = layer( + x=hidden_states, attention_mask=None, + freqs_cis=freqs_cis, optimized_attention=two_pass_attn, + past_key_value=past_kv_cold, + ) + if snapshots is not None: + K, V, _ = present_kv + snapshots.append((K[:, :, :cache_len].contiguous(), + V[:, :, :cache_len].contiguous())) + if snapshots is not None: + # Cap at 2 entries (cond + uncond). Multi-cond workflows LRU-evict. + new_entry = { + "input_ids": input_ids.clone(), + "cache_len": cache_len, + "kv": snapshots, + "ref_id": ref_id, + "position_ids": pos_ids_key.clone(), + } + self._kv_cache_entries = (cache_entries + [new_entry])[-2:] + + if self.language_model.norm is not None: + hidden_states = self.language_model.norm(hidden_states) + + # Slice target-image positions before the final projection so the Linear only runs on tgt_image_len tokens. + # In the hot path hidden_states starts at original position cache_len, so masks/indices shift by cache_len. + sliced_offset = cache_len if kv_cache is not None else 0 + if vinput_mask is not None: + vmask = vinput_mask.to(x.device).bool() + if sliced_offset > 0: + vmask = vmask[:, sliced_offset:] + target_hidden = hidden_states[vmask].view(B, -1, hidden_states.shape[-1])[:, :tgt_image_len] + else: + txt_seq_len = input_ids.shape[1] + start = txt_seq_len - sliced_offset + target_hidden = hidden_states[:, start:start + tgt_image_len] + x_pred_tgt = self.final_layer2(target_hidden) + + # fp32 final subtraction, bf16 here noticeably degrades samples. + x_pred_img = einops.rearrange( + x_pred_tgt, 'B (H W) (C p1 p2) -> B C (H p1) (W p2)', + H=h_p, W=w_p, p1=self.patch_size, p2=self.patch_size, + ) + return (x.float() - x_pred_img.float()) / sigma.view(B, 1, 1, 1).clamp_min(1e-3) diff --git a/comfy/ldm/hidream_o1/utils.py b/comfy/ldm/hidream_o1/utils.py new file mode 100644 index 000000000..5a1249c72 --- /dev/null +++ b/comfy/ldm/hidream_o1/utils.py @@ -0,0 +1,173 @@ +"""HiDream-O1 input-prep helpers: image/resolution math and unified-sequence +RoPE position-id assembly. The fix_point offset in get_rope_index_fix_point +lets the target image and patchified ref images share spatial RoPE positions +despite living at different sequence indices — same 2D image plane. +""" + +import math +from typing import Optional + +import torch + + +PATCH_SIZE = 32 +CONDITION_IMAGE_SIZE = 384 # ViT-side base size for ref images + + +def resize_tensor(img_t, image_size, patch_size=16): + """img_t: (1, 3, H, W) float [0, 1]. Fit to image_size**2 area, patch-aligned, center-cropped.""" + + while min(img_t.shape[-2], img_t.shape[-1]) >= 2 * image_size: # Pre-halves with 2x2 box averaging while the image is still very large + img_t = torch.nn.functional.avg_pool2d(img_t, kernel_size=2, stride=2) + + _, _, height, width = img_t.shape + m = patch_size + s_max = image_size * image_size + scale = math.sqrt(s_max / (width * height)) + + candidates = [ + (round(width * scale) // m * m, round(height * scale) // m * m), + (round(width * scale) // m * m, math.floor(height * scale) // m * m), + (math.floor(width * scale) // m * m, round(height * scale) // m * m), + (math.floor(width * scale) // m * m, math.floor(height * scale) // m * m), + ] + candidates = sorted(candidates, key=lambda x: x[0] * x[1], reverse=True) + new_size = candidates[-1] + for c in candidates: + if c[0] * c[1] <= s_max: + new_size = c + break + + new_w, new_h = new_size + s1 = width / new_w + s2 = height / new_h + if s1 < s2: + resize_w, resize_h = new_w, round(height / s1) + else: + resize_w, resize_h = round(width / s2), new_h + img_t = torch.nn.functional.interpolate(img_t, size=(resize_h, resize_w), mode="bicubic") + top = (resize_h - new_h) // 2 + left = (resize_w - new_w) // 2 + return img_t[..., top:top + new_h, left:left + new_w] + + +def calculate_dimensions(max_size, ratio): + """(W, H) for an aspect ratio fitting in max_size**2 area, 32-aligned.""" + width = math.sqrt(max_size * max_size * ratio) + height = width / ratio + width = int(width / 32) * 32 + height = int(height / 32) * 32 + return width, height + + +def ref_max_size(target_max_dim, k): + """K-dependent ref-image max dim before patchifying.""" + if k == 1: + return target_max_dim + if k == 2: + return target_max_dim * 48 // 64 + if k <= 4: + return target_max_dim // 2 + if k <= 8: + return target_max_dim * 24 // 64 + return target_max_dim // 4 + + +def cond_image_size(k): + """K-dependent ViT-side image size.""" + if k <= 4: + return CONDITION_IMAGE_SIZE + if k <= 8: + return CONDITION_IMAGE_SIZE * 48 // 64 + return CONDITION_IMAGE_SIZE // 2 + + +def get_rope_index_fix_point( + spatial_merge_size: int, + image_token_id: int, + vision_start_token_id: int, + input_ids: Optional[torch.LongTensor] = None, + image_grid_thw: Optional[torch.LongTensor] = None, + attention_mask: Optional[torch.Tensor] = None, + skip_vision_start_token=None, + fix_point: int = 4096, +): + mrope_position_deltas = [] + if input_ids is not None and image_grid_thw is not None: + total_input_ids = input_ids + if attention_mask is None: + attention_mask = torch.ones_like(total_input_ids) + position_ids = torch.ones( + 3, input_ids.shape[0], input_ids.shape[1], + dtype=input_ids.dtype, device=input_ids.device, + ) + attention_mask = attention_mask.to(total_input_ids.device) + for i, input_ids_b in enumerate(total_input_ids): + fp = fix_point + image_index = 0 + input_ids_b = input_ids_b[attention_mask[i] == 1] + vision_start_indices = torch.argwhere(input_ids_b == vision_start_token_id).squeeze(1) + vision_tokens = input_ids_b[vision_start_indices + 1] + image_nums = (vision_tokens == image_token_id).sum() + input_tokens = input_ids_b.tolist() + llm_pos_ids_list = [] + st = 0 + remain_images = image_nums + for _ in range(image_nums): + if image_token_id in input_tokens and remain_images > 0: + ed = input_tokens.index(image_token_id, st) + else: + ed = len(input_tokens) + 1 + t = image_grid_thw[image_index][0] + h = image_grid_thw[image_index][1] + w = image_grid_thw[image_index][2] + image_index += 1 + remain_images -= 1 + llm_grid_t = t.item() + llm_grid_h = h.item() // spatial_merge_size + llm_grid_w = w.item() // spatial_merge_size + text_len = ed - st + text_len -= skip_vision_start_token[image_index - 1] + text_len = max(0, text_len) + st_idx = llm_pos_ids_list[-1].max() + 1 if len(llm_pos_ids_list) > 0 else 0 + llm_pos_ids_list.append(torch.arange(text_len).view(1, -1).expand(3, -1) + st_idx) + + t_index = torch.arange(llm_grid_t).view(-1, 1).expand(-1, llm_grid_h * llm_grid_w).flatten() + h_index = torch.arange(llm_grid_h).view(1, -1, 1).expand(llm_grid_t, -1, llm_grid_w).flatten() + w_index = torch.arange(llm_grid_w).view(1, 1, -1).expand(llm_grid_t, llm_grid_h, -1).flatten() + + if skip_vision_start_token[image_index - 1]: + if fp > 0: + fp = fp - st_idx + llm_pos_ids_list.append(torch.stack([t_index, h_index, w_index]) + fp + st_idx) + fp = 0 + else: + llm_pos_ids_list.append(torch.stack([t_index, h_index, w_index]) + text_len + st_idx) + st = ed + llm_grid_t * llm_grid_h * llm_grid_w + + if st < len(input_tokens): + st_idx = llm_pos_ids_list[-1].max() + 1 if len(llm_pos_ids_list) > 0 else 0 + text_len = len(input_tokens) - st + llm_pos_ids_list.append(torch.arange(text_len).view(1, -1).expand(3, -1) + st_idx) + + llm_positions = torch.cat(llm_pos_ids_list, dim=1).reshape(3, -1) + position_ids[..., i, attention_mask[i] == 1] = llm_positions.to(position_ids.device) + mrope_position_deltas.append(llm_positions.max() + 1 - len(total_input_ids[i])) + mrope_position_deltas = torch.tensor(mrope_position_deltas, device=input_ids.device).unsqueeze(1) + return position_ids, mrope_position_deltas + + if attention_mask is not None: + position_ids = attention_mask.long().cumsum(-1) - 1 + position_ids.masked_fill_(attention_mask == 0, 1) + position_ids = position_ids.unsqueeze(0).expand(3, -1, -1).to(attention_mask.device) + max_position_ids = position_ids.max(0, keepdim=False)[0].max(-1, keepdim=True)[0] + mrope_position_deltas = max_position_ids + 1 - attention_mask.shape[-1] + else: + position_ids = ( + torch.arange(input_ids.shape[1], device=input_ids.device) + .view(1, 1, -1).expand(3, input_ids.shape[0], -1) + ) + mrope_position_deltas = torch.zeros( + [input_ids.shape[0], 1], device=input_ids.device, dtype=input_ids.dtype, + ) + return position_ids, mrope_position_deltas diff --git a/comfy/ldm/lightricks/av_model.py b/comfy/ldm/lightricks/av_model.py index 6f2ba41ef..3fb87b4a3 100644 --- a/comfy/ldm/lightricks/av_model.py +++ b/comfy/ldm/lightricks/av_model.py @@ -16,6 +16,7 @@ from comfy.ldm.lightricks.model import ( from comfy.ldm.lightricks.symmetric_patchifier import AudioPatchifier from comfy.ldm.lightricks.embeddings_connector import Embeddings1DConnector import comfy.ldm.common_dit +import comfy.model_prefetch class CompressedTimestep: """Store video timestep embeddings in compressed form using per-frame indexing.""" @@ -907,9 +908,11 @@ class LTXAVModel(LTXVModel): """Process transformer blocks for LTXAV.""" patches_replace = transformer_options.get("patches_replace", {}) blocks_replace = patches_replace.get("dit", {}) + prefetch_queue = comfy.model_prefetch.make_prefetch_queue(list(self.transformer_blocks), vx.device, transformer_options) # Process transformer blocks for i, block in enumerate(self.transformer_blocks): + comfy.model_prefetch.prefetch_queue_pop(prefetch_queue, vx.device, block) if ("double_block", i) in blocks_replace: def block_wrap(args): @@ -982,6 +985,8 @@ class LTXAVModel(LTXVModel): a_prompt_timestep=a_prompt_timestep, ) + comfy.model_prefetch.prefetch_queue_pop(prefetch_queue, vx.device, None) + return [vx, ax] def _process_output(self, x, embedded_timestep, keyframe_idxs, **kwargs): diff --git a/comfy/ldm/lightricks/vae/audio_vae.py b/comfy/ldm/lightricks/vae/audio_vae.py index fa0a00748..dd5320c8f 100644 --- a/comfy/ldm/lightricks/vae/audio_vae.py +++ b/comfy/ldm/lightricks/vae/audio_vae.py @@ -4,9 +4,6 @@ import math import torch import torchaudio -import comfy.model_management -import comfy.model_patcher -import comfy.utils as utils from comfy.ldm.mmaudio.vae.distributions import DiagonalGaussianDistribution from comfy.ldm.lightricks.symmetric_patchifier import AudioPatchifier from comfy.ldm.lightricks.vae.causal_audio_autoencoder import ( @@ -43,30 +40,6 @@ class AudioVAEComponentConfig: return cls(autoencoder=audio_config, vocoder=vocoder_config) - -class ModelDeviceManager: - """Manages device placement and GPU residency for the composed model.""" - - def __init__(self, module: torch.nn.Module): - load_device = comfy.model_management.get_torch_device() - offload_device = comfy.model_management.vae_offload_device() - self.patcher = comfy.model_patcher.ModelPatcher(module, load_device, offload_device) - - def ensure_model_loaded(self) -> None: - comfy.model_management.free_memory( - self.patcher.model_size(), - self.patcher.load_device, - ) - comfy.model_management.load_model_gpu(self.patcher) - - def move_to_load_device(self, tensor: torch.Tensor) -> torch.Tensor: - return tensor.to(self.patcher.load_device) - - @property - def load_device(self): - return self.patcher.load_device - - class AudioLatentNormalizer: """Applies per-channel statistics in patch space and restores original layout.""" @@ -132,23 +105,17 @@ class AudioPreprocessor: class AudioVAE(torch.nn.Module): """High-level Audio VAE wrapper exposing encode and decode entry points.""" - def __init__(self, state_dict: dict, metadata: dict): + def __init__(self, metadata: dict): super().__init__() component_config = AudioVAEComponentConfig.from_metadata(metadata) - vae_sd = utils.state_dict_prefix_replace(state_dict, {"audio_vae.": ""}, filter_keys=True) - vocoder_sd = utils.state_dict_prefix_replace(state_dict, {"vocoder.": ""}, filter_keys=True) - self.autoencoder = CausalAudioAutoencoder(config=component_config.autoencoder) if "bwe" in component_config.vocoder: self.vocoder = VocoderWithBWE(config=component_config.vocoder) else: self.vocoder = Vocoder(config=component_config.vocoder) - self.autoencoder.load_state_dict(vae_sd, strict=False) - self.vocoder.load_state_dict(vocoder_sd, strict=False) - autoencoder_config = self.autoencoder.get_config() self.normalizer = AudioLatentNormalizer( AudioPatchifier( @@ -168,18 +135,12 @@ class AudioVAE(torch.nn.Module): n_fft=autoencoder_config["n_fft"], ) - self.device_manager = ModelDeviceManager(self) - - def encode(self, audio: dict) -> torch.Tensor: + def encode(self, audio, sample_rate=44100) -> torch.Tensor: """Encode a waveform dictionary into normalized latent tensors.""" - waveform = audio["waveform"] - waveform_sample_rate = audio["sample_rate"] + waveform = audio + waveform_sample_rate = sample_rate input_device = waveform.device - # Ensure that Audio VAE is loaded on the correct device. - self.device_manager.ensure_model_loaded() - - waveform = self.device_manager.move_to_load_device(waveform) expected_channels = self.autoencoder.encoder.in_channels if waveform.shape[1] != expected_channels: if waveform.shape[1] == 1: @@ -190,7 +151,7 @@ class AudioVAE(torch.nn.Module): ) mel_spec = self.preprocessor.waveform_to_mel( - waveform, waveform_sample_rate, device=self.device_manager.load_device + waveform, waveform_sample_rate, device=waveform.device ) latents = self.autoencoder.encode(mel_spec) @@ -204,17 +165,13 @@ class AudioVAE(torch.nn.Module): """Decode normalized latent tensors into an audio waveform.""" original_shape = latents.shape - # Ensure that Audio VAE is loaded on the correct device. - self.device_manager.ensure_model_loaded() - - latents = self.device_manager.move_to_load_device(latents) latents = self.normalizer.denormalize(latents) target_shape = self.target_shape_from_latents(original_shape) mel_spec = self.autoencoder.decode(latents, target_shape=target_shape) waveform = self.run_vocoder(mel_spec) - return self.device_manager.move_to_load_device(waveform) + return waveform def target_shape_from_latents(self, latents_shape): batch, _, time, _ = latents_shape diff --git a/comfy/ldm/modules/attention.py b/comfy/ldm/modules/attention.py index b193fe5e8..a68cb8439 100644 --- a/comfy/ldm/modules/attention.py +++ b/comfy/ldm/modules/attention.py @@ -14,6 +14,8 @@ from .sub_quadratic_attention import efficient_dot_product_attention from comfy import model_management +TORCH_HAS_GQA = model_management.torch_version_numeric >= (2, 5) + if model_management.xformers_enabled(): import xformers import xformers.ops @@ -150,7 +152,12 @@ def attention_basic(q, k, v, heads, mask=None, attn_precision=None, skip_reshape b, _, dim_head = q.shape dim_head //= heads - scale = dim_head ** -0.5 + if kwargs.get("enable_gqa", False) and q.shape[-3] != k.shape[-3]: + n_rep = q.shape[-3] // k.shape[-3] + k = k.repeat_interleave(n_rep, dim=-3) + v = v.repeat_interleave(n_rep, dim=-3) + + scale = kwargs.get("scale", dim_head ** -0.5) h = heads if skip_reshape: @@ -219,6 +226,10 @@ def attention_sub_quad(query, key, value, heads, mask=None, attn_precision=None, b, _, dim_head = query.shape dim_head //= heads + if "scale" in kwargs: + # Pre-scale query to match requested scale (cancels internal 1/sqrt(dim_head)) + query = query * (kwargs["scale"] * dim_head ** 0.5) + if skip_reshape: query = query.reshape(b * heads, -1, dim_head) value = value.reshape(b * heads, -1, dim_head) @@ -290,7 +301,7 @@ def attention_split(q, k, v, heads, mask=None, attn_precision=None, skip_reshape b, _, dim_head = q.shape dim_head //= heads - scale = dim_head ** -0.5 + scale = kwargs.get("scale", dim_head ** -0.5) if skip_reshape: q, k, v = map( @@ -500,8 +511,13 @@ def attention_pytorch(q, k, v, heads, mask=None, attn_precision=None, skip_resha if mask.ndim == 3: mask = mask.unsqueeze(1) + # Pass through extra SDPA kwargs (scale, enable_gqa) if provided + # enable_gqa requires PyTorch 2.5+; older versions use manual KV expansion above + sdpa_keys = ("scale", "enable_gqa") if TORCH_HAS_GQA else ("scale",) + sdpa_extra = {k: v for k, v in kwargs.items() if k in sdpa_keys} + if SDP_BATCH_LIMIT >= b: - out = comfy.ops.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False) + out = comfy.ops.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False, **sdpa_extra) if not skip_output_reshape: out = ( out.transpose(1, 2).reshape(b, -1, heads * dim_head) @@ -519,7 +535,7 @@ def attention_pytorch(q, k, v, heads, mask=None, attn_precision=None, skip_resha k[i : i + SDP_BATCH_LIMIT], v[i : i + SDP_BATCH_LIMIT], attn_mask=m, - dropout_p=0.0, is_causal=False + dropout_p=0.0, is_causal=False, **sdpa_extra ).transpose(1, 2).reshape(-1, q.shape[2], heads * dim_head) return out diff --git a/comfy/ldm/modules/diffusionmodules/openaimodel.py b/comfy/ldm/modules/diffusionmodules/openaimodel.py index 295310df6..4b92c44cf 100644 --- a/comfy/ldm/modules/diffusionmodules/openaimodel.py +++ b/comfy/ldm/modules/diffusionmodules/openaimodel.py @@ -34,6 +34,16 @@ class TimestepBlock(nn.Module): #This is needed because accelerate makes a copy of transformer_options which breaks "transformer_index" def forward_timestep_embed(ts, x, emb, context=None, transformer_options={}, output_shape=None, time_context=None, num_video_frames=None, image_only_indicator=None): for layer in ts: + if "patches" in transformer_options and "forward_timestep_embed_patch" in transformer_options["patches"]: + found_patched = False + for class_type, handler in transformer_options["patches"]["forward_timestep_embed_patch"]: + if isinstance(layer, class_type): + x = handler(layer, x, emb, context, transformer_options, output_shape, time_context, num_video_frames, image_only_indicator) + found_patched = True + break + if found_patched: + continue + if isinstance(layer, VideoResBlock): x = layer(x, emb, num_video_frames, image_only_indicator) elif isinstance(layer, TimestepBlock): @@ -49,15 +59,6 @@ def forward_timestep_embed(ts, x, emb, context=None, transformer_options={}, out elif isinstance(layer, Upsample): x = layer(x, output_shape=output_shape) else: - if "patches" in transformer_options and "forward_timestep_embed_patch" in transformer_options["patches"]: - found_patched = False - for class_type, handler in transformer_options["patches"]["forward_timestep_embed_patch"]: - if isinstance(layer, class_type): - x = handler(layer, x, emb, context, transformer_options, output_shape, time_context, num_video_frames, image_only_indicator) - found_patched = True - break - if found_patched: - continue x = layer(x) return x @@ -894,6 +895,12 @@ class UNetModel(nn.Module): h = forward_timestep_embed(self.middle_block, h, emb, context, transformer_options, time_context=time_context, num_video_frames=num_video_frames, image_only_indicator=image_only_indicator) h = apply_control(h, control, 'middle') + if "middle_block_after_patch" in transformer_patches: + patch = transformer_patches["middle_block_after_patch"] + for p in patch: + out = p({"h": h, "x": x, "emb": emb, "context": context, "y": y, + "timesteps": timesteps, "transformer_options": transformer_options}) + h = out["h"] for id, module in enumerate(self.output_blocks): transformer_options["block"] = ("output", id) @@ -905,8 +912,9 @@ class UNetModel(nn.Module): for p in patch: h, hsp = p(h, hsp, transformer_options) - h = th.cat([h, hsp], dim=1) - del hsp + if hsp is not None: + h = th.cat([h, hsp], dim=1) + del hsp if len(hs) > 0: output_shape = hs[-1].shape else: diff --git a/comfy/ldm/modules/diffusionmodules/util.py b/comfy/ldm/modules/diffusionmodules/util.py index 233011dc9..aed5c149c 100644 --- a/comfy/ldm/modules/diffusionmodules/util.py +++ b/comfy/ldm/modules/diffusionmodules/util.py @@ -140,7 +140,7 @@ def make_ddim_sampling_parameters(alphacums, ddim_timesteps, eta, verbose=True): alphas = alphacums[ddim_timesteps] alphas_prev = np.asarray([alphacums[0]] + alphacums[ddim_timesteps[:-1]].tolist()) - # according the the formula provided in https://arxiv.org/abs/2010.02502 + # according to the formula provided in https://arxiv.org/abs/2010.02502 sigmas = eta * np.sqrt((1 - alphas_prev) / (1 - alphas) * (1 - alphas / alphas_prev)) if verbose: logging.info(f'Selected alphas for ddim sampler: a_t: {alphas}; a_(t-1): {alphas_prev}') diff --git a/comfy/ldm/modules/sdpose.py b/comfy/ldm/modules/sdpose.py index d67b60b76..1a9585fc2 100644 --- a/comfy/ldm/modules/sdpose.py +++ b/comfy/ldm/modules/sdpose.py @@ -90,7 +90,7 @@ class HeatmapHead(torch.nn.Module): origin_max = np.max(hm[k]) dr = np.zeros((H + 2 * border, W + 2 * border), dtype=np.float32) dr[border:-border, border:-border] = hm[k].copy() - dr = gaussian_filter(dr, sigma=2.0) + dr = gaussian_filter(dr, sigma=2.0, truncate=2.5) hm[k] = dr[border:-border, border:-border].copy() cur_max = np.max(hm[k]) if cur_max > 0: diff --git a/comfy/ldm/sam3/detector.py b/comfy/ldm/sam3/detector.py new file mode 100644 index 000000000..23a972ac7 --- /dev/null +++ b/comfy/ldm/sam3/detector.py @@ -0,0 +1,599 @@ +# SAM3 detector: transformer encoder-decoder, segmentation head, geometry encoder, scoring. + +import math + +import torch +import torch.nn as nn +import torch.nn.functional as F +from torchvision.ops import roi_align + +from comfy.ldm.modules.attention import optimized_attention +from comfy.ldm.sam3.tracker import SAM3Tracker, SAM31Tracker +from comfy.ldm.sam3.sam import SAM3VisionBackbone # noqa: used in __init__ +from comfy.ldm.sam3.sam import MLP, PositionEmbeddingSine + +TRACKER_CLASSES = {"SAM3": SAM3Tracker, "SAM31": SAM31Tracker} +from comfy.ops import cast_to_input + + +def box_cxcywh_to_xyxy(x): + cx, cy, w, h = x.unbind(-1) + return torch.stack([cx - 0.5 * w, cy - 0.5 * h, cx + 0.5 * w, cy + 0.5 * h], dim=-1) + + +def gen_sineembed_for_position(pos_tensor, num_feats=256): + """Per-coordinate sinusoidal embedding: (..., N) -> (..., N * num_feats).""" + assert num_feats % 2 == 0 + hdim = num_feats // 2 + freqs = 10000.0 ** (2 * (torch.arange(hdim, dtype=torch.float32, device=pos_tensor.device) // 2) / hdim) + embeds = [] + for c in range(pos_tensor.shape[-1]): + raw = (pos_tensor[..., c].float() * 2 * math.pi).unsqueeze(-1) / freqs + embeds.append(torch.stack([raw[..., 0::2].sin(), raw[..., 1::2].cos()], dim=-1).flatten(-2)) + return torch.cat(embeds, dim=-1).to(pos_tensor.dtype) + + +class SplitMHA(nn.Module): + """Multi-head attention with separate Q/K/V projections (split from fused in_proj_weight).""" + def __init__(self, d_model, num_heads=8, device=None, dtype=None, operations=None): + super().__init__() + self.num_heads = num_heads + self.q_proj = operations.Linear(d_model, d_model, device=device, dtype=dtype) + self.k_proj = operations.Linear(d_model, d_model, device=device, dtype=dtype) + self.v_proj = operations.Linear(d_model, d_model, device=device, dtype=dtype) + self.out_proj = operations.Linear(d_model, d_model, device=device, dtype=dtype) + + def forward(self, q_input, k_input=None, v_input=None, mask=None): + q = self.q_proj(q_input) + if k_input is None: + k = self.k_proj(q_input) + v = self.v_proj(q_input) + else: + k = self.k_proj(k_input) + v = self.v_proj(v_input if v_input is not None else k_input) + if mask is not None and mask.ndim == 2: + mask = mask[:, None, None, :] # [B, T] -> [B, 1, 1, T] for SDPA broadcast + dtype = q.dtype # manual_cast may produce mixed dtypes + out = optimized_attention(q, k.to(dtype), v.to(dtype), self.num_heads, mask=mask, low_precision_attention=False) + return self.out_proj(out) + + +class MLPWithNorm(nn.Module): + """MLP with residual connection and output LayerNorm.""" + def __init__(self, input_dim, hidden_dim, output_dim, num_layers, residual=True, device=None, dtype=None, operations=None): + super().__init__() + dims = [input_dim] + [hidden_dim] * (num_layers - 1) + [output_dim] + self.layers = nn.ModuleList([ + operations.Linear(dims[i], dims[i + 1], device=device, dtype=dtype) + for i in range(num_layers) + ]) + self.out_norm = operations.LayerNorm(output_dim, device=device, dtype=dtype) + self.residual = residual and (input_dim == output_dim) + + def forward(self, x): + orig = x + for i, layer in enumerate(self.layers): + x = layer(x) + if i < len(self.layers) - 1: + x = F.relu(x) + if self.residual: + x = x + orig + return self.out_norm(x) + + +class EncoderLayer(nn.Module): + def __init__(self, d_model=256, num_heads=8, dim_ff=2048, device=None, dtype=None, operations=None): + super().__init__() + self.self_attn = SplitMHA(d_model, num_heads, device=device, dtype=dtype, operations=operations) + self.cross_attn_image = SplitMHA(d_model, num_heads, device=device, dtype=dtype, operations=operations) + self.linear1 = operations.Linear(d_model, dim_ff, device=device, dtype=dtype) + self.linear2 = operations.Linear(dim_ff, d_model, device=device, dtype=dtype) + self.norm1 = operations.LayerNorm(d_model, device=device, dtype=dtype) + self.norm2 = operations.LayerNorm(d_model, device=device, dtype=dtype) + self.norm3 = operations.LayerNorm(d_model, device=device, dtype=dtype) + + def forward(self, x, pos, text_memory=None, text_mask=None): + normed = self.norm1(x) + q_k = normed + pos + x = x + self.self_attn(q_k, q_k, normed) + if text_memory is not None: + normed = self.norm2(x) + x = x + self.cross_attn_image(normed, text_memory, text_memory, mask=text_mask) + normed = self.norm3(x) + x = x + self.linear2(F.relu(self.linear1(normed))) + return x + + +class TransformerEncoder(nn.Module): + """Checkpoint: transformer.encoder.layers.N.*""" + def __init__(self, d_model=256, num_heads=8, dim_ff=2048, num_layers=6, device=None, dtype=None, operations=None): + super().__init__() + self.layers = nn.ModuleList([ + EncoderLayer(d_model, num_heads, dim_ff, device=device, dtype=dtype, operations=operations) + for _ in range(num_layers) + ]) + + def forward(self, x, pos, text_memory=None, text_mask=None): + for layer in self.layers: + x = layer(x, pos, text_memory, text_mask) + return x + + +class DecoderLayer(nn.Module): + def __init__(self, d_model=256, num_heads=8, dim_ff=2048, device=None, dtype=None, operations=None): + super().__init__() + self.self_attn = SplitMHA(d_model, num_heads, device=device, dtype=dtype, operations=operations) + self.cross_attn = SplitMHA(d_model, num_heads, device=device, dtype=dtype, operations=operations) + self.ca_text = SplitMHA(d_model, num_heads, device=device, dtype=dtype, operations=operations) + self.norm1 = operations.LayerNorm(d_model, device=device, dtype=dtype) + self.norm2 = operations.LayerNorm(d_model, device=device, dtype=dtype) + self.norm3 = operations.LayerNorm(d_model, device=device, dtype=dtype) + self.catext_norm = operations.LayerNorm(d_model, device=device, dtype=dtype) + self.linear1 = operations.Linear(d_model, dim_ff, device=device, dtype=dtype) + self.linear2 = operations.Linear(dim_ff, d_model, device=device, dtype=dtype) + + def forward(self, x, memory, x_pos, memory_pos, text_memory=None, text_mask=None, cross_attn_bias=None): + q_k = x + x_pos + x = self.norm2(x + self.self_attn(q_k, q_k, x)) + if text_memory is not None: + x = self.catext_norm(x + self.ca_text(x + x_pos, text_memory, text_memory, mask=text_mask)) + x = self.norm1(x + self.cross_attn(x + x_pos, memory + memory_pos, memory, mask=cross_attn_bias)) + x = self.norm3(x + self.linear2(F.relu(self.linear1(x)))) + return x + + +class TransformerDecoder(nn.Module): + def __init__(self, d_model=256, num_heads=8, dim_ff=2048, num_layers=6, + num_queries=200, device=None, dtype=None, operations=None): + super().__init__() + self.d_model = d_model + self.num_queries = num_queries + + self.layers = nn.ModuleList([ + DecoderLayer(d_model, num_heads, dim_ff, device=device, dtype=dtype, operations=operations) + for _ in range(num_layers) + ]) + self.norm = operations.LayerNorm(d_model, device=device, dtype=dtype) + self.query_embed = operations.Embedding(num_queries, d_model, device=device, dtype=dtype) + self.reference_points = operations.Embedding(num_queries, 4, device=device, dtype=dtype) # Reference points: Embedding(num_queries, 4) — learned anchor boxes + self.ref_point_head = MLP(d_model * 2, d_model, d_model, 2, device=device, dtype=dtype, operations=operations) # ref_point_head input: 512 (4 coords * 128 sine features each) + self.bbox_embed = MLP(d_model, d_model, 4, 3, device=device, dtype=dtype, operations=operations) + + self.boxRPB_embed_x = MLP(2, d_model, num_heads, 2, device=device, dtype=dtype, operations=operations) + self.boxRPB_embed_y = MLP(2, d_model, num_heads, 2, device=device, dtype=dtype, operations=operations) + + self.presence_token = operations.Embedding(1, d_model, device=device, dtype=dtype) + self.presence_token_head = MLP(d_model, d_model, 1, 3, device=device, dtype=dtype, operations=operations) + self.presence_token_out_norm = operations.LayerNorm(d_model, device=device, dtype=dtype) + + @staticmethod + def _inverse_sigmoid(x): + return torch.log(x / (1 - x + 1e-6) + 1e-6) + + def _compute_box_rpb(self, ref_points, H, W): + """Box rotary position bias: (B, Q, 4) cxcywh -> (B, n_heads, Q+1, H*W) bias.""" + boxes_xyxy = box_cxcywh_to_xyxy(ref_points) + B, Q, _ = boxes_xyxy.shape + coords_h = torch.arange(H, device=ref_points.device, dtype=torch.float32) / H + coords_w = torch.arange(W, device=ref_points.device, dtype=torch.float32) / W + deltas_x = coords_w.view(1, 1, -1, 1) - boxes_xyxy[:, :, None, 0:3:2] + deltas_y = coords_h.view(1, 1, -1, 1) - boxes_xyxy[:, :, None, 1:4:2] + + log2_8 = float(math.log2(8)) + def log_scale(d): + return torch.sign(d * 8) * torch.log2(torch.abs(d * 8) + 1.0) / log2_8 + + rpb_x = self.boxRPB_embed_x(log_scale(deltas_x).to(ref_points.dtype)) + rpb_y = self.boxRPB_embed_y(log_scale(deltas_y).to(ref_points.dtype)) + + bias = (rpb_y.unsqueeze(3) + rpb_x.unsqueeze(2)).flatten(2, 3).permute(0, 3, 1, 2) + pres_bias = torch.zeros(B, bias.shape[1], 1, bias.shape[3], device=bias.device, dtype=bias.dtype) + return torch.cat([pres_bias, bias], dim=2) + + def forward(self, memory, memory_pos, text_memory=None, text_mask=None, H=72, W=72): + B = memory.shape[0] + tgt = cast_to_input(self.query_embed.weight, memory).unsqueeze(0).expand(B, -1, -1) + presence_out = cast_to_input(self.presence_token.weight, memory)[None].expand(B, -1, -1) + ref_points = cast_to_input(self.reference_points.weight, memory).unsqueeze(0).expand(B, -1, -1).sigmoid() + + for layer_idx, layer in enumerate(self.layers): + query_pos = self.ref_point_head(gen_sineembed_for_position(ref_points, self.d_model)) + tgt_with_pres = torch.cat([presence_out, tgt], dim=1) + pos_with_pres = torch.cat([torch.zeros_like(presence_out), query_pos], dim=1) + tgt_with_pres = layer(tgt_with_pres, memory, pos_with_pres, memory_pos, + text_memory, text_mask, self._compute_box_rpb(ref_points, H, W)) + presence_out, tgt = tgt_with_pres[:, :1], tgt_with_pres[:, 1:] + if layer_idx < len(self.layers) - 1: + ref_inv = self._inverse_sigmoid(ref_points) + ref_points = (ref_inv + self.bbox_embed(self.norm(tgt))).sigmoid().detach() + + query_out = self.norm(tgt) + ref_inv = self._inverse_sigmoid(ref_points) + boxes = (ref_inv + self.bbox_embed(query_out)).sigmoid() + presence = self.presence_token_head(self.presence_token_out_norm(presence_out)).squeeze(-1) + return {"decoder_output": query_out, "pred_boxes": boxes, "presence": presence} + + +class Transformer(nn.Module): + def __init__(self, d_model=256, num_heads=8, dim_ff=2048, enc_layers=6, dec_layers=6, + num_queries=200, device=None, dtype=None, operations=None): + super().__init__() + self.encoder = TransformerEncoder(d_model, num_heads, dim_ff, enc_layers, device=device, dtype=dtype, operations=operations) + self.decoder = TransformerDecoder(d_model, num_heads, dim_ff, dec_layers, num_queries, device=device, dtype=dtype, operations=operations) + + +class GeometryEncoder(nn.Module): + def __init__(self, d_model=256, num_heads=8, num_layers=3, roi_size=7, device=None, dtype=None, operations=None): + super().__init__() + self.d_model = d_model + self.roi_size = roi_size + self.pos_enc = PositionEmbeddingSine(num_pos_feats=d_model, normalize=True) + self.points_direct_project = operations.Linear(2, d_model, device=device, dtype=dtype) + self.points_pool_project = operations.Linear(d_model, d_model, device=device, dtype=dtype) + self.points_pos_enc_project = operations.Linear(d_model, d_model, device=device, dtype=dtype) + self.boxes_direct_project = operations.Linear(4, d_model, device=device, dtype=dtype) + self.boxes_pool_project = operations.Conv2d(d_model, d_model, kernel_size=roi_size, device=device, dtype=dtype) + self.boxes_pos_enc_project = operations.Linear(d_model + 2, d_model, device=device, dtype=dtype) + self.label_embed = operations.Embedding(2, d_model, device=device, dtype=dtype) + self.cls_embed = operations.Embedding(1, d_model, device=device, dtype=dtype) + self.norm = operations.LayerNorm(d_model, device=device, dtype=dtype) + self.img_pre_norm = operations.LayerNorm(d_model, device=device, dtype=dtype) + self.encode = nn.ModuleList([ + EncoderLayer(d_model, num_heads, 2048, device=device, dtype=dtype, operations=operations) + for _ in range(num_layers) + ]) + self.encode_norm = operations.LayerNorm(d_model, device=device, dtype=dtype) + self.final_proj = operations.Linear(d_model, d_model, device=device, dtype=dtype) + + def _encode_points(self, coords, labels, img_feat_2d): + """Encode point prompts: direct + pool + pos_enc + label. coords: [B, N, 2] normalized.""" + B, N, _ = coords.shape + embed = self.points_direct_project(coords) + # Pool features from backbone at point locations via grid_sample + grid = (coords * 2 - 1).unsqueeze(2) # [B, N, 1, 2] in [-1, 1] + sampled = F.grid_sample(img_feat_2d, grid, align_corners=False) # [B, C, N, 1] + embed = embed + self.points_pool_project(sampled.squeeze(-1).permute(0, 2, 1)) # [B, N, C] + # Positional encoding of coordinates + x, y = coords[:, :, 0], coords[:, :, 1] # [B, N] + pos_x, pos_y = self.pos_enc._encode_xy(x.flatten(), y.flatten()) + enc = torch.cat([pos_x, pos_y], dim=-1).view(B, N, -1) + embed = embed + self.points_pos_enc_project(cast_to_input(enc, embed)) + embed = embed + cast_to_input(self.label_embed(labels.long()), embed) + return embed + + def _encode_boxes(self, boxes, labels, img_feat_2d): + """Encode box prompts: direct + pool + pos_enc + label. boxes: [B, N, 4] normalized cxcywh.""" + B, N, _ = boxes.shape + embed = self.boxes_direct_project(boxes) + # ROI align from backbone at box regions + H, W = img_feat_2d.shape[-2:] + boxes_xyxy = box_cxcywh_to_xyxy(boxes) + scale = torch.tensor([W, H, W, H], dtype=boxes_xyxy.dtype, device=boxes_xyxy.device) + boxes_scaled = boxes_xyxy * scale + sampled = roi_align(img_feat_2d, boxes_scaled.view(-1, 4).split(N), self.roi_size) + proj = self.boxes_pool_project(sampled).view(B, N, -1) # Conv2d(roi_size) -> [B*N, C, 1, 1] -> [B, N, C] + embed = embed + proj + # Positional encoding of box center + size + cx, cy, w, h = boxes[:, :, 0], boxes[:, :, 1], boxes[:, :, 2], boxes[:, :, 3] + enc = self.pos_enc.encode_boxes(cx.flatten(), cy.flatten(), w.flatten(), h.flatten()) + enc = enc.view(B, N, -1) + embed = embed + self.boxes_pos_enc_project(cast_to_input(enc, embed)) + embed = embed + cast_to_input(self.label_embed(labels.long()), embed) + return embed + + def forward(self, points=None, boxes=None, image_features=None): + """Encode geometry prompts. image_features: [B, HW, C] flattened backbone features.""" + # Prepare 2D image features for pooling + img_feat_2d = None + if image_features is not None: + B = image_features.shape[0] + HW, C = image_features.shape[1], image_features.shape[2] + hw = int(math.sqrt(HW)) + img_normed = self.img_pre_norm(image_features) + img_feat_2d = img_normed.permute(0, 2, 1).view(B, C, hw, hw) + + embeddings = [] + if points is not None: + coords, labels = points + embeddings.append(self._encode_points(coords, labels, img_feat_2d)) + if boxes is not None: + B = boxes.shape[0] + box_labels = torch.ones(B, boxes.shape[1], dtype=torch.long, device=boxes.device) + embeddings.append(self._encode_boxes(boxes, box_labels, img_feat_2d)) + if not embeddings: + return None + geo = torch.cat(embeddings, dim=1) + geo = self.norm(geo) + if image_features is not None: + for layer in self.encode: + geo = layer(geo, torch.zeros_like(geo), image_features) + geo = self.encode_norm(geo) + return self.final_proj(geo) + + +class PixelDecoder(nn.Module): + """Top-down FPN pixel decoder with GroupNorm + ReLU + nearest interpolation.""" + def __init__(self, d_model=256, num_stages=3, device=None, dtype=None, operations=None): + super().__init__() + self.conv_layers = nn.ModuleList([operations.Conv2d(d_model, d_model, kernel_size=3, padding=1, device=device, dtype=dtype) for _ in range(num_stages)]) + self.norms = nn.ModuleList([operations.GroupNorm(8, d_model, device=device, dtype=dtype) for _ in range(num_stages)]) + + def forward(self, backbone_features): + prev = backbone_features[-1] + for i, feat in enumerate(backbone_features[:-1][::-1]): + prev = F.relu(self.norms[i](self.conv_layers[i](feat + F.interpolate(prev, size=feat.shape[-2:], mode="nearest")))) + return prev + + +class MaskPredictor(nn.Module): + def __init__(self, d_model=256, device=None, dtype=None, operations=None): + super().__init__() + self.mask_embed = MLP(d_model, d_model, d_model, 3, device=device, dtype=dtype, operations=operations) + + def forward(self, query_embeddings, pixel_features): + mask_embed = self.mask_embed(query_embeddings) + return torch.einsum("bqc,bchw->bqhw", mask_embed, pixel_features) + + +class SegmentationHead(nn.Module): + def __init__(self, d_model=256, num_heads=8, device=None, dtype=None, operations=None): + super().__init__() + self.d_model = d_model + self.pixel_decoder = PixelDecoder(d_model, 3, device=device, dtype=dtype, operations=operations) + self.mask_predictor = MaskPredictor(d_model, device=device, dtype=dtype, operations=operations) + self.cross_attend_prompt = SplitMHA(d_model, num_heads, device=device, dtype=dtype, operations=operations) + self.cross_attn_norm = operations.LayerNorm(d_model, device=device, dtype=dtype) + self.instance_seg_head = operations.Conv2d(d_model, d_model, kernel_size=1, device=device, dtype=dtype) + self.semantic_seg_head = operations.Conv2d(d_model, 1, kernel_size=1, device=device, dtype=dtype) + + def forward(self, query_embeddings, backbone_features, encoder_hidden_states=None, prompt=None, prompt_mask=None): + if encoder_hidden_states is not None and prompt is not None: + enc_normed = self.cross_attn_norm(encoder_hidden_states) + enc_cross = self.cross_attend_prompt(enc_normed, prompt, prompt, mask=prompt_mask) + encoder_hidden_states = enc_cross + encoder_hidden_states + + if encoder_hidden_states is not None: + B, H, W = encoder_hidden_states.shape[0], backbone_features[-1].shape[-2], backbone_features[-1].shape[-1] + encoder_visual = encoder_hidden_states[:, :H * W].permute(0, 2, 1).view(B, self.d_model, H, W) + backbone_features = list(backbone_features) + backbone_features[-1] = encoder_visual + + pixel_features = self.pixel_decoder(backbone_features) + instance_features = self.instance_seg_head(pixel_features) + masks = self.mask_predictor(query_embeddings, instance_features) + return masks + + +class DotProductScoring(nn.Module): + def __init__(self, d_model=256, device=None, dtype=None, operations=None): + super().__init__() + self.hs_proj = operations.Linear(d_model, d_model, device=device, dtype=dtype) + self.prompt_proj = operations.Linear(d_model, d_model, device=device, dtype=dtype) + self.prompt_mlp = MLPWithNorm(d_model, 2048, d_model, 2, device=device, dtype=dtype, operations=operations) + self.scale = 1.0 / (d_model ** 0.5) + + def forward(self, query_embeddings, prompt_embeddings, prompt_mask=None): + prompt = self.prompt_mlp(prompt_embeddings) + if prompt_mask is not None: + weight = prompt_mask.unsqueeze(-1).to(dtype=prompt.dtype) + pooled = (prompt * weight).sum(dim=1) / weight.sum(dim=1).clamp(min=1) + else: + pooled = prompt.mean(dim=1) + hs = self.hs_proj(query_embeddings) + pp = self.prompt_proj(pooled).unsqueeze(-1).to(hs.dtype) + scores = torch.matmul(hs, pp) + return (scores * self.scale).clamp(-12.0, 12.0).squeeze(-1) + + +class SAM3Detector(nn.Module): + def __init__(self, d_model=256, embed_dim=1024, num_queries=200, device=None, dtype=None, operations=None, **kwargs): + super().__init__() + image_model = kwargs.pop("image_model", "SAM3") + for k in ("num_heads", "num_head_channels"): + kwargs.pop(k, None) + multiplex = image_model == "SAM31" + # SAM3: 4 FPN levels, drop last (scalp=1); SAM3.1: 3 levels, use all (scalp=0) + self.scalp = 0 if multiplex else 1 + self.backbone = nn.ModuleDict({ + "vision_backbone": SAM3VisionBackbone(embed_dim=embed_dim, d_model=d_model, multiplex=multiplex, device=device, dtype=dtype, operations=operations, **kwargs), + "language_backbone": nn.ModuleDict({"resizer": operations.Linear(embed_dim, d_model, device=device, dtype=dtype)}), + }) + self.transformer = Transformer(d_model=d_model, num_queries=num_queries, device=device, dtype=dtype, operations=operations) + self.segmentation_head = SegmentationHead(d_model=d_model, device=device, dtype=dtype, operations=operations) + self.geometry_encoder = GeometryEncoder(d_model=d_model, device=device, dtype=dtype, operations=operations) + self.dot_prod_scoring = DotProductScoring(d_model=d_model, device=device, dtype=dtype, operations=operations) + + def _get_backbone_features(self, images): + """Run backbone and return (detector_features, detector_positions, tracker_features, tracker_positions).""" + bb = self.backbone["vision_backbone"] + if bb.multiplex: + all_f, all_p, tf, tp = bb(images, tracker_mode="propagation") + else: + all_f, all_p, tf, tp = bb(images, need_tracker=True) + return all_f, all_p, tf, tp + + @staticmethod + def _run_geo_layer(layer, x, memory, memory_pos): + x = x + layer.self_attn(layer.norm1(x)) + x = x + layer.cross_attn_image(layer.norm2(x), memory + memory_pos, memory) + x = x + layer.linear2(F.relu(layer.linear1(layer.norm3(x)))) + return x + + def _detect(self, features, positions, text_embeddings=None, text_mask=None, + points=None, boxes=None): + """Shared detection: geometry encoding, transformer, scoring, segmentation.""" + B = features[0].shape[0] + # Scalp for encoder (use top-level feature), but keep all levels for segmentation head + seg_features = features + if self.scalp > 0: + features = features[:-self.scalp] + positions = positions[:-self.scalp] + enc_feat, enc_pos = features[-1], positions[-1] + _, _, H, W = enc_feat.shape + img_flat = enc_feat.flatten(2).permute(0, 2, 1) + pos_flat = enc_pos.flatten(2).permute(0, 2, 1) + + has_prompts = text_embeddings is not None or points is not None or boxes is not None + if has_prompts: + geo_enc = self.geometry_encoder + geo_prompts = geo_enc(points=points, boxes=boxes, image_features=img_flat) + geo_cls = geo_enc.norm(geo_enc.final_proj(cast_to_input(geo_enc.cls_embed.weight, img_flat).view(1, 1, -1).expand(B, -1, -1))) + for layer in geo_enc.encode: + geo_cls = self._run_geo_layer(layer, geo_cls, img_flat, pos_flat) + geo_cls = geo_enc.encode_norm(geo_cls) + if text_embeddings is not None and text_embeddings.shape[0] != B: + text_embeddings = text_embeddings.expand(B, -1, -1) + if text_mask is not None and text_mask.shape[0] != B: + text_mask = text_mask.expand(B, -1) + parts = [t for t in [text_embeddings, geo_prompts, geo_cls] if t is not None] + text_embeddings = torch.cat(parts, dim=1) + n_new = text_embeddings.shape[1] - (text_mask.shape[1] if text_mask is not None else 0) + if text_mask is not None: + text_mask = torch.cat([text_mask, torch.ones(B, n_new, dtype=torch.bool, device=text_mask.device)], dim=1) + else: + text_mask = torch.ones(B, text_embeddings.shape[1], dtype=torch.bool, device=text_embeddings.device) + + memory = self.transformer.encoder(img_flat, pos_flat, text_embeddings, text_mask) + dec_out = self.transformer.decoder(memory, pos_flat, text_embeddings, text_mask, H, W) + query_out, pred_boxes = dec_out["decoder_output"], dec_out["pred_boxes"] + + if text_embeddings is not None: + scores = self.dot_prod_scoring(query_out, text_embeddings, text_mask) + else: + scores = torch.zeros(B, query_out.shape[1], device=query_out.device) + + masks = self.segmentation_head(query_out, seg_features, encoder_hidden_states=memory, prompt=text_embeddings, prompt_mask=text_mask) + return box_cxcywh_to_xyxy(pred_boxes), scores, masks, dec_out + + def forward(self, images, text_embeddings=None, text_mask=None, points=None, boxes=None, threshold=0.3, orig_size=None): + features, positions, _, _ = self._get_backbone_features(images) + + if text_embeddings is not None: + text_embeddings = self.backbone["language_backbone"]["resizer"](text_embeddings) + if text_mask is not None: + text_mask = text_mask.bool() + + boxes_xyxy, scores, masks, dec_out = self._detect( + features, positions, text_embeddings, text_mask, points, boxes) + + if orig_size is not None: + oh, ow = orig_size + boxes_xyxy = boxes_xyxy * torch.tensor([ow, oh, ow, oh], device=boxes_xyxy.device, dtype=boxes_xyxy.dtype) + masks = F.interpolate(masks, size=orig_size, mode="bilinear", align_corners=False) + + return { + "boxes": boxes_xyxy, + "scores": scores, + "masks": masks, + "presence": dec_out.get("presence"), + } + + def forward_from_trunk(self, trunk_out, text_embeddings, text_mask): + """Run detection using a pre-computed ViTDet trunk output. + + text_embeddings must already be resized through language_backbone.resizer. + Returns dict with boxes (normalized xyxy), scores, masks at detector resolution. + """ + bb = self.backbone["vision_backbone"] + features = [conv(trunk_out) for conv in bb.convs] + positions = [cast_to_input(bb.position_encoding(f), f) for f in features] + + if text_mask is not None: + text_mask = text_mask.bool() + + boxes_xyxy, scores, masks, _ = self._detect(features, positions, text_embeddings, text_mask) + return {"boxes": boxes_xyxy, "scores": scores, "masks": masks} + + +class SAM3Model(nn.Module): + def __init__(self, device=None, dtype=None, operations=None, **kwargs): + super().__init__() + self.dtype = dtype + image_model = kwargs.get("image_model", "SAM3") + tracker_cls = TRACKER_CLASSES[image_model] + self.detector = SAM3Detector(device=device, dtype=dtype, operations=operations, **kwargs) + self.tracker = tracker_cls(device=device, dtype=dtype, operations=operations, **kwargs) + + def forward(self, images, **kwargs): + return self.detector(images, **kwargs) + + def forward_segment(self, images, point_inputs=None, box_inputs=None, mask_inputs=None): + """Interactive segmentation using SAM decoder with point/box/mask prompts. + + Args: + images: [B, 3, 1008, 1008] preprocessed images + point_inputs: {"point_coords": [B, N, 2], "point_labels": [B, N]} in 1008x1008 pixel space + box_inputs: [B, 2, 2] box corners (top-left, bottom-right) in 1008x1008 pixel space + mask_inputs: [B, 1, H, W] coarse mask logits to refine + Returns: + [B, 1, image_size, image_size] high-res mask logits + """ + bb = self.detector.backbone["vision_backbone"] + if bb.multiplex: + _, _, tracker_features, tracker_positions = bb(images, tracker_mode="interactive") + else: + _, _, tracker_features, tracker_positions = bb(images, need_tracker=True) + if self.detector.scalp > 0: + tracker_features = tracker_features[:-self.detector.scalp] + tracker_positions = tracker_positions[:-self.detector.scalp] + + high_res = list(tracker_features[:-1]) + backbone_feat = tracker_features[-1] + B, C, H, W = backbone_feat.shape + # Add no-memory embedding (init frame path) + no_mem = getattr(self.tracker, 'interactivity_no_mem_embed', None) + if no_mem is None: + no_mem = getattr(self.tracker, 'no_mem_embed', None) + if no_mem is not None: + feat_flat = backbone_feat.flatten(2).permute(0, 2, 1) + feat_flat = feat_flat + cast_to_input(no_mem, feat_flat) + backbone_feat = feat_flat.view(B, H, W, C).permute(0, 3, 1, 2) + + num_pts = 0 if point_inputs is None else point_inputs["point_labels"].size(1) + _, high_res_masks, _, _ = self.tracker._forward_sam_heads( + backbone_features=backbone_feat, + point_inputs=point_inputs, + mask_inputs=mask_inputs, + box_inputs=box_inputs, + high_res_features=high_res, + multimask_output=(0 < num_pts <= 1), + ) + return high_res_masks + + def forward_video(self, images, initial_masks, pbar=None, text_prompts=None, + new_det_thresh=0.5, max_objects=0, detect_interval=1, + target_device=None, target_dtype=None): + """Track video with optional per-frame text-prompted detection.""" + bb = self.detector.backbone["vision_backbone"] + + def backbone_fn(frame, frame_idx=None): + trunk_out = bb.trunk(frame) + if bb.multiplex: + _, _, tf, tp = bb(frame, tracker_mode="propagation", cached_trunk=trunk_out, tracker_only=True) + else: + _, _, tf, tp = bb(frame, need_tracker=True, cached_trunk=trunk_out, tracker_only=True) + return tf, tp, trunk_out + + detect_fn = None + if text_prompts: + resizer = self.detector.backbone["language_backbone"]["resizer"] + resized = [(resizer(emb), m.bool() if m is not None else None) for emb, m in text_prompts] + def detect_fn(trunk_out): + all_scores, all_masks = [], [] + for emb, mask in resized: + det = self.detector.forward_from_trunk(trunk_out, emb, mask) + all_scores.append(det["scores"]) + all_masks.append(det["masks"]) + return {"scores": torch.cat(all_scores, dim=1), "masks": torch.cat(all_masks, dim=1)} + + if hasattr(self.tracker, 'track_video_with_detection'): + return self.tracker.track_video_with_detection( + backbone_fn, images, initial_masks, detect_fn, + new_det_thresh=new_det_thresh, max_objects=max_objects, + detect_interval=detect_interval, backbone_obj=bb, pbar=pbar, + target_device=target_device, target_dtype=target_dtype) + # SAM3 (non-multiplex) — no detection support, requires initial masks + if initial_masks is None: + raise ValueError("SAM3 (non-multiplex) requires initial_mask for video tracking") + return self.tracker.track_video(backbone_fn, images, initial_masks, pbar=pbar, backbone_obj=bb, + target_device=target_device, target_dtype=target_dtype) diff --git a/comfy/ldm/sam3/sam.py b/comfy/ldm/sam3/sam.py new file mode 100644 index 000000000..75cb457cf --- /dev/null +++ b/comfy/ldm/sam3/sam.py @@ -0,0 +1,425 @@ +# SAM3 shared components: primitives, ViTDet backbone, FPN neck, position encodings. + +import math + +import torch +import torch.nn as nn +import torch.nn.functional as F + +from comfy.ldm.modules.attention import optimized_attention +from comfy.ldm.flux.math import apply_rope +from comfy.ldm.flux.layers import EmbedND +from comfy.ops import cast_to_input + + +class MLP(nn.Module): + def __init__(self, input_dim, hidden_dim, output_dim, num_layers, sigmoid_output=False, device=None, dtype=None, operations=None): + super().__init__() + dims = [input_dim] + [hidden_dim] * (num_layers - 1) + [output_dim] + self.layers = nn.ModuleList([operations.Linear(dims[i], dims[i + 1], device=device, dtype=dtype) for i in range(num_layers)]) + self.sigmoid_output = sigmoid_output + + def forward(self, x): + for i, layer in enumerate(self.layers): + x = F.relu(layer(x)) if i < len(self.layers) - 1 else layer(x) + return torch.sigmoid(x) if self.sigmoid_output else x + + +class SAMAttention(nn.Module): + def __init__(self, embedding_dim, num_heads, downsample_rate=1, kv_in_dim=None, device=None, dtype=None, operations=None): + super().__init__() + self.num_heads = num_heads + internal_dim = embedding_dim // downsample_rate + kv_dim = kv_in_dim if kv_in_dim is not None else embedding_dim + self.q_proj = operations.Linear(embedding_dim, internal_dim, device=device, dtype=dtype) + self.k_proj = operations.Linear(kv_dim, internal_dim, device=device, dtype=dtype) + self.v_proj = operations.Linear(kv_dim, internal_dim, device=device, dtype=dtype) + self.out_proj = operations.Linear(internal_dim, embedding_dim, device=device, dtype=dtype) + + def forward(self, q, k, v): + q = self.q_proj(q) + k = self.k_proj(k) + v = self.v_proj(v) + return self.out_proj(optimized_attention(q, k, v, self.num_heads, low_precision_attention=False)) + + +class TwoWayAttentionBlock(nn.Module): + def __init__(self, embedding_dim, num_heads, mlp_dim=2048, attention_downsample_rate=2, skip_first_layer_pe=False, device=None, dtype=None, operations=None): + super().__init__() + self.skip_first_layer_pe = skip_first_layer_pe + self.self_attn = SAMAttention(embedding_dim, num_heads, device=device, dtype=dtype, operations=operations) + self.cross_attn_token_to_image = SAMAttention(embedding_dim, num_heads, downsample_rate=attention_downsample_rate, device=device, dtype=dtype, operations=operations) + self.cross_attn_image_to_token = SAMAttention(embedding_dim, num_heads, downsample_rate=attention_downsample_rate, device=device, dtype=dtype, operations=operations) + self.mlp = nn.Sequential(operations.Linear(embedding_dim, mlp_dim, device=device, dtype=dtype), nn.ReLU(), operations.Linear(mlp_dim, embedding_dim, device=device, dtype=dtype)) + self.norm1 = operations.LayerNorm(embedding_dim, device=device, dtype=dtype) + self.norm2 = operations.LayerNorm(embedding_dim, device=device, dtype=dtype) + self.norm3 = operations.LayerNorm(embedding_dim, device=device, dtype=dtype) + self.norm4 = operations.LayerNorm(embedding_dim, device=device, dtype=dtype) + + def forward(self, queries, keys, query_pe, key_pe): + if self.skip_first_layer_pe: + queries = self.norm1(self.self_attn(queries, queries, queries)) + else: + q = queries + query_pe + queries = self.norm1(queries + self.self_attn(q, q, queries)) + q, k = queries + query_pe, keys + key_pe + queries = self.norm2(queries + self.cross_attn_token_to_image(q, k, keys)) + queries = self.norm3(queries + self.mlp(queries)) + q, k = queries + query_pe, keys + key_pe + keys = self.norm4(keys + self.cross_attn_image_to_token(k, q, queries)) + return queries, keys + + +class TwoWayTransformer(nn.Module): + def __init__(self, depth=2, embedding_dim=256, num_heads=8, mlp_dim=2048, attention_downsample_rate=2, device=None, dtype=None, operations=None): + super().__init__() + self.layers = nn.ModuleList([ + TwoWayAttentionBlock(embedding_dim, num_heads, mlp_dim, attention_downsample_rate, + skip_first_layer_pe=(i == 0), device=device, dtype=dtype, operations=operations) + for i in range(depth) + ]) + self.final_attn_token_to_image = SAMAttention(embedding_dim, num_heads, downsample_rate=attention_downsample_rate, device=device, dtype=dtype, operations=operations) + self.norm_final = operations.LayerNorm(embedding_dim, device=device, dtype=dtype) + + def forward(self, image_embedding, image_pe, point_embedding): + queries, keys = point_embedding, image_embedding + for layer in self.layers: + queries, keys = layer(queries, keys, point_embedding, image_pe) + q, k = queries + point_embedding, keys + image_pe + queries = self.norm_final(queries + self.final_attn_token_to_image(q, k, keys)) + return queries, keys + + +class PositionEmbeddingRandom(nn.Module): + """Fourier feature positional encoding with random gaussian projection.""" + def __init__(self, num_pos_feats=64, scale=None): + super().__init__() + self.register_buffer("positional_encoding_gaussian_matrix", (scale or 1.0) * torch.randn(2, num_pos_feats)) + + def _encode(self, normalized_coords): + """Map normalized [0,1] coordinates to fourier features via random projection. Computes in fp32.""" + orig_dtype = normalized_coords.dtype + proj_matrix = self.positional_encoding_gaussian_matrix.to(device=normalized_coords.device, dtype=torch.float32) + projected = 2 * math.pi * (2 * normalized_coords.float() - 1) @ proj_matrix + return torch.cat([projected.sin(), projected.cos()], dim=-1).to(orig_dtype) + + def forward(self, size, device=None): + h, w = size + dev = device if device is not None else self.positional_encoding_gaussian_matrix.device + ones = torch.ones((h, w), device=dev, dtype=torch.float32) + norm_xy = torch.stack([(ones.cumsum(1) - 0.5) / w, (ones.cumsum(0) - 0.5) / h], dim=-1) + return self._encode(norm_xy).permute(2, 0, 1).unsqueeze(0) + + def forward_with_coords(self, pixel_coords, image_size): + norm = pixel_coords.clone() + norm[:, :, 0] /= image_size[1] + norm[:, :, 1] /= image_size[0] + return self._encode(norm) + + +# ViTDet backbone + FPN neck + +def window_partition(x: torch.Tensor, window_size: int): + B, H, W, C = x.shape + pad_h = (window_size - H % window_size) % window_size + pad_w = (window_size - W % window_size) % window_size + if pad_h > 0 or pad_w > 0: + x = F.pad(x, (0, 0, 0, pad_w, 0, pad_h)) + Hp, Wp = H + pad_h, W + pad_w + x = x.view(B, Hp // window_size, window_size, Wp // window_size, window_size, C) + windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C) + return windows, (Hp, Wp) + + +def window_unpartition(windows: torch.Tensor, window_size: int, pad_hw, hw): + Hp, Wp = pad_hw + H, W = hw + B = windows.shape[0] // (Hp * Wp // window_size // window_size) + x = windows.view(B, Hp // window_size, Wp // window_size, window_size, window_size, -1) + x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, Hp, Wp, -1) + if Hp > H or Wp > W: + x = x[:, :H, :W, :].contiguous() + return x + + +def rope_2d(end_x: int, end_y: int, dim: int, theta: float = 10000.0, scale_pos: float = 1.0): + """Generate 2D axial RoPE using flux EmbedND. Returns [1, 1, HW, dim//2, 2, 2].""" + t = torch.arange(end_x * end_y, dtype=torch.float32) + ids = torch.stack([(t % end_x) * scale_pos, + torch.div(t, end_x, rounding_mode="floor") * scale_pos], dim=-1) + return EmbedND(dim=dim, theta=theta, axes_dim=[dim // 2, dim // 2])(ids.unsqueeze(0)) + + +class _ViTMLP(nn.Module): + def __init__(self, dim, mlp_ratio=4.0, device=None, dtype=None, operations=None): + super().__init__() + hidden = int(dim * mlp_ratio) + self.fc1 = operations.Linear(dim, hidden, device=device, dtype=dtype) + self.act = nn.GELU() + self.fc2 = operations.Linear(hidden, dim, device=device, dtype=dtype) + + def forward(self, x): + return self.fc2(self.act(self.fc1(x))) + + +class Attention(nn.Module): + """ViTDet multi-head attention with fused QKV projection.""" + + def __init__(self, dim, num_heads=8, qkv_bias=True, use_rope=False, device=None, dtype=None, operations=None): + super().__init__() + self.num_heads = num_heads + self.head_dim = dim // num_heads + self.use_rope = use_rope + self.qkv = operations.Linear(dim, dim * 3, bias=qkv_bias, device=device, dtype=dtype) + self.proj = operations.Linear(dim, dim, device=device, dtype=dtype) + + def forward(self, x, freqs_cis=None): + B, N, C = x.shape + qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, self.head_dim) + q, k, v = qkv.permute(2, 0, 3, 1, 4).unbind(dim=0) + if self.use_rope and freqs_cis is not None: + q, k = apply_rope(q, k, freqs_cis) + return self.proj(optimized_attention(q, k, v, self.num_heads, skip_reshape=True, low_precision_attention=False)) + + +class Block(nn.Module): + def __init__(self, dim, num_heads, mlp_ratio=4.0, qkv_bias=True, window_size=0, use_rope=False, device=None, dtype=None, operations=None): + super().__init__() + self.window_size = window_size + self.norm1 = operations.LayerNorm(dim, device=device, dtype=dtype) + self.attn = Attention(dim, num_heads, qkv_bias, use_rope, device=device, dtype=dtype, operations=operations) + self.norm2 = operations.LayerNorm(dim, device=device, dtype=dtype) + self.mlp = _ViTMLP(dim, mlp_ratio, device=device, dtype=dtype, operations=operations) + + def forward(self, x, freqs_cis=None): + shortcut = x + x = self.norm1(x) + if self.window_size > 0: + H, W = x.shape[1], x.shape[2] + x, pad_hw = window_partition(x, self.window_size) + x = x.view(x.shape[0], self.window_size * self.window_size, -1) + x = self.attn(x, freqs_cis=freqs_cis) + x = x.view(-1, self.window_size, self.window_size, x.shape[-1]) + x = window_unpartition(x, self.window_size, pad_hw, (H, W)) + else: + B, H, W, C = x.shape + x = x.view(B, H * W, C) + x = self.attn(x, freqs_cis=freqs_cis) + x = x.view(B, H, W, C) + x = shortcut + x + x = x + self.mlp(self.norm2(x)) + return x + + +class PatchEmbed(nn.Module): + def __init__(self, patch_size=14, in_chans=3, embed_dim=1024, device=None, dtype=None, operations=None): + super().__init__() + self.proj = operations.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size, bias=False, device=device, dtype=dtype) + + def forward(self, x): + return self.proj(x) + + +class ViTDet(nn.Module): + def __init__(self, img_size=1008, patch_size=14, embed_dim=1024, depth=32, num_heads=16, mlp_ratio=4.625, qkv_bias=True, window_size=24, + global_att_blocks=(7, 15, 23, 31), use_rope=True, pretrain_img_size=336, device=None, dtype=None, operations=None, **kwargs): + super().__init__() + self.img_size = img_size + self.patch_size = patch_size + self.embed_dim = embed_dim + self.num_heads = num_heads + self.global_att_blocks = set(global_att_blocks) + + self.patch_embed = PatchEmbed(patch_size, 3, embed_dim, device=device, dtype=dtype, operations=operations) + + num_patches = (pretrain_img_size // patch_size) ** 2 + 1 # +1 for cls token + self.pos_embed = nn.Parameter(torch.zeros(1, num_patches, embed_dim, device=device, dtype=dtype)) + + self.ln_pre = operations.LayerNorm(embed_dim, device=device, dtype=dtype) + + grid_size = img_size // patch_size + pretrain_grid = pretrain_img_size // patch_size + + self.blocks = nn.ModuleList() + for i in range(depth): + is_global = i in self.global_att_blocks + self.blocks.append(Block( + embed_dim, num_heads, mlp_ratio, qkv_bias, + window_size=0 if is_global else window_size, + use_rope=use_rope, + device=device, dtype=dtype, operations=operations, + )) + + if use_rope: + rope_scale = pretrain_grid / grid_size + self.register_buffer("freqs_cis", rope_2d(grid_size, grid_size, embed_dim // num_heads, scale_pos=rope_scale), persistent=False) + self.register_buffer("freqs_cis_window", rope_2d(window_size, window_size, embed_dim // num_heads), persistent=False) + else: + self.freqs_cis = None + self.freqs_cis_window = None + + def _get_pos_embed(self, num_tokens): + pos = self.pos_embed + if pos.shape[1] == num_tokens: + return pos + cls_pos = pos[:, :1] + spatial_pos = pos[:, 1:] + old_size = int(math.sqrt(spatial_pos.shape[1])) + new_size = int(math.sqrt(num_tokens - 1)) if num_tokens > 1 else old_size + spatial_2d = spatial_pos.reshape(1, old_size, old_size, -1).permute(0, 3, 1, 2) + tiles_h = new_size // old_size + 1 + tiles_w = new_size // old_size + 1 + tiled = spatial_2d.tile([1, 1, tiles_h, tiles_w])[:, :, :new_size, :new_size] + tiled = tiled.permute(0, 2, 3, 1).reshape(1, new_size * new_size, -1) + return torch.cat([cls_pos, tiled], dim=1) + + def forward(self, x): + x = self.patch_embed(x) + B, C, Hp, Wp = x.shape + x = x.permute(0, 2, 3, 1).reshape(B, Hp * Wp, C) + + pos = cast_to_input(self._get_pos_embed(Hp * Wp + 1), x) + x = x + pos[:, 1:Hp * Wp + 1] + + x = x.view(B, Hp, Wp, C) + x = self.ln_pre(x) + + freqs_cis_global = self.freqs_cis + freqs_cis_win = self.freqs_cis_window + if freqs_cis_global is not None: + freqs_cis_global = cast_to_input(freqs_cis_global, x) + if freqs_cis_win is not None: + freqs_cis_win = cast_to_input(freqs_cis_win, x) + + for block in self.blocks: + fc = freqs_cis_win if block.window_size > 0 else freqs_cis_global + x = block(x, freqs_cis=fc) + + return x.permute(0, 3, 1, 2) + + +class FPNScaleConv(nn.Module): + def __init__(self, in_dim, out_dim, scale, device=None, dtype=None, operations=None): + super().__init__() + if scale == 4.0: + self.dconv_2x2_0 = operations.ConvTranspose2d(in_dim, in_dim // 2, kernel_size=2, stride=2, device=device, dtype=dtype) + self.dconv_2x2_1 = operations.ConvTranspose2d(in_dim // 2, in_dim // 4, kernel_size=2, stride=2, device=device, dtype=dtype) + proj_in = in_dim // 4 + elif scale == 2.0: + self.dconv_2x2 = operations.ConvTranspose2d(in_dim, in_dim // 2, kernel_size=2, stride=2, device=device, dtype=dtype) + proj_in = in_dim // 2 + elif scale == 1.0: + proj_in = in_dim + elif scale == 0.5: + self.pool = nn.MaxPool2d(kernel_size=2, stride=2) + proj_in = in_dim + self.scale = scale + self.conv_1x1 = operations.Conv2d(proj_in, out_dim, kernel_size=1, device=device, dtype=dtype) + self.conv_3x3 = operations.Conv2d(out_dim, out_dim, kernel_size=3, padding=1, device=device, dtype=dtype) + + def forward(self, x): + if self.scale == 4.0: + x = F.gelu(self.dconv_2x2_0(x)) + x = self.dconv_2x2_1(x) + elif self.scale == 2.0: + x = self.dconv_2x2(x) + elif self.scale == 0.5: + x = self.pool(x) + x = self.conv_1x1(x) + x = self.conv_3x3(x) + return x + + +class PositionEmbeddingSine(nn.Module): + """2D sinusoidal position encoding (DETR-style) with result caching.""" + def __init__(self, num_pos_feats=256, temperature=10000.0, normalize=True, scale=None): + super().__init__() + assert num_pos_feats % 2 == 0 + self.half_dim = num_pos_feats // 2 + self.temperature = temperature + self.normalize = normalize + self.scale = scale if scale is not None else 2 * math.pi + self._cache = {} + + def _sincos(self, vals): + """Encode 1D values to interleaved sin/cos features.""" + freqs = self.temperature ** (2 * (torch.arange(self.half_dim, dtype=torch.float32, device=vals.device) // 2) / self.half_dim) + raw = vals[..., None] * self.scale / freqs + return torch.stack((raw[..., 0::2].sin(), raw[..., 1::2].cos()), dim=-1).flatten(-2) + + def _encode_xy(self, x, y): + """Encode normalized x, y coordinates to sinusoidal features. Returns (pos_x, pos_y) each [N, half_dim].""" + dim_t = self.temperature ** (2 * (torch.arange(self.half_dim, dtype=torch.float32, device=x.device) // 2) / self.half_dim) + pos_x = x[:, None] * self.scale / dim_t + pos_y = y[:, None] * self.scale / dim_t + pos_x = torch.stack((pos_x[:, 0::2].sin(), pos_x[:, 1::2].cos()), dim=2).flatten(1) + pos_y = torch.stack((pos_y[:, 0::2].sin(), pos_y[:, 1::2].cos()), dim=2).flatten(1) + return pos_x, pos_y + + def encode_boxes(self, cx, cy, w, h): + """Encode box center + size to [N, d_model+2] features.""" + pos_x, pos_y = self._encode_xy(cx, cy) + return torch.cat((pos_y, pos_x, h[:, None], w[:, None]), dim=1) + + def forward(self, x): + B, C, H, W = x.shape + key = (H, W, x.device) + if key not in self._cache: + gy = torch.arange(H, dtype=torch.float32, device=x.device) + gx = torch.arange(W, dtype=torch.float32, device=x.device) + if self.normalize: + gy, gx = gy / (H - 1 + 1e-6), gx / (W - 1 + 1e-6) + yy, xx = torch.meshgrid(gy, gx, indexing="ij") + self._cache[key] = torch.cat((self._sincos(yy), self._sincos(xx)), dim=-1).permute(2, 0, 1).unsqueeze(0) + return self._cache[key].expand(B, -1, -1, -1) + + +class SAM3VisionBackbone(nn.Module): + def __init__(self, embed_dim=1024, d_model=256, multiplex=False, device=None, dtype=None, operations=None, **kwargs): + super().__init__() + self.trunk = ViTDet(embed_dim=embed_dim, device=device, dtype=dtype, operations=operations, **kwargs) + self.position_encoding = PositionEmbeddingSine(num_pos_feats=d_model, normalize=True) + self.multiplex = multiplex + + fpn_args = dict(device=device, dtype=dtype, operations=operations) + if multiplex: + scales = [4.0, 2.0, 1.0] + self.convs = nn.ModuleList([FPNScaleConv(embed_dim, d_model, s, **fpn_args) for s in scales]) + self.propagation_convs = nn.ModuleList([FPNScaleConv(embed_dim, d_model, s, **fpn_args) for s in scales]) + self.interactive_convs = nn.ModuleList([FPNScaleConv(embed_dim, d_model, s, **fpn_args) for s in scales]) + else: + scales = [4.0, 2.0, 1.0, 0.5] + self.convs = nn.ModuleList([FPNScaleConv(embed_dim, d_model, s, **fpn_args) for s in scales]) + self.sam2_convs = nn.ModuleList([FPNScaleConv(embed_dim, d_model, s, **fpn_args) for s in scales]) + + def forward(self, images, need_tracker=False, tracker_mode=None, cached_trunk=None, tracker_only=False): + backbone_out = cached_trunk if cached_trunk is not None else self.trunk(images) + + if tracker_only: + # Skip detector FPN when only tracker features are needed (video tracking) + if self.multiplex: + tracker_convs = self.propagation_convs if tracker_mode == "propagation" else self.interactive_convs + else: + tracker_convs = self.sam2_convs + tracker_features = [conv(backbone_out) for conv in tracker_convs] + tracker_positions = [cast_to_input(self.position_encoding(f), f) for f in tracker_features] + return None, None, tracker_features, tracker_positions + + features = [conv(backbone_out) for conv in self.convs] + positions = [cast_to_input(self.position_encoding(f), f) for f in features] + + if self.multiplex: + if tracker_mode == "propagation": + tracker_convs = self.propagation_convs + elif tracker_mode == "interactive": + tracker_convs = self.interactive_convs + else: + return features, positions, None, None + elif need_tracker: + tracker_convs = self.sam2_convs + else: + return features, positions, None, None + + tracker_features = [conv(backbone_out) for conv in tracker_convs] + tracker_positions = [cast_to_input(self.position_encoding(f), f) for f in tracker_features] + return features, positions, tracker_features, tracker_positions diff --git a/comfy/ldm/sam3/tracker.py b/comfy/ldm/sam3/tracker.py new file mode 100644 index 000000000..8456e90a6 --- /dev/null +++ b/comfy/ldm/sam3/tracker.py @@ -0,0 +1,1802 @@ +# SAM3 video tracker: memory encoder, memory attention, SAM mask decoder/prompt encoder. + +import numpy as np +import torch +import torch.nn as nn +import torch.nn.functional as F +from tqdm import tqdm + +try: + import cv2 + _HAS_CV2 = True +except ImportError: + from scipy import ndimage + _HAS_CV2 = False + +import comfy.model_management +from comfy.ldm.modules.attention import optimized_attention +from comfy.ldm.sam3.sam import rope_2d, PositionEmbeddingSine +from comfy.ops import cast_to_input +from comfy.ldm.flux.math import apply_rope1 +from comfy.ldm.cascade.common import LayerNorm2d_op +from comfy.ldm.sam3.sam import MLP, PositionEmbeddingRandom +from comfy.ldm.sam3.sam import TwoWayTransformer as SAMTwoWayTransformer + +NO_OBJ_SCORE = -1024.0 + + +def to_spatial(x, H, W): + """Reshape (B, H*W, C) → (B, C, H, W).""" + return x.view(x.shape[0], H, W, -1).permute(0, 3, 1, 2) + +class MultiplexState: + """Tracks object-to-slot assignments for multiplex tracking. Provides mux/demux operations.""" + + def __init__(self, num_objects, multiplex_count, device, dtype): + self.multiplex_count = multiplex_count + self.device = device + self.dtype = dtype + self._build(num_objects) + + def mux(self, x): + """[N_obj, ...] -> [num_buckets, multiplex_count, ...]""" + out_shape = (self.num_buckets, self.multiplex_count) + x.shape[1:] + return (self.mux_matrix.to(device=x.device, dtype=x.dtype) @ x.reshape(self.total_valid_entries, -1)).view(out_shape) + + def demux(self, x): + """[num_buckets, multiplex_count, ...] -> [N_obj, ...]""" + out_shape = (self.total_valid_entries,) + x.shape[2:] + flat = x.reshape(self.num_buckets * self.multiplex_count, -1) + return (self.demux_matrix.to(device=x.device, dtype=x.dtype) @ flat).view(out_shape) + + def get_valid_object_mask(self): + """[num_buckets, multiplex_count] bool tensor, True for valid slots.""" + return (self.mux_matrix.sum(dim=1) > 0).reshape(self.num_buckets, self.multiplex_count) + + def _build(self, num_objects): + M = self.multiplex_count + self.num_buckets = (num_objects + M - 1) // M + self.total_valid_entries = num_objects + total_slots = self.num_buckets * M + self.mux_matrix = torch.zeros(total_slots, num_objects, device=self.device, dtype=self.dtype) + self.demux_matrix = torch.zeros(num_objects, total_slots, device=self.device, dtype=self.dtype) + oids = torch.arange(num_objects, device=self.device) + slots = (oids // M) * M + (oids % M) + self.mux_matrix[slots, oids] = 1.0 + self.demux_matrix[oids, slots] = 1.0 + + def add_objects(self, n_new): + """Grow multiplex state for n_new additional objects.""" + self._build(self.total_valid_entries + n_new) + +def _compute_mask_overlap(masks_a, masks_b): + """Max of IoU and IoM (intersection over minimum area). More robust to size differences.""" + a_flat = (masks_a > 0).float().flatten(1) + b_flat = (masks_b > 0).float().flatten(1) + intersection = a_flat @ b_flat.T + area_a = a_flat.sum(1, keepdim=True) + area_b = b_flat.sum(1, keepdim=True).T + iou = intersection / (area_a + area_b - intersection).clamp(min=1) + iom = intersection / torch.min(area_a.expand_as(iou), area_b.expand_as(iou)).clamp(min=1) + return torch.max(iou, iom) + + +def _nms_masks(masks, scores, thresh=0.5): + """Mask-based NMS using IoU+IoM overlap. Returns (filtered_masks, filtered_scores).""" + order = scores.argsort(descending=True) + masks, scores = masks[order], scores[order] + keep = [] + for i in range(masks.shape[0]): + if keep: + if _compute_mask_overlap(masks[i:i+1], masks[torch.tensor(keep, device=masks.device)]).max() >= thresh: + continue + keep.append(i) + return masks[keep], scores[keep] + + +def _get_connected_components(mask_bin): + """Get connected component labels and areas. mask_bin: [B, 1, H, W] uint8.""" + labels_list, areas_list = [], [] + for i in range(mask_bin.shape[0]): + m = mask_bin[i, 0].cpu().numpy() + if _HAS_CV2: + _, labeled, stats, _ = cv2.connectedComponentsWithStats(m, connectivity=8) + areas = stats[labeled, cv2.CC_STAT_AREA].astype('int32') + else: + labeled, num_features = ndimage.label(m) + areas = np.zeros_like(m, dtype=np.int32) + for c in range(1, num_features + 1): + component = labeled == c + areas[component] = component.sum() + labels_list.append(torch.from_numpy(labeled).to(mask_bin.device)) + areas_list.append(torch.from_numpy(areas).to(device=mask_bin.device, dtype=torch.int32)) + return torch.stack(labels_list).unsqueeze(1), torch.stack(areas_list).unsqueeze(1) + + +def fill_holes_in_mask_scores(mask, max_area=0): + """Remove small foreground sprinkles and fill small background holes using connected components.""" + if max_area <= 0: + return mask + + # Fill holes: small connected components in background → foreground + mask_bg = (mask <= 0).to(torch.uint8) + _, areas_bg = _get_connected_components(mask_bg) + small_bg = mask_bg.bool() & (areas_bg <= max_area) + mask = torch.where(small_bg, 0.1, mask) + + # Remove sprinkles: small connected components in foreground → background + # Only remove if area < min(max_area, half of total foreground area) + mask_fg = (mask > 0).to(torch.uint8) + fg_area_thresh = mask_fg.sum(dim=(2, 3), keepdim=True, dtype=torch.int32) + fg_area_thresh.floor_divide_(2).clamp_(max=max_area) + _, areas_fg = _get_connected_components(mask_fg) + small_fg = mask_fg.bool() & (areas_fg <= fg_area_thresh) + mask = torch.where(small_fg, -0.1, mask) + + return mask + + +def apply_rope_memory(q, k, freqs, num_heads, num_k_exclude_rope=0): + """Apply 2D axial RoPE to memory attention using flux rope format. + + Args: + q: [B, Nq, C] projected queries (current frame features) + k: [B, Nk, C] projected keys (memory tokens) + freqs: [1, Nq, dim//2, 2, 2] flux-format rotation matrices for one frame + num_heads: number of attention heads + num_k_exclude_rope: number of trailing k tokens to skip RoPE (object pointers) + """ + B, Nq, C = q.shape + head_dim = C // num_heads + + # freqs shape: [1, 1, Nq, dim//2, 2, 2] (heads broadcast dim already included) + q_h = q.view(B, Nq, num_heads, head_dim).transpose(1, 2) + q_h = apply_rope1(q_h, freqs) + q = q_h.transpose(1, 2).reshape(B, Nq, C) + + # Apply RoPE to k (excluding last num_k_exclude_rope tokens) + Nk = k.shape[1] + num_k_rope = Nk - num_k_exclude_rope + if num_k_rope > 0: + # Repeat freqs for multiple frames of spatial memory + Nf = freqs.shape[2] # spatial positions in one frame + if num_k_rope > Nf: + r = (num_k_rope + Nf - 1) // Nf + pe_k = freqs.repeat(1, 1, r, 1, 1, 1)[:, :, :num_k_rope] + else: + pe_k = freqs[:, :, :num_k_rope] + + k_h = k[:, :num_k_rope].view(B, num_k_rope, num_heads, head_dim).transpose(1, 2) + k_h = apply_rope1(k_h, pe_k) + k = k.clone() + k[:, :num_k_rope] = k_h.transpose(1, 2).reshape(B, num_k_rope, C) + + return q, k + + +def get_1d_sine_pe(pos_inds, dim, temperature=10000): + """1D sinusoidal positional encoding for temporal positions.""" + pe_dim = dim // 2 + dim_t = torch.arange(pe_dim, dtype=torch.float32, device=pos_inds.device) + dim_t = temperature ** (2 * (dim_t // 2) / pe_dim) + pos_embed = pos_inds.unsqueeze(-1) / dim_t + return torch.cat([pos_embed.sin(), pos_embed.cos()], dim=-1) + + +def _pad_to_buckets(tensor, target_buckets): + """Pad a [num_buckets, ...] tensor to target_buckets along dim 0 if needed.""" + if tensor.shape[0] >= target_buckets: + return tensor + pad_shape = (target_buckets - tensor.shape[0],) + tensor.shape[1:] + return torch.cat([tensor, torch.zeros(pad_shape, device=tensor.device, dtype=tensor.dtype)], dim=0) + + +def pack_masks(masks): + """Pack binary masks [*, H, W] to bit-packed [*, H, W//8] uint8. W must be divisible by 8.""" + binary = masks > 0 + shifts = torch.arange(8, device=masks.device) + return (binary.view(*masks.shape[:-1], -1, 8) * (1 << shifts)).sum(-1).byte() + + +def unpack_masks(packed): + """Unpack bit-packed [*, H, W//8] uint8 to bool [*, H, W*8].""" + bits = torch.tensor([1, 2, 4, 8, 16, 32, 64, 128], dtype=torch.uint8, device=packed.device) + return (packed.unsqueeze(-1) & bits).bool().view(*packed.shape[:-1], -1) + + +def _prep_frame(images, idx, device, dt, size): + """Slice CPU full-res frames, transfer to GPU in target dtype, and resize to (size, size).""" + return comfy.utils.common_upscale(images[idx].to(device=device, dtype=dt), size, size, "bicubic", crop="disabled") + + +def _compute_backbone(backbone_fn, frame, frame_idx=None): + """Compute backbone features for a single frame. Returns (vision_feats, vision_pos, feat_sizes, features, trunk_out).""" + features, positions, trunk_out = backbone_fn(frame, frame_idx=frame_idx) + feat_sizes = [(x.shape[-2], x.shape[-1]) for x in features] + vision_feats = [x.flatten(2).permute(0, 2, 1) for x in features] + vision_pos = [x.flatten(2).permute(0, 2, 1) for x in positions] + return vision_feats, vision_pos, feat_sizes, features, trunk_out + + +def collect_memory_tokens(output_dict, frame_idx, num_maskmem, maskmem_tpos_enc, device, + collect_image_feats=False, tpos_v2=False, num_buckets=None): + """Collect spatial memory, position encodings, and optionally image features from past frames.""" + to_cat_memory, to_cat_memory_pos = [], [] + to_cat_image_feat, to_cat_image_pos = [], [] + + def _append(out, tpos_idx): + feats = out["maskmem_features"].to(device) + if num_buckets is not None: + feats = _pad_to_buckets(feats, num_buckets) + to_cat_memory.append(feats.flatten(2).permute(0, 2, 1)) + enc = out["maskmem_pos_enc"][-1].to(device).flatten(2).permute(0, 2, 1) + if num_buckets is not None: + enc = _pad_to_buckets(enc, num_buckets) + tpos = cast_to_input(maskmem_tpos_enc[tpos_idx], enc) + to_cat_memory_pos.append(enc + tpos) + if collect_image_feats and "image_features" in out: + to_cat_image_feat.append(out["image_features"].to(device)) + to_cat_image_pos.append(out["image_pos_enc"].to(device) + tpos) + + cond_outputs = output_dict["cond_frame_outputs"] + for t, out in cond_outputs.items(): + if tpos_v2: + t_pos = frame_idx - t + tpos_idx = num_maskmem - t_pos - 1 if 0 < t_pos < num_maskmem else num_maskmem - 1 + else: + tpos_idx = num_maskmem - 1 + _append(out, tpos_idx) + + for t_pos in range(1, num_maskmem): + out = output_dict["non_cond_frame_outputs"].get(frame_idx - (num_maskmem - t_pos), None) + if out is None or out.get("maskmem_features") is None: + continue + _append(out, num_maskmem - t_pos - 1) + + return to_cat_memory, to_cat_memory_pos, to_cat_image_feat, to_cat_image_pos, cond_outputs + + +def compute_tpos_enc(rel_pos_list, device, d_model, proj_layer, dtype=None, max_abs_pos=None): + """Temporal position encoding for object pointers.""" + pos_enc = torch.tensor(rel_pos_list, dtype=torch.float32, device=device) / max((max_abs_pos or 2) - 1, 1) + pos_enc = get_1d_sine_pe(pos_enc, dim=d_model) + if dtype is not None: + pos_enc = pos_enc.to(dtype) + return proj_layer(pos_enc) + + +def forward_sam_heads(backbone_features, prompt_encoder, mask_decoder, obj_ptr_proj, no_obj_fn, + image_size, point_inputs=None, mask_inputs=None, box_inputs=None, + high_res_features=None, multimask_output=False): + """Shared SAM prompt encoder + mask decoder forward for both SAM3 and SAM3.1 trackers.""" + device = backbone_features.device + # Batch size from inputs (mask_inputs may have N_obj > 1 while backbone is batch 1) + if mask_inputs is not None: + B = mask_inputs.shape[0] + elif box_inputs is not None: + B = box_inputs.shape[0] + elif point_inputs is not None: + B = point_inputs["point_coords"].shape[0] + else: + B = backbone_features.shape[0] + + if point_inputs is not None: + sam_point_coords = point_inputs["point_coords"] + sam_point_labels = point_inputs["point_labels"] + else: + sam_point_coords = torch.zeros(B, 1, 2, device=device) + sam_point_labels = -torch.ones(B, 1, dtype=torch.int32, device=device) + + if mask_inputs is not None: + prompt_size = (prompt_encoder.image_embedding_size[0] * 4, prompt_encoder.image_embedding_size[1] * 4) + if mask_inputs.shape[-2:] != prompt_size: + sam_mask_prompt = F.interpolate(mask_inputs, size=prompt_size, mode="bilinear", align_corners=False, antialias=True) + else: + sam_mask_prompt = mask_inputs + else: + sam_mask_prompt = None + + sparse, dense = prompt_encoder(points=(sam_point_coords, sam_point_labels), boxes=box_inputs, masks=sam_mask_prompt) + sparse = cast_to_input(sparse, backbone_features) + dense = cast_to_input(dense, backbone_features) + image_pe = cast_to_input(prompt_encoder.get_dense_pe(), backbone_features) + + low_res_multimasks, ious, sam_output_tokens, object_score_logits = mask_decoder( + image_embeddings=backbone_features, image_pe=image_pe, + sparse_prompt_embeddings=sparse, dense_prompt_embeddings=dense, + high_res_features=high_res_features, multimask_output=multimask_output, return_all=True, + ) + + is_obj_appearing = object_score_logits > 0 + low_res_multimasks = torch.where(is_obj_appearing[:, None, None], low_res_multimasks, + torch.tensor(NO_OBJ_SCORE, device=device, dtype=low_res_multimasks.dtype)) + high_res_multimasks = F.interpolate(low_res_multimasks, size=(image_size, image_size), mode="bilinear", align_corners=False) + + sam_output_token = sam_output_tokens[:, 0] + if multimask_output: + best_iou_inds = torch.argmax(ious, dim=-1) + batch_inds = torch.arange(B, device=device) + low_res_masks = low_res_multimasks[batch_inds, best_iou_inds].unsqueeze(1) + high_res_masks = high_res_multimasks[batch_inds, best_iou_inds].unsqueeze(1) + if sam_output_tokens.size(1) > 1: + sam_output_token = sam_output_tokens[batch_inds, best_iou_inds] + else: + low_res_masks, high_res_masks = low_res_multimasks, high_res_multimasks + + obj_ptr = obj_ptr_proj(sam_output_token) + obj_ptr = no_obj_fn(obj_ptr, is_obj_appearing) + + return low_res_masks, high_res_masks, obj_ptr, object_score_logits + + +def use_mask_as_output(backbone_features, high_res_features, mask_inputs, mask_downsample, + prompt_encoder, mask_decoder, obj_ptr_proj, no_obj_fn, image_size, backbone_stride): + """Shared mask-as-output for both SAM3 and SAM3.1 trackers.""" + out_scale, out_bias = 20.0, -10.0 + mask_inputs_float = cast_to_input(mask_inputs, backbone_features) + high_res_masks = mask_inputs_float * out_scale + out_bias + low_res_masks = F.interpolate(high_res_masks, size=(image_size // backbone_stride * 4,) * 2, + mode="bilinear", align_corners=False, antialias=True) + _, _, obj_ptr, _ = forward_sam_heads( + backbone_features, prompt_encoder, mask_decoder, obj_ptr_proj, no_obj_fn, + image_size, mask_inputs=mask_downsample(mask_inputs_float), high_res_features=high_res_features, + ) + is_obj_appearing = torch.any(mask_inputs.flatten(1) > 0.0, dim=1)[..., None] + alpha = is_obj_appearing.to(obj_ptr.dtype) + object_score_logits = out_scale * alpha + out_bias + return low_res_masks, high_res_masks, obj_ptr, object_score_logits + + +# Split attention with configurable input dims (for asymmetric cross-attention) +class SplitAttn(nn.Module): + def __init__(self, embed_dim, num_heads=1, kv_dim=None, internal_dim=None, device=None, dtype=None, operations=None): + super().__init__() + self.num_heads = num_heads + kv_dim = kv_dim or embed_dim + internal_dim = internal_dim or embed_dim + self.q_proj = operations.Linear(embed_dim, internal_dim, device=device, dtype=dtype) + self.k_proj = operations.Linear(kv_dim, internal_dim, device=device, dtype=dtype) + self.v_proj = operations.Linear(kv_dim, internal_dim, device=device, dtype=dtype) + self.out_proj = operations.Linear(internal_dim, embed_dim, device=device, dtype=dtype) + + def forward(self, q, k=None, v=None, rope=None, num_k_exclude_rope=0): + if k is None: + k = q + if v is None: + v = k + q = self.q_proj(q) + k = self.k_proj(k) + v = self.v_proj(v) + if rope is not None: + q, k = apply_rope_memory(q, k, rope, self.num_heads, num_k_exclude_rope) + out = optimized_attention(q, k, v, self.num_heads, low_precision_attention=False) + return self.out_proj(out) + + +class MemoryAttnLayer(nn.Module): + def __init__(self, d_model=256, num_heads=1, kv_dim=64, dim_ff=2048, device=None, dtype=None, operations=None): + super().__init__() + self.num_heads = num_heads + self.self_attn = SplitAttn(d_model, num_heads, device=device, dtype=dtype, operations=operations) + self.cross_attn_image = SplitAttn(d_model, num_heads, kv_dim=kv_dim, device=device, dtype=dtype, operations=operations) + self.linear1 = operations.Linear(d_model, dim_ff, device=device, dtype=dtype) + self.linear2 = operations.Linear(dim_ff, d_model, device=device, dtype=dtype) + self.norm1 = operations.LayerNorm(d_model, device=device, dtype=dtype) + self.norm2 = operations.LayerNorm(d_model, device=device, dtype=dtype) + self.norm3 = operations.LayerNorm(d_model, device=device, dtype=dtype) + + def forward(self, x, memory, memory_pos=None, rope=None, num_k_exclude_rope=0): + x = x + self.self_attn(self.norm1(x), rope=rope) + mem_k = memory + memory_pos if memory_pos is not None else memory + x = x + self.cross_attn_image(self.norm2(x), mem_k, memory, rope=rope, num_k_exclude_rope=num_k_exclude_rope) + normed = self.norm3(x) + x = x + self.linear2(F.relu(self.linear1(normed))) + return x + + +class MemoryAttnEncoder(nn.Module): + def __init__(self, d_model=256, num_heads=1, kv_dim=64, dim_ff=2048, num_layers=4, image_size=1008, patch_size=14, + device=None, dtype=None, operations=None): + super().__init__() + self.layers = nn.ModuleList([ + MemoryAttnLayer(d_model, num_heads, kv_dim, dim_ff, device=device, dtype=dtype, operations=operations) + for _ in range(num_layers) + ]) + self.norm = operations.LayerNorm(d_model, device=device, dtype=dtype) + hw = image_size // patch_size + self.register_buffer("_rope", rope_2d(hw, hw, d_model // num_heads), persistent=False) + + def forward(self, x, memory, src_pos=None, memory_pos=None, num_k_exclude_rope=0): + if src_pos is not None: + x = x + 0.1 * src_pos + + rope = self._rope.to(device=x.device) + for layer in self.layers: + x = layer(x, memory, memory_pos=memory_pos, rope=rope, num_k_exclude_rope=num_k_exclude_rope) + return self.norm(x) + + +class MemoryTransformer(nn.Module): + def __init__(self, d_model=256, num_heads=1, kv_dim=64, dim_ff=2048, num_layers=4, device=None, dtype=None, operations=None): + super().__init__() + self.encoder = MemoryAttnEncoder(d_model, num_heads, kv_dim, dim_ff, num_layers, device=device, dtype=dtype, operations=operations) + + +def _upscale_masks(output_upscaling, conv_s0, conv_s1, src_out, high_res_features): + """Shared upscaling for SAM mask decoders: deconv + high-res feature integration.""" + dc1, ln1, act1, dc2, act2 = output_upscaling + if high_res_features is not None: + upscaled = act1(ln1(dc1(src_out) + conv_s1(high_res_features[1]))) + upscaled = act2(dc2(upscaled) + conv_s0(high_res_features[0])) + else: + upscaled = act2(dc2(act1(ln1(dc1(src_out))))) + return upscaled + + +class SAMMaskDecoder(nn.Module): + def __init__(self, d_model=256, num_multimask_outputs=3, device=None, dtype=None, operations=None): + super().__init__() + self.num_mask_tokens = num_multimask_outputs + 1 + + self.transformer = SAMTwoWayTransformer(depth=2, embedding_dim=d_model, num_heads=8, mlp_dim=2048, device=device, dtype=dtype, operations=operations) + + self.iou_token = operations.Embedding(1, d_model, device=device, dtype=dtype) + self.mask_tokens = operations.Embedding(self.num_mask_tokens, d_model, device=device, dtype=dtype) + self.obj_score_token = operations.Embedding(1, d_model, device=device, dtype=dtype) + + # Output upscaling: d_model -> d_model//4 -> d_model//8 at 4x resolution + LN2d = LayerNorm2d_op(operations) + self.output_upscaling = nn.Sequential( + operations.ConvTranspose2d(d_model, d_model // 4, kernel_size=2, stride=2, device=device, dtype=dtype), LN2d(d_model // 4, device=device, dtype=dtype), nn.GELU(), + operations.ConvTranspose2d(d_model // 4, d_model // 8, kernel_size=2, stride=2, device=device, dtype=dtype), nn.GELU(), + ) + + # High-res feature integration + self.conv_s0 = operations.Conv2d(d_model, d_model // 8, kernel_size=1, device=device, dtype=dtype) + self.conv_s1 = operations.Conv2d(d_model, d_model // 4, kernel_size=1, device=device, dtype=dtype) + + # Per-mask hypernetwork MLPs + self.output_hypernetworks_mlps = nn.ModuleList([ + MLP(d_model, d_model, d_model // 8, 3, device=device, dtype=dtype, operations=operations) + for _ in range(self.num_mask_tokens) + ]) + + self.iou_prediction_head = MLP(d_model, d_model, self.num_mask_tokens, 3, device=device, dtype=dtype, operations=operations) + self.pred_obj_score_head = MLP(d_model, d_model, 1, 3, device=device, dtype=dtype, operations=operations) + + def forward(self, image_embeddings, image_pe, sparse_prompt_embeddings, dense_prompt_embeddings, + high_res_features=None, multimask_output=False, return_all=False): + B = sparse_prompt_embeddings.shape[0] + ref = sparse_prompt_embeddings + # Token order: [obj_score(1), iou(1), mask(num_mask_tokens)] + tokens = torch.cat([cast_to_input(self.obj_score_token.weight, ref), + cast_to_input(self.iou_token.weight, ref), + cast_to_input(self.mask_tokens.weight, ref)], dim=0) + tokens = torch.cat([tokens.unsqueeze(0).expand(B, -1, -1), sparse_prompt_embeddings], dim=1) + + src = image_embeddings + if src.shape[0] != B: + src = src.expand(B, -1, -1, -1) + src = src + dense_prompt_embeddings + pos_src = image_pe.expand(B, -1, -1, -1) + + b, c, h, w = src.shape + src_flat = src.flatten(2).permute(0, 2, 1) + pos_flat = pos_src.flatten(2).permute(0, 2, 1) + + hs, src_out = self.transformer(src_flat, pos_flat, tokens) + + obj_score_token_out = hs[:, 0, :] + iou_token_out = hs[:, 1, :] + mask_tokens_out = hs[:, 2:2 + self.num_mask_tokens, :] + + src_out = src_out.permute(0, 2, 1).view(b, c, h, w) + upscaled = _upscale_masks(self.output_upscaling, self.conv_s0, self.conv_s1, src_out, high_res_features) + + hyper_in = torch.stack([ + mlp(mask_tokens_out[:, i, :]) for i, mlp in enumerate(self.output_hypernetworks_mlps) + ], dim=1) + + masks = (hyper_in @ upscaled.flatten(2)).view(B, self.num_mask_tokens, upscaled.shape[2], upscaled.shape[3]) + iou_pred = self.iou_prediction_head(iou_token_out) + object_score_logits = self.pred_obj_score_head(obj_score_token_out) + + if multimask_output: + out_masks = masks[:, 1:] + out_iou = iou_pred[:, 1:] + out_tokens = mask_tokens_out[:, 1:] + else: + out_masks = masks[:, 0:1] + out_iou = iou_pred[:, 0:1] + out_tokens = mask_tokens_out[:, 0:1] + + if return_all: + return out_masks, out_iou, out_tokens, object_score_logits + return out_masks, out_iou + + +class SAMPromptEncoder(nn.Module): + def __init__(self, d_model=256, image_embedding_size=(72, 72), input_image_size=(1008, 1008), device=None, dtype=None, operations=None): + super().__init__() + self.embed_dim = d_model + self.image_embedding_size = image_embedding_size + self.input_image_size = input_image_size + + self.pe_layer = PositionEmbeddingRandom(d_model // 2) + self.point_embeddings = nn.ModuleList([ + operations.Embedding(1, d_model, device=device, dtype=dtype) for _ in range(4) + ]) + self.not_a_point_embed = operations.Embedding(1, d_model, device=device, dtype=dtype) + + LN2d = LayerNorm2d_op(operations) + self.mask_downscaling = nn.Sequential( + operations.Conv2d(1, 4, kernel_size=2, stride=2, device=device, dtype=dtype), + LN2d(4, device=device, dtype=dtype), nn.GELU(), + operations.Conv2d(4, 16, kernel_size=2, stride=2, device=device, dtype=dtype), + LN2d(16, device=device, dtype=dtype), nn.GELU(), + operations.Conv2d(16, d_model, kernel_size=1, device=device, dtype=dtype), + ) + self.no_mask_embed = operations.Embedding(1, d_model, device=device, dtype=dtype) + + def get_dense_pe(self): + return self.pe_layer(self.image_embedding_size) + + def forward(self, points=None, boxes=None, masks=None): + ref = points[0] if points is not None else boxes if boxes is not None else masks + B = 1 + sparse = torch.empty((B, 0, self.embed_dim), device=ref.device, dtype=ref.dtype) + + if points is not None: + coords, labels = points + B = coords.shape[0] + # Pad with an extra point (label=-1) when no boxes are provided (matching reference) + if boxes is None: + coords = torch.cat([coords, torch.zeros(B, 1, 2, device=coords.device, dtype=coords.dtype)], dim=1) + labels = torch.cat([labels, -torch.ones(B, 1, device=labels.device, dtype=labels.dtype)], dim=1) + pe = self.pe_layer.forward_with_coords(coords + 0.5, self.input_image_size) + for i in range(4): + pe[labels == i] += cast_to_input(self.point_embeddings[i].weight, ref) + invalid = (labels == -1) + pe[invalid] = 0.0 + pe[invalid] += cast_to_input(self.not_a_point_embed.weight, ref) + sparse = torch.cat([sparse.expand(B, -1, -1), pe], dim=1) + + if boxes is not None: + B = boxes.shape[0] + corners = self.pe_layer.forward_with_coords((boxes.reshape(-1, 2, 2) + 0.5), self.input_image_size) + corners[:, 0] += cast_to_input(self.point_embeddings[2].weight, ref) + corners[:, 1] += cast_to_input(self.point_embeddings[3].weight, ref) + sparse = torch.cat([sparse.expand(B, -1, -1), corners], dim=1) + + if masks is not None: + dense = self.mask_downscaling(masks) + else: + dense = cast_to_input(self.no_mask_embed.weight, ref).reshape(1, -1, 1, 1).expand( + B, -1, self.image_embedding_size[0], self.image_embedding_size[1]) + + return sparse, dense + + +class CXBlock(nn.Module): + def __init__(self, dim=256, kernel_size=7, device=None, dtype=None, operations=None): + super().__init__() + self.dwconv = operations.Conv2d(dim, dim, kernel_size=kernel_size, padding=kernel_size // 2, groups=dim, device=device, dtype=dtype) + self.norm = operations.LayerNorm(dim, device=device, dtype=dtype) + self.pwconv1 = operations.Linear(dim, 4 * dim, device=device, dtype=dtype) + self.pwconv2 = operations.Linear(4 * dim, dim, device=device, dtype=dtype) + self.gamma = nn.Parameter(torch.ones(dim, device=device, dtype=dtype)) + + def forward(self, x): + residual = x + x = self.dwconv(x).permute(0, 2, 3, 1) + x = self.pwconv2(F.gelu(self.pwconv1(self.norm(x)))) + x.mul_(cast_to_input(self.gamma, x)) + return residual + x.permute(0, 3, 1, 2) + + +class MaskDownSampler(nn.Module): + def __init__(self, out_dim=256, in_chans=1, channels=None, interpol_size=(1152, 1152), device=None, dtype=None, operations=None): + super().__init__() + self.interpol_size = list(interpol_size) if interpol_size else None + if channels is None: + channels = [4, 16, 64, out_dim] # SAM3 default + LN2d = LayerNorm2d_op(operations) + layers = [] + prev = in_chans + for ch in channels: + layers += [operations.Conv2d(prev, ch, kernel_size=3, stride=2, padding=1, device=device, dtype=dtype), + LN2d(ch, device=device, dtype=dtype), nn.GELU()] + prev = ch + layers.append(operations.Conv2d(prev, out_dim, kernel_size=1, device=device, dtype=dtype)) + self.encoder = nn.Sequential(*layers) + + def forward(self, x): + if self.interpol_size is not None and list(x.shape[-2:]) != self.interpol_size: + x = F.interpolate(x, size=self.interpol_size, mode="bilinear", align_corners=False, antialias=True) + return self.encoder(x) + + +class Fuser(nn.Module): + def __init__(self, dim=256, num_layers=2, device=None, dtype=None, operations=None): + super().__init__() + self.layers = nn.Sequential(*[CXBlock(dim, device=device, dtype=dtype, operations=operations) for _ in range(num_layers)]) + + def forward(self, x): + return self.layers(x) + + +# --- SAM3.1 Multiplex components --- + +class DecoupledMemoryAttnLayer(nn.Module): + """Decoupled cross-attention layer for SAM3.1: fuses image and memory projections.""" + + def __init__(self, d_model=256, num_heads=1, dim_ff=2048, device=None, dtype=None, operations=None): + super().__init__() + self.num_heads = num_heads + # Self-attention projections (flat, not nested) + self.self_attn_q_proj = operations.Linear(d_model, d_model, device=device, dtype=dtype) + self.self_attn_k_proj = operations.Linear(d_model, d_model, device=device, dtype=dtype) + self.self_attn_v_proj = operations.Linear(d_model, d_model, device=device, dtype=dtype) + self.self_attn_out_proj = operations.Linear(d_model, d_model, device=device, dtype=dtype) + # Cross-attention projections + self.cross_attn_q_proj = operations.Linear(d_model, d_model, device=device, dtype=dtype) + self.cross_attn_k_proj = operations.Linear(d_model, d_model, device=device, dtype=dtype) + self.cross_attn_v_proj = operations.Linear(d_model, d_model, device=device, dtype=dtype) + self.cross_attn_out_proj = operations.Linear(d_model, d_model, device=device, dtype=dtype) + # Image cross-attention (q/k only, fused with cross_attn) + self.image_cross_attn_q_proj = operations.Linear(d_model, d_model, device=device, dtype=dtype) + self.image_cross_attn_k_proj = operations.Linear(d_model, d_model, device=device, dtype=dtype) + # FFN + self.linear1 = operations.Linear(d_model, dim_ff, device=device, dtype=dtype) + self.linear2 = operations.Linear(dim_ff, d_model, device=device, dtype=dtype) + self.norm1 = operations.LayerNorm(d_model, device=device, dtype=dtype) + self.norm2 = operations.LayerNorm(d_model, device=device, dtype=dtype) + self.norm3 = operations.LayerNorm(d_model, device=device, dtype=dtype) + + def forward(self, image, x, memory_image, memory, memory_image_pos=None, + rope=None, num_k_exclude_rope=0): + # Self-attention with RoPE + normed = self.norm1(x) + q = self.self_attn_q_proj(normed) + k = self.self_attn_k_proj(normed) + v = self.self_attn_v_proj(normed) + if rope is not None: + q, k = apply_rope_memory(q, k, rope, self.num_heads, 0) + x = x + self.self_attn_out_proj(optimized_attention(q, k, v, self.num_heads, low_precision_attention=False)) + + # Decoupled cross-attention: fuse image and memory projections + normed = self.norm2(x) + q = self.image_cross_attn_q_proj(image) + self.cross_attn_q_proj(normed) + k = self.image_cross_attn_k_proj(memory_image) + self.cross_attn_k_proj(memory) + if memory_image_pos is not None: + k = k + memory_image_pos + v = self.cross_attn_v_proj(memory) + if rope is not None: + q, k = apply_rope_memory(q, k, rope, self.num_heads, num_k_exclude_rope) + x = x + self.cross_attn_out_proj(optimized_attention(q, k, v, self.num_heads, low_precision_attention=False)) + + # FFN + x = x + self.linear2(F.gelu(self.linear1(self.norm3(x)))) + return image, x + + +class DecoupledMemoryEncoder(nn.Module): + """Memory attention encoder for SAM3.1 with decoupled cross-attention.""" + + def __init__(self, d_model=256, num_heads=1, dim_ff=2048, num_layers=4, image_size=1008, patch_size=14, + device=None, dtype=None, operations=None): + super().__init__() + self.layers = nn.ModuleList([ + DecoupledMemoryAttnLayer(d_model, num_heads, dim_ff, device=device, dtype=dtype, operations=operations) + for _ in range(num_layers) + ]) + self.norm = operations.LayerNorm(d_model, device=device, dtype=dtype) + hw = image_size // patch_size + self.register_buffer("_rope", rope_2d(hw, hw, d_model // num_heads), persistent=False) + + def forward(self, x, memory, memory_pos=None, src_pos=None, num_k_exclude_rope=0, + memory_image=None, memory_image_pos=None): + image = x # constant residual for decoupled cross-attention + output = x + if src_pos is not None: + output = output + 0.1 * src_pos + + B, _, C = x.shape + rope = self._rope.to(device=x.device) + + # memory_image: raw backbone features from past frames for decoupled cross-attention + if memory_image is None: + # Fallback: use spatial portion of memory (without obj pointers) + num_spatial = memory.shape[1] - num_k_exclude_rope + memory_image = memory[:, :num_spatial] + memory_image_pos = memory_pos[:, :num_spatial] if memory_pos is not None else None + # Pad memory_image to match memory length (zeros for obj pointer tokens) + if memory_image.shape[1] < memory.shape[1]: + pad_len = memory.shape[1] - memory_image.shape[1] + pad = torch.zeros(B, pad_len, C, device=memory.device, dtype=memory.dtype) + memory_image = torch.cat([memory_image, pad], dim=1) + if memory_image_pos is not None: + ptr_pos = memory_pos[:, -pad_len:] if memory_pos is not None else torch.zeros_like(pad) + memory_image_pos = torch.cat([memory_image_pos, ptr_pos], dim=1) + + for layer in self.layers: + image, output = layer(image, output, memory_image, memory, + memory_image_pos=memory_image_pos, rope=rope, + num_k_exclude_rope=num_k_exclude_rope) + + return self.norm(output) + + +class DecoupledMemoryTransformer(nn.Module): + def __init__(self, d_model=256, num_heads=1, dim_ff=2048, num_layers=4, device=None, dtype=None, operations=None): + super().__init__() + self.encoder = DecoupledMemoryEncoder(d_model, num_heads, dim_ff, num_layers, + device=device, dtype=dtype, operations=operations) + + +class MemoryBackbone(nn.Module): + """Memory encoder: downsamples mask, fuses with pixel features, optionally compresses.""" + + def __init__(self, d_model=256, out_dim=None, in_chans=1, channels=None, device=None, dtype=None, operations=None): + super().__init__() + self.mask_downsampler = MaskDownSampler(d_model, in_chans=in_chans, channels=channels, device=device, dtype=dtype, operations=operations) + self.pix_feat_proj = operations.Conv2d(d_model, d_model, kernel_size=1, device=device, dtype=dtype) + self.fuser = Fuser(d_model, num_layers=2, device=device, dtype=dtype, operations=operations) + self.has_out_proj = out_dim is not None and out_dim != d_model + if self.has_out_proj: + self.out_proj = operations.Conv2d(d_model, out_dim, kernel_size=1, device=device, dtype=dtype) + feat_dim = out_dim + else: + feat_dim = d_model + self.position_encoding = PositionEmbeddingSine(num_pos_feats=feat_dim, normalize=True) + + def forward(self, image_features, mask_for_mem, skip_mask_sigmoid=False): + if not skip_mask_sigmoid: + mask_for_mem = mask_for_mem.sigmoid() + mask_features = self.mask_downsampler(cast_to_input(mask_for_mem, image_features)) + if mask_features.shape[-2:] != image_features.shape[-2:]: + mask_features = F.interpolate(mask_features, size=image_features.shape[-2:], mode="bilinear", align_corners=False) + features = self.pix_feat_proj(image_features) + mask_features + features = self.fuser(features) + if self.has_out_proj: + features = self.out_proj(features) + pos = cast_to_input(self.position_encoding(features), features) + return {"vision_features": features, "vision_pos_enc": [pos]} + + +class MultiplexMaskDecoder(nn.Module): + """SAM mask decoder for SAM3.1 multiplex: predicts masks for num_multiplex objects simultaneously. + + Uses multimask_outputs_only=True: num_mask_output_per_object = num_multimask_outputs (no +1). + Hypernetwork MLPs are shared across multiplex objects. + Token order: [obj_score_token(M), iou_token(M), mask_tokens(M*T)]. + """ + + def __init__(self, d_model=256, num_multiplex=16, num_multimask_outputs=3, device=None, dtype=None, operations=None): + super().__init__() + self.num_multiplex = num_multiplex + self.num_mask_output_per_object = num_multimask_outputs # 3 (multimask_outputs_only) + total_mask_tokens = num_multiplex * self.num_mask_output_per_object # 48 + + self.transformer = SAMTwoWayTransformer(depth=2, embedding_dim=d_model, num_heads=8, mlp_dim=2048, device=device, dtype=dtype, operations=operations) + + self.obj_score_token = operations.Embedding(num_multiplex, d_model, device=device, dtype=dtype) + self.iou_token = operations.Embedding(num_multiplex, d_model, device=device, dtype=dtype) + self.mask_tokens = operations.Embedding(total_mask_tokens, d_model, device=device, dtype=dtype) + + LN2d = LayerNorm2d_op(operations) + self.output_upscaling = nn.Sequential( + operations.ConvTranspose2d(d_model, d_model // 4, kernel_size=2, stride=2, device=device, dtype=dtype), + LN2d(d_model // 4, device=device, dtype=dtype), nn.GELU(), + operations.ConvTranspose2d(d_model // 4, d_model // 8, kernel_size=2, stride=2, device=device, dtype=dtype), nn.GELU(), + ) + self.conv_s0 = operations.Conv2d(d_model, d_model // 8, kernel_size=1, device=device, dtype=dtype) + self.conv_s1 = operations.Conv2d(d_model, d_model // 4, kernel_size=1, device=device, dtype=dtype) + + # Shared across all multiplex objects (one per mask output) + self.output_hypernetworks_mlps = nn.ModuleList([ + MLP(d_model, d_model, d_model // 8, 3, device=device, dtype=dtype, operations=operations) + for _ in range(self.num_mask_output_per_object) + ]) + self.iou_prediction_head = MLP(d_model, d_model, self.num_mask_output_per_object, 3, device=device, dtype=dtype, operations=operations) + self.pred_obj_score_head = MLP(d_model, d_model, 1, 3, device=device, dtype=dtype, operations=operations) + + def forward(self, image_embeddings, image_pe, sparse_prompt_embeddings, dense_prompt_embeddings, + high_res_features=None, multimask_output=False, return_all=False, extra_per_object_embeddings=None): + B = sparse_prompt_embeddings.shape[0] + M = self.num_multiplex + T = self.num_mask_output_per_object + + # Token order: [obj_score(M), iou(M), mask(M*T)] + ref = sparse_prompt_embeddings + mask_tokens = cast_to_input(self.mask_tokens.weight, ref) + if extra_per_object_embeddings is not None: + mask_tokens = mask_tokens.view(1, M, T, -1).expand(B, -1, -1, -1) + extra_per_object_embeddings.unsqueeze(2) + mask_tokens = mask_tokens.flatten(1, 2) # [B, M*T, C] + other_tokens = torch.cat([cast_to_input(self.obj_score_token.weight, ref), + cast_to_input(self.iou_token.weight, ref)], dim=0).unsqueeze(0).expand(B, -1, -1) + tokens = torch.cat([other_tokens, mask_tokens, sparse_prompt_embeddings], dim=1) + else: + tokens = torch.cat([cast_to_input(self.obj_score_token.weight, ref), + cast_to_input(self.iou_token.weight, ref), mask_tokens], dim=0) + tokens = torch.cat([tokens.unsqueeze(0).expand(B, -1, -1), sparse_prompt_embeddings], dim=1) + + src = image_embeddings + if src.shape[0] != B: + src = src.expand(B, -1, -1, -1) + src = src + dense_prompt_embeddings + pos_src = image_pe.expand(B, -1, -1, -1) + + b, c, h, w = src.shape + hs, src_out = self.transformer(src.flatten(2).permute(0, 2, 1), pos_src.flatten(2).permute(0, 2, 1), tokens) + + # Parse output tokens + obj_score_token_out = hs[:, :M] + iou_token_out = hs[:, M:2 * M] + mask_tokens_out = hs[:, 2 * M:2 * M + M * T] + + src_out = src_out.permute(0, 2, 1).view(b, c, h, w) + upscaled = _upscale_masks(self.output_upscaling, self.conv_s0, self.conv_s1, src_out, high_res_features) + + # Reshape mask tokens to [B, M, T, C] and apply shared hypernetwork MLPs per mask output index + mask_tokens_2d = mask_tokens_out.view(B, M, T, -1) + hyper_in = torch.stack([ + self.output_hypernetworks_mlps[i](mask_tokens_2d[:, :, i, :]) # [B, M, C//8] + for i in range(T) + ], dim=2) # [B, M, T, C//8] + + # Generate masks: [B, M*T, H*W] -> [B, M, T, H, W] + masks = torch.bmm(hyper_in.flatten(1, 2), upscaled.flatten(2)).view(b, M, T, upscaled.shape[2], upscaled.shape[3]) + + # IoU and object scores + iou_pred = self.iou_prediction_head(iou_token_out).view(b, M, T) + object_score_logits = self.pred_obj_score_head(obj_score_token_out) # [B, M, 1] + + # multimask_outputs_only: always output all T masks (no singlemask token) + sam_tokens_out = mask_tokens_2d[:, :, 0:1] # [B, M, 1, C] + + if return_all: + return masks, iou_pred, sam_tokens_out, object_score_logits + return masks, iou_pred + + +class SAM3Tracker(nn.Module): + def __init__(self, d_model=256, mem_dim=64, num_maskmem=7, device=None, dtype=None, operations=None, **kwargs): + super().__init__() + + # Memory attention transformer + self.transformer = MemoryTransformer(d_model, num_heads=1, kv_dim=mem_dim, dim_ff=2048, num_layers=4, + device=device, dtype=dtype, operations=operations) + # SAM components + self.sam_mask_decoder = SAMMaskDecoder(d_model, device=device, dtype=dtype, operations=operations) + self.sam_prompt_encoder = SAMPromptEncoder(d_model, device=device, dtype=dtype, operations=operations) + + # Memory backbone + self.maskmem_backbone = MemoryBackbone(d_model, out_dim=mem_dim, device=device, dtype=dtype, operations=operations) + + # Standalone parameters + self.maskmem_tpos_enc = nn.Parameter(torch.zeros(num_maskmem, 1, 1, mem_dim, device=device, dtype=dtype)) + self.no_mem_embed = nn.Parameter(torch.zeros(1, 1, d_model, device=device, dtype=dtype)) + self.register_buffer("no_mem_pos_enc", torch.zeros(1, 1, d_model, device=device, dtype=dtype)) # checkpoint key, unused in forward + self.no_obj_embed_spatial = nn.Parameter(torch.zeros(1, mem_dim, device=device, dtype=dtype)) + self.no_obj_ptr = nn.Parameter(torch.zeros(1, d_model, device=device, dtype=dtype)) + + # Object pointer projection + self.obj_ptr_proj = MLP(d_model, d_model, d_model, 3, device=device, dtype=dtype, operations=operations) + self.obj_ptr_tpos_proj = operations.Linear(d_model, mem_dim, device=device, dtype=dtype) + + # Mask downsample: Conv2d stride 4 to reduce GT mask to SAM logit scale + self.mask_downsample = operations.Conv2d(1, 1, kernel_size=4, stride=4, device=device, dtype=dtype) + + # Config + self.d_model = d_model + self.mem_dim = mem_dim + self.num_maskmem = num_maskmem + self.image_size = 1008 + self.backbone_stride = 14 + self.max_obj_ptrs_in_encoder = 16 + self.sigmoid_scale_for_mem_enc = 20.0 + self.sigmoid_bias_for_mem_enc = -10.0 + + def _no_obj_blend(self, obj_ptr, is_obj): + alpha = is_obj.to(obj_ptr.dtype) + return torch.lerp(cast_to_input(self.no_obj_ptr, obj_ptr), obj_ptr, alpha) + + def _forward_sam_heads(self, backbone_features, point_inputs=None, mask_inputs=None, box_inputs=None, + high_res_features=None, multimask_output=False): + return forward_sam_heads(backbone_features, self.sam_prompt_encoder, self.sam_mask_decoder, + self.obj_ptr_proj, self._no_obj_blend, self.image_size, + point_inputs, mask_inputs, box_inputs, high_res_features, multimask_output) + + def _use_mask_as_output(self, backbone_features, high_res_features, mask_inputs): + return use_mask_as_output(backbone_features, high_res_features, mask_inputs, + self.mask_downsample, self.sam_prompt_encoder, self.sam_mask_decoder, + self.obj_ptr_proj, self._no_obj_blend, self.image_size, self.backbone_stride) + + def _prepare_memory_conditioned_features(self, frame_idx, is_init_cond_frame, current_vision_feats, current_vision_pos_embeds, feat_sizes, output_dict, num_frames): + """Fuse current frame features with memory from previous frames.""" + B = current_vision_feats[-1].shape[0] + C = self.d_model + H, W = feat_sizes[-1] + device = current_vision_feats[-1].device + + if self.num_maskmem == 0: + return current_vision_feats[-1].permute(0, 2, 1).view(B, C, H, W) + + if is_init_cond_frame: + # First conditioning frame: no memory yet, add no_mem_embed + pix_feat = current_vision_feats[-1] + cast_to_input(self.no_mem_embed, current_vision_feats[-1]) + return to_spatial(pix_feat, H, W) + + to_cat_memory, to_cat_memory_pos, _, _, cond_outputs = collect_memory_tokens( + output_dict, frame_idx, self.num_maskmem, self.maskmem_tpos_enc, device) + + max_obj_ptrs = min(num_frames, self.max_obj_ptrs_in_encoder) + pos_and_ptrs = [] + for t, out in cond_outputs.items(): + if t <= frame_idx: + pos_and_ptrs.append(((frame_idx - t), out["obj_ptr"].to(device))) + for t_diff in range(1, max_obj_ptrs): + t = frame_idx - t_diff + if t < 0: + break + out = output_dict["non_cond_frame_outputs"].get(t, None) + if out is not None: + pos_and_ptrs.append((t_diff, out["obj_ptr"].to(device))) + + num_obj_ptr_tokens = 0 + if len(pos_and_ptrs) > 0: + pos_list, ptrs_list = zip(*pos_and_ptrs) + obj_ptrs = torch.stack(ptrs_list, dim=1) # [B, N, C=256] + + # Temporal position encoding for pointers + obj_pos = compute_tpos_enc( + list(pos_list), device, self.d_model, self.obj_ptr_tpos_proj, + max_abs_pos=max_obj_ptrs, dtype=current_vision_feats[-1].dtype + ) # [N, mem_dim=64] + obj_pos = obj_pos.unsqueeze(0).expand(B, -1, -1) # [B, N, 64] + + # Split each 256-dim pointer into 4 x 64-dim tokens + if self.mem_dim < C: + N = obj_ptrs.shape[1] + obj_ptrs = obj_ptrs.view(B, N, C // self.mem_dim, self.mem_dim) # [B, N, 4, 64] + obj_ptrs = obj_ptrs.reshape(B, N * (C // self.mem_dim), self.mem_dim) # [B, N*4, 64] + obj_pos = obj_pos.unsqueeze(2).expand(-1, -1, C // self.mem_dim, -1) + obj_pos = obj_pos.reshape(B, N * (C // self.mem_dim), self.mem_dim) # [B, N*4, 64] + + to_cat_memory.append(obj_ptrs) + to_cat_memory_pos.append(obj_pos) + num_obj_ptr_tokens = obj_ptrs.shape[1] + + if len(to_cat_memory) == 0: + # No memory available yet, add no_mem_embed + pix_feat = current_vision_feats[-1] + cast_to_input(self.no_mem_embed, current_vision_feats[-1]) + return to_spatial(pix_feat, H, W) + + # Concatenate all memory and position encodings [B, total_mem, mem_dim=64] + memory = torch.cat(to_cat_memory, dim=1) + memory_pos = torch.cat(to_cat_memory_pos, dim=1) + + # Run memory attention encoder + pix_feat = current_vision_feats[-1] # [B, HW, C] + src_pos = current_vision_pos_embeds[-1] # [B, HW, C] + + pix_feat_with_mem = self.transformer.encoder( + x=pix_feat, + memory=memory, + src_pos=src_pos, + memory_pos=memory_pos, + num_k_exclude_rope=num_obj_ptr_tokens, + ) + return to_spatial(pix_feat_with_mem, H, W) + + def _encode_new_memory(self, pix_feat, pred_masks_high_res, object_score_logits, is_mask_from_pts=False): + """Encode predicted mask into memory features.""" + if is_mask_from_pts: + mask_for_mem = (pred_masks_high_res > 0).to(pix_feat.dtype) + else: + mask_for_mem = torch.sigmoid(pred_masks_high_res) + + mask_for_mem.mul_(self.sigmoid_scale_for_mem_enc).add_(self.sigmoid_bias_for_mem_enc) + + maskmem_out = self.maskmem_backbone(pix_feat, mask_for_mem, skip_mask_sigmoid=True) + maskmem_features = maskmem_out["vision_features"] + maskmem_pos_enc = maskmem_out["vision_pos_enc"] + + # Add no_obj_embed for occluded objects + alpha = (object_score_logits > 0).to(maskmem_features.dtype)[..., None, None] + no_obj = cast_to_input(self.no_obj_embed_spatial, maskmem_features)[..., None, None].expand_as(maskmem_features) + return maskmem_features + (1 - alpha) * no_obj, maskmem_pos_enc + + def track_step(self, frame_idx, is_init_cond_frame, current_vision_feats, current_vision_pos_embeds, feat_sizes, mask_inputs, output_dict, + num_frames, point_inputs=None): + """Track one frame: fuse with memory, predict mask, encode memory.""" + current_out = {} + + # High-res features for SAM head [stride-8, stride-4] + if len(current_vision_feats) > 1: + high_res_features = [ + x.view(x.shape[0], feat_sizes[i][0], feat_sizes[i][1], -1).permute(0, 3, 1, 2) + for i, x in enumerate(current_vision_feats[:-1]) + ] + else: + high_res_features = None + + # Top-level feature for memory + H, W = feat_sizes[-1] + + if mask_inputs is not None: + # Conditioning frame: use mask directly + pix_feat = to_spatial(current_vision_feats[-1], H, W) + sam_outputs = self._use_mask_as_output(pix_feat, high_res_features, mask_inputs) + else: + # Track frame: fuse with memory, then SAM decoder + pix_feat_with_mem = self._prepare_memory_conditioned_features( + frame_idx=frame_idx, + is_init_cond_frame=is_init_cond_frame, + current_vision_feats=current_vision_feats, + current_vision_pos_embeds=current_vision_pos_embeds, + feat_sizes=feat_sizes, + output_dict=output_dict, + num_frames=num_frames, + ) + # Use multimask for point prompts on init frames (picks best of 3 candidates) + num_pts = 0 if point_inputs is None else point_inputs["point_labels"].size(1) + multimask_output = is_init_cond_frame and 0 < num_pts <= 1 + sam_outputs = self._forward_sam_heads( + backbone_features=pix_feat_with_mem, + point_inputs=point_inputs, + high_res_features=high_res_features, + multimask_output=multimask_output, + ) + + (low_res_masks, high_res_masks, obj_ptr, object_score_logits) = sam_outputs + + # Clean low-res masks: remove sprinkles and fill holes + low_res_masks = fill_holes_in_mask_scores(low_res_masks, max_area=200) + high_res_masks = F.interpolate(low_res_masks, size=(self.image_size, self.image_size), mode="bilinear", align_corners=False) + + current_out["pred_masks"] = low_res_masks + current_out["pred_masks_high_res"] = high_res_masks + current_out["obj_ptr"] = obj_ptr + current_out["object_score_logits"] = object_score_logits + + # Encode memory + if self.num_maskmem > 0: + pix_feat = to_spatial(current_vision_feats[-1], H, W) + maskmem_features, maskmem_pos_enc = self._encode_new_memory( + pix_feat=pix_feat, + pred_masks_high_res=high_res_masks, + object_score_logits=object_score_logits, + is_mask_from_pts=(point_inputs is not None), + ) + current_out["maskmem_features"] = maskmem_features + current_out["maskmem_pos_enc"] = maskmem_pos_enc + else: + current_out["maskmem_features"] = None + current_out["maskmem_pos_enc"] = None + + return current_out + + def _compute_backbone_frame(self, backbone_fn, frame, frame_idx=None): + vision_feats, vision_pos, feat_sizes, _, _ = _compute_backbone(backbone_fn, frame, frame_idx) + # SAM3: drop last FPN level + return vision_feats[:-1], vision_pos[:-1], feat_sizes[:-1] + + def _track_single_object(self, backbone_fn, images, initial_mask, pbar=None, + target_device=None, target_dtype=None): + """Track one object, computing backbone per frame to save VRAM.""" + N = images.shape[0] + device = target_device if target_device is not None else images.device + dt = target_dtype if target_dtype is not None else images.dtype + size = self.image_size + output_dict = {"cond_frame_outputs": {}, "non_cond_frame_outputs": {}} + all_masks = [] + + for frame_idx in tqdm(range(N), desc="tracking"): + vision_feats, vision_pos, feat_sizes = self._compute_backbone_frame( + backbone_fn, _prep_frame(images, slice(frame_idx, frame_idx + 1), device, dt, size), frame_idx=frame_idx) + mask_input = None + if frame_idx == 0: + mask_input = F.interpolate(initial_mask.to(device=device, dtype=dt), + size=(self.image_size, self.image_size), mode="bilinear", align_corners=False) + mask_input = (mask_input > 0.5).to(dt) + + current_out = self.track_step( + frame_idx=frame_idx, is_init_cond_frame=(frame_idx == 0), + current_vision_feats=vision_feats, current_vision_pos_embeds=vision_pos, + feat_sizes=feat_sizes, mask_inputs=mask_input, output_dict=output_dict, num_frames=N) + + if frame_idx == 0: + output_dict["cond_frame_outputs"][frame_idx] = current_out + else: + output_dict["non_cond_frame_outputs"][frame_idx] = current_out + lookback = max(self.num_maskmem, self.max_obj_ptrs_in_encoder) + for old_idx in list(output_dict["non_cond_frame_outputs"]): + if old_idx < frame_idx - lookback: + del output_dict["non_cond_frame_outputs"][old_idx] + # Move masks to CPU immediately to free VRAM + all_masks.append(current_out["pred_masks_high_res"].to(comfy.model_management.intermediate_device())) + if pbar is not None: + pbar.update(1) + + return torch.cat(all_masks, dim=0) # [N, 1, H, W] + + def track_video(self, backbone_fn, images, initial_masks, pbar=None, + target_device=None, target_dtype=None, **kwargs): + """Track one or more objects across video frames. + + Args: + backbone_fn: callable that returns (sam2_features, sam2_positions, trunk_out) for a frame + images: [N, 3, H, W] CPU full-res video frames (resized per-frame to self.image_size) + initial_masks: [N_obj, 1, H, W] binary masks for first frame (one per object) + pbar: optional progress bar + + Returns: + [N, N_obj, image_size, image_size] predicted mask logits per frame per object + """ + N_obj = initial_masks.shape[0] + per_object = [] + for obj_idx in range(N_obj): + obj_masks = self._track_single_object( + backbone_fn, images, initial_masks[obj_idx:obj_idx + 1], pbar=pbar, + target_device=target_device, target_dtype=target_dtype) + per_object.append(obj_masks) + + return torch.cat(per_object, dim=1) # [N, N_obj, H, W] + + +class SAM31Tracker(nn.Module): + """SAM3.1 multiplex tracker: decoupled memory attention, dual decoder, 16-object multiplex.""" + + def __init__(self, d_model=256, mem_dim=256, num_maskmem=7, num_multiplex=16, device=None, dtype=None, operations=None, **kwargs): + super().__init__() + self.d_model = d_model + self.mem_dim = mem_dim + self.num_maskmem = num_maskmem + self.num_multiplex = num_multiplex + self.image_size = 1008 + self.backbone_stride = 14 + self.max_obj_ptrs_in_encoder = 16 + self.sigmoid_scale_for_mem_enc = 2.0 + self.sigmoid_bias_for_mem_enc = -1.0 + + # Memory attention (decoupled cross-attention, 8 heads matching reference) + self.transformer = DecoupledMemoryTransformer(d_model, num_heads=8, dim_ff=2048, num_layers=4, + device=device, dtype=dtype, operations=operations) + + # Propagation decoder (multiplex: 16 objects, multimask_outputs_only) + self.sam_mask_decoder = MultiplexMaskDecoder(d_model, num_multiplex, num_multimask_outputs=3, + device=device, dtype=dtype, operations=operations) + # Interactive decoder (single object, same as SAM3) + self.interactive_sam_mask_decoder = SAMMaskDecoder(d_model, num_multimask_outputs=3, + device=device, dtype=dtype, operations=operations) + self.interactive_sam_prompt_encoder = SAMPromptEncoder(d_model, device=device, dtype=dtype, operations=operations) + + # Memory backbone (mem_dim=256, no out_proj compression) + self.maskmem_backbone = MemoryBackbone(d_model, in_chans=num_multiplex * 2, channels=[16, 64, 256, 1024], + device=device, dtype=dtype, operations=operations) + + # Standalone parameters + self.maskmem_tpos_enc = nn.Parameter(torch.zeros(num_maskmem, 1, 1, mem_dim, device=device, dtype=dtype)) + self.no_obj_embed_spatial = nn.Parameter(torch.zeros(num_multiplex, mem_dim, device=device, dtype=dtype)) + self.interactivity_no_mem_embed = nn.Parameter(torch.zeros(1, 1, d_model, device=device, dtype=dtype)) + + # Object pointer projection + self.obj_ptr_proj = MLP(d_model, d_model, d_model, 3, device=device, dtype=dtype, operations=operations) + self.obj_ptr_tpos_proj = operations.Linear(d_model, mem_dim, device=device, dtype=dtype) + self.no_obj_ptr_linear = operations.Linear(d_model, d_model, device=device, dtype=dtype) + self.interactive_obj_ptr_proj = MLP(d_model, d_model, d_model, 3, device=device, dtype=dtype, operations=operations) + + # Interactive mask downsample + self.interactive_mask_downsample = operations.Conv2d(1, 1, kernel_size=4, stride=4, device=device, dtype=dtype) + + # Multiplex validity embeddings + self.output_valid_embed = nn.Parameter(torch.zeros(num_multiplex, d_model, device=device, dtype=dtype)) + self.output_invalid_embed = nn.Parameter(torch.zeros(num_multiplex, d_model, device=device, dtype=dtype)) + + # Position encoding for image (used by multiplex decoder) + self.image_pe_layer = PositionEmbeddingRandom(d_model // 2) + + def _no_obj_blend(self, obj_ptr, is_obj): + alpha = is_obj.to(obj_ptr.dtype) + return torch.lerp(self.no_obj_ptr_linear(obj_ptr), obj_ptr, alpha) + + def _forward_sam_heads(self, backbone_features, point_inputs=None, mask_inputs=None, box_inputs=None, + high_res_features=None, multimask_output=False): + return forward_sam_heads(backbone_features, self.interactive_sam_prompt_encoder, self.interactive_sam_mask_decoder, + self.interactive_obj_ptr_proj, self._no_obj_blend, self.image_size, + point_inputs, mask_inputs, box_inputs, high_res_features, multimask_output) + + def _use_mask_as_output(self, backbone_features, high_res_features, mask_inputs): + return use_mask_as_output(backbone_features, high_res_features, mask_inputs, + self.interactive_mask_downsample, self.interactive_sam_prompt_encoder, + self.interactive_sam_mask_decoder, self.interactive_obj_ptr_proj, + self._no_obj_blend, self.image_size, self.backbone_stride) + + def _prepare_memory_conditioned_features(self, frame_idx, is_init_cond_frame, current_vision_feats, + current_vision_pos_embeds, feat_sizes, output_dict, num_frames, + multiplex_state=None): + B = current_vision_feats[-1].shape[0] + C = self.d_model + H, W = feat_sizes[-1] + device = current_vision_feats[-1].device + num_buc = multiplex_state.num_buckets if multiplex_state is not None else None + + if self.num_maskmem == 0: + return current_vision_feats[-1].permute(0, 2, 1).view(B, C, H, W) + + if is_init_cond_frame: + pix_feat = current_vision_feats[-1] + cast_to_input(self.interactivity_no_mem_embed, current_vision_feats[-1]) + return to_spatial(pix_feat, H, W) + + to_cat_memory, to_cat_memory_pos, to_cat_image_feat, to_cat_image_pos, cond_outputs = collect_memory_tokens( + output_dict, frame_idx, self.num_maskmem, self.maskmem_tpos_enc, device, + collect_image_feats=True, tpos_v2=True, num_buckets=num_buc) + + max_obj_ptrs = min(num_frames, self.max_obj_ptrs_in_encoder) + pos_and_ptrs = [] + for t, out in cond_outputs.items(): + if t <= frame_idx and "obj_ptr" in out: + ptr = out["obj_ptr"].to(device) + if num_buc is not None: + ptr = _pad_to_buckets(ptr, num_buc) + pos_and_ptrs.append(((frame_idx - t), ptr)) + for t_diff in range(1, max_obj_ptrs): + t = frame_idx - t_diff + if t < 0: + break + out = output_dict["non_cond_frame_outputs"].get(t, None) + if out is not None and "obj_ptr" in out: + ptr = out["obj_ptr"].to(device) + if num_buc is not None: + ptr = _pad_to_buckets(ptr, num_buc) + pos_and_ptrs.append((t_diff, ptr)) + + num_obj_ptr_tokens = 0 + if len(pos_and_ptrs) > 0: + pos_list, ptrs_list = zip(*pos_and_ptrs) + obj_ptrs = torch.stack(ptrs_list, dim=1) # [num_buckets, N, M, C] + B_ptr = obj_ptrs.shape[0] + N_ptrs = obj_ptrs.shape[1] + M = obj_ptrs.shape[2] + obj_ptrs = obj_ptrs.reshape(B_ptr, N_ptrs * M, -1) + obj_pos = compute_tpos_enc(list(pos_list), device, self.d_model, self.obj_ptr_tpos_proj, + max_abs_pos=max_obj_ptrs, dtype=current_vision_feats[-1].dtype) + obj_pos = obj_pos.unsqueeze(0).expand(B_ptr, -1, -1) + obj_pos = obj_pos.unsqueeze(2).expand(-1, -1, M, -1).reshape(B_ptr, N_ptrs * M, -1) + to_cat_memory.append(obj_ptrs) + to_cat_memory_pos.append(obj_pos) + num_obj_ptr_tokens = obj_ptrs.shape[1] + + if len(to_cat_memory) == 0: + pix_feat = current_vision_feats[-1] + cast_to_input(self.interactivity_no_mem_embed, current_vision_feats[-1]) + return to_spatial(pix_feat, H, W) + + memory = torch.cat(to_cat_memory, dim=1) + memory_pos = torch.cat(to_cat_memory_pos, dim=1) + + # Expand vision features to num_buckets if memory has more buckets than B + mem_B = memory.shape[0] + x = current_vision_feats[-1] + x_pos = current_vision_pos_embeds[-1] + if x.shape[0] < mem_B: + x = x.expand(mem_B, -1, -1) + x_pos = x_pos.expand(mem_B, -1, -1) + + if len(to_cat_image_feat) > 0: + # Decoupled cross-attention: separate image features from memory + memory_image = cast_to_input(torch.cat(to_cat_image_feat, dim=1), x) + memory_image_pos = cast_to_input(torch.cat(to_cat_image_pos, dim=1), x) + if memory_image.shape[0] < mem_B: + memory_image = memory_image.expand(mem_B, -1, -1) + memory_image_pos = memory_image_pos.expand(mem_B, -1, -1) + pix_feat_with_mem = self.transformer.encoder( + x=x, + memory=cast_to_input(memory, x), + memory_pos=cast_to_input(memory_pos, x), + src_pos=cast_to_input(x_pos, x), + num_k_exclude_rope=num_obj_ptr_tokens, + memory_image=memory_image, + memory_image_pos=memory_image_pos, + ) + else: + pix_feat_with_mem = self.transformer.encoder( + x=x, + memory=memory, + memory_pos=memory_pos, + src_pos=x_pos, + num_k_exclude_rope=num_obj_ptr_tokens, + ) + return to_spatial(pix_feat_with_mem, H, W) + + def _encode_new_memory(self, pix_feat, pred_masks_high_res, object_score_logits, is_mask_from_pts=False, + multiplex_state=None, is_conditioning=False, cond_obj_mask=None): + if is_mask_from_pts: + mask_for_mem = (pred_masks_high_res > 0).to(pix_feat.dtype) + else: + mask_for_mem = torch.sigmoid(pred_masks_high_res) + mask_for_mem.mul_(self.sigmoid_scale_for_mem_enc).add_(self.sigmoid_bias_for_mem_enc) + + # Mux masks: [N_obj, 1, H, W] -> [num_buckets, M, H, W] + mux_masks = multiplex_state.mux(mask_for_mem[:, 0]) + + # Conditioning channel: 1.0 = clean mask (trust it), 0.0 = propagation (noisy) + N_obj = mask_for_mem.shape[0] + cond_values = torch.full((N_obj,), 0.0, device=mask_for_mem.device, dtype=mask_for_mem.dtype) + if is_conditioning: + cond_values[:] = 1.0 + elif cond_obj_mask is not None: + cond_values[cond_obj_mask] = 1.0 + cond_spatial = cond_values.view(-1, 1, 1, 1).expand_as(mask_for_mem[:, 0:1, :, :]).squeeze(1) + mux_cond = multiplex_state.mux(cond_spatial) # [num_buckets, M, H, W] + mux_input = torch.cat([mux_masks, mux_cond], dim=1) # [num_buckets, 2*M, H, W] + + maskmem_out = self.maskmem_backbone(pix_feat, mux_input, skip_mask_sigmoid=True) + maskmem_features = maskmem_out["vision_features"] + maskmem_pos_enc = maskmem_out["vision_pos_enc"] + + # Add no_obj_embed_spatial for occluded objects + is_obj = (object_score_logits > 0).float() # [N_obj, 1] + mux_is_obj = multiplex_state.mux(is_obj) # [num_buckets, M, 1] + no_obj_embed = cast_to_input(self.no_obj_embed_spatial, maskmem_features) # [M, C] + no_obj_spatial = no_obj_embed.unsqueeze(0)[..., None, None] # [1, M, C, 1, 1] + # Expand and sum across multiplex slots weighted by (1 - is_obj) + alpha = mux_is_obj[..., None, None] # [num_buckets, M, 1, 1, 1] + per_slot_no_obj = ((1 - alpha) * no_obj_spatial).sum(dim=1) # [num_buckets, C, 1, 1] + maskmem_features = maskmem_features + per_slot_no_obj.expand_as(maskmem_features) + + return maskmem_features, maskmem_pos_enc + + def _forward_propagation(self, backbone_features, high_res_features=None, multiplex_state=None): + """Propagation path using the multiplex SAM decoder (no prompts).""" + B = backbone_features.shape[0] + device = backbone_features.device + + # Suppression embeddings from valid object mask + valid_mask = cast_to_input(multiplex_state.get_valid_object_mask().unsqueeze(-1).float(), backbone_features) + output_valid = cast_to_input(self.output_valid_embed, backbone_features).unsqueeze(0) + output_invalid = cast_to_input(self.output_invalid_embed, backbone_features).unsqueeze(0) + extra_embed = valid_mask * output_valid + (1 - valid_mask) * output_invalid + + image_pe = self.image_pe_layer((backbone_features.shape[-2], backbone_features.shape[-1]), device=backbone_features.device) + image_pe = cast_to_input(image_pe, backbone_features) + + masks, iou_pred, sam_tokens_out, object_score_logits = self.sam_mask_decoder( + image_embeddings=backbone_features, image_pe=image_pe, + sparse_prompt_embeddings=torch.empty(B, 0, self.d_model, device=device, dtype=backbone_features.dtype), + dense_prompt_embeddings=torch.zeros(B, self.d_model, *backbone_features.shape[-2:], device=device, dtype=backbone_features.dtype), + high_res_features=high_res_features, multimask_output=True, return_all=True, + extra_per_object_embeddings=extra_embed.expand(B, -1, -1), + ) + # masks: [B=num_buckets, M, T, H, W] + # Demux to per-object: [N_obj, T, H, W] + masks_obj = multiplex_state.demux(masks) + iou_obj = multiplex_state.demux(iou_pred) + score_obj = multiplex_state.demux(object_score_logits) + tokens_obj = multiplex_state.demux(sam_tokens_out) + + # Select best mask by IoU for each object + best_idx = torch.argmax(iou_obj, dim=-1) # [N_obj] + N_obj = masks_obj.shape[0] + obj_range = torch.arange(N_obj, device=device) + low_res_masks = masks_obj[obj_range, best_idx].unsqueeze(1) # [N_obj, 1, H, W] + # Suppress masks for objects with low confidence + is_obj = score_obj > 0 + low_res_masks = torch.where(is_obj[:, :, None, None], low_res_masks, + torch.tensor(NO_OBJ_SCORE, device=device, dtype=low_res_masks.dtype)) + high_res_masks = F.interpolate(low_res_masks.float(), size=(self.image_size, self.image_size), mode="bilinear", align_corners=False) + + # Object pointer: compute per-object, mux for storage as [num_buckets, M, C] + sam_token = tokens_obj[:, 0] # [N_obj, C] + obj_ptr = self.obj_ptr_proj(sam_token) + is_obj = (score_obj > 0).float() + no_obj = self.no_obj_ptr_linear(obj_ptr) + obj_ptr = is_obj * obj_ptr + (1 - is_obj) * no_obj + obj_ptr_muxed = multiplex_state.mux(obj_ptr) # [num_buckets, M, C] + + return low_res_masks, high_res_masks, obj_ptr_muxed, score_obj + + def track_step(self, frame_idx, is_init_cond_frame, current_vision_feats, current_vision_pos_embeds, + feat_sizes, mask_inputs, output_dict, num_frames, point_inputs=None, + interactive_high_res=None, interactive_backbone=None, propagation_high_res=None, + multiplex_state=None, run_mem_encoder=True): + current_out = {} + H, W = feat_sizes[-1] + + if mask_inputs is not None: + # Conditioning frame: use interactive features if available, else propagation + if interactive_backbone is not None: + pix_feat = interactive_backbone + # Add no_mem_embed for interactive path + pix_flat = pix_feat.flatten(2) + bf = pix_flat.permute(0, 2, 1) + cast_to_input(self.interactivity_no_mem_embed, pix_flat) + pix_feat = to_spatial(bf, H, W) + hi_res = interactive_high_res + else: + # Fallback: interactive backbone not available (e.g. called outside track_video). + # Propagation features work but may produce lower-quality conditioning. + pix_feat = to_spatial(current_vision_feats[-1], H, W) + hi_res = propagation_high_res + sam_outputs = self._use_mask_as_output(pix_feat, hi_res, mask_inputs) + elif point_inputs is not None: + # Interactive path: use interactive SAM decoder + pix_feat_with_mem = self._prepare_memory_conditioned_features( + frame_idx=frame_idx, is_init_cond_frame=is_init_cond_frame, + current_vision_feats=current_vision_feats, current_vision_pos_embeds=current_vision_pos_embeds, + feat_sizes=feat_sizes, output_dict=output_dict, num_frames=num_frames, + multiplex_state=multiplex_state, + ) + hi_res = interactive_high_res if interactive_high_res is not None else propagation_high_res + num_pts = point_inputs["point_labels"].size(1) + multimask_output = is_init_cond_frame and 0 < num_pts <= 1 + sam_outputs = self._forward_sam_heads( + backbone_features=pix_feat_with_mem, point_inputs=point_inputs, + high_res_features=hi_res, multimask_output=multimask_output, + ) + else: + # Propagation path: use multiplex SAM decoder with propagation features + pix_feat_with_mem = self._prepare_memory_conditioned_features( + frame_idx=frame_idx, is_init_cond_frame=is_init_cond_frame, + current_vision_feats=current_vision_feats, current_vision_pos_embeds=current_vision_pos_embeds, + feat_sizes=feat_sizes, output_dict=output_dict, num_frames=num_frames, + multiplex_state=multiplex_state, + ) + sam_outputs = self._forward_propagation(pix_feat_with_mem, propagation_high_res, + multiplex_state=multiplex_state) + + (low_res_masks, high_res_masks, obj_ptr, object_score_logits) = sam_outputs + + # Mux obj_ptr if it came from interactive path (shape [B, C]) vs propagation ([num_buckets, M, C]) + if multiplex_state is not None and obj_ptr.dim() == 2: + obj_ptr = multiplex_state.mux(obj_ptr) # [N_obj, C] -> [num_buckets, M, C] + + # Encode memory (can be deferred with run_mem_encoder=False) + if run_mem_encoder and self.num_maskmem > 0: + pix_feat = to_spatial(current_vision_feats[-1], H, W) + maskmem_features, maskmem_pos_enc = self._encode_new_memory( + pix_feat=pix_feat, pred_masks_high_res=high_res_masks, + object_score_logits=object_score_logits, + is_mask_from_pts=(point_inputs is not None), + multiplex_state=multiplex_state, + is_conditioning=(mask_inputs is not None), + ) + current_out["maskmem_features"] = maskmem_features + current_out["maskmem_pos_enc"] = maskmem_pos_enc + else: + current_out["maskmem_features"] = None + current_out["maskmem_pos_enc"] = None + + # Store propagation image features for decoupled memory attention + current_out["image_features"] = current_vision_feats[-1] # [B, HW, C] + current_out["image_pos_enc"] = current_vision_pos_embeds[-1] # [B, HW, C] + + current_out["pred_masks"] = low_res_masks + current_out["pred_masks_high_res"] = high_res_masks + current_out["obj_ptr"] = obj_ptr + current_out["object_score_logits"] = object_score_logits + + return current_out + + def _compute_backbone_frame(self, backbone_fn, frame, frame_idx=None): + vision_feats, vision_pos, feat_sizes, features, trunk_out = _compute_backbone(backbone_fn, frame, frame_idx) + return vision_feats, vision_pos, feat_sizes, list(features[:-1]), trunk_out + + @staticmethod + def _suppress_recently_occluded(low_res_masks, last_occluded, frame_idx, threshold=0.3): + """Suppress overlapping masks for objects that were most recently occluded. + Prevents corrupted masks from occluded objects from contaminating other objects.""" + N_obj = low_res_masks.shape[0] + if N_obj <= 1: + return low_res_masks + binary = low_res_masks[:, 0] > 0 # [N_obj, H, W] + iou = _compute_mask_overlap(low_res_masks[:, 0], low_res_masks[:, 0]) + overlapping = torch.triu(iou >= threshold, diagonal=1) # [N, N] upper triangle + last_occ_i = last_occluded.unsqueeze(1) # [N, 1] + last_occ_j = last_occluded.unsqueeze(0) # [1, N] + # Suppress the more recently occluded object in each overlapping pair + suppress_i = overlapping & (last_occ_i > last_occ_j) & (last_occ_j > -1) + suppress_j = overlapping & (last_occ_j > last_occ_i) & (last_occ_i > -1) + to_suppress = suppress_i.any(dim=1) | suppress_j.any(dim=0) + # Update last_occluded for occluded/suppressed objects + is_empty = ~binary.any(dim=(-1, -2)) + newly_occluded = is_empty | to_suppress + last_occluded[newly_occluded] = frame_idx + # Suppress masks + low_res_masks[to_suppress] = -10.0 + return low_res_masks + + def _deferred_memory_encode(self, current_out, N_obj, vision_feats, feat_sizes, mux_state, device, + cond_obj_mask=None): + """Deferred memory encoding for propagation frames. cond_obj_mask: per-object bool for conditioning.""" + low_res_masks = current_out["pred_masks"] # [N_obj, 1, H_low, W_low] + + if N_obj > 1: + lr = low_res_masks.squeeze(1) # [N_obj, H, W] + max_obj = torch.argmax(lr, dim=0, keepdim=True) + batch_inds = torch.arange(N_obj, device=device)[:, None, None] + pixel_nol = torch.where(max_obj == batch_inds, lr, torch.clamp(lr, max=-10.0)) + area_before = (lr > 0).sum(dim=(-1, -2)).float().clamp(min=1) + area_after = (pixel_nol > 0).sum(dim=(-1, -2)).float() + shrink_ok = (area_after / area_before) >= 0.3 + low_res_masks = torch.where( + shrink_ok[:, None, None, None].expand_as(low_res_masks), + low_res_masks, torch.clamp(low_res_masks, max=-10.0)) + + interpol_size = self.maskmem_backbone.mask_downsampler.interpol_size + mem_masks = F.interpolate(low_res_masks, size=interpol_size, + mode="bilinear", align_corners=False) + + obj_scores = torch.where( + (mem_masks > 0).any(dim=(-1, -2)), 10.0, -10.0) + + pix_feat = to_spatial(vision_feats[-1], feat_sizes[-1][0], feat_sizes[-1][1]) + maskmem_features, maskmem_pos_enc = self._encode_new_memory( + pix_feat=pix_feat, pred_masks_high_res=mem_masks, + object_score_logits=obj_scores, + multiplex_state=mux_state, cond_obj_mask=cond_obj_mask) + current_out["maskmem_features"] = maskmem_features + current_out["maskmem_pos_enc"] = maskmem_pos_enc + + def _add_detected_objects(self, new_masks, mux_state, vision_feats, feat_sizes, current_out): + """Grow MultiplexState with new detections, merge masks, re-encode memory. Modifies current_out.""" + n_old = mux_state.total_valid_entries + mux_state.add_objects(new_masks.shape[0]) + N_obj = mux_state.total_valid_entries + # Stored memory with old bucket counts is padded at read time by _pad_to_buckets + for k in ("pred_masks", "pred_masks_high_res"): + det = F.interpolate(new_masks.unsqueeze(1), size=current_out[k].shape[-2:], + mode="bilinear", align_corners=False) + current_out[k] = torch.cat([current_out[k], det], dim=0) + if self.num_maskmem > 0: + # Mark new objects as conditioning (clean detection masks) so model trusts them + cond_mask = torch.zeros(N_obj, dtype=torch.bool, device=new_masks.device) + cond_mask[n_old:] = True + self._deferred_memory_encode(current_out, N_obj, vision_feats, feat_sizes, + mux_state, new_masks.device, cond_obj_mask=cond_mask) + + def _condition_with_masks(self, masks, frame_idx, vision_feats, vision_pos, feat_sizes, + high_res_prop, output_dict, N, mux_state, backbone_obj, frame, + trunk_out, threshold=0.5): + """Condition tracker with masks on a frame.""" + mask_input = F.interpolate(masks if masks.dim() == 4 else masks.unsqueeze(1), + size=(self.image_size, self.image_size), mode="bilinear", align_corners=False) + mask_input = (mask_input > threshold).to(masks.dtype) + hi_res = lo_feat = None + if backbone_obj is not None and backbone_obj.multiplex: + _, _, itf, _ = backbone_obj(frame, tracker_mode="interactive", cached_trunk=trunk_out, tracker_only=True) + hi_res, lo_feat = itf[:-1], itf[-1] + current_out = self.track_step( + frame_idx=frame_idx, is_init_cond_frame=True, current_vision_feats=vision_feats, + current_vision_pos_embeds=vision_pos, feat_sizes=feat_sizes, mask_inputs=mask_input, + output_dict=output_dict, num_frames=N, interactive_high_res=hi_res, + interactive_backbone=lo_feat, propagation_high_res=high_res_prop, + multiplex_state=mux_state, run_mem_encoder=True) + output_dict["cond_frame_outputs"][frame_idx] = current_out + return current_out + + def _match_and_add_detections(self, det_masks, det_scores, current_out, mux_state, + vision_feats, feat_sizes, device, max_objects=0, + keep_alive=None): + """Match detections against tracked masks, add new objects, recondition degraded tracks. + Updates keep_alive counters: +1 for matched tracks, -1 for unmatched.""" + N_obj = mux_state.total_valid_entries + if det_masks.shape[0] == 0: + if keep_alive is not None: + for i in range(N_obj): + keep_alive[i] = max(-4, keep_alive.get(i, 0) - 1) + return [] + + # Match at low-res (like reference) + trk_masks = current_out["pred_masks"][:, 0] # [N_obj, H_low, W_low] + det_resized = F.interpolate(det_masks.unsqueeze(1), size=trk_masks.shape[-2:], + mode="bilinear", align_corners=False)[:, 0] + overlap = _compute_mask_overlap(det_resized, trk_masks) + + # Update keep_alive and find matched tracks + matched = set() + if overlap.shape[1] > 0: + matched = set((overlap >= 0.5).any(dim=0).nonzero(as_tuple=True)[0].tolist()) + if keep_alive is not None: + for i in range(N_obj): + if i in matched: + keep_alive[i] = min(8, keep_alive.get(i, 0) + 1) + else: + keep_alive[i] = max(-4, keep_alive.get(i, 0) - 1) + + # Recondition: high-confidence detections (>=0.8) with high overlap refresh tracked masks + reconditioned = False + if det_scores is not None and overlap.shape[1] > 0: + HIGH_CONF = 0.8 + for det_idx in range(overlap.shape[0]): + if det_scores[det_idx] < HIGH_CONF: + continue + best_trk = overlap[det_idx].argmax().item() + if overlap[det_idx, best_trk] >= 0.5: + # Replace tracked mask with fresh detection mask + current_out["pred_masks"][best_trk] = det_resized[det_idx].unsqueeze(0) + det_hr = F.interpolate(det_masks[det_idx:det_idx+1].unsqueeze(1), + size=current_out["pred_masks_high_res"].shape[-2:], + mode="bilinear", align_corners=False) + current_out["pred_masks_high_res"][best_trk] = det_hr[0] + reconditioned = True + + # Re-encode memory if any tracks were reconditioned + if reconditioned and self.num_maskmem > 0: + self._deferred_memory_encode(current_out, N_obj, vision_feats, feat_sizes, mux_state, device) + + # Add new detections (not matching any track) + if max_objects > 0 and N_obj >= max_objects: + return [] + max_overlap = overlap.max(dim=1)[0] if overlap.shape[1] > 0 else torch.zeros(overlap.shape[0], device=device) + new_dets = max_overlap < 0.5 + if new_dets.any(): + if max_objects > 0: + slots = max_objects - N_obj + new_dets = new_dets & (torch.cumsum(new_dets.int(), 0) <= slots) + self._add_detected_objects(det_masks[new_dets], mux_state, + vision_feats, feat_sizes, current_out) + if keep_alive is not None: + for i in range(N_obj, mux_state.total_valid_entries): + keep_alive[i] = 1 + return det_scores[new_dets].tolist() if det_scores is not None else [0.0] * new_dets.sum().item() + return [] + + INTERNAL_MAX_OBJECTS = 64 # Hard ceiling on accumulated tracks; max_objects=0 or any value above this is clamped here. + + def track_video_with_detection(self, backbone_fn, images, initial_masks, detect_fn=None, + new_det_thresh=0.5, max_objects=0, detect_interval=1, + backbone_obj=None, pbar=None, target_device=None, target_dtype=None): + """Track with optional per-frame detection. Returns [N, max_N_obj, H, W] mask logits.""" + if max_objects <= 0 or max_objects > self.INTERNAL_MAX_OBJECTS: + max_objects = self.INTERNAL_MAX_OBJECTS + N = images.shape[0] + device = target_device if target_device is not None else images.device + dt = target_dtype if target_dtype is not None else images.dtype + size = self.image_size + output_dict = {"cond_frame_outputs": {}, "non_cond_frame_outputs": {}} + all_masks = [] + idev = comfy.model_management.intermediate_device() + mux_state = None + if initial_masks is not None: + mux_state = MultiplexState(initial_masks.shape[0], self.num_multiplex, device, dt) + obj_scores = [] # per-object detection score (1.0 for initial masks) + keep_alive = {} if detect_fn is not None else None + last_occluded = torch.empty(0, device=device, dtype=torch.long) # per-object last occluded frame + + # Prefetch next frame's backbone on a separate CUDA stream + prefetch = False + backbone_stream = None + if comfy.model_management.is_device_cuda(device): + try: + backbone_stream = torch.cuda.Stream(device=device) + prefetch = True + except RuntimeError: + pass + cur_bb = self._compute_backbone_frame(backbone_fn, _prep_frame(images, slice(0, 1), device, dt, size), frame_idx=0) + + for frame_idx in tqdm(range(N), desc="tracking"): + vision_feats, vision_pos, feat_sizes, high_res_prop, trunk_out = cur_bb + + # Start next frame's backbone on separate stream (overlaps with current frame's work) + if prefetch and frame_idx + 1 < N: + backbone_stream.wait_stream(torch.cuda.current_stream(device)) + with torch.cuda.stream(backbone_stream): + next_bb = self._compute_backbone_frame( + backbone_fn, _prep_frame(images, slice(frame_idx + 1, frame_idx + 2), device, dt, size), frame_idx=frame_idx + 1) + + # Per-frame detection with NMS (skip if no detect_fn, or interval/max not met) + det_masks = torch.empty(0, device=device) + det_scores = None + run_det = (detect_fn is not None + and frame_idx % max(detect_interval, 1) == 0 + and not (max_objects > 0 and mux_state is not None + and mux_state.total_valid_entries >= max_objects)) + if run_det: + det_out = detect_fn(trunk_out) + scores = det_out["scores"][0].sigmoid() + keep = scores > new_det_thresh + det_masks, det_scores = det_out["masks"][0][keep], scores[keep] + if det_masks.shape[0] > 1: + det_masks, det_scores = _nms_masks(det_masks, det_scores) + + if frame_idx == 0 and initial_masks is not None: + current_out = self._condition_with_masks( + initial_masks.to(device=device, dtype=dt), frame_idx, vision_feats, vision_pos, + feat_sizes, high_res_prop, output_dict, N, mux_state, backbone_obj, + _prep_frame(images, slice(frame_idx, frame_idx + 1), device, dt, size), trunk_out) + last_occluded = torch.full((mux_state.total_valid_entries,), -1, device=device, dtype=torch.long) + obj_scores = [1.0] * mux_state.total_valid_entries + if keep_alive is not None: + for i in range(mux_state.total_valid_entries): + keep_alive[i] = 8 + elif mux_state is None or mux_state.total_valid_entries == 0: + if det_masks.shape[0] > 0: + if max_objects > 0: + det_scores = det_scores[:max_objects] + det_masks = det_masks[:max_objects] + mux_state = MultiplexState(det_masks.shape[0], self.num_multiplex, device, dt) + current_out = self._condition_with_masks( + det_masks, frame_idx, vision_feats, vision_pos, feat_sizes, high_res_prop, + output_dict, N, mux_state, backbone_obj, + _prep_frame(images, slice(frame_idx, frame_idx + 1), device, dt, size), trunk_out, threshold=0.0) + last_occluded = torch.full((mux_state.total_valid_entries,), -1, device=device, dtype=torch.long) + obj_scores = det_scores[:mux_state.total_valid_entries].tolist() + if keep_alive is not None: + for i in range(mux_state.total_valid_entries): + keep_alive[i] = 1 + else: + all_masks.append(None) + if pbar is not None: + pbar.update(1) + # Skip to backbone advance at end of loop + if frame_idx + 1 < N: + if prefetch: + torch.cuda.current_stream(device).wait_stream(backbone_stream) + cur_bb = next_bb + else: + cur_bb = self._compute_backbone_frame(backbone_fn, _prep_frame(images, slice(frame_idx + 1, frame_idx + 2), device, dt, size), frame_idx=frame_idx + 1) + continue + else: + N_obj = mux_state.total_valid_entries + current_out = self.track_step( + frame_idx=frame_idx, is_init_cond_frame=False, current_vision_feats=vision_feats, + current_vision_pos_embeds=vision_pos, feat_sizes=feat_sizes, mask_inputs=None, + output_dict=output_dict, num_frames=N, propagation_high_res=high_res_prop, + multiplex_state=mux_state, run_mem_encoder=False) + current_out["pred_masks"] = fill_holes_in_mask_scores( + current_out["pred_masks"], max_area=16) + if last_occluded.shape[0] == N_obj and N_obj > 1: + self._suppress_recently_occluded( + current_out["pred_masks"], last_occluded, frame_idx) + if self.num_maskmem > 0: + self._deferred_memory_encode(current_out, N_obj, vision_feats, feat_sizes, mux_state, device) + output_dict["non_cond_frame_outputs"][frame_idx] = current_out + lookback = max(self.num_maskmem, self.max_obj_ptrs_in_encoder) + for old_idx in list(output_dict["non_cond_frame_outputs"]): + if old_idx < frame_idx - lookback: + del output_dict["non_cond_frame_outputs"][old_idx] + n_before = mux_state.total_valid_entries + new_obj_scores = self._match_and_add_detections(det_masks, det_scores, current_out, mux_state, + vision_feats, feat_sizes, device, max_objects, + keep_alive if run_det else None) + n_added = mux_state.total_valid_entries - n_before + if n_added > 0: + last_occluded = torch.cat([last_occluded, + torch.full((n_added,), -1, device=device, dtype=torch.long)]) + obj_scores.extend(new_obj_scores) + + masks_out = current_out["pred_masks_high_res"][:, 0] + if keep_alive is not None: + for i in range(masks_out.shape[0]): + if keep_alive.get(i, 0) <= 0: + masks_out[i] = NO_OBJ_SCORE + N_obj_now = mux_state.total_valid_entries if mux_state is not None else 0 + if N_obj_now > 0: + all_masks.append(pack_masks(masks_out).to(idev)) + else: + all_masks.append(None) + if pbar is not None: + pbar.update(1) + + # Next frame's backbone + if frame_idx + 1 < N: + if prefetch: + torch.cuda.current_stream(device).wait_stream(backbone_stream) + cur_bb = next_bb + else: + cur_bb = self._compute_backbone_frame(backbone_fn, _prep_frame(images, slice(frame_idx + 1, frame_idx + 2), device, dt, size), frame_idx=frame_idx + 1) + + if not all_masks or all(m is None for m in all_masks): + return {"packed_masks": None, "n_frames": N, "scores": []} + + max_obj = max(m.shape[0] for m in all_masks if m is not None) + sample = next(m for m in all_masks if m is not None) + empty_packed = torch.zeros(max_obj, *sample.shape[1:], dtype=torch.uint8, device=sample.device) + for i, m in enumerate(all_masks): + if m is None: + all_masks[i] = empty_packed + elif m.shape[0] < max_obj: + pad = torch.zeros(max_obj - m.shape[0], *m.shape[1:], dtype=torch.uint8, device=m.device) + all_masks[i] = torch.cat([m, pad], dim=0) + return {"packed_masks": torch.stack(all_masks, dim=0), "n_frames": N, "scores": obj_scores} diff --git a/comfy/ldm/supir/__init__.py b/comfy/ldm/supir/__init__.py new file mode 100644 index 000000000..e69de29bb diff --git a/comfy/ldm/supir/supir_modules.py b/comfy/ldm/supir/supir_modules.py new file mode 100644 index 000000000..7389b01d2 --- /dev/null +++ b/comfy/ldm/supir/supir_modules.py @@ -0,0 +1,226 @@ +import torch +import torch.nn as nn + +from comfy.ldm.modules.diffusionmodules.util import timestep_embedding +from comfy.ldm.modules.diffusionmodules.openaimodel import Downsample, TimestepEmbedSequential, ResBlock, SpatialTransformer +from comfy.ldm.modules.attention import optimized_attention + + +class ZeroSFT(nn.Module): + def __init__(self, label_nc, norm_nc, concat_channels=0, dtype=None, device=None, operations=None): + super().__init__() + + ks = 3 + pw = ks // 2 + + self.param_free_norm = operations.GroupNorm(32, norm_nc + concat_channels, dtype=dtype, device=device) + + nhidden = 128 + + self.mlp_shared = nn.Sequential( + operations.Conv2d(label_nc, nhidden, kernel_size=ks, padding=pw, dtype=dtype, device=device), + nn.SiLU() + ) + self.zero_mul = operations.Conv2d(nhidden, norm_nc + concat_channels, kernel_size=ks, padding=pw, dtype=dtype, device=device) + self.zero_add = operations.Conv2d(nhidden, norm_nc + concat_channels, kernel_size=ks, padding=pw, dtype=dtype, device=device) + + self.zero_conv = operations.Conv2d(label_nc, norm_nc, 1, 1, 0, dtype=dtype, device=device) + self.pre_concat = bool(concat_channels != 0) + + def forward(self, c, h, h_ori=None, control_scale=1): + if h_ori is not None and self.pre_concat: + h_raw = torch.cat([h_ori, h], dim=1) + else: + h_raw = h + + h = h + self.zero_conv(c) + if h_ori is not None and self.pre_concat: + h = torch.cat([h_ori, h], dim=1) + actv = self.mlp_shared(c) + gamma = self.zero_mul(actv) + beta = self.zero_add(actv) + h = self.param_free_norm(h) + h = torch.addcmul(h + beta, h, gamma) + if h_ori is not None and not self.pre_concat: + h = torch.cat([h_ori, h], dim=1) + return torch.lerp(h_raw, h, control_scale) + + +class _CrossAttnInner(nn.Module): + """Inner cross-attention module matching the state_dict layout of the original CrossAttention.""" + def __init__(self, query_dim, context_dim, heads, dim_head, dtype=None, device=None, operations=None): + super().__init__() + inner_dim = dim_head * heads + self.heads = heads + self.to_q = operations.Linear(query_dim, inner_dim, bias=False, dtype=dtype, device=device) + self.to_k = operations.Linear(context_dim, inner_dim, bias=False, dtype=dtype, device=device) + self.to_v = operations.Linear(context_dim, inner_dim, bias=False, dtype=dtype, device=device) + self.to_out = nn.Sequential( + operations.Linear(inner_dim, query_dim, dtype=dtype, device=device), + ) + + def forward(self, x, context): + q = self.to_q(x) + k = self.to_k(context) + v = self.to_v(context) + return self.to_out(optimized_attention(q, k, v, self.heads)) + + +class ZeroCrossAttn(nn.Module): + def __init__(self, context_dim, query_dim, dtype=None, device=None, operations=None): + super().__init__() + heads = query_dim // 64 + dim_head = 64 + self.attn = _CrossAttnInner(query_dim, context_dim, heads, dim_head, dtype=dtype, device=device, operations=operations) + self.norm1 = operations.GroupNorm(32, query_dim, dtype=dtype, device=device) + self.norm2 = operations.GroupNorm(32, context_dim, dtype=dtype, device=device) + + def forward(self, context, x, control_scale=1): + b, c, h, w = x.shape + x_in = x + + x = self.attn( + self.norm1(x).flatten(2).transpose(1, 2), + self.norm2(context).flatten(2).transpose(1, 2), + ).transpose(1, 2).unflatten(2, (h, w)) + + return x_in + x * control_scale + + +class GLVControl(nn.Module): + """SUPIR's Guided Latent Vector control encoder. Truncated UNet (input + middle blocks only).""" + def __init__( + self, + in_channels=4, + model_channels=320, + num_res_blocks=2, + attention_resolutions=(4, 2), + channel_mult=(1, 2, 4), + num_head_channels=64, + transformer_depth=(1, 2, 10), + context_dim=2048, + adm_in_channels=2816, + use_linear_in_transformer=True, + use_checkpoint=False, + dtype=None, + device=None, + operations=None, + **kwargs, + ): + super().__init__() + self.model_channels = model_channels + time_embed_dim = model_channels * 4 + + self.time_embed = nn.Sequential( + operations.Linear(model_channels, time_embed_dim, dtype=dtype, device=device), + nn.SiLU(), + operations.Linear(time_embed_dim, time_embed_dim, dtype=dtype, device=device), + ) + + self.label_emb = nn.Sequential( + nn.Sequential( + operations.Linear(adm_in_channels, time_embed_dim, dtype=dtype, device=device), + nn.SiLU(), + operations.Linear(time_embed_dim, time_embed_dim, dtype=dtype, device=device), + ) + ) + + self.input_blocks = nn.ModuleList([ + TimestepEmbedSequential( + operations.Conv2d(in_channels, model_channels, 3, padding=1, dtype=dtype, device=device) + ) + ]) + ch = model_channels + ds = 1 + for level, mult in enumerate(channel_mult): + for nr in range(num_res_blocks): + layers = [ + ResBlock(ch, time_embed_dim, 0, out_channels=mult * model_channels, + dtype=dtype, device=device, operations=operations) + ] + ch = mult * model_channels + if ds in attention_resolutions: + num_heads = ch // num_head_channels + layers.append( + SpatialTransformer(ch, num_heads, num_head_channels, + depth=transformer_depth[level], context_dim=context_dim, + use_linear=use_linear_in_transformer, + use_checkpoint=use_checkpoint, + dtype=dtype, device=device, operations=operations) + ) + self.input_blocks.append(TimestepEmbedSequential(*layers)) + if level != len(channel_mult) - 1: + self.input_blocks.append( + TimestepEmbedSequential( + Downsample(ch, True, out_channels=ch, dtype=dtype, device=device, operations=operations) + ) + ) + ds *= 2 + + num_heads = ch // num_head_channels + self.middle_block = TimestepEmbedSequential( + ResBlock(ch, time_embed_dim, 0, dtype=dtype, device=device, operations=operations), + SpatialTransformer(ch, num_heads, num_head_channels, + depth=transformer_depth[-1], context_dim=context_dim, + use_linear=use_linear_in_transformer, + use_checkpoint=use_checkpoint, + dtype=dtype, device=device, operations=operations), + ResBlock(ch, time_embed_dim, 0, dtype=dtype, device=device, operations=operations), + ) + + self.input_hint_block = TimestepEmbedSequential( + operations.Conv2d(in_channels, model_channels, 3, padding=1, dtype=dtype, device=device) + ) + + def forward(self, x, timesteps, xt, context=None, y=None, **kwargs): + t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False).to(x.dtype) + emb = self.time_embed(t_emb) + self.label_emb(y) + + guided_hint = self.input_hint_block(x, emb, context) + + hs = [] + h = xt + for module in self.input_blocks: + if guided_hint is not None: + h = module(h, emb, context) + h += guided_hint + guided_hint = None + else: + h = module(h, emb, context) + hs.append(h) + h = self.middle_block(h, emb, context) + hs.append(h) + return hs + + +class SUPIR(nn.Module): + """ + SUPIR model containing GLVControl (control encoder) and project_modules (adapters). + State dict keys match the original SUPIR checkpoint layout: + control_model.* -> GLVControl + project_modules.* -> nn.ModuleList of ZeroSFT/ZeroCrossAttn + """ + def __init__(self, device=None, dtype=None, operations=None): + super().__init__() + + self.control_model = GLVControl(dtype=dtype, device=device, operations=operations) + + project_channel_scale = 2 + cond_output_channels = [320] * 4 + [640] * 3 + [1280] * 3 + project_channels = [int(c * project_channel_scale) for c in [160] * 4 + [320] * 3 + [640] * 3] + concat_channels = [320] * 2 + [640] * 3 + [1280] * 4 + [0] + cross_attn_insert_idx = [6, 3] + + self.project_modules = nn.ModuleList() + for i in range(len(cond_output_channels)): + self.project_modules.append(ZeroSFT( + project_channels[i], cond_output_channels[i], + concat_channels=concat_channels[i], + dtype=dtype, device=device, operations=operations, + )) + + for i in cross_attn_insert_idx: + self.project_modules.insert(i, ZeroCrossAttn( + cond_output_channels[i], concat_channels[i], + dtype=dtype, device=device, operations=operations, + )) diff --git a/comfy/ldm/supir/supir_patch.py b/comfy/ldm/supir/supir_patch.py new file mode 100644 index 000000000..b67ab4cd8 --- /dev/null +++ b/comfy/ldm/supir/supir_patch.py @@ -0,0 +1,103 @@ +import torch +from comfy.ldm.modules.diffusionmodules.openaimodel import Upsample + + +class SUPIRPatch: + """ + Holds GLVControl (control encoder) + project_modules (ZeroSFT/ZeroCrossAttn adapters). + Runs GLVControl lazily on first patch invocation per step, applies adapters through + middle_block_after_patch, output_block_merge_patch, and forward_timestep_embed_patch. + """ + SIGMA_MAX = 14.6146 + + def __init__(self, model_patch, project_modules, hint_latent, strength_start, strength_end): + self.model_patch = model_patch # CoreModelPatcher wrapping GLVControl + self.project_modules = project_modules # nn.ModuleList of ZeroSFT/ZeroCrossAttn + self.hint_latent = hint_latent # encoded LQ image latent + self.strength_start = strength_start + self.strength_end = strength_end + self.cached_features = None + self.adapter_idx = 0 + self.control_idx = 0 + self.current_control_idx = 0 + self.active = True + + def _ensure_features(self, kwargs): + """Run GLVControl on first call per step, cache results.""" + if self.cached_features is not None: + return + x = kwargs["x"] + b = x.shape[0] + hint = self.hint_latent.to(device=x.device, dtype=x.dtype) + if hint.shape[0] != b: + hint = hint.expand(b, -1, -1, -1) if hint.shape[0] == 1 else hint.repeat((b + hint.shape[0] - 1) // hint.shape[0], 1, 1, 1)[:b] + self.cached_features = self.model_patch.model.control_model( + hint, kwargs["timesteps"], x, + kwargs["context"], kwargs["y"] + ) + self.adapter_idx = len(self.project_modules) - 1 + self.control_idx = len(self.cached_features) - 1 + + def _get_control_scale(self, kwargs): + if self.strength_start == self.strength_end: + return self.strength_end + sigma = kwargs["transformer_options"].get("sigmas") + if sigma is None: + return self.strength_end + s = sigma[0].item() if sigma.dim() > 0 else sigma.item() + t = min(s / self.SIGMA_MAX, 1.0) + return t * (self.strength_start - self.strength_end) + self.strength_end + + def middle_after(self, kwargs): + """middle_block_after_patch: run GLVControl lazily, apply last adapter after middle block.""" + self.cached_features = None # reset from previous step + self.current_scale = self._get_control_scale(kwargs) + self.active = self.current_scale > 0 + if not self.active: + return {"h": kwargs["h"]} + self._ensure_features(kwargs) + h = kwargs["h"] + h = self.project_modules[self.adapter_idx]( + self.cached_features[self.control_idx], h, control_scale=self.current_scale + ) + self.adapter_idx -= 1 + self.control_idx -= 1 + return {"h": h} + + def output_block(self, h, hsp, transformer_options): + """output_block_patch: ZeroSFT adapter fusion replaces cat([h, hsp]). Returns (h, None) to skip cat.""" + if not self.active: + return h, hsp + self.current_control_idx = self.control_idx + h = self.project_modules[self.adapter_idx]( + self.cached_features[self.control_idx], hsp, h, control_scale=self.current_scale + ) + self.adapter_idx -= 1 + self.control_idx -= 1 + return h, None + + def pre_upsample(self, layer, x, emb, context, transformer_options, output_shape, *args, **kw): + """forward_timestep_embed_patch for Upsample: extra cross-attn adapter before upsample.""" + block_type, _ = transformer_options["block"] + if block_type == "output" and self.active and self.cached_features is not None: + x = self.project_modules[self.adapter_idx]( + self.cached_features[self.current_control_idx], x, control_scale=self.current_scale + ) + self.adapter_idx -= 1 + return layer(x, output_shape=output_shape) + + def to(self, device_or_dtype): + if isinstance(device_or_dtype, torch.device): + self.cached_features = None + if self.hint_latent is not None: + self.hint_latent = self.hint_latent.to(device_or_dtype) + return self + + def models(self): + return [self.model_patch] + + def register(self, model_patcher): + """Register all patches on a cloned model patcher.""" + model_patcher.set_model_patch(self.middle_after, "middle_block_after_patch") + model_patcher.set_model_output_block_patch(self.output_block) + model_patcher.set_model_patch((Upsample, self.pre_upsample), "forward_timestep_embed_patch") diff --git a/comfy/ldm/wan/ar_model.py b/comfy/ldm/wan/ar_model.py new file mode 100644 index 000000000..d72f53602 --- /dev/null +++ b/comfy/ldm/wan/ar_model.py @@ -0,0 +1,276 @@ +""" +CausalWanModel: Wan 2.1 backbone with KV-cached causal self-attention for +autoregressive (frame-by-frame) video generation via Causal Forcing. + +Weight-compatible with the standard WanModel -- same layer names, same shapes. +The difference is purely in the forward pass: this model processes one temporal +block at a time and maintains a KV cache across blocks. + +Reference: https://github.com/thu-ml/Causal-Forcing +""" + +import torch +import torch.nn as nn + +from comfy.ldm.modules.attention import optimized_attention +from comfy.ldm.flux.math import apply_rope1 +from comfy.ldm.wan.model import ( + sinusoidal_embedding_1d, + repeat_e, + WanModel, + WanAttentionBlock, +) +import comfy.ldm.common_dit +import comfy.model_management + + +class CausalWanSelfAttention(nn.Module): + """Self-attention with KV cache support for autoregressive inference.""" + + def __init__(self, dim, num_heads, window_size=(-1, -1), qk_norm=True, + eps=1e-6, operation_settings={}): + assert dim % num_heads == 0 + super().__init__() + self.dim = dim + self.num_heads = num_heads + self.head_dim = dim // num_heads + self.qk_norm = qk_norm + self.eps = eps + + ops = operation_settings.get("operations") + device = operation_settings.get("device") + dtype = operation_settings.get("dtype") + + self.q = ops.Linear(dim, dim, device=device, dtype=dtype) + self.k = ops.Linear(dim, dim, device=device, dtype=dtype) + self.v = ops.Linear(dim, dim, device=device, dtype=dtype) + self.o = ops.Linear(dim, dim, device=device, dtype=dtype) + self.norm_q = ops.RMSNorm(dim, eps=eps, elementwise_affine=True, device=device, dtype=dtype) if qk_norm else nn.Identity() + self.norm_k = ops.RMSNorm(dim, eps=eps, elementwise_affine=True, device=device, dtype=dtype) if qk_norm else nn.Identity() + + def forward(self, x, freqs, kv_cache=None, transformer_options={}): + b, s, n, d = *x.shape[:2], self.num_heads, self.head_dim + + q = apply_rope1(self.norm_q(self.q(x)).view(b, s, n, d), freqs) + k = apply_rope1(self.norm_k(self.k(x)).view(b, s, n, d), freqs) + v = self.v(x).view(b, s, n, d) + + if kv_cache is None: + x = optimized_attention( + q.view(b, s, n * d), + k.view(b, s, n * d), + v.view(b, s, n * d), + heads=self.num_heads, + transformer_options=transformer_options, + ) + else: + end = kv_cache["end"] + new_end = end + s + + # Roped K and plain V go into cache + kv_cache["k"][:, end:new_end] = k + kv_cache["v"][:, end:new_end] = v + kv_cache["end"] = new_end + + x = optimized_attention( + q.view(b, s, n * d), + kv_cache["k"][:, :new_end].view(b, new_end, n * d), + kv_cache["v"][:, :new_end].view(b, new_end, n * d), + heads=self.num_heads, + transformer_options=transformer_options, + ) + + x = self.o(x) + return x + + +class CausalWanAttentionBlock(WanAttentionBlock): + """Transformer block with KV-cached self-attention and cross-attention caching.""" + + def __init__(self, cross_attn_type, dim, ffn_dim, num_heads, + window_size=(-1, -1), qk_norm=True, cross_attn_norm=False, + eps=1e-6, operation_settings={}): + super().__init__(cross_attn_type, dim, ffn_dim, num_heads, + window_size, qk_norm, cross_attn_norm, eps, + operation_settings=operation_settings) + self.self_attn = CausalWanSelfAttention( + dim, num_heads, window_size, qk_norm, eps, + operation_settings=operation_settings) + + def forward(self, x, e, freqs, context, context_img_len=257, + kv_cache=None, crossattn_cache=None, transformer_options={}): + if e.ndim < 4: + e = (comfy.model_management.cast_to(self.modulation, dtype=x.dtype, device=x.device) + e).chunk(6, dim=1) + else: + e = (comfy.model_management.cast_to(self.modulation, dtype=x.dtype, device=x.device).unsqueeze(0) + e).unbind(2) + + # Self-attention with optional KV cache + x = x.contiguous() + y = self.self_attn( + torch.addcmul(repeat_e(e[0], x), self.norm1(x), 1 + repeat_e(e[1], x)), + freqs, kv_cache=kv_cache, transformer_options=transformer_options) + x = torch.addcmul(x, y, repeat_e(e[2], x)) + del y + + # Cross-attention with optional caching + if crossattn_cache is not None and crossattn_cache.get("is_init"): + q = self.cross_attn.norm_q(self.cross_attn.q(self.norm3(x))) + x_ca = optimized_attention( + q, crossattn_cache["k"], crossattn_cache["v"], + heads=self.num_heads, transformer_options=transformer_options) + x = x + self.cross_attn.o(x_ca) + else: + x = x + self.cross_attn(self.norm3(x), context, context_img_len=context_img_len, transformer_options=transformer_options) + if crossattn_cache is not None: + crossattn_cache["k"] = self.cross_attn.norm_k(self.cross_attn.k(context)) + crossattn_cache["v"] = self.cross_attn.v(context) + crossattn_cache["is_init"] = True + + # FFN + y = self.ffn(torch.addcmul(repeat_e(e[3], x), self.norm2(x), 1 + repeat_e(e[4], x))) + x = torch.addcmul(x, y, repeat_e(e[5], x)) + return x + + +class CausalWanModel(WanModel): + """ + Wan 2.1 diffusion backbone with causal KV-cache support. + + Same weight structure as WanModel -- loads identical state dicts. + Adds forward_block() for frame-by-frame autoregressive inference. + """ + + def __init__(self, + model_type='t2v', + patch_size=(1, 2, 2), + text_len=512, + in_dim=16, + dim=2048, + ffn_dim=8192, + freq_dim=256, + text_dim=4096, + out_dim=16, + num_heads=16, + num_layers=32, + window_size=(-1, -1), + qk_norm=True, + cross_attn_norm=True, + eps=1e-6, + image_model=None, + device=None, + dtype=None, + operations=None): + super().__init__( + model_type=model_type, patch_size=patch_size, text_len=text_len, + in_dim=in_dim, dim=dim, ffn_dim=ffn_dim, freq_dim=freq_dim, + text_dim=text_dim, out_dim=out_dim, num_heads=num_heads, + num_layers=num_layers, window_size=window_size, qk_norm=qk_norm, + cross_attn_norm=cross_attn_norm, eps=eps, image_model=image_model, + wan_attn_block_class=CausalWanAttentionBlock, + device=device, dtype=dtype, operations=operations) + + def forward_block(self, x, timestep, context, start_frame, + kv_caches, crossattn_caches, clip_fea=None): + """ + Forward one temporal block for autoregressive inference. + + Args: + x: [B, C, block_frames, H, W] input latent for the current block + timestep: [B, block_frames] per-frame timesteps + context: [B, L, text_dim] raw text embeddings (pre-text_embedding) + start_frame: temporal frame index for RoPE offset + kv_caches: list of per-layer KV cache dicts + crossattn_caches: list of per-layer cross-attention cache dicts + clip_fea: optional CLIP features for I2V + + Returns: + flow_pred: [B, C_out, block_frames, H, W] flow prediction + """ + x = comfy.ldm.common_dit.pad_to_patch_size(x, self.patch_size) + bs, c, t, h, w = x.shape + + x = self.patch_embedding(x.float()).to(x.dtype) + grid_sizes = x.shape[2:] + x = x.flatten(2).transpose(1, 2) + + # Per-frame time embedding + e = self.time_embedding( + sinusoidal_embedding_1d(self.freq_dim, timestep.flatten()).to(dtype=x.dtype)) + e = e.reshape(timestep.shape[0], -1, e.shape[-1]) + e0 = self.time_projection(e).unflatten(2, (6, self.dim)) + + # Text embedding (reuses crossattn_cache after first block) + context = self.text_embedding(context) + + context_img_len = None + if clip_fea is not None and self.img_emb is not None: + context_clip = self.img_emb(clip_fea) + context = torch.concat([context_clip, context], dim=1) + context_img_len = clip_fea.shape[-2] + + # RoPE for current block's temporal position + freqs = self.rope_encode(t, h, w, t_start=start_frame, device=x.device, dtype=x.dtype) + + # Transformer blocks + for i, block in enumerate(self.blocks): + x = block(x, e=e0, freqs=freqs, context=context, + context_img_len=context_img_len, + kv_cache=kv_caches[i], + crossattn_cache=crossattn_caches[i]) + + # Head + x = self.head(x, e) + + # Unpatchify + x = self.unpatchify(x, grid_sizes) + return x[:, :, :t, :h, :w] + + def init_kv_caches(self, batch_size, max_seq_len, device, dtype): + """Create fresh KV caches for all layers.""" + caches = [] + for _ in range(self.num_layers): + caches.append({ + "k": torch.zeros(batch_size, max_seq_len, self.num_heads, self.head_dim, device=device, dtype=dtype), + "v": torch.zeros(batch_size, max_seq_len, self.num_heads, self.head_dim, device=device, dtype=dtype), + "end": 0, + }) + return caches + + def init_crossattn_caches(self, batch_size, device, dtype): + """Create fresh cross-attention caches for all layers.""" + caches = [] + for _ in range(self.num_layers): + caches.append({"is_init": False}) + return caches + + def reset_kv_caches(self, kv_caches): + """Reset KV caches to empty (reuse allocated memory).""" + for cache in kv_caches: + cache["end"] = 0 + + def reset_crossattn_caches(self, crossattn_caches): + """Reset cross-attention caches.""" + for cache in crossattn_caches: + cache["is_init"] = False + + @property + def head_dim(self): + return self.dim // self.num_heads + + def forward(self, x, timestep, context, clip_fea=None, time_dim_concat=None, transformer_options={}, **kwargs): + ar_state = transformer_options.get("ar_state") + if ar_state is not None: + bs = x.shape[0] + block_frames = x.shape[2] + t_per_frame = timestep.unsqueeze(1).expand(bs, block_frames) + return self.forward_block( + x=x, timestep=t_per_frame, context=context, + start_frame=ar_state["start_frame"], + kv_caches=ar_state["kv_caches"], + crossattn_caches=ar_state["crossattn_caches"], + clip_fea=clip_fea, + ) + + return super().forward(x, timestep, context, clip_fea=clip_fea, + time_dim_concat=time_dim_concat, + transformer_options=transformer_options, **kwargs) diff --git a/comfy/ldm/wan/model.py b/comfy/ldm/wan/model.py index b2287dba9..70dfe7b16 100644 --- a/comfy/ldm/wan/model.py +++ b/comfy/ldm/wan/model.py @@ -1135,7 +1135,7 @@ class AudioInjector_WAN(nn.Module): self.injector_adain_output_layers = nn.ModuleList( [operations.Linear(dim, dim, dtype=dtype, device=device) for _ in range(audio_injector_id)]) - def forward(self, x, block_id, audio_emb, audio_emb_global, seq_len): + def forward(self, x, block_id, audio_emb, audio_emb_global, seq_len, scale=1.0): audio_attn_id = self.injected_block_id.get(block_id, None) if audio_attn_id is None: return x @@ -1148,12 +1148,15 @@ class AudioInjector_WAN(nn.Module): attn_hidden_states = adain_hidden_states else: attn_hidden_states = self.injector_pre_norm_feat[audio_attn_id](input_hidden_states) - audio_emb = rearrange(audio_emb, "b t n c -> (b t) n c", t=num_frames) - attn_audio_emb = audio_emb + + if audio_emb.dim() == 3: # WanDancer case + attn_audio_emb = rearrange(audio_emb, "b t c -> (b t) 1 c", t=num_frames) + else: # S2V case + attn_audio_emb = rearrange(audio_emb, "b t n c -> (b t) n c", t=num_frames) + residual_out = self.injector[audio_attn_id](x=attn_hidden_states, context=attn_audio_emb) - residual_out = rearrange( - residual_out, "(b t) n c -> b (t n) c", t=num_frames) - x[:, :seq_len] = x[:, :seq_len] + residual_out + residual_out = rearrange(residual_out, "(b t) n c -> b (t n) c", t=num_frames) + x[:, :seq_len] = x[:, :seq_len] + residual_out * scale return x diff --git a/comfy/ldm/wan/model_wandancer.py b/comfy/ldm/wan/model_wandancer.py new file mode 100644 index 000000000..3caef6dc5 --- /dev/null +++ b/comfy/ldm/wan/model_wandancer.py @@ -0,0 +1,251 @@ +import torch +import torch.nn as nn +import comfy +from comfy.ldm.modules.attention import optimized_attention +from comfy.ldm.flux.math import apply_rope1 +from comfy.ldm.flux.layers import EmbedND + +from .model import AudioInjector_WAN, WanModel, MLPProj, Head, sinusoidal_embedding_1d + + +class MusicSelfAttention(nn.Module): + def __init__(self, dim, num_heads, device=None, dtype=None, operations=None): + assert dim % num_heads == 0 + super().__init__() + self.embed_dim = dim + self.num_heads = num_heads + self.head_dim = dim // num_heads + + self.q_proj = operations.Linear(dim, dim, device=device, dtype=dtype) + self.k_proj = operations.Linear(dim, dim, device=device, dtype=dtype) + self.v_proj = operations.Linear(dim, dim, device=device, dtype=dtype) + self.out_proj = operations.Linear(dim, dim, device=device, dtype=dtype) + + def forward(self, x, freqs): + b, s, n, d = *x.shape[:2], self.num_heads, self.head_dim + + q = self.q_proj(x).view(b, s, n, d) + q = apply_rope1(q, freqs) + + k = self.k_proj(x).view(b, s, n, d) + k = apply_rope1(k, freqs) + + x = optimized_attention( + q.view(b, s, n * d), + k.view(b, s, n * d), + self.v_proj(x).view(b, s, n * d), + heads=self.num_heads, + ) + + return self.out_proj(x) + + +class MusicEncoderLayer(nn.Module): + def __init__(self, dim: int, num_heads: int, ffn_dim: int, device=None, dtype=None, operations=None): + super().__init__() + self.self_attn = MusicSelfAttention(dim, num_heads, device=device, dtype=dtype, operations=operations) + + self.linear1 = operations.Linear(dim, ffn_dim, device=device, dtype=dtype) + self.linear2 = operations.Linear(ffn_dim, dim, device=device, dtype=dtype) + + self.norm1 = operations.LayerNorm(dim, device=device, dtype=dtype) + self.norm2 = operations.LayerNorm(dim, device=device, dtype=dtype) + + def forward(self, x: torch.Tensor, freqs: torch.Tensor) -> torch.Tensor: + x = x + self.self_attn(self.norm1(x), freqs=freqs) + x = x + self.linear2(torch.nn.functional.gelu(self.linear1(self.norm2(x)))) # ffn + return x + + +class WanDancerModel(WanModel): + def __init__(self, + model_type='wandancer', + patch_size=(1, 2, 2), + text_len=512, + in_dim=16, + dim=5120, + ffn_dim=8192, + freq_dim=256, + text_dim=4096, + out_dim=16, + num_heads=16, + num_layers=40, + window_size=(-1, -1), + qk_norm=True, + cross_attn_norm=True, + eps=1e-6, + in_dim_ref_conv=None, + image_model=None, + device=None, dtype=None, operations=None, + audio_inject_layers=[0, 4, 8, 12, 16, 20, 24, 27], + music_dim = 256, + music_heads = 4, + music_feature_dim = 35, + music_latent_dim = 256 + ): + + super().__init__(model_type='i2v', patch_size=patch_size, text_len=text_len, in_dim=in_dim, dim=dim, ffn_dim=ffn_dim, freq_dim=freq_dim, text_dim=text_dim, out_dim=out_dim, + num_heads=num_heads, num_layers=num_layers, window_size=window_size, qk_norm=qk_norm, cross_attn_norm=cross_attn_norm, eps=eps, image_model=image_model, in_dim_ref_conv=in_dim_ref_conv, + device=device, dtype=dtype, operations=operations) + + self.dtype = dtype + operation_settings = {"operations": operations, "device": device, "dtype": dtype} + + self.patch_embedding_global = operations.Conv3d(in_dim, dim, kernel_size=patch_size, stride=patch_size, device=operation_settings.get("device"), dtype=torch.float32) + self.img_emb_refimage = MLPProj(1280, dim, operation_settings=operation_settings) + self.head_global = Head(dim, out_dim, patch_size, eps, operation_settings=operation_settings) + + self.music_injector = AudioInjector_WAN( + dim=self.dim, + num_heads=self.num_heads, + inject_layer=audio_inject_layers, + root_net=self, + enable_adain=False, + dtype=dtype, device=device, operations=operations + ) + + self.music_projection = operations.Linear(music_feature_dim, music_latent_dim, device=device, dtype=dtype) + self.music_encoder = nn.ModuleList([MusicEncoderLayer(dim=music_dim, num_heads=music_heads, ffn_dim=1024, device=device, dtype=dtype, operations=operations) for _ in range(2)]) + music_head_dim = music_dim // music_heads + self.music_rope_embedder = EmbedND(dim=music_head_dim, theta=10000.0, axes_dim=[music_head_dim]) + + def forward_orig(self, x, t, context, clip_fea=None, clip_fea_ref=None, freqs=None, audio_embed=None, fps=30, audio_inject_scale=1.0, transformer_options={}, **kwargs): + # embeddings + if int(fps + 0.5) != 30: + x = self.patch_embedding_global(x.float()).to(x.dtype) + else: + x = self.patch_embedding(x.float()).to(x.dtype) + + grid_sizes = x.shape[2:] + latent_frames = grid_sizes[0] + transformer_options["grid_sizes"] = grid_sizes + x = x.flatten(2).transpose(1, 2) + seq_len = x.size(1) + + # time embeddings + e = self.time_embedding(sinusoidal_embedding_1d(self.freq_dim, t.flatten()).to(dtype=x[0].dtype)) + e = e.reshape(t.shape[0], -1, e.shape[-1]) + e0 = self.time_projection(e).unflatten(2, (6, self.dim)) + + full_ref = None + if self.ref_conv is not None: # model has the weight, but this wasn't used in the original pipeline + full_ref = kwargs.get("reference_latent", None) + if full_ref is not None: + full_ref = self.ref_conv(full_ref).flatten(2).transpose(1, 2) + x = torch.concat((full_ref, x), dim=1) + + # context + context = self.text_embedding(context) + + audio_emb = None + if audio_embed is not None: # encode music feature，[1, frame_num, 35] -> [1, F*8, dim] + music_feature = self.music_projection(audio_embed) + + music_seq_len = music_feature.shape[1] + music_ids = torch.arange(music_seq_len, device=music_feature.device, dtype=music_feature.dtype).reshape(1, -1, 1) # create 1D position IDs + music_freqs = self.music_rope_embedder(music_ids).movedim(1, 2) + + # apply encoder layers + for layer in self.music_encoder: + music_feature = layer(music_feature, music_freqs) + + # interpolate + audio_emb = torch.nn.functional.interpolate(music_feature.unsqueeze(1), size=(latent_frames * 8, self.dim), mode='bilinear').squeeze(1) + + context_img_len = 0 + if self.img_emb is not None and clip_fea is not None: + context_clip = self.img_emb(clip_fea) # bs x 257 x dim + context = torch.cat([context_clip, context], dim=1) + context_img_len += clip_fea.shape[-2] + if self.img_emb_refimage is not None and clip_fea_ref is not None: + context_clip_ref = self.img_emb_refimage(clip_fea_ref) + context = torch.cat([context_clip_ref, context], dim=1) + context_img_len += clip_fea_ref.shape[-2] + + patches_replace = transformer_options.get("patches_replace", {}) + blocks_replace = patches_replace.get("dit", {}) + transformer_options["total_blocks"] = len(self.blocks) + transformer_options["block_type"] = "double" + for i, block in enumerate(self.blocks): + transformer_options["block_index"] = i + if ("double_block", i) in blocks_replace: + def block_wrap(args): + out = {} + out["img"] = block(args["img"], context=args["txt"], e=args["vec"], freqs=args["pe"], context_img_len=context_img_len, transformer_options=args["transformer_options"]) + return out + out = blocks_replace[("double_block", i)]({"img": x, "txt": context, "vec": e0, "pe": freqs, "transformer_options": transformer_options}, {"original_block": block_wrap}) + x = out["img"] + else: + x = block(x, e=e0, freqs=freqs, context=context, context_img_len=context_img_len, transformer_options=transformer_options) + if audio_emb is not None: + x = self.music_injector(x, i, audio_emb, audio_emb_global=None, seq_len=seq_len, scale=audio_inject_scale) + + # head + if int(fps + 0.5) != 30: + x = self.head_global(x, e) + else: + x = self.head(x, e) + + if full_ref is not None: + x = x[:, full_ref.shape[1]:] + + # unpatchify + x = self.unpatchify(x, grid_sizes) + return x + + def _forward(self, x, timestep, context, clip_fea=None, time_dim_concat=None, transformer_options={}, clip_fea_ref=None, fps=30, audio_inject_scale=1.0, **kwargs): + bs, c, t, h, w = x.shape + x = comfy.ldm.common_dit.pad_to_patch_size(x, self.patch_size) + + t_len = t + if time_dim_concat is not None: + time_dim_concat = comfy.ldm.common_dit.pad_to_patch_size(time_dim_concat, self.patch_size) + x = torch.cat([x, time_dim_concat], dim=2) + t_len = x.shape[2] + + freqs = self.rope_encode(t_len, h, w, device=x.device, dtype=x.dtype, fps=fps, transformer_options=transformer_options) + return self.forward_orig(x, timestep, context, clip_fea=clip_fea, clip_fea_ref=clip_fea_ref, freqs=freqs, fps=fps, audio_inject_scale=audio_inject_scale, transformer_options=transformer_options, **kwargs)[:, :, :t, :h, :w] + + def rope_encode(self, t, h, w, t_start=0, steps_t=None, steps_h=None, steps_w=None, fps=30, device=None, dtype=None, transformer_options={}): + patch_size = self.patch_size + t_len = ((t + (patch_size[0] // 2)) // patch_size[0]) + h_len = ((h + (patch_size[1] // 2)) // patch_size[1]) + w_len = ((w + (patch_size[2] // 2)) // patch_size[2]) + + if steps_t is None: + steps_t = t_len + if steps_h is None: + steps_h = h_len + if steps_w is None: + steps_w = w_len + + h_start = 0 + w_start = 0 + rope_options = transformer_options.get("rope_options", None) + if rope_options is not None: + t_len = (t_len - 1.0) * rope_options.get("scale_t", 1.0) + 1.0 + h_len = (h_len - 1.0) * rope_options.get("scale_y", 1.0) + 1.0 + w_len = (w_len - 1.0) * rope_options.get("scale_x", 1.0) + 1.0 + + t_start += rope_options.get("shift_t", 0.0) + h_start += rope_options.get("shift_y", 0.0) + w_start += rope_options.get("shift_x", 0.0) + + img_ids = torch.zeros((steps_t, steps_h, steps_w, 3), device=device, dtype=dtype) + + if int(fps + 0.5) != 30: + time_scale = 30.0 / fps # how many time units each frame represents relative to 30fps + positions_new = torch.arange(steps_t, device=device, dtype=dtype) * time_scale + t_start + total_frames_at_30fps = int(time_scale * steps_t + 0.5) + positions_new[-1] = t_start + (total_frames_at_30fps - 1) + + img_ids[:, :, :, 0] = img_ids[:, :, :, 0] + positions_new.reshape(-1, 1, 1) + else: + img_ids[:, :, :, 0] = img_ids[:, :, :, 0] + torch.linspace(t_start, t_start + (t_len - 1), steps=steps_t, device=device, dtype=dtype).reshape(-1, 1, 1) + + img_ids[:, :, :, 1] = img_ids[:, :, :, 1] + torch.linspace(h_start, h_start + (h_len - 1), steps=steps_h, device=device, dtype=dtype).reshape(1, -1, 1) + img_ids[:, :, :, 2] = img_ids[:, :, :, 2] + torch.linspace(w_start, w_start + (w_len - 1), steps=steps_w, device=device, dtype=dtype).reshape(1, 1, -1) + img_ids = img_ids.reshape(1, -1, img_ids.shape[-1]) + + freqs = self.rope_embedder(img_ids).movedim(1, 2) + return freqs diff --git a/comfy/lora.py b/comfy/lora.py index 63ee85323..f11e26ec9 100644 --- a/comfy/lora.py +++ b/comfy/lora.py @@ -17,6 +17,7 @@ """ from __future__ import annotations +import comfy.memory_management import comfy.utils import comfy.model_management import comfy.model_base @@ -96,12 +97,14 @@ def load_lora(lora, to_load, log_missing=True): def model_lora_keys_clip(model, key_map={}): sdk = model.state_dict().keys() + prefix_set = set() for k in sdk: if k.endswith(".weight"): key_map["text_encoders.{}".format(k[:-len(".weight")])] = k #generic lora format without any weird key names tp = k.find(".transformer.") #also map without wrapper prefix for composite text encoder models if tp > 0 and not k.startswith("clip_"): key_map["text_encoders.{}".format(k[tp + 1:-len(".weight")])] = k + prefix_set.add(k.split('.')[0]) text_model_lora_key = "lora_te_text_model_encoder_layers_{}_{}" clip_l_present = False @@ -162,6 +165,13 @@ def model_lora_keys_clip(model, key_map={}): lora_key = "lora_te1_{}".format(l_key.replace(".", "_")) key_map[lora_key] = k + if len(prefix_set) == 1: + full_prefix = "{}.transformer.model.".format(next(iter(prefix_set))) # kohya anima and maybe other single TE models that use a single llama arch based te + for k in sdk: + if k.endswith(".weight"): + if k.startswith(full_prefix): + l_key = k[len(full_prefix):-len(".weight")] + key_map["lora_te_{}".format(l_key.replace(".", "_"))] = k k = "clip_g.transformer.text_projection.weight" if k in sdk: @@ -342,6 +352,12 @@ def model_lora_keys_unet(model, key_map={}): key_map["base_model.model.{}".format(key_lora)] = k # Official base model loras key_map["lycoris_{}".format(key_lora.replace(".", "_"))] = k # LyCORIS/LoKR format + if isinstance(model, comfy.model_base.ErnieImage): + for k in sdk: + if k.startswith("diffusion_model.") and k.endswith(".weight"): + key_lora = k[len("diffusion_model."):-len(".weight")] + key_map["transformer.{}".format(key_lora)] = k + return key_map @@ -467,3 +483,17 @@ def calculate_weight(patches, weight, key, intermediate_dtype=torch.float32, ori weight = old_weight return weight + +def prefetch_prepared_value(value, allocate_buffer, stream): + if isinstance(value, torch.Tensor): + dest = allocate_buffer(comfy.memory_management.vram_aligned_size(value)) + comfy.model_management.cast_to_gathered([value], dest, non_blocking=True, stream=stream) + return comfy.memory_management.interpret_gathered_like([value], dest)[0] + elif isinstance(value, weight_adapter.WeightAdapterBase): + return type(value)(value.loaded_keys, prefetch_prepared_value(value.weights, allocate_buffer, stream)) + elif isinstance(value, tuple): + return tuple(prefetch_prepared_value(item, allocate_buffer, stream) for item in value) + elif isinstance(value, list): + return [prefetch_prepared_value(item, allocate_buffer, stream) for item in value] + + return value diff --git a/comfy/model_base.py b/comfy/model_base.py index c2ae646aa..0736321b3 100644 --- a/comfy/model_base.py +++ b/comfy/model_base.py @@ -42,6 +42,8 @@ import comfy.ldm.cosmos.predict2 import comfy.ldm.lumina.model import comfy.ldm.wan.model import comfy.ldm.wan.model_animate +import comfy.ldm.wan.ar_model +import comfy.ldm.wan.model_wandancer import comfy.ldm.hunyuan3d.model import comfy.ldm.hidream.model import comfy.ldm.chroma.model @@ -52,7 +54,12 @@ import comfy.ldm.qwen_image.model import comfy.ldm.kandinsky5.model import comfy.ldm.anima.model import comfy.ldm.ace.ace_step15 +import comfy.ldm.cogvideo.model import comfy.ldm.rt_detr.rtdetr_v4 +import comfy.ldm.ernie.model +import comfy.ldm.sam3.detector +import comfy.ldm.hidream_o1.model +from comfy.ldm.hidream_o1.conditioning import build_extra_conds import comfy.model_management import comfy.patcher_extension @@ -79,6 +86,7 @@ class ModelType(Enum): IMG_TO_IMG = 9 FLOW_COSMOS = 10 IMG_TO_IMG_FLOW = 11 + V_PREDICTION_DDPM = 12 def model_sampling(model_config, model_type): @@ -113,6 +121,8 @@ def model_sampling(model_config, model_type): s = comfy.model_sampling.ModelSamplingCosmosRFlow elif model_type == ModelType.IMG_TO_IMG_FLOW: c = comfy.model_sampling.IMG_TO_IMG_FLOW + elif model_type == ModelType.V_PREDICTION_DDPM: + c = comfy.model_sampling.V_PREDICTION_DDPM class ModelSampling(s, c): pass @@ -208,6 +218,11 @@ class BaseModel(torch.nn.Module): if "latent_shapes" in extra_conds: xc = utils.unpack_latents(xc, extra_conds.pop("latent_shapes")) + transformer_options = transformer_options.copy() + transformer_options["prefetch_dynamic_vbars"] = ( + self.current_patcher is not None and self.current_patcher.is_dynamic() + ) + model_output = self.diffusion_model(xc, t, context=context, control=control, transformer_options=transformer_options, **extra_conds) if len(model_output) > 1 and not torch.is_tensor(model_output): model_output, _ = utils.pack_latents(model_output) @@ -577,8 +592,8 @@ class Stable_Zero123(BaseModel): def __init__(self, model_config, model_type=ModelType.EPS, device=None, cc_projection_weight=None, cc_projection_bias=None): super().__init__(model_config, model_type, device=device) self.cc_projection = comfy.ops.manual_cast.Linear(cc_projection_weight.shape[1], cc_projection_weight.shape[0], dtype=self.get_dtype(), device=device) - self.cc_projection.weight.copy_(cc_projection_weight) - self.cc_projection.bias.copy_(cc_projection_bias) + self.cc_projection.weight = torch.nn.Parameter(cc_projection_weight.clone()) + self.cc_projection.bias = torch.nn.Parameter(cc_projection_bias.clone()) def extra_conds(self, **kwargs): out = {} @@ -1354,6 +1369,13 @@ class WAN21(BaseModel): return out +class WAN21_CausalAR(WAN21): + def __init__(self, model_config, model_type=ModelType.FLOW, device=None): + super(WAN21, self).__init__(model_config, model_type, device=device, + unet_model=comfy.ldm.wan.ar_model.CausalWanModel) + self.image_to_video = False + + class WAN21_Vace(WAN21): def __init__(self, model_config, model_type=ModelType.FLOW, image_to_video=False, device=None): super(WAN21, self).__init__(model_config, model_type, device=device, unet_model=comfy.ldm.wan.model.VaceWanModel) @@ -1580,6 +1602,30 @@ class WAN21_SCAIL(WAN21): return out +class WAN22_WanDancer(WAN21): + def __init__(self, model_config, model_type=ModelType.FLOW, image_to_video=True, device=None): + super(WAN21, self).__init__(model_config, model_type, device=device, unet_model=comfy.ldm.wan.model_wandancer.WanDancerModel) + self.image_to_video = image_to_video + + def extra_conds(self, **kwargs): + out = super().extra_conds(**kwargs) + audio_embed = kwargs.get("audio_embed", None) + if audio_embed is not None: + out['audio_embed'] = comfy.conds.CONDRegular(audio_embed) + + clip_vision_output_ref = kwargs.get("clip_vision_output_ref", None) + if clip_vision_output_ref is not None: + out['clip_fea_ref'] = comfy.conds.CONDRegular(clip_vision_output_ref.penultimate_hidden_states) + + fps = kwargs.get("fps", None) + if fps is not None: + out['fps'] = comfy.conds.CONDRegular(torch.FloatTensor([fps])) + + audio_inject_scale = kwargs.get("audio_inject_scale", None) + if audio_inject_scale is not None: + out['audio_inject_scale'] = comfy.conds.CONDRegular(torch.FloatTensor([audio_inject_scale])) + return out + class Hunyuan3Dv2(BaseModel): def __init__(self, model_config, model_type=ModelType.FLOW, device=None): super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.hunyuan3d.model.Hunyuan3Dv2) @@ -1630,6 +1676,32 @@ class HiDream(BaseModel): out['image_cond'] = comfy.conds.CONDNoiseShape(self.process_latent_in(image_cond)) return out +class HiDreamO1(BaseModel): + """HiDream-O1-Image: pixel-space DiT (no VAE). Refs from HiDreamO1ReferenceImages and tokens from the stub TE flow through + extra_conds; the heavy preprocessing lives in comfy.ldm.hidream_o1.conditioning.""" + PATCH_SIZE = 32 + + def __init__(self, model_config, model_type=ModelType.FLOW, device=None): + super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.hidream_o1.model.HiDreamO1Transformer) + + def extra_conds(self, **kwargs): + out = super().extra_conds(**kwargs) + text_input_ids = kwargs.get("text_input_ids", None) + noise = kwargs.get("noise", None) + if text_input_ids is None or noise is None: + return out + + conds = build_extra_conds( + text_input_ids, noise, + ref_images=kwargs.get("reference_latents", None), + target_patch_size=self.PATCH_SIZE, + ) + for k, v in conds.items(): + # ar_len is a Python int (precomputed to avoid a GPU sync in forward). + cls = comfy.conds.CONDConstant if k == "ar_len" else comfy.conds.CONDRegular + out[k] = cls(v) + return out + class Chroma(Flux): def __init__(self, model_config, model_type=ModelType.FLUX, device=None, unet_model=comfy.ldm.chroma.model.Chroma): super().__init__(model_config, model_type, device=device, unet_model=unet_model) @@ -1962,3 +2034,74 @@ class Kandinsky5Image(Kandinsky5): class RT_DETR_v4(BaseModel): def __init__(self, model_config, model_type=ModelType.FLOW, device=None): super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.rt_detr.rtdetr_v4.RTv4) + +class ErnieImage(BaseModel): + def __init__(self, model_config, model_type=ModelType.FLOW, device=None): + super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.ernie.model.ErnieImageModel) + + def extra_conds(self, **kwargs): + out = super().extra_conds(**kwargs) + cross_attn = kwargs.get("cross_attn", None) + if cross_attn is not None: + out['c_crossattn'] = comfy.conds.CONDRegular(cross_attn) + return out + +class SAM3(BaseModel): + def __init__(self, model_config, model_type=ModelType.FLOW, device=None): + super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.sam3.detector.SAM3Model) + +class CogVideoX(BaseModel): + def __init__(self, model_config, model_type=ModelType.V_PREDICTION_DDPM, image_to_video=False, device=None): + super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.cogvideo.model.CogVideoXTransformer3DModel) + self.image_to_video = image_to_video + + def concat_cond(self, **kwargs): + noise = kwargs.get("noise", None) + # Detect extra channels needed (e.g. 32 - 16 = 16 for ref latent) + extra_channels = self.diffusion_model.in_channels - noise.shape[1] + if extra_channels == 0: + return None + + image = kwargs.get("concat_latent_image", None) + device = kwargs["device"] + + if image is None: + shape = list(noise.shape) + shape[1] = extra_channels + return torch.zeros(shape, dtype=noise.dtype, layout=noise.layout, device=noise.device) + + latent_dim = self.latent_format.latent_channels + image = utils.common_upscale(image.to(device), noise.shape[-1], noise.shape[-2], "bilinear", "center") + + if noise.ndim == 5 and image.ndim == 5: + if image.shape[-3] < noise.shape[-3]: + image = torch.nn.functional.pad(image, (0, 0, 0, 0, 0, noise.shape[-3] - image.shape[-3]), "constant", 0) + elif image.shape[-3] > noise.shape[-3]: + image = image[:, :, :noise.shape[-3]] + + for i in range(0, image.shape[1], latent_dim): + image[:, i:i + latent_dim] = self.process_latent_in(image[:, i:i + latent_dim]) + image = utils.resize_to_batch_size(image, noise.shape[0]) + + if image.shape[1] > extra_channels: + image = image[:, :extra_channels] + elif image.shape[1] < extra_channels: + repeats = extra_channels // image.shape[1] + remainder = extra_channels % image.shape[1] + parts = [image] * repeats + if remainder > 0: + parts.append(image[:, :remainder]) + image = torch.cat(parts, dim=1) + + return image + + def extra_conds(self, **kwargs): + out = super().extra_conds(**kwargs) + # OFS embedding (CogVideoX 1.5 I2V), default 2.0 as used by SparkVSR + if self.diffusion_model.ofs_proj_dim is not None: + ofs = kwargs.get("ofs", None) + if ofs is None: + noise = kwargs.get("noise", None) + ofs = torch.full((noise.shape[0],), 2.0, device=noise.device, dtype=noise.dtype) + out['ofs'] = comfy.conds.CONDRegular(ofs) + return out diff --git a/comfy/model_detection.py b/comfy/model_detection.py index 8bed6828d..bc0b933bc 100644 --- a/comfy/model_detection.py +++ b/comfy/model_detection.py @@ -490,6 +490,54 @@ def detect_unet_config(state_dict, key_prefix, metadata=None): return dit_config + if '{}blocks.0.norm1.linear.weight'.format(key_prefix) in state_dict_keys: # CogVideoX + dit_config = {} + dit_config["image_model"] = "cogvideox" + + # Extract config from weight shapes + norm1_weight = state_dict['{}blocks.0.norm1.linear.weight'.format(key_prefix)] + time_embed_dim = norm1_weight.shape[1] + dim = norm1_weight.shape[0] // 6 + + dit_config["num_attention_heads"] = dim // 64 + dit_config["attention_head_dim"] = 64 + dit_config["time_embed_dim"] = time_embed_dim + dit_config["num_layers"] = count_blocks(state_dict_keys, '{}blocks.'.format(key_prefix) + '{}.') + + # Detect in_channels from patch_embed + patch_proj_key = '{}patch_embed.proj.weight'.format(key_prefix) + if patch_proj_key in state_dict_keys: + w = state_dict[patch_proj_key] + if w.ndim == 4: + # Conv2d: [out, in, kh, kw] — CogVideoX 1.0 + dit_config["in_channels"] = w.shape[1] + dit_config["patch_size"] = w.shape[2] + elif w.ndim == 2: + # Linear: [out, in_channels * patch_size * patch_size * patch_size_t] — CogVideoX 1.5 + dit_config["patch_size"] = 2 + dit_config["patch_size_t"] = 2 + dit_config["in_channels"] = w.shape[1] // (2 * 2 * 2) # 256 // 8 = 32 + + text_proj_key = '{}patch_embed.text_proj.weight'.format(key_prefix) + if text_proj_key in state_dict_keys: + dit_config["text_embed_dim"] = state_dict[text_proj_key].shape[1] + + # Detect OFS embedding + ofs_key = '{}ofs_embedding_linear_1.weight'.format(key_prefix) + if ofs_key in state_dict_keys: + dit_config["ofs_embed_dim"] = state_dict[ofs_key].shape[1] + + # Detect positional embedding type + pos_key = '{}patch_embed.pos_embedding'.format(key_prefix) + if pos_key in state_dict_keys: + dit_config["use_learned_positional_embeddings"] = True + dit_config["use_rotary_positional_embeddings"] = False + else: + dit_config["use_learned_positional_embeddings"] = False + dit_config["use_rotary_positional_embeddings"] = True + + return dit_config + if '{}head.modulation'.format(key_prefix) in state_dict_keys: # Wan 2.1 dit_config = {} dit_config["image_model"] = "wan2.1" @@ -524,6 +572,8 @@ def detect_unet_config(state_dict, key_prefix, metadata=None): dit_config["model_type"] = "animate" elif '{}patch_embedding_pose.weight'.format(key_prefix) in state_dict_keys: dit_config["model_type"] = "scail" + elif '{}patch_embedding_global.weight'.format(key_prefix) in state_dict_keys: + dit_config["model_type"] = "wandancer" else: if '{}img_emb.proj.0.bias'.format(key_prefix) in state_dict_keys: dit_config["model_type"] = "i2v" @@ -570,6 +620,9 @@ def detect_unet_config(state_dict, key_prefix, metadata=None): dit_config["guidance_cond_proj_dim"] = None#f"{key_prefix}t_embedder.cond_proj.weight" in state_dict_keys return dit_config + if '{}t_embedder1.mlp.0.weight'.format(key_prefix) in state_dict_keys and '{}x_embedder.proj1.weight'.format(key_prefix) in state_dict_keys: # HiDream-O1 + return {"image_model": "hidream_o1"} + if '{}caption_projection.0.linear.weight'.format(key_prefix) in state_dict_keys: # HiDream dit_config = {} dit_config["image_model"] = "hidream" @@ -713,6 +766,19 @@ def detect_unet_config(state_dict, key_prefix, metadata=None): dit_config["enc_h"] = state_dict['{}encoder.pan_blocks.1.cv4.conv.weight'.format(key_prefix)].shape[0] return dit_config + if '{}layers.0.mlp.linear_fc2.weight'.format(key_prefix) in state_dict_keys: # Ernie Image + dit_config = {} + dit_config["image_model"] = "ernie" + return dit_config + + if 'detector.backbone.vision_backbone.trunk.blocks.0.attn.qkv.weight' in state_dict_keys: # SAM3 / SAM3.1 + if 'detector.transformer.decoder.query_embed.weight' in state_dict_keys: + dit_config = {} + dit_config["image_model"] = "SAM3" + if 'detector.backbone.vision_backbone.propagation_convs.0.conv_1x1.weight' in state_dict_keys: + dit_config["image_model"] = "SAM31" + return dit_config + if '{}input_blocks.0.0.weight'.format(key_prefix) not in state_dict_keys: return None @@ -868,6 +934,10 @@ def model_config_from_unet(state_dict, unet_key_prefix, use_base_if_no_match=Fal return model_config def unet_prefix_from_state_dict(state_dict): + # SAM3: detector.* and tracker.* at top level, no common prefix + if any(k.startswith("detector.") for k in state_dict) and any(k.startswith("tracker.") for k in state_dict): + return "" + candidates = ["model.diffusion_model.", #ldm/sgm models "model.model.", #audio models "net.", #cosmos diff --git a/comfy/model_management.py b/comfy/model_management.py index 0eebf1ded..21738a4c7 100644 --- a/comfy/model_management.py +++ b/comfy/model_management.py @@ -31,6 +31,7 @@ from contextlib import nullcontext import comfy.memory_management import comfy.utils import comfy.quant_ops +import comfy_aimdo.vram_buffer class VRAMState(Enum): DISABLED = 0 #No vram present: no need to move models to vram @@ -112,10 +113,6 @@ if args.directml is not None: # torch_directml.disable_tiled_resources(True) lowvram_available = False #TODO: need to find a way to get free memory in directml before this can be enabled by default. -try: - import intel_extension_for_pytorch as ipex # noqa: F401 -except: - pass try: _ = torch.xpu.device_count() @@ -583,9 +580,6 @@ class LoadedModel: real_model = self.model.model - if is_intel_xpu() and not args.disable_ipex_optimize and 'ipex' in globals() and real_model is not None: - with torch.no_grad(): - real_model = ipex.optimize(real_model.eval(), inplace=True, graph_mode=True, concat_linear=True) self.real_model = weakref.ref(real_model) self.model_finalizer = weakref.finalize(real_model, cleanup_models) @@ -663,6 +657,7 @@ def minimum_inference_memory(): def free_memory(memory_required, device, keep_loaded=[], for_dynamic=False, pins_required=0, ram_required=0): cleanup_models_gc() + comfy.memory_management.extra_ram_release(max(pins_required, ram_required)) unloaded_model = [] can_unload = [] unloaded_models = [] @@ -726,13 +721,15 @@ def load_models_gpu(models, memory_required=0, force_patch_weights=False, minimu else: minimum_memory_required = max(inference_memory, minimum_memory_required + extra_reserved_memory()) - models_temp = set() + # Order-preserving dedup. A plain set() would randomize iteration order across runs + models_temp = {} for m in models: - models_temp.add(m) + models_temp[m] = None for mm in m.model_patches_models(): - models_temp.add(mm) + models_temp[mm] = None - models = models_temp + models = list(models_temp) + models.reverse() models_to_load = [] @@ -1181,6 +1178,10 @@ stream_counters = {} STREAM_CAST_BUFFERS = {} LARGEST_CASTED_WEIGHT = (None, 0) +STREAM_AIMDO_CAST_BUFFERS = {} +LARGEST_AIMDO_CASTED_WEIGHT = (None, 0) + +DEFAULT_AIMDO_CAST_BUFFER_RESERVATION_SIZE = 16 * 1024 ** 3 def get_cast_buffer(offload_stream, device, size, ref): global LARGEST_CASTED_WEIGHT @@ -1214,13 +1215,26 @@ def get_cast_buffer(offload_stream, device, size, ref): return cast_buffer +def get_aimdo_cast_buffer(offload_stream, device): + cast_buffer = STREAM_AIMDO_CAST_BUFFERS.get(offload_stream, None) + if cast_buffer is None: + cast_buffer = comfy_aimdo.vram_buffer.VRAMBuffer(DEFAULT_AIMDO_CAST_BUFFER_RESERVATION_SIZE, device.index) + STREAM_AIMDO_CAST_BUFFERS[offload_stream] = cast_buffer + + return cast_buffer def reset_cast_buffers(): global LARGEST_CASTED_WEIGHT + global LARGEST_AIMDO_CASTED_WEIGHT + LARGEST_CASTED_WEIGHT = (None, 0) - for offload_stream in STREAM_CAST_BUFFERS: - offload_stream.synchronize() + LARGEST_AIMDO_CASTED_WEIGHT = (None, 0) + for offload_stream in set(STREAM_CAST_BUFFERS) | set(STREAM_AIMDO_CAST_BUFFERS): + if offload_stream is not None: + offload_stream.synchronize() synchronize() + STREAM_CAST_BUFFERS.clear() + STREAM_AIMDO_CAST_BUFFERS.clear() soft_empty_cache() def get_offload_stream(device): @@ -1580,10 +1594,7 @@ def should_use_fp16(device=None, model_params=0, prioritize_performance=True, ma return False if is_intel_xpu(): - if torch_version_numeric < (2, 3): - return True - else: - return torch.xpu.get_device_properties(device).has_fp16 + return torch.xpu.get_device_properties(device).has_fp16 if is_ascend_npu(): return True @@ -1649,10 +1660,7 @@ def should_use_bf16(device=None, model_params=0, prioritize_performance=True, ma return False if is_intel_xpu(): - if torch_version_numeric < (2, 3): - return True - else: - return torch.xpu.is_bf16_supported() + return torch.xpu.is_bf16_supported() if is_ascend_npu(): return True @@ -1732,6 +1740,21 @@ def supports_mxfp8_compute(device=None): return True +def supports_fp64(device=None): + if is_device_mps(device): + return False + + if is_intel_xpu(): + return False + + if is_directml_enabled(): + return False + + if is_ixuca(): + return False + + return True + def extended_fp16_support(): # TODO: check why some models work with fp16 on newer torch versions but not on older if torch_version_numeric < (2, 7): @@ -1768,6 +1791,7 @@ def soft_empty_cache(force=False): if cpu_state == CPUState.MPS: torch.mps.empty_cache() elif is_intel_xpu(): + torch.xpu.synchronize() torch.xpu.empty_cache() elif is_ascend_npu(): torch.npu.empty_cache() @@ -1786,7 +1810,7 @@ def debug_memory_summary(): return torch.cuda.memory.memory_summary() return "" -class InterruptProcessingException(Exception): +class InterruptProcessingException(BaseException): pass interrupt_processing_mutex = threading.RLock() diff --git a/comfy/model_patcher.py b/comfy/model_patcher.py index 6deb71e12..2ea14bc2c 100644 --- a/comfy/model_patcher.py +++ b/comfy/model_patcher.py @@ -26,11 +26,13 @@ import uuid from typing import Callable, Optional import torch +import tqdm import comfy.float import comfy.hooks import comfy.lora import comfy.model_management +import comfy.ops import comfy.patcher_extension import comfy.utils from comfy.comfy_types import UnetWrapperFunction @@ -120,9 +122,20 @@ class LowVramPatch: self.patches = patches self.convert_func = convert_func # TODO: remove self.set_func = set_func + self.prepared_patches = None + + def prepare(self, allocate_buffer, stream): + self.prepared_patches = [ + (patch[0], comfy.lora.prefetch_prepared_value(patch[1], allocate_buffer, stream), patch[2], patch[3], patch[4]) + for patch in self.patches[self.key] + ] + + def clear_prepared(self): + self.prepared_patches = None def __call__(self, weight): - return comfy.lora.calculate_weight(self.patches[self.key], weight, self.key, intermediate_dtype=weight.dtype) + patches = self.prepared_patches if self.prepared_patches is not None else self.patches[self.key] + return comfy.lora.calculate_weight(patches, weight, self.key, intermediate_dtype=weight.dtype) LOWVRAM_PATCH_ESTIMATE_MATH_FACTOR = 2 @@ -229,6 +242,37 @@ class LazyCastingParam(torch.nn.Parameter): return self.model.patch_weight_to_device(self.key, device_to=self.model.load_device, return_weight=True).to("cpu") +class LazyCastingQuantizedParam: + def __init__(self, model, key): + self.model = model + self.key = key + self.cpu_state_dict = None + + def state_dict_tensor(self, state_dict_key): + if self.cpu_state_dict is None: + weight = self.model.patch_weight_to_device(self.key, device_to=self.model.load_device, return_weight=True) + self.cpu_state_dict = {k: v.to("cpu") for k, v in weight.state_dict(self.key).items()} + return self.cpu_state_dict[state_dict_key] + + +class LazyCastingParamPiece(torch.nn.Parameter): + def __new__(cls, caster, state_dict_key, tensor): + return super().__new__(cls, tensor) + + def __init__(self, caster, state_dict_key, tensor): + self.caster = caster + self.state_dict_key = state_dict_key + + @property + def device(self): + return CustomTorchDevice + + def to(self, *args, **kwargs): + caster = self.caster + del self.caster + return caster.state_dict_tensor(self.state_dict_key) + + class ModelPatcher: def __init__(self, model, load_device, offload_device, size=0, weight_inplace_update=False): self.size = size @@ -506,6 +550,10 @@ class ModelPatcher: def set_model_noise_refiner_patch(self, patch): self.set_model_patch(patch, "noise_refiner") + def set_model_middle_block_after_patch(self, patch): + self.set_model_patch(patch, "middle_block_after_patch") + + def set_model_rope_options(self, scale_x, shift_x, scale_y, shift_y, scale_t, shift_t, **kwargs): rope_options = self.model_options["transformer_options"].get("rope_options", {}) rope_options["scale_x"] = scale_x @@ -681,9 +729,9 @@ class ModelPatcher: sd.pop(k) return sd - def patch_weight_to_device(self, key, device_to=None, inplace_update=False, return_weight=False): + def patch_weight_to_device(self, key, device_to=None, inplace_update=False, return_weight=False, force_cast=False): weight, set_func, convert_func = get_key_weight(self.model, key) - if key not in self.patches: + if key not in self.patches and not force_cast: return weight inplace_update = self.weight_inplace_update or inplace_update @@ -691,7 +739,7 @@ class ModelPatcher: if key not in self.backup and not return_weight: self.backup[key] = collections.namedtuple('Dimension', ['weight', 'inplace_update'])(weight.to(device=self.offload_device, copy=inplace_update), inplace_update) - temp_dtype = comfy.model_management.lora_compute_dtype(device_to) + temp_dtype = comfy.model_management.lora_compute_dtype(device_to) if key in self.patches else None if device_to is not None: temp_weight = comfy.model_management.cast_to_device(weight, device_to, temp_dtype, copy=True) else: @@ -699,9 +747,10 @@ class ModelPatcher: if convert_func is not None: temp_weight = convert_func(temp_weight, inplace=True) - out_weight = comfy.lora.calculate_weight(self.patches[key], temp_weight, key) + out_weight = comfy.lora.calculate_weight(self.patches[key], temp_weight, key) if key in self.patches else temp_weight if set_func is None: - out_weight = comfy.float.stochastic_rounding(out_weight, weight.dtype, seed=comfy.utils.string_to_seed(key)) + if key in self.patches: + out_weight = comfy.float.stochastic_rounding(out_weight, weight.dtype, seed=comfy.utils.string_to_seed(key)) if return_weight: return out_weight elif inplace_update: @@ -851,7 +900,9 @@ class ModelPatcher: if m.comfy_patched_weights == True: continue - for param in params: + for param, param_value in params.items(): + if hasattr(m, "comfy_cast_weights") and getattr(param_value, "is_meta", False): + comfy.ops.disable_weight_init._zero_init_parameter(m, param) key = key_param_name_to_key(n, param) self.unpin_weight(key) self.patch_weight_to_device(key, device_to=device_to) @@ -1443,20 +1494,37 @@ class ModelPatcher: self.clear_cached_hook_weights() def state_dict_for_saving(self, clip_state_dict=None, vae_state_dict=None, clip_vision_state_dict=None): - unet_state_dict = self.model.diffusion_model.state_dict() - for k, v in unet_state_dict.items(): + original_state_dict = self.model.diffusion_model.state_dict() + unet_state_dict = {} + keys = list(original_state_dict) + while len(keys) > 0: + k = keys.pop(0) + v = original_state_dict[k] op_keys = k.rsplit('.', 1) if (len(op_keys) < 2) or op_keys[1] not in ["weight", "bias"]: + unet_state_dict[k] = v continue try: op = comfy.utils.get_attr(self.model.diffusion_model, op_keys[0]) except: + unet_state_dict[k] = v continue if not op or not hasattr(op, "comfy_cast_weights") or \ (hasattr(op, "comfy_patched_weights") and op.comfy_patched_weights == True): + unet_state_dict[k] = v continue key = "diffusion_model." + k - unet_state_dict[k] = LazyCastingParam(self, key, comfy.utils.get_attr(self.model, key)) + weight = comfy.utils.get_attr(self.model, key) + if isinstance(weight, QuantizedTensor) and k in original_state_dict: + qt_state_dict = weight.state_dict(k) + caster = LazyCastingQuantizedParam(self, key) + for group_key in (x for x in qt_state_dict if x in original_state_dict): + if group_key in keys: + keys.remove(group_key) + unet_state_dict.pop(group_key, "") + unet_state_dict[group_key] = LazyCastingParamPiece(caster, "diffusion_model." + group_key, original_state_dict[group_key]) + continue + unet_state_dict[k] = LazyCastingParam(self, key, weight) return self.model.state_dict_for_saving(unet_state_dict, clip_state_dict=clip_state_dict, vae_state_dict=vae_state_dict, clip_vision_state_dict=clip_vision_state_dict) def __del__(self): @@ -1580,7 +1648,7 @@ class ModelPatcherDynamic(ModelPatcher): key = key_param_name_to_key(n, param_key) if key in self.backup: comfy.utils.set_attr_param(self.model, key, self.backup[key].weight) - self.patch_weight_to_device(key, device_to=device_to) + self.patch_weight_to_device(key, device_to=device_to, force_cast=True) weight, _, _ = get_key_weight(self.model, key) if weight is not None: self.model.model_loaded_weight_memory += weight.numel() * weight.element_size() @@ -1605,6 +1673,10 @@ class ModelPatcherDynamic(ModelPatcher): m._v = vbar.alloc(v_weight_size) allocated_size += v_weight_size + for param in params: + if param not in ("weight", "bias"): + force_load_param(self, param, device_to) + else: for param in params: key = key_param_name_to_key(n, param) @@ -1628,7 +1700,11 @@ class ModelPatcherDynamic(ModelPatcher): self.model.model_loaded_weight_memory += casted_buf.numel() * casted_buf.element_size() force_load_stat = f" Force pre-loaded {len(self.backup)} weights: {self.model.model_loaded_weight_memory // 1024} KB." if len(self.backup) > 0 else "" - logging.info(f"Model {self.model.__class__.__name__} prepared for dynamic VRAM loading. {allocated_size // (1024 ** 2)}MB Staged. {num_patches} patches attached.{force_load_stat}") + log_key = (self.patches_uuid, allocated_size, num_patches, len(self.backup), self.model.model_loaded_weight_memory) + in_loop = bool(getattr(tqdm.tqdm, "_instances", None)) + level = logging.DEBUG if in_loop and getattr(self, "_last_prepare_log_key", None) == log_key else logging.INFO + self._last_prepare_log_key = log_key + logging.log(level, f"Model {self.model.__class__.__name__} prepared for dynamic VRAM loading. {allocated_size // (1024 ** 2)}MB Staged. {num_patches} patches attached.{force_load_stat}") self.model.device = device_to self.model.current_weight_patches_uuid = self.patches_uuid diff --git a/comfy/model_prefetch.py b/comfy/model_prefetch.py new file mode 100644 index 000000000..72e11dec6 --- /dev/null +++ b/comfy/model_prefetch.py @@ -0,0 +1,66 @@ +import comfy_aimdo.model_vbar +import comfy.model_management +import comfy.ops + +PREFETCH_QUEUES = [] + +def cleanup_prefetched_modules(comfy_modules): + for s in comfy_modules: + prefetch = getattr(s, "_prefetch", None) + if prefetch is None: + continue + for param_key in ("weight", "bias"): + lowvram_fn = getattr(s, param_key + "_lowvram_function", None) + if lowvram_fn is not None: + lowvram_fn.clear_prepared() + if prefetch["signature"] is not None: + comfy_aimdo.model_vbar.vbar_unpin(s._v) + delattr(s, "_prefetch") + +def cleanup_prefetch_queues(): + global PREFETCH_QUEUES + + for queue in PREFETCH_QUEUES: + for entry in queue: + if entry is None or not isinstance(entry, tuple): + continue + _, prefetch_state = entry + comfy_modules = prefetch_state[1] + if comfy_modules is not None: + cleanup_prefetched_modules(comfy_modules) + PREFETCH_QUEUES = [] + +def prefetch_queue_pop(queue, device, module): + if queue is None: + return + + consumed = queue.pop(0) + if consumed is not None: + offload_stream, prefetch_state = consumed + if offload_stream is not None: + offload_stream.wait_stream(comfy.model_management.current_stream(device)) + _, comfy_modules = prefetch_state + if comfy_modules is not None: + cleanup_prefetched_modules(comfy_modules) + + prefetch = queue[0] + if prefetch is not None: + comfy_modules = [] + for s in prefetch.modules(): + if hasattr(s, "_v"): + comfy_modules.append(s) + + offload_stream = comfy.ops.cast_modules_with_vbar(comfy_modules, None, device, None, True) + comfy.model_management.sync_stream(device, offload_stream) + queue[0] = (offload_stream, (prefetch, comfy_modules)) + +def make_prefetch_queue(queue, device, transformer_options): + if (not transformer_options.get("prefetch_dynamic_vbars", False) + or comfy.model_management.NUM_STREAMS == 0 + or comfy.model_management.is_device_cpu(device) + or not comfy.model_management.device_supports_non_blocking(device)): + return None + + queue = [None] + queue + [None] + PREFETCH_QUEUES.append(queue) + return queue diff --git a/comfy/model_sampling.py b/comfy/model_sampling.py index 13860e6a2..5af336e76 100644 --- a/comfy/model_sampling.py +++ b/comfy/model_sampling.py @@ -54,6 +54,30 @@ class V_PREDICTION(EPS): sigma = reshape_sigma(sigma, model_output.ndim) return model_input * self.sigma_data ** 2 / (sigma ** 2 + self.sigma_data ** 2) - model_output * sigma * self.sigma_data / (sigma ** 2 + self.sigma_data ** 2) ** 0.5 +class V_PREDICTION_DDPM: + """CogVideoX v-prediction: model receives raw x_t (unscaled), predicts velocity v. + x_0 = sqrt(alpha) * x_t - sqrt(1-alpha) * v + = x_t / sqrt(sigma^2 + 1) - v * sigma / sqrt(sigma^2 + 1) + """ + def calculate_input(self, sigma, noise): + return noise + + def calculate_denoised(self, sigma, model_output, model_input): + sigma = reshape_sigma(sigma, model_output.ndim) + return model_input / (sigma ** 2 + 1.0) ** 0.5 - model_output * sigma / (sigma ** 2 + 1.0) ** 0.5 + + def noise_scaling(self, sigma, noise, latent_image, max_denoise=False): + sigma = reshape_sigma(sigma, noise.ndim) + if max_denoise: + noise = noise * torch.sqrt(1.0 + sigma ** 2.0) + else: + noise = noise * sigma + noise += latent_image + return noise + + def inverse_noise_scaling(self, sigma, latent): + return latent + class EDM(V_PREDICTION): def calculate_denoised(self, sigma, model_output, model_input): sigma = reshape_sigma(sigma, model_output.ndim) @@ -69,7 +93,8 @@ class CONST: def noise_scaling(self, sigma, noise, latent_image, max_denoise=False): sigma = reshape_sigma(sigma, noise.ndim) - return sigma * noise + (1.0 - sigma) * latent_image + s = getattr(self, "noise_scale", 1.0) + return sigma * (s * noise) + (1.0 - sigma) * latent_image def inverse_noise_scaling(self, sigma, latent): sigma = reshape_sigma(sigma, latent.ndim) @@ -264,7 +289,11 @@ class ModelSamplingDiscreteFlow(torch.nn.Module): else: sampling_settings = {} - self.set_parameters(shift=sampling_settings.get("shift", 1.0), multiplier=sampling_settings.get("multiplier", 1000)) + self.set_noise_scale(sampling_settings.get("noise_scale", 1.0)) + self.set_parameters( + shift=sampling_settings.get("shift", 1.0), + multiplier=sampling_settings.get("multiplier", 1000), + ) def set_parameters(self, shift=1.0, timesteps=1000, multiplier=1000): self.shift = shift @@ -272,6 +301,9 @@ class ModelSamplingDiscreteFlow(torch.nn.Module): ts = self.sigma((torch.arange(1, timesteps + 1, 1) / timesteps) * multiplier) self.register_buffer('sigmas', ts) + def set_noise_scale(self, noise_scale): + self.noise_scale = float(noise_scale) + @property def sigma_min(self): return self.sigmas[0] diff --git a/comfy/ops.py b/comfy/ops.py index b5cd1d47e..117cdd327 100644 --- a/comfy/ops.py +++ b/comfy/ops.py @@ -79,37 +79,68 @@ def cast_to_input(weight, input, non_blocking=False, copy=True): return comfy.model_management.cast_to(weight, input.dtype, input.device, non_blocking=non_blocking, copy=copy) -def cast_bias_weight_with_vbar(s, dtype, device, bias_dtype, non_blocking, compute_dtype, want_requant): +def materialize_meta_param(s, param_keys): + for param_key in param_keys: + param = getattr(s, param_key, None) + if param is not None and getattr(param, "is_meta", False): + setattr(s, param_key, torch.nn.Parameter(torch.zeros(param.shape, dtype=param.dtype), requires_grad=param.requires_grad)) - #vbar doesn't support CPU weights, but some custom nodes have weird paths - #that might switch the layer to the CPU and expect it to work. We have to take - #a clone conservatively as we are mmapped and some SFT files are packed misaligned - #If you are a custom node author reading this, please move your layer to the GPU - #or declare your ModelPatcher as CPU in the first place. - if comfy.model_management.is_device_cpu(device): - weight = s.weight.to(dtype=dtype, copy=True) - if isinstance(weight, QuantizedTensor): - weight = weight.dequantize() - bias = None - if s.bias is not None: - bias = s.bias.to(dtype=bias_dtype, copy=True) - return weight, bias, (None, None, None) +# FIXME: add n=1 cache hit fast path +def cast_modules_with_vbar(comfy_modules, dtype, device, bias_dtype, non_blocking): offload_stream = None - xfer_dest = None + cast_buffer = None + cast_buffer_offset = 0 + + def ensure_offload_stream(module, required_size, check_largest): + nonlocal offload_stream + nonlocal cast_buffer + + if offload_stream is None: + offload_stream = comfy.model_management.get_offload_stream(device) + if offload_stream is None or not check_largest or len(comfy_modules) != 1: + return + + current_size = 0 if cast_buffer is None else cast_buffer.size() + if current_size < required_size and module is comfy.model_management.LARGEST_AIMDO_CASTED_WEIGHT[0]: + offload_stream = comfy.model_management.get_offload_stream(device) + cast_buffer = None + if required_size > comfy.model_management.LARGEST_AIMDO_CASTED_WEIGHT[1]: + comfy.model_management.LARGEST_AIMDO_CASTED_WEIGHT = (module, required_size) + + def get_cast_buffer(buffer_size): + nonlocal offload_stream + nonlocal cast_buffer + nonlocal cast_buffer_offset + + if buffer_size == 0: + return None + + if offload_stream is None: + return torch.empty((buffer_size,), dtype=torch.uint8, device=device) + + cast_buffer = comfy.model_management.get_aimdo_cast_buffer(offload_stream, device) + buffer = comfy_aimdo.torch.aimdo_to_tensor(cast_buffer.get(buffer_size, cast_buffer_offset), device) + cast_buffer_offset += buffer_size + return buffer + + for s in comfy_modules: + signature = comfy_aimdo.model_vbar.vbar_fault(s._v) + resident = comfy_aimdo.model_vbar.vbar_signature_compare(signature, s._v_signature) + prefetch = { + "signature": signature, + "resident": resident, + } - signature = comfy_aimdo.model_vbar.vbar_fault(s._v) - resident = comfy_aimdo.model_vbar.vbar_signature_compare(signature, s._v_signature) - if signature is not None: if resident: - weight = s._v_weight - bias = s._v_bias - else: - xfer_dest = comfy_aimdo.torch.aimdo_to_tensor(s._v, device) + s._prefetch = prefetch + continue - if not resident: + materialize_meta_param(s, ["weight", "bias"]) + xfer_dest = comfy_aimdo.torch.aimdo_to_tensor(s._v, device) if signature is not None else None cast_geometry = comfy.memory_management.tensors_to_geometries([ s.weight, s.bias ]) cast_dest = None + needs_cast = False xfer_source = [ s.weight, s.bias ] @@ -121,22 +152,15 @@ def cast_bias_weight_with_vbar(s, dtype, device, bias_dtype, non_blocking, compu if data is None: continue if data.dtype != geometry.dtype: + needs_cast = True cast_dest = xfer_dest - if cast_dest is None: - cast_dest = torch.empty((comfy.memory_management.vram_aligned_size(cast_geometry),), dtype=torch.uint8, device=device) xfer_dest = None break dest_size = comfy.memory_management.vram_aligned_size(xfer_source) - offload_stream = comfy.model_management.get_offload_stream(device) - if xfer_dest is None and offload_stream is not None: - xfer_dest = comfy.model_management.get_cast_buffer(offload_stream, device, dest_size, s) - if xfer_dest is None: - offload_stream = comfy.model_management.get_offload_stream(device) - xfer_dest = comfy.model_management.get_cast_buffer(offload_stream, device, dest_size, s) + ensure_offload_stream(s, dest_size if xfer_dest is None else 0, True) if xfer_dest is None: - xfer_dest = torch.empty((dest_size,), dtype=torch.uint8, device=device) - offload_stream = None + xfer_dest = get_cast_buffer(dest_size) if signature is None and pin is None: comfy.pinned_memory.pin_memory(s) @@ -149,27 +173,54 @@ def cast_bias_weight_with_vbar(s, dtype, device, bias_dtype, non_blocking, compu xfer_source = [ pin ] #send it over comfy.model_management.cast_to_gathered(xfer_source, xfer_dest, non_blocking=non_blocking, stream=offload_stream) - comfy.model_management.sync_stream(device, offload_stream) - if cast_dest is not None: + for param_key in ("weight", "bias"): + lowvram_fn = getattr(s, param_key + "_lowvram_function", None) + if lowvram_fn is not None: + ensure_offload_stream(s, cast_buffer_offset, False) + lowvram_fn.prepare(lambda size: get_cast_buffer(size), offload_stream) + + prefetch["xfer_dest"] = xfer_dest + prefetch["cast_dest"] = cast_dest + prefetch["cast_geometry"] = cast_geometry + prefetch["needs_cast"] = needs_cast + s._prefetch = prefetch + + return offload_stream + + +def resolve_cast_module_with_vbar(s, dtype, device, bias_dtype, compute_dtype, want_requant): + + prefetch = getattr(s, "_prefetch", None) + + if prefetch["resident"]: + weight = s._v_weight + bias = s._v_bias + else: + xfer_dest = prefetch["xfer_dest"] + if prefetch["needs_cast"]: + cast_dest = prefetch["cast_dest"] if prefetch["cast_dest"] is not None else torch.empty((comfy.memory_management.vram_aligned_size(prefetch["cast_geometry"]),), dtype=torch.uint8, device=device) for pre_cast, post_cast in zip(comfy.memory_management.interpret_gathered_like([s.weight, s.bias ], xfer_dest), - comfy.memory_management.interpret_gathered_like(cast_geometry, cast_dest)): + comfy.memory_management.interpret_gathered_like(prefetch["cast_geometry"], cast_dest)): if post_cast is not None: post_cast.copy_(pre_cast) xfer_dest = cast_dest - params = comfy.memory_management.interpret_gathered_like(cast_geometry, xfer_dest) + params = comfy.memory_management.interpret_gathered_like(prefetch["cast_geometry"], xfer_dest) weight = params[0] bias = params[1] - if signature is not None: + if prefetch["signature"] is not None: s._v_weight = weight s._v_bias = bias - s._v_signature=signature + s._v_signature = prefetch["signature"] def post_cast(s, param_key, x, dtype, resident, update_weight): lowvram_fn = getattr(s, param_key + "_lowvram_function", None) fns = getattr(s, param_key + "_function", []) + if x is None: + return None + orig = x def to_dequant(tensor, dtype): @@ -197,14 +248,15 @@ def cast_bias_weight_with_vbar(s, dtype, device, bias_dtype, non_blocking, compu x = f(x) return x - update_weight = signature is not None + update_weight = prefetch["signature"] is not None + weight = post_cast(s, "weight", weight, dtype, prefetch["resident"], update_weight) + if bias is not None: + bias = post_cast(s, "bias", bias, bias_dtype, prefetch["resident"], update_weight) - weight = post_cast(s, "weight", weight, dtype, resident, update_weight) - if s.bias is not None: - bias = post_cast(s, "bias", bias, bias_dtype, resident, update_weight) + if prefetch["signature"] is not None: + prefetch["resident"] = True - #FIXME: weird offload return protocol - return weight, bias, (offload_stream, device if signature is not None else None, None) + return weight, bias def cast_bias_weight(s, input=None, dtype=None, device=None, bias_dtype=None, offloadable=False, compute_dtype=None, want_requant=False): @@ -222,10 +274,46 @@ def cast_bias_weight(s, input=None, dtype=None, device=None, bias_dtype=None, of if device is None: device = input.device + def format_return(result, offloadable): + weight, bias, offload_stream = result + return (weight, bias, offload_stream) if offloadable else (weight, bias) + non_blocking = comfy.model_management.device_supports_non_blocking(device) if hasattr(s, "_v"): - return cast_bias_weight_with_vbar(s, dtype, device, bias_dtype, non_blocking, compute_dtype, want_requant) + + #vbar doesn't support CPU weights, but some custom nodes have weird paths + #that might switch the layer to the CPU and expect it to work. We have to take + #a clone conservatively as we are mmapped and some SFT files are packed misaligned + #If you are a custom node author reading this, please move your layer to the GPU + #or declare your ModelPatcher as CPU in the first place. + if comfy.model_management.is_device_cpu(device): + materialize_meta_param(s, ["weight", "bias"]) + weight = s.weight.to(dtype=dtype, copy=True) + if isinstance(weight, QuantizedTensor): + weight = weight.dequantize() + bias = s.bias.to(dtype=bias_dtype, copy=True) if s.bias is not None else None + return format_return((weight, bias, (None, None, None)), offloadable) + + prefetched = hasattr(s, "_prefetch") + offload_stream = None + offload_device = None + if not prefetched: + offload_stream = cast_modules_with_vbar([s], dtype, device, bias_dtype, non_blocking) + comfy.model_management.sync_stream(device, offload_stream) + + weight, bias = resolve_cast_module_with_vbar(s, dtype, device, bias_dtype, compute_dtype, want_requant) + + if not prefetched: + if getattr(s, "_prefetch")["signature"] is not None: + offload_device = device + for param_key in ("weight", "bias"): + lowvram_fn = getattr(s, param_key + "_lowvram_function", None) + if lowvram_fn is not None: + lowvram_fn.clear_prepared() + delattr(s, "_prefetch") + return format_return((weight, bias, (offload_stream, offload_device, None)), offloadable) + if offloadable and (device != s.weight.device or (s.bias is not None and device != s.bias.device)): @@ -272,11 +360,7 @@ def cast_bias_weight(s, input=None, dtype=None, device=None, bias_dtype=None, of for f in s.weight_function: weight = f(weight) - if offloadable: - return weight, bias, (offload_stream, weight_a, bias_a) - else: - #Legacy function signature - return weight, bias + return format_return((weight, bias, (offload_stream, weight_a, bias_a)), offloadable) def uncast_bias_weight(s, weight, bias, offload_stream): @@ -306,6 +390,12 @@ class CastWeightBiasOp: bias_function = [] class disable_weight_init: + @staticmethod + def _zero_init_parameter(module, name): + param = getattr(module, name) + device = None if getattr(param, "is_meta", False) else param.device + setattr(module, name, torch.nn.Parameter(torch.zeros(param.shape, device=device, dtype=param.dtype), requires_grad=False)) + @staticmethod def _lazy_load_from_state_dict(module, state_dict, prefix, local_metadata, missing_keys, unexpected_keys, weight_shape, @@ -472,6 +562,25 @@ class disable_weight_init: else: return super().forward(*args, **kwargs) + class BatchNorm2d(torch.nn.BatchNorm2d, CastWeightBiasOp): + def reset_parameters(self): + return None + + def forward_comfy_cast_weights(self, input): + weight, bias, offload_stream = cast_bias_weight(self, input, offloadable=True) + running_mean = self.running_mean.to(device=input.device, dtype=weight.dtype) if self.running_mean is not None else None + running_var = self.running_var.to(device=input.device, dtype=weight.dtype) if self.running_var is not None else None + x = torch.nn.functional.batch_norm(input, running_mean, running_var, weight, bias, self.training, self.momentum, self.eps) + uncast_bias_weight(self, weight, bias, offload_stream) + return x + + def forward(self, *args, **kwargs): + run_every_op() + if self.comfy_cast_weights or len(self.weight_function) > 0 or len(self.bias_function) > 0: + return self.forward_comfy_cast_weights(*args, **kwargs) + else: + return super().forward(*args, **kwargs) + class LayerNorm(torch.nn.LayerNorm, CastWeightBiasOp): def reset_parameters(self): return None @@ -659,6 +768,9 @@ class manual_cast(disable_weight_init): class Conv3d(disable_weight_init.Conv3d): comfy_cast_weights = True + class BatchNorm2d(disable_weight_init.BatchNorm2d): + comfy_cast_weights = True + class GroupNorm(disable_weight_init.GroupNorm): comfy_cast_weights = True @@ -1151,7 +1263,7 @@ def mixed_precision_ops(quant_config={}, compute_dtype=torch.bfloat16, full_prec if param is None: continue p = fn(param) - if p.is_inference(): + if (not torch.is_inference_mode_enabled()) and p.is_inference(): p = p.clone() self.register_parameter(key, torch.nn.Parameter(p, requires_grad=False)) for key, buf in self._buffers.items(): @@ -1159,6 +1271,94 @@ def mixed_precision_ops(quant_config={}, compute_dtype=torch.bfloat16, full_prec self._buffers[key] = fn(buf) return self + class Embedding(manual_cast.Embedding): + def _load_from_state_dict(self, state_dict, prefix, local_metadata, + strict, missing_keys, unexpected_keys, error_msgs): + weight_key = f"{prefix}weight" + layer_conf = state_dict.pop(f"{prefix}comfy_quant", None) + if layer_conf is not None: + layer_conf = json.loads(layer_conf.numpy().tobytes()) + + # Only fp8 makes sense for embeddings (per-row dequant via index select). + # Block-scaled formats (NVFP4, MXFP8) can't do per-row lookup efficiently. + quant_format = layer_conf.get("format", None) if layer_conf is not None else None + if quant_format in ["float8_e4m3fn", "float8_e5m2"] and weight_key in state_dict: + self.quant_format = quant_format + qconfig = QUANT_ALGOS[quant_format] + self.layout_type = qconfig["comfy_tensor_layout"] + layout_cls = get_layout_class(self.layout_type) + weight = state_dict.pop(weight_key) + manually_loaded_keys = [weight_key] + + scale_key = f"{prefix}weight_scale" + scale = state_dict.pop(scale_key, None) + if scale is not None: + scale = scale.float() + manually_loaded_keys.append(scale_key) + + params = layout_cls.Params( + scale=scale if scale is not None else torch.ones((), dtype=torch.float32), + orig_dtype=MixedPrecisionOps._compute_dtype, + orig_shape=(self.num_embeddings, self.embedding_dim), + ) + self.weight = torch.nn.Parameter( + QuantizedTensor(weight.to(dtype=qconfig["storage_t"]), qconfig["comfy_tensor_layout"], params), + requires_grad=False) + + super()._load_from_state_dict(state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs) + for k in manually_loaded_keys: + if k in missing_keys: + missing_keys.remove(k) + else: + if layer_conf is not None: + state_dict[f"{prefix}comfy_quant"] = torch.tensor(list(json.dumps(layer_conf).encode('utf-8')), dtype=torch.uint8) + super()._load_from_state_dict(state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs) + + def state_dict(self, *args, destination=None, prefix="", **kwargs): + if destination is not None: + sd = destination + else: + sd = {} + + if not hasattr(self, 'weight') or self.weight is None: + return sd + + if isinstance(self.weight, QuantizedTensor): + sd_out = self.weight.state_dict("{}weight".format(prefix)) + for k in sd_out: + sd[k] = sd_out[k] + + quant_conf = {"format": self.quant_format} + sd["{}comfy_quant".format(prefix)] = torch.tensor(list(json.dumps(quant_conf).encode('utf-8')), dtype=torch.uint8) + else: + sd["{}weight".format(prefix)] = self.weight + return sd + + def forward_comfy_cast_weights(self, input, out_dtype=None): + weight = self.weight + + # Optimized path: lookup in fp8, dequantize only the selected rows. + if isinstance(weight, QuantizedTensor) and len(self.weight_function) == 0: + qdata, _, offload_stream = cast_bias_weight(self, device=input.device, dtype=weight.dtype, offloadable=True) + if isinstance(qdata, QuantizedTensor): + scale = qdata._params.scale + qdata = qdata._qdata + else: + scale = None + + x = torch.nn.functional.embedding( + input, qdata, self.padding_idx, self.max_norm, + self.norm_type, self.scale_grad_by_freq, self.sparse) + uncast_bias_weight(self, qdata, None, offload_stream) + target_dtype = out_dtype if out_dtype is not None else weight._params.orig_dtype + x = x.to(dtype=target_dtype) + if scale is not None and scale != 1.0: + x = x * scale.to(dtype=target_dtype) + return x + + # Fallback for non-quantized or weight_function (LoRA) case + return super().forward_comfy_cast_weights(input, out_dtype=out_dtype) + return MixedPrecisionOps def pick_operations(weight_dtype, compute_dtype, load_device=None, disable_fast_fp8=False, fp8_optimizations=False, model_config=None): diff --git a/comfy/pinned_memory.py b/comfy/pinned_memory.py index 6f142282d..6d3ba367a 100644 --- a/comfy/pinned_memory.py +++ b/comfy/pinned_memory.py @@ -2,7 +2,6 @@ import comfy.model_management import comfy.memory_management import comfy_aimdo.host_buffer import comfy_aimdo.torch -import psutil from comfy.cli_args import args @@ -12,11 +11,6 @@ def get_pin(module): def pin_memory(module): if module.pin_failed or args.disable_pinned_memory or get_pin(module) is not None: return - #FIXME: This is a RAM cache trigger event - ram_headroom = comfy.memory_management.RAM_CACHE_HEADROOM - #we split the difference and assume half the RAM cache headroom is for us - if ram_headroom > 0 and psutil.virtual_memory().available < (ram_headroom * 0.5): - comfy.memory_management.extra_ram_release(ram_headroom) size = comfy.memory_management.vram_aligned_size([ module.weight, module.bias ]) diff --git a/comfy/quant_ops.py b/comfy/quant_ops.py index 42ee08fb2..b90bcfd25 100644 --- a/comfy/quant_ops.py +++ b/comfy/quant_ops.py @@ -1,6 +1,8 @@ import torch import logging +from comfy.cli_args import args + try: import comfy_kitchen as ck from comfy_kitchen.tensor import ( @@ -21,7 +23,15 @@ try: ck.registry.disable("cuda") logging.warning("WARNING: You need pytorch with cu130 or higher to use optimized CUDA operations.") - ck.registry.disable("triton") + if args.enable_triton_backend: + try: + import triton + logging.info("Found triton %s. Enabling comfy-kitchen triton backend.", triton.__version__) + except ImportError as e: + logging.error(f"Failed to import triton, Error: {e}, the comfy-kitchen triton backend will not be available.") + ck.registry.disable("triton") + else: + ck.registry.disable("triton") for k, v in ck.list_backends().items(): logging.info(f"Found comfy_kitchen backend {k}: {v}") except ImportError as e: diff --git a/comfy/rmsnorm.py b/comfy/rmsnorm.py index ab7cf14fa..e54be98d6 100644 --- a/comfy/rmsnorm.py +++ b/comfy/rmsnorm.py @@ -3,6 +3,7 @@ import comfy.model_management RMSNorm = torch.nn.RMSNorm +# Note: torch's fused F.rms_norm is faster but produces slightly different output than manual implementations (rsqrt/reduction rounding). def rms_norm(x, weight=None, eps=1e-6): if weight is None: return torch.nn.functional.rms_norm(x, (x.shape[-1],), eps=eps) diff --git a/comfy/sampler_helpers.py b/comfy/sampler_helpers.py index bbba09e26..3782fd2d5 100644 --- a/comfy/sampler_helpers.py +++ b/comfy/sampler_helpers.py @@ -89,7 +89,8 @@ def get_additional_models(conds, dtype): gligen += get_models_from_cond(conds[k], "gligen") add_models += get_models_from_cond(conds[k], "additional_models") - control_nets = set(cnets) + # Order-preserving dedup. A plain set() would randomize iteration order across runs + control_nets = list(dict.fromkeys(cnets)) inference_memory = 0 control_models = [] diff --git a/comfy/sd.py b/comfy/sd.py index f331feefb..ab2718892 100644 --- a/comfy/sd.py +++ b/comfy/sd.py @@ -12,11 +12,13 @@ from .ldm.cascade.stage_c_coder import StageC_coder from .ldm.audio.autoencoder import AudioOobleckVAE import comfy.ldm.genmo.vae.model import comfy.ldm.lightricks.vae.causal_video_autoencoder +import comfy.ldm.lightricks.vae.audio_vae import comfy.ldm.cosmos.vae import comfy.ldm.wan.vae import comfy.ldm.wan.vae2_2 import comfy.ldm.hunyuan3d.vae import comfy.ldm.ace.vae.music_dcae_pipeline +import comfy.ldm.cogvideo.vae import comfy.ldm.hunyuan_video.vae import comfy.ldm.mmaudio.vae.autoencoder import comfy.pixel_space_convert @@ -62,6 +64,9 @@ import comfy.text_encoders.anima import comfy.text_encoders.ace15 import comfy.text_encoders.longcat_image import comfy.text_encoders.qwen35 +import comfy.text_encoders.ernie +import comfy.text_encoders.gemma4 +import comfy.text_encoders.cogvideo import comfy.model_patcher import comfy.lora @@ -234,7 +239,8 @@ class CLIP: model_management.archive_model_dtypes(self.cond_stage_model) self.tokenizer = tokenizer(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data) - ModelPatcher = comfy.model_patcher.ModelPatcher if disable_dynamic else comfy.model_patcher.CoreModelPatcher + te_disable_dynamic = disable_dynamic or getattr(self.cond_stage_model, "disable_offload", False) + ModelPatcher = comfy.model_patcher.ModelPatcher if te_disable_dynamic else comfy.model_patcher.CoreModelPatcher self.patcher = ModelPatcher(self.cond_stage_model, load_device=load_device, offload_device=offload_device) #Match torch.float32 hardcode upcast in TE implemention self.patcher.set_model_compute_dtype(torch.float32) @@ -476,7 +482,10 @@ class VAE: encoder_config={'target': "comfy.ldm.modules.diffusionmodules.model.Encoder", 'params': encoder_config}, decoder_config={'target': "comfy.ldm.modules.temporal_ae.VideoDecoder", 'params': decoder_config}) elif "taesd_decoder.1.weight" in sd: - self.latent_channels = sd["taesd_decoder.1.weight"].shape[1] + if isinstance(metadata, dict) and "tae_latent_channels" in metadata: + self.latent_channels = metadata["tae_latent_channels"] + else: + self.latent_channels = sd["taesd_decoder.1.weight"].shape[1] self.first_stage_model = comfy.taesd.taesd.TAESD(latent_channels=self.latent_channels) elif "vquantizer.codebook.weight" in sd: #VQGan: stage a of stable cascade self.first_stage_model = StageA() @@ -650,6 +659,17 @@ class VAE: self.memory_used_encode = lambda shape, dtype: (1400 * 9 * shape[-2] * shape[-1]) * model_management.dtype_size(dtype) self.memory_used_decode = lambda shape, dtype: (3600 * 4 * shape[-2] * shape[-1] * 16 * 16) * model_management.dtype_size(dtype) + elif "decoder.conv_in.conv.weight" in sd and "decoder.mid_block.resnets.0.norm1.norm_layer.weight" in sd: # CogVideoX VAE + self.upscale_ratio = (lambda a: max(0, a * 4 - 3), 8, 8) + self.upscale_index_formula = (4, 8, 8) + self.downscale_ratio = (lambda a: max(0, math.floor((a + 3) / 4)), 8, 8) + self.downscale_index_formula = (4, 8, 8) + self.latent_dim = 3 + self.latent_channels = sd["encoder.conv_out.conv.weight"].shape[0] // 2 + self.first_stage_model = comfy.ldm.cogvideo.vae.AutoencoderKLCogVideoX(latent_channels=self.latent_channels) + self.memory_used_decode = lambda shape, dtype: (2800 * max(2, ((shape[2] - 1) * 4) + 1) * shape[3] * shape[4] * (8 * 8)) * model_management.dtype_size(dtype) + self.memory_used_encode = lambda shape, dtype: (1400 * max(1, shape[2]) * shape[3] * shape[4]) * model_management.dtype_size(dtype) + self.working_dtypes = [torch.bfloat16, torch.float16, torch.float32] elif "decoder.conv_in.conv.weight" in sd: ddconfig = {'double_z': True, 'z_channels': 4, 'resolution': 256, 'in_channels': 3, 'out_ch': 3, 'ch': 128, 'ch_mult': [1, 2, 4, 4], 'num_res_blocks': 2, 'attn_resolutions': [], 'dropout': 0.0} ddconfig["conv3d"] = True @@ -757,6 +777,7 @@ class VAE: self.latent_channels = 3 self.latent_dim = 2 self.output_channels = 3 + self.disable_offload = True elif "vocoder.activation_post.downsample.lowpass.filter" in sd: #MMAudio VAE sample_rate = 16000 if sample_rate == 16000: @@ -804,6 +825,24 @@ class VAE: self.downscale_index_formula = (4, 8, 8) self.memory_used_encode = lambda shape, dtype: (700 * (max(1, (shape[-3] ** 0.66 * 0.11)) * shape[-2] * shape[-1]) * model_management.dtype_size(dtype)) self.memory_used_decode = lambda shape, dtype: (50 * (max(1, (shape[-3] ** 0.65 * 0.26)) * shape[-2] * shape[-1] * 32 * 32) * model_management.dtype_size(dtype)) + elif "vocoder.resblocks.0.convs1.0.weight" in sd or "vocoder.vocoder.resblocks.0.convs1.0.weight" in sd: # LTX Audio + sd = comfy.utils.state_dict_prefix_replace(sd, {"audio_vae.": "autoencoder."}) + self.first_stage_model = comfy.ldm.lightricks.vae.audio_vae.AudioVAE(metadata=metadata) + self.memory_used_encode = lambda shape, dtype: (shape[2] * 330) * model_management.dtype_size(dtype) + self.memory_used_decode = lambda shape, dtype: (shape[2] * shape[3] * 87000) * model_management.dtype_size(dtype) + self.latent_channels = self.first_stage_model.latent_channels + self.audio_sample_rate_output = self.first_stage_model.output_sample_rate + self.autoencoder = self.first_stage_model.autoencoder # TODO: remove hack for ltxv custom nodes + self.output_channels = 2 + self.pad_channel_value = "replicate" + self.upscale_ratio = 4096 + self.downscale_ratio = 4096 + self.latent_dim = 2 + self.process_output = lambda audio: audio + self.process_input = lambda audio: audio + self.working_dtypes = [torch.float32] + self.disable_offload = True + self.extra_1d_channel = 16 else: logging.warning("WARNING: No VAE weights detected, VAE not initalized.") self.first_stage_model = None @@ -1188,6 +1227,7 @@ class CLIPType(Enum): NEWBIE = 24 FLUX2 = 25 LONGCAT_IMAGE = 26 + COGVIDEOX = 27 @@ -1235,6 +1275,10 @@ class TEModel(Enum): QWEN35_4B = 25 QWEN35_9B = 26 QWEN35_27B = 27 + MINISTRAL_3_3B = 28 + GEMMA_4_E4B = 29 + GEMMA_4_E2B = 30 + GEMMA_4_31B = 31 def detect_te_model(sd): @@ -1260,6 +1304,12 @@ def detect_te_model(sd): return TEModel.BYT5_SMALL_GLYPH return TEModel.T5_BASE if 'model.layers.0.post_feedforward_layernorm.weight' in sd: + if 'model.layers.59.self_attn.q_norm.weight' in sd: + return TEModel.GEMMA_4_31B + if 'model.layers.41.self_attn.q_norm.weight' in sd and 'model.layers.47.self_attn.q_norm.weight' not in sd: + return TEModel.GEMMA_4_E4B + if 'model.layers.34.self_attn.q_norm.weight' in sd and 'model.layers.41.self_attn.q_norm.weight' not in sd: + return TEModel.GEMMA_4_E2B if 'model.layers.47.self_attn.q_norm.weight' in sd: return TEModel.GEMMA_3_12B if 'model.layers.0.self_attn.q_norm.weight' in sd: @@ -1301,6 +1351,8 @@ def detect_te_model(sd): return TEModel.MISTRAL3_24B else: return TEModel.MISTRAL3_24B_PRUNED_FLUX2 + if weight.shape[0] == 3072: + return TEModel.MINISTRAL_3_3B return TEModel.LLAMA3_8 return None @@ -1380,6 +1432,9 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip clip_target.clip = comfy.text_encoders.hidream.hidream_clip(**t5xxl_detect(clip_data), clip_l=False, clip_g=False, t5=True, llama=False, dtype_llama=None) clip_target.tokenizer = comfy.text_encoders.hidream.HiDreamTokenizer + elif clip_type == CLIPType.COGVIDEOX: + clip_target.clip = comfy.text_encoders.cogvideo.cogvideo_te(**t5xxl_detect(clip_data)) + clip_target.tokenizer = comfy.text_encoders.cogvideo.CogVideoXTokenizer else: #CLIPType.MOCHI clip_target.clip = comfy.text_encoders.genmo.mochi_te(**t5xxl_detect(clip_data)) clip_target.tokenizer = comfy.text_encoders.genmo.MochiT5Tokenizer @@ -1397,6 +1452,13 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip else: clip_target.clip = comfy.text_encoders.sa_t5.SAT5Model clip_target.tokenizer = comfy.text_encoders.sa_t5.SAT5Tokenizer + elif te_model in (TEModel.GEMMA_4_E4B, TEModel.GEMMA_4_E2B, TEModel.GEMMA_4_31B): + variant = {TEModel.GEMMA_4_E4B: comfy.text_encoders.gemma4.Gemma4_E4B, + TEModel.GEMMA_4_E2B: comfy.text_encoders.gemma4.Gemma4_E2B, + TEModel.GEMMA_4_31B: comfy.text_encoders.gemma4.Gemma4_31B}[te_model] + clip_target.clip = comfy.text_encoders.gemma4.gemma4_te(**llama_detect(clip_data), model_class=variant) + clip_target.tokenizer = variant.tokenizer + tokenizer_data["tokenizer_json"] = clip_data[0].get("tokenizer_json", None) elif te_model == TEModel.GEMMA_2_2B: clip_target.clip = comfy.text_encoders.lumina2.te(**llama_detect(clip_data)) clip_target.tokenizer = comfy.text_encoders.lumina2.LuminaTokenizer @@ -1458,6 +1520,10 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip elif te_model == TEModel.QWEN3_06B: clip_target.clip = comfy.text_encoders.anima.te(**llama_detect(clip_data)) clip_target.tokenizer = comfy.text_encoders.anima.AnimaTokenizer + elif te_model == TEModel.MINISTRAL_3_3B: + clip_target.clip = comfy.text_encoders.ernie.te(**llama_detect(clip_data)) + clip_target.tokenizer = comfy.text_encoders.ernie.ErnieTokenizer + tokenizer_data["tekken_model"] = clip_data[0].get("tekken_model", None) else: # clip_l if clip_type == CLIPType.SD3: diff --git a/comfy/supported_models.py b/comfy/supported_models.py index 9a5612716..1e4434fd5 100644 --- a/comfy/supported_models.py +++ b/comfy/supported_models.py @@ -26,6 +26,9 @@ import comfy.text_encoders.z_image import comfy.text_encoders.anima import comfy.text_encoders.ace15 import comfy.text_encoders.longcat_image +import comfy.text_encoders.ernie +import comfy.text_encoders.cogvideo +import comfy.text_encoders.hidream_o1 from . import supported_models_base from . import latent_formats @@ -1165,6 +1168,25 @@ class WAN21_T2V(supported_models_base.BASE): t5_detect = comfy.text_encoders.sd3_clip.t5_xxl_detect(state_dict, "{}umt5xxl.transformer.".format(pref)) return supported_models_base.ClipTarget(comfy.text_encoders.wan.WanT5Tokenizer, comfy.text_encoders.wan.te(**t5_detect)) +class WAN21_CausalAR_T2V(WAN21_T2V): + unet_config = { + "image_model": "wan2.1", + "model_type": "t2v", + "causal_ar": True, + } + + sampling_settings = { + "shift": 5.0, + } + + def __init__(self, unet_config): + super().__init__(unet_config) + self.unet_config.pop("causal_ar", None) + + def get_model(self, state_dict, prefix="", device=None): + return model_base.WAN21_CausalAR(self, device=device) + + class WAN21_I2V(WAN21_T2V): unet_config = { "image_model": "wan2.1", @@ -1292,6 +1314,37 @@ class WAN21_SCAIL(WAN21_T2V): out = model_base.WAN21_SCAIL(self, image_to_video=False, device=device) return out +class WAN22_WanDancer(WAN21_T2V): + unet_config = { + "image_model": "wan2.1", + "model_type": "wandancer", + "in_dim": 36, + } + + def __init__(self, unet_config): + super().__init__(unet_config) + self.memory_usage_factor = 1.8 + + def get_model(self, state_dict, prefix="", device=None): + out = model_base.WAN22_WanDancer(self, image_to_video=True, device=device) + return out + + def process_unet_state_dict(self, state_dict): + out_sd = {} + for k in list(state_dict.keys()): + # split music_encoder in_proj into q_proj, k_proj, v_proj + if "music_encoder" in k and "self_attn.in_proj" in k: + suffix = "weight" if k.endswith("weight") else "bias" + tensor = state_dict[k] + d = tensor.shape[0] // 3 + prefix = k.replace(f"in_proj_{suffix}", "") + out_sd[f"{prefix}q_proj.{suffix}"] = tensor[:d] + out_sd[f"{prefix}k_proj.{suffix}"] = tensor[d:2*d] + out_sd[f"{prefix}v_proj.{suffix}"] = tensor[2*d:] + else: + out_sd[k] = state_dict[k] + return out_sd + class Hunyuan3Dv2(supported_models_base.BASE): unet_config = { "image_model": "hunyuan3d2", @@ -1379,6 +1432,50 @@ class HiDream(supported_models_base.BASE): def clip_target(self, state_dict={}): return None # TODO +class HiDreamO1(supported_models_base.BASE): + unet_config = { + "image_model": "hidream_o1", + } + + sampling_settings = { + "shift": 3.0, + "noise_scale": 8.0, + } + + latent_format = latent_formats.HiDreamO1Pixel + memory_usage_factor = 0.033 + # fp16 not supported: LM MLP down_proj activations fp16 overflow, causing NaNs + supported_inference_dtypes = [torch.bfloat16, torch.float32] + + vae_key_prefix = ["vae."] + text_encoder_key_prefix = ["text_encoders."] + + optimizations = {"fp8": False} + + def get_model(self, state_dict, prefix="", device=None): + return model_base.HiDreamO1(self, device=device) + + def process_unet_state_dict(self, state_dict): + # Drop unused Qwen3-VL deepstack merger weights; upstream discards them at inference. + for key in list(state_dict.keys()): + if "visual.deepstack_merger_list" in key: + del state_dict[key] + return state_dict + + def process_vae_state_dict(self, state_dict): + # Pixel-space model: inject sentinel so VAE construction picks PixelspaceConversionVAE. + return {"pixel_space_vae": torch.tensor(1.0)} + + def process_clip_state_dict(self, state_dict): + # Tokenizer-only TE: inject sentinel so load_state_dict_guess_config triggers CLIP init. + return {"_hidream_o1_te_sentinel": torch.zeros(1)} + + def clip_target(self, state_dict={}): + return supported_models_base.ClipTarget( + comfy.text_encoders.hidream_o1.HiDreamO1Tokenizer, + comfy.text_encoders.hidream_o1.HiDreamO1TE, + ) + class Chroma(supported_models_base.BASE): unet_config = { "image_model": "chroma", @@ -1749,6 +1846,240 @@ class RT_DETR_v4(supported_models_base.BASE): def clip_target(self, state_dict={}): return None -models = [LotusD, Stable_Zero123, SD15_instructpix2pix, SD15, SD20, SD21UnclipL, SD21UnclipH, SDXL_instructpix2pix, SDXLRefiner, SDXL, SSD1B, KOALA_700M, KOALA_1B, Segmind_Vega, SD_X4Upscaler, Stable_Cascade_C, Stable_Cascade_B, SV3D_u, SV3D_p, SD3, StableAudio, AuraFlow, PixArtAlpha, PixArtSigma, HunyuanDiT, HunyuanDiT1, FluxInpaint, Flux, LongCatImage, FluxSchnell, GenmoMochi, LTXV, LTXAV, HunyuanVideo15_SR_Distilled, HunyuanVideo15, HunyuanImage21Refiner, HunyuanImage21, HunyuanVideoSkyreelsI2V, HunyuanVideoI2V, HunyuanVideo, CosmosT2V, CosmosI2V, CosmosT2IPredict2, CosmosI2VPredict2, ZImagePixelSpace, ZImage, Lumina2, WAN22_T2V, WAN21_T2V, WAN21_I2V, WAN21_FunControl2V, WAN21_Vace, WAN21_Camera, WAN22_Camera, WAN22_S2V, WAN21_HuMo, WAN22_Animate, WAN21_FlowRVS, WAN21_SCAIL, Hunyuan3Dv2mini, Hunyuan3Dv2, Hunyuan3Dv2_1, HiDream, Chroma, ChromaRadiance, ACEStep, ACEStep15, Omnigen2, QwenImage, Flux2, Kandinsky5Image, Kandinsky5, Anima, RT_DETR_v4] -models += [SVD_img2vid] +class ErnieImage(supported_models_base.BASE): + unet_config = { + "image_model": "ernie", + } + + sampling_settings = { + "multiplier": 1000.0, + "shift": 3.0, + } + + memory_usage_factor = 10.0 + + unet_extra_config = {} + latent_format = latent_formats.Flux2 + + supported_inference_dtypes = [torch.bfloat16, torch.float32] + + vae_key_prefix = ["vae."] + text_encoder_key_prefix = ["text_encoders."] + + def get_model(self, state_dict, prefix="", device=None): + out = model_base.ErnieImage(self, device=device) + return out + + def clip_target(self, state_dict={}): + pref = self.text_encoder_key_prefix[0] + hunyuan_detect = comfy.text_encoders.hunyuan_video.llama_detect(state_dict, "{}ministral3_3b.transformer.".format(pref)) + return supported_models_base.ClipTarget(comfy.text_encoders.ernie.ErnieTokenizer, comfy.text_encoders.ernie.te(**hunyuan_detect)) + + +class SAM3(supported_models_base.BASE): + unet_config = {"image_model": "SAM3"} + supported_inference_dtypes = [torch.float16, torch.bfloat16, torch.float32] + text_encoder_key_prefix = ["detector.backbone.language_backbone."] + unet_extra_prefix = "" + + def process_clip_state_dict(self, state_dict): + clip_keys = getattr(self, "_clip_stash", {}) + clip_keys = utils.state_dict_prefix_replace(clip_keys, {"detector.backbone.language_backbone.": "", "backbone.language_backbone.": ""}, filter_keys=True) + clip_keys = utils.clip_text_transformers_convert(clip_keys, "encoder.", "sam3_clip.transformer.") + return {k: v for k, v in clip_keys.items() if not k.startswith("resizer.")} + + def process_unet_state_dict(self, state_dict): + self._clip_stash = {k: state_dict.pop(k) for k in list(state_dict.keys()) if "language_backbone" in k and "resizer" not in k} + # SAM3.1: remap tracker.model.* -> tracker.* + for k in list(state_dict.keys()): + if k.startswith("tracker.model."): + state_dict["tracker." + k[len("tracker.model."):]] = state_dict.pop(k) + # SAM3.1: remove per-block freqs_cis buffers (computed dynamically) + for k in [k for k in list(state_dict.keys()) if ".attn.freqs_cis" in k]: + state_dict.pop(k) + # Split fused QKV projections + for k in [k for k in list(state_dict.keys()) if k.endswith((".in_proj_weight", ".in_proj_bias"))]: + t = state_dict.pop(k) + base, suffix = k.rsplit(".in_proj_", 1) + s = ".weight" if suffix == "weight" else ".bias" + d = t.shape[0] // 3 + state_dict[base + ".q_proj" + s] = t[:d] + state_dict[base + ".k_proj" + s] = t[d:2*d] + state_dict[base + ".v_proj" + s] = t[2*d:] + # Remap tracker SAM decoder transformer key names to match sam.py TwoWayTransformer + for k in list(state_dict.keys()): + if "sam_mask_decoder.transformer." not in k: + continue + new_k = k.replace(".mlp.lin1.", ".mlp.0.").replace(".mlp.lin2.", ".mlp.2.").replace(".norm_final_attn.", ".norm_final.") + if new_k != k: + state_dict[new_k] = state_dict.pop(k) + return state_dict + + def get_model(self, state_dict, prefix="", device=None): + return model_base.SAM3(self, device=device) + + def clip_target(self, state_dict={}): + import comfy.text_encoders.sam3_clip + return supported_models_base.ClipTarget(comfy.text_encoders.sam3_clip.SAM3TokenizerWrapper, comfy.text_encoders.sam3_clip.SAM3ClipModelWrapper) + + +class SAM31(SAM3): + unet_config = {"image_model": "SAM31"} + + +class CogVideoX_T2V(supported_models_base.BASE): + unet_config = { + "image_model": "cogvideox", + } + + sampling_settings = { + "linear_start": 0.00085, + "linear_end": 0.012, + "beta_schedule": "linear", + "zsnr": True, + } + + unet_extra_config = {} + latent_format = latent_formats.CogVideoX + + supported_inference_dtypes = [torch.bfloat16, torch.float16, torch.float32] + + vae_key_prefix = ["vae."] + text_encoder_key_prefix = ["text_encoders."] + + def __init__(self, unet_config): + # 2b-class (dim=1920, heads=30) uses scale_factor=1.15258426. + # 5b-class (dim=3072, heads=48) — incl. CogVideoX-5b, 1.5-5B, and + # Fun-V1.5 inpainting — uses scale_factor=0.7 per vae/config.json. + if unet_config.get("num_attention_heads", 0) >= 48: + self.latent_format = latent_formats.CogVideoX1_5 + super().__init__(unet_config) + + def get_model(self, state_dict, prefix="", device=None): + # CogVideoX 1.5 (patch_size_t=2) has different training base dimensions for RoPE + if self.unet_config.get("patch_size_t") is not None: + self.unet_config.setdefault("sample_height", 96) + self.unet_config.setdefault("sample_width", 170) + self.unet_config.setdefault("sample_frames", 81) + out = model_base.CogVideoX(self, device=device) + return out + + def clip_target(self, state_dict={}): + return supported_models_base.ClipTarget(comfy.text_encoders.cogvideo.CogVideoXT5Tokenizer, comfy.text_encoders.sd3_clip.T5XXLModel) + +class CogVideoX_I2V(CogVideoX_T2V): + unet_config = { + "image_model": "cogvideox", + "in_channels": 32, + } + + def get_model(self, state_dict, prefix="", device=None): + if self.unet_config.get("patch_size_t") is not None: + self.unet_config.setdefault("sample_height", 96) + self.unet_config.setdefault("sample_width", 170) + self.unet_config.setdefault("sample_frames", 81) + out = model_base.CogVideoX(self, image_to_video=True, device=device) + return out + +class CogVideoX_Inpaint(CogVideoX_T2V): + unet_config = { + "image_model": "cogvideox", + "in_channels": 48, + } + + def get_model(self, state_dict, prefix="", device=None): + if self.unet_config.get("patch_size_t") is not None: + self.unet_config.setdefault("sample_height", 96) + self.unet_config.setdefault("sample_width", 170) + self.unet_config.setdefault("sample_frames", 81) + out = model_base.CogVideoX(self, image_to_video=True, device=device) + return out + + +models = [ + LotusD, + Stable_Zero123, + SD15_instructpix2pix, + SD15, + SD20, + SD21UnclipL, + SD21UnclipH, + SDXL_instructpix2pix, + SDXLRefiner, + SDXL, + SSD1B, + KOALA_700M, + KOALA_1B, + Segmind_Vega, + SD_X4Upscaler, + Stable_Cascade_C, + Stable_Cascade_B, + SV3D_u, + SV3D_p, + SD3, + StableAudio, + AuraFlow, + PixArtAlpha, + PixArtSigma, + HunyuanDiT, + HunyuanDiT1, + FluxInpaint, + Flux, + LongCatImage, + FluxSchnell, + GenmoMochi, + LTXV, + LTXAV, + HunyuanVideo15_SR_Distilled, + HunyuanVideo15, + HunyuanImage21Refiner, + HunyuanImage21, + HunyuanVideoSkyreelsI2V, + HunyuanVideoI2V, + HunyuanVideo, + CosmosT2V, + CosmosI2V, + CosmosT2IPredict2, + CosmosI2VPredict2, + ZImagePixelSpace, + ZImage, + Lumina2, + WAN22_T2V, + WAN21_CausalAR_T2V, + WAN21_T2V, + WAN21_I2V, + WAN21_FunControl2V, + WAN21_Vace, + WAN21_Camera, + WAN22_Camera, + WAN22_S2V, + WAN21_HuMo, + WAN22_Animate, + WAN21_FlowRVS, + WAN21_SCAIL, + WAN22_WanDancer, + Hunyuan3Dv2mini, + Hunyuan3Dv2, + Hunyuan3Dv2_1, + HiDream, + HiDreamO1, + Chroma, + ChromaRadiance, + ACEStep, + ACEStep15, + Omnigen2, + QwenImage, + Flux2, + Kandinsky5Image, + Kandinsky5, + Anima, + RT_DETR_v4, + ErnieImage, + SAM3, + SAM31, + CogVideoX_Inpaint, + CogVideoX_I2V, + CogVideoX_T2V, + SVD_img2vid, +] diff --git a/comfy/taesd/taehv.py b/comfy/taesd/taehv.py index 6c06ce19d..696013200 100644 --- a/comfy/taesd/taehv.py +++ b/comfy/taesd/taehv.py @@ -7,6 +7,7 @@ from tqdm.auto import tqdm from collections import namedtuple, deque import comfy.ops +import comfy.model_management operations=comfy.ops.disable_weight_init DecoderResult = namedtuple("DecoderResult", ("frame", "memory")) @@ -47,11 +48,14 @@ class TGrow(nn.Module): x = self.conv(x) return x.reshape(-1, C, H, W) -def apply_model_with_memblocks(model, x, parallel, show_progress_bar): +def apply_model_with_memblocks(model, x, parallel, show_progress_bar, output_device=None, + patch_size=1, decode=False): B, T, C, H, W = x.shape if parallel: x = x.reshape(B*T, C, H, W) + if not decode and patch_size > 1: + x = F.pixel_unshuffle(x, patch_size) # parallel over input timesteps, iterate over blocks for b in tqdm(model, disable=not show_progress_bar): if isinstance(b, MemBlock): @@ -62,20 +66,27 @@ def apply_model_with_memblocks(model, x, parallel, show_progress_bar): x = b(x, mem) else: x = b(x) - BT, C, H, W = x.shape - T = BT // B - x = x.view(B, T, C, H, W) + if decode and patch_size > 1: + x = F.pixel_shuffle(x, patch_size) + x = x.view(B, x.shape[0] // B, *x.shape[1:]) + x = x.to(output_device) else: out = [] - work_queue = deque([TWorkItem(xt, 0) for t, xt in enumerate(x.reshape(B, T * C, H, W).chunk(T, dim=1))]) + # Chunk along the time dim directly (chunks are [B,1,C,H,W] views, squeeze to [B,C,H,W] views). + # Avoids forcing a contiguous copy when x is non-contiguous (e.g. after movedim in encode/decode). + work_queue = deque([TWorkItem(xt.squeeze(1), 0) for xt in x.chunk(T, dim=1)]) progress_bar = tqdm(range(T), disable=not show_progress_bar) mem = [None] * len(model) while work_queue: xt, i = work_queue.popleft() if i == 0: progress_bar.update(1) + if not decode and patch_size > 1: + xt = F.pixel_unshuffle(xt, patch_size) if i == len(model): - out.append(xt) + if decode and patch_size > 1: + xt = F.pixel_shuffle(xt, patch_size) + out.append(xt.to(output_device)) del xt else: b = model[i] @@ -165,24 +176,20 @@ class TAEHV(nn.Module): def encode(self, x, **kwargs): x = x.movedim(2, 1) # [B, C, T, H, W] -> [B, T, C, H, W] - if self.patch_size > 1: - B, T, C, H, W = x.shape - x = x.reshape(B * T, C, H, W) - x = F.pixel_unshuffle(x, self.patch_size) - x = x.reshape(B, T, C * self.patch_size ** 2, H // self.patch_size, W // self.patch_size) if x.shape[1] % self.t_downscale != 0: # pad at end to multiple of t_downscale n_pad = self.t_downscale - x.shape[1] % self.t_downscale padding = x[:, -1:].repeat_interleave(n_pad, dim=1) x = torch.cat([x, padding], 1) - x = apply_model_with_memblocks(self.encoder, x, self.parallel, self.show_progress_bar).movedim(2, 1) + x = apply_model_with_memblocks(self.encoder, x, self.parallel, self.show_progress_bar, + patch_size=self.patch_size).movedim(2, 1) return self.process_out(x) def decode(self, x, **kwargs): x = x.unsqueeze(0) if x.ndim == 4 else x # [T, C, H, W] -> [1, T, C, H, W] x = x.movedim(1, 2) if x.shape[1] != self.latent_channels else x # [B, T, C, H, W] or [B, C, T, H, W] x = self.process_in(x).movedim(2, 1) # [B, C, T, H, W] -> [B, T, C, H, W] - x = apply_model_with_memblocks(self.decoder, x, self.parallel, self.show_progress_bar) - if self.patch_size > 1: - x = F.pixel_shuffle(x, self.patch_size) + x = apply_model_with_memblocks(self.decoder, x, self.parallel, self.show_progress_bar, + output_device=comfy.model_management.intermediate_device(), + patch_size=self.patch_size, decode=True) return x[:, self.frames_to_trim:].movedim(2, 1) diff --git a/comfy/taesd/taesd.py b/comfy/taesd/taesd.py index ce36f1a84..05d370209 100644 --- a/comfy/taesd/taesd.py +++ b/comfy/taesd/taesd.py @@ -17,32 +17,79 @@ class Clamp(nn.Module): return torch.tanh(x / 3) * 3 class Block(nn.Module): - def __init__(self, n_in, n_out): + def __init__(self, n_in: int, n_out: int, use_midblock_gn: bool = False): super().__init__() self.conv = nn.Sequential(conv(n_in, n_out), nn.ReLU(), conv(n_out, n_out), nn.ReLU(), conv(n_out, n_out)) self.skip = comfy.ops.disable_weight_init.Conv2d(n_in, n_out, 1, bias=False) if n_in != n_out else nn.Identity() self.fuse = nn.ReLU() - def forward(self, x): + if not use_midblock_gn: + self.pool = None + return + n_gn = n_in * 4 + self.pool = nn.Sequential( + comfy.ops.disable_weight_init.Conv2d(n_in, n_gn, 1, bias=False), + comfy.ops.disable_weight_init.GroupNorm(4, n_gn), + nn.ReLU(inplace=True), + comfy.ops.disable_weight_init.Conv2d(n_gn, n_in, 1, bias=False), + ) + + def forward(self, x: torch.Tensor) -> torch.Tensor: + if self.pool is not None: + x = x + self.pool(x) return self.fuse(self.conv(x) + self.skip(x)) -def Encoder(latent_channels=4): - return nn.Sequential( - conv(3, 64), Block(64, 64), - conv(64, 64, stride=2, bias=False), Block(64, 64), Block(64, 64), Block(64, 64), - conv(64, 64, stride=2, bias=False), Block(64, 64), Block(64, 64), Block(64, 64), - conv(64, 64, stride=2, bias=False), Block(64, 64), Block(64, 64), Block(64, 64), - conv(64, latent_channels), - ) +class Encoder(nn.Sequential): + def __init__(self, latent_channels: int = 4, use_gn: bool = False): + super().__init__( + conv(3, 64), Block(64, 64), + conv(64, 64, stride=2, bias=False), Block(64, 64), Block(64, 64), Block(64, 64), + conv(64, 64, stride=2, bias=False), Block(64, 64), Block(64, 64), Block(64, 64), + conv(64, 64, stride=2, bias=False), Block(64, 64, use_gn), Block(64, 64, use_gn), Block(64, 64, use_gn), + conv(64, latent_channels), + ) +class Decoder(nn.Sequential): + def __init__(self, latent_channels: int = 4, use_gn: bool = False): + super().__init__( + Clamp(), conv(latent_channels, 64), nn.ReLU(), + Block(64, 64, use_gn), Block(64, 64, use_gn), Block(64, 64, use_gn), nn.Upsample(scale_factor=2), conv(64, 64, bias=False), + Block(64, 64), Block(64, 64), Block(64, 64), nn.Upsample(scale_factor=2), conv(64, 64, bias=False), + Block(64, 64), Block(64, 64), Block(64, 64), nn.Upsample(scale_factor=2), conv(64, 64, bias=False), + Block(64, 64), conv(64, 3), + ) + +class DecoderFlux2(Decoder): + def __init__(self, latent_channels: int = 128, use_gn: bool = True): + if latent_channels != 128 or not use_gn: + raise ValueError("Unexpected parameters for Flux2 TAE module") + super().__init__(latent_channels=32, use_gn=True) + + def forward(self, x: torch.Tensor) -> torch.Tensor: + B, C, H, W = x.shape + x = ( + x + .reshape(B, 32, 2, 2, H, W) + .permute(0, 1, 4, 2, 5, 3) + .reshape(B, 32, H * 2, W * 2) + ) + return super().forward(x) + +class EncoderFlux2(Encoder): + def __init__(self, latent_channels: int = 128, use_gn: bool = True): + if latent_channels != 128 or not use_gn: + raise ValueError("Unexpected parameters for Flux2 TAE module") + super().__init__(latent_channels=32, use_gn=True) + + def forward(self, x: torch.Tensor) -> torch.Tensor: + result = super().forward(x) + B, C, H, W = result.shape + return ( + result + .reshape(B, C, H // 2, 2, W // 2, 2) + .permute(0, 1, 3, 5, 2, 4) + .reshape(B, 128, H // 2, W // 2) + ) -def Decoder(latent_channels=4): - return nn.Sequential( - Clamp(), conv(latent_channels, 64), nn.ReLU(), - Block(64, 64), Block(64, 64), Block(64, 64), nn.Upsample(scale_factor=2), conv(64, 64, bias=False), - Block(64, 64), Block(64, 64), Block(64, 64), nn.Upsample(scale_factor=2), conv(64, 64, bias=False), - Block(64, 64), Block(64, 64), Block(64, 64), nn.Upsample(scale_factor=2), conv(64, 64, bias=False), - Block(64, 64), conv(64, 3), - ) class TAESD(nn.Module): latent_magnitude = 3 @@ -51,8 +98,15 @@ class TAESD(nn.Module): def __init__(self, encoder_path=None, decoder_path=None, latent_channels=4): """Initialize pretrained TAESD on the given device from the given checkpoints.""" super().__init__() - self.taesd_encoder = Encoder(latent_channels=latent_channels) - self.taesd_decoder = Decoder(latent_channels=latent_channels) + if latent_channels == 128: + encoder_class = EncoderFlux2 + decoder_class = DecoderFlux2 + else: + encoder_class = Encoder + decoder_class = Decoder + self.taesd_encoder = encoder_class(latent_channels=latent_channels) + self.taesd_decoder = decoder_class(latent_channels=latent_channels) + self.vae_scale = torch.nn.Parameter(torch.tensor(1.0)) self.vae_shift = torch.nn.Parameter(torch.tensor(0.0)) if encoder_path is not None: @@ -61,19 +115,19 @@ class TAESD(nn.Module): self.taesd_decoder.load_state_dict(comfy.utils.load_torch_file(decoder_path, safe_load=True)) @staticmethod - def scale_latents(x): + def scale_latents(x: torch.Tensor) -> torch.Tensor: """raw latents -> [0, 1]""" return x.div(2 * TAESD.latent_magnitude).add(TAESD.latent_shift).clamp(0, 1) @staticmethod - def unscale_latents(x): + def unscale_latents(x: torch.Tensor) -> torch.Tensor: """[0, 1] -> raw latents""" return x.sub(TAESD.latent_shift).mul(2 * TAESD.latent_magnitude) - def decode(self, x): + def decode(self, x: torch.Tensor) -> torch.Tensor: x_sample = self.taesd_decoder((x - self.vae_shift) * self.vae_scale) x_sample = x_sample.sub(0.5).mul(2) return x_sample - def encode(self, x): + def encode(self, x: torch.Tensor) -> torch.Tensor: return (self.taesd_encoder(x * 0.5 + 0.5) / self.vae_scale) + self.vae_shift diff --git a/comfy/text_encoders/cogvideo.py b/comfy/text_encoders/cogvideo.py new file mode 100644 index 000000000..b97310709 --- /dev/null +++ b/comfy/text_encoders/cogvideo.py @@ -0,0 +1,48 @@ +import comfy.text_encoders.sd3_clip +from comfy import sd1_clip + + +class CogVideoXT5Tokenizer(comfy.text_encoders.sd3_clip.T5XXLTokenizer): + """Inner T5 tokenizer for CogVideoX. + + CogVideoX was trained with T5 embeddings padded to 226 tokens (not 77 like SD3). + Used both directly by supported_models.CogVideoX_T2V.clip_target (paired with + the raw T5XXLModel) and by the CogVideoXTokenizer outer wrapper below. + """ + def __init__(self, embedding_directory=None, tokenizer_data={}): + super().__init__(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data, min_length=226) + + +class CogVideoXTokenizer(sd1_clip.SD1Tokenizer): + """Outer tokenizer wrapper for CLIPLoader (type="cogvideox").""" + def __init__(self, embedding_directory=None, tokenizer_data={}): + super().__init__(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data, + clip_name="t5xxl", tokenizer=CogVideoXT5Tokenizer) + + +class CogVideoXT5XXL(sd1_clip.SD1ClipModel): + """Outer T5XXL model wrapper for CLIPLoader (type="cogvideox"). + + Wraps the raw T5XXL model in the SD1ClipModel interface so that CLIP.__init__ + (which reads self.dtypes) works correctly. The inner model is the standard + sd3_clip.T5XXLModel (no attention_mask change needed for CogVideoX). + """ + def __init__(self, device="cpu", dtype=None, model_options={}): + super().__init__(device=device, dtype=dtype, name="t5xxl", + clip_model=comfy.text_encoders.sd3_clip.T5XXLModel, + model_options=model_options) + + +def cogvideo_te(dtype_t5=None, t5_quantization_metadata=None): + """Factory that returns a CogVideoXT5XXL class configured with the detected + T5 dtype and optional quantization metadata, for use in load_text_encoder_state_dicts. + """ + class CogVideoXTEModel_(CogVideoXT5XXL): + def __init__(self, device="cpu", dtype=None, model_options={}): + if t5_quantization_metadata is not None: + model_options = model_options.copy() + model_options["t5xxl_quantization_metadata"] = t5_quantization_metadata + if dtype_t5 is not None: + dtype = dtype_t5 + super().__init__(device=device, dtype=dtype, model_options=model_options) + return CogVideoXTEModel_ diff --git a/comfy/text_encoders/ernie.py b/comfy/text_encoders/ernie.py new file mode 100644 index 000000000..46d24d222 --- /dev/null +++ b/comfy/text_encoders/ernie.py @@ -0,0 +1,38 @@ +from .flux import Mistral3Tokenizer +from comfy import sd1_clip +import comfy.text_encoders.llama + +class Ministral3_3BTokenizer(Mistral3Tokenizer): + def __init__(self, embedding_directory=None, embedding_size=5120, embedding_key='ministral3_3b', tokenizer_data={}): + return super().__init__(embedding_directory=embedding_directory, embedding_size=embedding_size, embedding_key=embedding_key, tokenizer_data=tokenizer_data) + +class ErnieTokenizer(sd1_clip.SD1Tokenizer): + def __init__(self, embedding_directory=None, tokenizer_data={}): + super().__init__(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data, name="ministral3_3b", tokenizer=Mistral3Tokenizer) + + def tokenize_with_weights(self, text, return_word_ids=False, llama_template=None, **kwargs): + tokens = super().tokenize_with_weights(text, return_word_ids=return_word_ids, disable_weights=True, **kwargs) + return tokens + + +class Ministral3_3BModel(sd1_clip.SDClipModel): + def __init__(self, device="cpu", layer="hidden", layer_idx=-2, dtype=None, attention_mask=True, model_options={}): + textmodel_json_config = {} + super().__init__(device=device, layer=layer, layer_idx=layer_idx, textmodel_json_config=textmodel_json_config, dtype=dtype, special_tokens={"start": 1, "pad": 0}, layer_norm_hidden_state=False, model_class=comfy.text_encoders.llama.Ministral3_3B, enable_attention_masks=attention_mask, return_attention_masks=attention_mask, model_options=model_options) + + +class ErnieTEModel(sd1_clip.SD1ClipModel): + def __init__(self, device="cpu", dtype=None, model_options={}, name="ministral3_3b", clip_model=Ministral3_3BModel): + super().__init__(device=device, dtype=dtype, name=name, clip_model=clip_model, model_options=model_options) + + +def te(dtype_llama=None, llama_quantization_metadata=None): + class ErnieTEModel_(ErnieTEModel): + def __init__(self, device="cpu", dtype=None, model_options={}): + if dtype_llama is not None: + dtype = dtype_llama + if llama_quantization_metadata is not None: + model_options = model_options.copy() + model_options["quantization_metadata"] = llama_quantization_metadata + super().__init__(device=device, dtype=dtype, model_options=model_options) + return ErnieTEModel_ diff --git a/comfy/text_encoders/flux.py b/comfy/text_encoders/flux.py index 1ae398789..d5eb91dcb 100644 --- a/comfy/text_encoders/flux.py +++ b/comfy/text_encoders/flux.py @@ -116,9 +116,9 @@ class MistralTokenizerClass: return LlamaTokenizerFast(**kwargs) class Mistral3Tokenizer(sd1_clip.SDTokenizer): - def __init__(self, embedding_directory=None, tokenizer_data={}): + def __init__(self, embedding_directory=None, embedding_size=5120, embedding_key='mistral3_24b', tokenizer_data={}): self.tekken_data = tokenizer_data.get("tekken_model", None) - super().__init__("", pad_with_end=False, embedding_directory=embedding_directory, embedding_size=5120, embedding_key='mistral3_24b', tokenizer_class=MistralTokenizerClass, has_end_token=False, pad_to_max_length=False, pad_token=11, start_token=1, max_length=99999999, min_length=1, pad_left=True, tokenizer_args=load_mistral_tokenizer(self.tekken_data), tokenizer_data=tokenizer_data) + super().__init__("", pad_with_end=False, embedding_directory=embedding_directory, embedding_size=embedding_size, embedding_key=embedding_key, tokenizer_class=MistralTokenizerClass, has_end_token=False, pad_to_max_length=False, pad_token=11, start_token=1, max_length=99999999, min_length=1, pad_left=True, disable_weights=True, tokenizer_args=load_mistral_tokenizer(self.tekken_data), tokenizer_data=tokenizer_data) def state_dict(self): return {"tekken_model": self.tekken_data} diff --git a/comfy/text_encoders/gemma4.py b/comfy/text_encoders/gemma4.py new file mode 100644 index 000000000..f050061ed --- /dev/null +++ b/comfy/text_encoders/gemma4.py @@ -0,0 +1,1298 @@ +import torch +import torch.nn as nn +import numpy as np +from dataclasses import dataclass +import math + +from comfy import sd1_clip +import comfy.model_management +from comfy.ldm.modules.attention import optimized_attention_for_device +from comfy.rmsnorm import rms_norm +from comfy.text_encoders.llama import RMSNorm, MLP, BaseLlama, BaseGenerate, _make_scaled_embedding + + +# Intentional minor divergences from transformers -reference implementation: +# - Embedding sqrt(hidden_size) scale applied as a Python scalar (full precision) instead of dtype-matched buffer tensor. +# - RMSNorm uses torch fused F.rms_norm, very slight numerical differences, but considerably faster +# - Input image and audio resizing/resampling slightly different numerically + + +GEMMA4_VISION_CONFIG = {"hidden_size": 768, "image_size": 896, "intermediate_size": 3072, "num_attention_heads": 12, "num_hidden_layers": 16, "patch_size": 16, "head_dim": 64, "rms_norm_eps": 1e-6, "position_embedding_size": 10240, "pooling_kernel_size": 3} +GEMMA4_VISION_31B_CONFIG = {"hidden_size": 1152, "image_size": 896, "intermediate_size": 4304, "num_attention_heads": 16, "num_hidden_layers": 27, "patch_size": 16, "head_dim": 72, "rms_norm_eps": 1e-6, "position_embedding_size": 10240, "pooling_kernel_size": 3} +GEMMA4_AUDIO_CONFIG = {"hidden_size": 1024, "num_hidden_layers": 12, "num_attention_heads": 8, "intermediate_size": 4096, "conv_kernel_size": 5, "attention_chunk_size": 12, "attention_context_left": 13, "attention_context_right": 0, "attention_logit_cap": 50.0, "output_proj_dims": 1536, "rms_norm_eps": 1e-6, "residual_weight": 0.5} + +@dataclass +class Gemma4Config: + vocab_size: int = 262144 + hidden_size: int = 2560 + intermediate_size: int = 10240 + num_hidden_layers: int = 42 + num_attention_heads: int = 8 + num_key_value_heads: int = 2 + max_position_embeddings: int = 131072 + rms_norm_eps: float = 1e-6 + rope_theta = [1000000.0, 10000.0] + transformer_type: str = "gemma4" + head_dim = 256 + global_head_dim = 512 + rms_norm_add = False + mlp_activation = "gelu_pytorch_tanh" + qkv_bias = False + rope_dims = None + q_norm = "gemma3" + k_norm = "gemma3" + sliding_attention = [512, 512, 512, 512, 512, False] + rope_scale = None + partial_rotary_factor: float = 0.25 + final_norm: bool = True + lm_head: bool = False + final_logit_softcapping: float = 30.0 + hidden_size_per_layer_input: int = 256 + num_kv_shared_layers: int = 18 + use_double_wide_mlp: bool = False + stop_tokens = [1, 50, 106] + vision_config = GEMMA4_VISION_CONFIG + audio_config = GEMMA4_AUDIO_CONFIG + mm_tokens_per_image = 280 + +@dataclass +class Gemma4_E2B_Config(Gemma4Config): + hidden_size: int = 1536 + intermediate_size: int = 6144 + num_hidden_layers: int = 35 + num_key_value_heads: int = 1 + sliding_attention = [512, 512, 512, 512, False] + num_kv_shared_layers: int = 20 + use_double_wide_mlp: bool = True + +@dataclass +class Gemma4_31B_Config(Gemma4Config): + hidden_size: int = 5376 + intermediate_size: int = 21504 + num_hidden_layers: int = 60 + num_attention_heads: int = 32 + num_key_value_heads: int = 16 + sliding_attention = [1024, 1024, 1024, 1024, 1024, False] + hidden_size_per_layer_input: int = 0 + num_kv_shared_layers: int = 0 + audio_config = None + vision_config = GEMMA4_VISION_31B_CONFIG + + +# unfused RoPE as addcmul_ RoPE diverges from reference code +def _apply_rotary_pos_emb(x, freqs_cis): + cos, sin = freqs_cis[0], freqs_cis[1] + half = x.shape[-1] // 2 + out = x * cos + out[..., :half] -= x[..., half:] * sin[..., :half] + out[..., half:] += x[..., :half] * sin[..., half:] + return out + +class Gemma4Attention(nn.Module): + def __init__(self, config, head_dim, device=None, dtype=None, ops=None): + super().__init__() + self.num_heads = config.num_attention_heads + self.num_kv_heads = config.num_key_value_heads + self.hidden_size = config.hidden_size + self.head_dim = head_dim + self.inner_size = self.num_heads * head_dim + + self.q_proj = ops.Linear(config.hidden_size, self.inner_size, bias=config.qkv_bias, device=device, dtype=dtype) + self.k_proj = ops.Linear(config.hidden_size, self.num_kv_heads * head_dim, bias=config.qkv_bias, device=device, dtype=dtype) + self.v_proj = ops.Linear(config.hidden_size, self.num_kv_heads * head_dim, bias=config.qkv_bias, device=device, dtype=dtype) + self.o_proj = ops.Linear(self.inner_size, config.hidden_size, bias=False, device=device, dtype=dtype) + + self.q_norm = None + self.k_norm = None + if config.q_norm == "gemma3": + self.q_norm = RMSNorm(head_dim, eps=config.rms_norm_eps, device=device, dtype=dtype) + if config.k_norm == "gemma3": + self.k_norm = RMSNorm(head_dim, eps=config.rms_norm_eps, device=device, dtype=dtype) + + def forward( + self, + hidden_states: torch.Tensor, + attention_mask=None, + freqs_cis=None, + past_key_value=None, + sliding_window=None, + shared_kv=None, + ): + batch_size, seq_length, _ = hidden_states.shape + + xq = self.q_proj(hidden_states) + xq = xq.view(batch_size, seq_length, self.num_heads, self.head_dim).transpose(1, 2) + if self.q_norm is not None: + xq = self.q_norm(xq) + + if shared_kv is not None: + xk, xv = shared_kv + # Apply RoPE to Q only (K already has RoPE from source layer) + xq = _apply_rotary_pos_emb(xq, freqs_cis) + present_key_value = None + shareable_kv = None + else: + xk = self.k_proj(hidden_states).view(batch_size, seq_length, self.num_kv_heads, self.head_dim) + xv = self.v_proj(hidden_states).view(batch_size, seq_length, self.num_kv_heads, self.head_dim) + if self.k_norm is not None: + xk = self.k_norm(xk) + xv = rms_norm(xv) + xk = xk.transpose(1, 2) + xv = xv.transpose(1, 2) + xq = _apply_rotary_pos_emb(xq, freqs_cis) + xk = _apply_rotary_pos_emb(xk, freqs_cis) + + present_key_value = None + if past_key_value is not None: + cumulative_len = 0 + if len(past_key_value) > 0: + past_key, past_value, cumulative_len = past_key_value + xk = torch.cat((past_key, xk), dim=2) + xv = torch.cat((past_value, xv), dim=2) + new_cumulative = cumulative_len + seq_length + if sliding_window is not None and xk.shape[2] > sliding_window - 1: + cache_k = xk[:, :, -(sliding_window - 1):] + cache_v = xv[:, :, -(sliding_window - 1):] + else: + cache_k = xk + cache_v = xv + present_key_value = (cache_k, cache_v, new_cumulative) + + # KV for sharing: full xk/xv that SDPA sees (not evicted cache) + shareable_kv = (xk, xv) + + # GQA: pass unexpanded KV with enable_gqa when no sliding mask, + # expand heads when sliding mask is present + # has to be done within SDPA itself to match the reference code, pre-scaling expansion causes numerical differences + expand_kv = (self.num_heads != self.num_kv_heads and + sliding_window is not None and + xk.shape[2] >= sliding_window) + if expand_kv: + xk = xk.repeat_interleave(self.num_heads // self.num_kv_heads, dim=1) + xv = xv.repeat_interleave(self.num_heads // self.num_kv_heads, dim=1) + gqa_kwargs = {} if expand_kv else ({"enable_gqa": True} if self.num_heads != self.num_kv_heads else {}) + output = optimized_attention_for_device(xq.device, mask=attention_mask is not None, small_input=True)(xq, xk, xv, self.num_heads, mask=attention_mask, skip_reshape=True, scale=1.0, **gqa_kwargs) + + return self.o_proj(output), present_key_value, shareable_kv + + +class TransformerBlockGemma4(nn.Module): + def __init__(self, config, index, device=None, dtype=None, ops=None): + super().__init__() + if config.sliding_attention is not None: + self.sliding_attention = config.sliding_attention[index % len(config.sliding_attention)] + else: + self.sliding_attention = False + + head_dim = config.head_dim if self.sliding_attention else config.global_head_dim + + self.self_attn = Gemma4Attention(config, head_dim=head_dim, device=device, dtype=dtype, ops=ops) + + num_kv_shared = config.num_kv_shared_layers + first_kv_shared = config.num_hidden_layers - num_kv_shared + mlp_size = config.intermediate_size * 2 if config.use_double_wide_mlp and index >= first_kv_shared else None + self.mlp = MLP(config, device=device, dtype=dtype, ops=ops, intermediate_size=mlp_size) + + self.input_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, device=device, dtype=dtype) + self.post_attention_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, device=device, dtype=dtype) + self.pre_feedforward_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, device=device, dtype=dtype) + self.post_feedforward_layernorm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, device=device, dtype=dtype) + + self.hidden_size_per_layer_input = config.hidden_size_per_layer_input + if self.hidden_size_per_layer_input: + self.per_layer_input_gate = ops.Linear(config.hidden_size, self.hidden_size_per_layer_input, bias=False, device=device, dtype=dtype) + self.per_layer_projection = ops.Linear(self.hidden_size_per_layer_input, config.hidden_size, bias=False, device=device, dtype=dtype) + self.post_per_layer_input_norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, device=device, dtype=dtype) + self.register_buffer("layer_scalar", torch.ones(1, device=device, dtype=dtype)) + else: + self.layer_scalar = None + + def forward(self, x, attention_mask=None, freqs_cis=None, past_key_value=None, per_layer_input=None, shared_kv=None): + sliding_window = None + if self.sliding_attention: + sliding_window = self.sliding_attention + # For prefill > sliding window, add sliding window restriction to the causal mask. + if x.shape[1] > self.sliding_attention: + sw_mask = torch.zeros(x.shape[1], x.shape[1], dtype=x.dtype, device=x.device) + sw_mask.masked_fill_(torch.ones_like(sw_mask, dtype=torch.bool).tril_(-self.sliding_attention), torch.finfo(x.dtype).min) + attention_mask = attention_mask + sw_mask if attention_mask is not None else sw_mask + freqs_cis = freqs_cis[1] + else: + freqs_cis = freqs_cis[0] + + residual = x + x = self.input_layernorm(x) + x, present_key_value, shareable_kv = self.self_attn( + hidden_states=x, attention_mask=attention_mask, freqs_cis=freqs_cis, + past_key_value=past_key_value, sliding_window=sliding_window, shared_kv=shared_kv, + ) + x = self.post_attention_layernorm(x) + x = residual + x + + residual = x + x = self.pre_feedforward_layernorm(x) + x = self.mlp(x) + x = self.post_feedforward_layernorm(x) + x = residual + x + + if self.hidden_size_per_layer_input and per_layer_input is not None: + residual = x + x = self.per_layer_input_gate(x) + x = torch.nn.functional.gelu(x, approximate="tanh") + x = x * per_layer_input + x = self.per_layer_projection(x) + x = self.post_per_layer_input_norm(x) + x = residual + x + + if self.layer_scalar is not None: + x = x * self.layer_scalar + + return x, present_key_value, shareable_kv + + +class Gemma4Transformer(nn.Module): + def __init__(self, config, device=None, dtype=None, ops=None): + super().__init__() + self.config = config + + self.embed_tokens = _make_scaled_embedding(ops, config.vocab_size, config.hidden_size, config.hidden_size ** 0.5, device, dtype) + + self.layers = nn.ModuleList([ + TransformerBlockGemma4(config, index=i, device=device, dtype=dtype, ops=ops) + for i in range(config.num_hidden_layers) + ]) + + self.norm = RMSNorm(config.hidden_size, eps=config.rms_norm_eps, device=device, dtype=dtype) if config.final_norm else None + + # Precompute RoPE inv_freq on CPU to match reference code's exact value + rope_angles_global = int(config.partial_rotary_factor * config.global_head_dim // 2) + nope_global = config.global_head_dim // 2 - rope_angles_global + global_inv = 1.0 / (config.rope_theta[0] ** (torch.arange(0, 2 * rope_angles_global, 2).float() / config.global_head_dim)) + if nope_global > 0: + global_inv = torch.cat([global_inv, torch.zeros(nope_global)]) + self.register_buffer("_global_inv_freq", global_inv, persistent=False) + + sliding_inv = 1.0 / (config.rope_theta[1] ** (torch.arange(0, config.head_dim, 2).float() / config.head_dim)) + self.register_buffer("_sliding_inv_freq", sliding_inv, persistent=False) + + # Per-layer input mechanism + self.hidden_size_per_layer_input = config.hidden_size_per_layer_input + if self.hidden_size_per_layer_input: + self.embed_tokens_per_layer = _make_scaled_embedding(ops, config.vocab_size, config.num_hidden_layers * self.hidden_size_per_layer_input, self.hidden_size_per_layer_input ** 0.5, device, dtype) + self.per_layer_model_projection = ops.Linear( + config.hidden_size, config.num_hidden_layers * self.hidden_size_per_layer_input, + bias=False, device=device, dtype=dtype) + self.per_layer_projection_norm = RMSNorm( + self.hidden_size_per_layer_input, eps=config.rms_norm_eps, + device=device, dtype=dtype) + + def get_past_len(self, past_key_values): + for kv in past_key_values: + if len(kv) >= 3: + return kv[2] + return 0 + + def _freqs_from_inv(self, inv_freq, position_ids, device, dtype): + """Compute cos/sin from stored inv_freq""" + inv_exp = inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1).to(device) + pos_exp = position_ids[:, None, :].float() + freqs = (inv_exp @ pos_exp).transpose(1, 2) + emb = torch.cat((freqs, freqs), dim=-1) + return emb.cos().unsqueeze(1).to(dtype), emb.sin().unsqueeze(1).to(dtype) + + def compute_freqs_cis(self, position_ids, device, dtype=None): + global_freqs = self._freqs_from_inv(self._global_inv_freq, position_ids, device, dtype) + sliding_freqs = self._freqs_from_inv(self._sliding_inv_freq, position_ids, device, dtype) + return [global_freqs, sliding_freqs] + + def forward(self, x, attention_mask=None, embeds=None, num_tokens=None, intermediate_output=None, + final_layer_norm_intermediate=True, dtype=None, position_ids=None, embeds_info=None, + past_key_values=None, input_ids=None): + if embeds is not None: + x = embeds + else: + x = self.embed_tokens(x, out_dtype=dtype) + + seq_len = x.shape[1] + past_len = 0 + if past_key_values is not None and len(past_key_values) > 0: + past_len = self.get_past_len(past_key_values) + + if position_ids is None: + position_ids = torch.arange(past_len, past_len + seq_len, device=x.device).unsqueeze(0) + + freqs_cis = self.compute_freqs_cis(position_ids, x.device, dtype=x.dtype) + + mask = None + min_val = torch.finfo(x.dtype).min + if attention_mask is not None: + mask = 1.0 - attention_mask.to(x.dtype).reshape((attention_mask.shape[0], 1, -1, attention_mask.shape[-1])).expand(attention_mask.shape[0], 1, seq_len, attention_mask.shape[-1]) + mask = mask.masked_fill(mask.to(torch.bool), min_val) + + if seq_len > 1: + causal_mask = torch.zeros(past_len + seq_len, past_len + seq_len, dtype=x.dtype, device=x.device) + causal_mask.masked_fill_(torch.ones_like(causal_mask, dtype=torch.bool).triu_(1), min_val) + mask = mask + causal_mask if mask is not None else causal_mask + + # Per-layer inputs + per_layer_inputs = None + if self.hidden_size_per_layer_input: + num_layers = self.config.num_hidden_layers + hpl = self.hidden_size_per_layer_input + per_layer_proj = self.per_layer_model_projection(x) * (1.0 / (self.config.hidden_size ** 0.5)) + per_layer_proj = self.per_layer_projection_norm(per_layer_proj.reshape(*x.shape[:-1], num_layers, hpl)) + if input_ids is not None and input_ids.shape[1] == x.shape[1]: + per_layer_emb = self.embed_tokens_per_layer(input_ids).reshape(*input_ids.shape, num_layers, hpl) + per_layer_inputs = (per_layer_proj + per_layer_emb) * (0.5 ** 0.5) + else: + per_layer_inputs = per_layer_proj + + # KV sharing: later layers reuse KV from the last non-shared sliding/global layer + num_kv_shared = self.config.num_kv_shared_layers + first_kv_shared = self.config.num_hidden_layers - num_kv_shared if num_kv_shared > 0 else self.config.num_hidden_layers + shared_sliding_kv = None # KV from last non-shared sliding layer + shared_global_kv = None # KV from last non-shared global layer + + intermediate = None + next_key_values = [] + for i, layer in enumerate(self.layers): + past_kv = past_key_values[i] if past_key_values is not None and len(past_key_values) > 0 else None + + layer_kwargs = {} + if per_layer_inputs is not None: + layer_kwargs['per_layer_input'] = per_layer_inputs[:, :, i, :] + + is_sliding = hasattr(layer, 'sliding_attention') and layer.sliding_attention + if i >= first_kv_shared and num_kv_shared > 0: + shared = shared_sliding_kv if is_sliding else shared_global_kv + if shared is not None: + layer_kwargs['shared_kv'] = shared + + x, current_kv, shareable_kv = layer(x=x, attention_mask=mask, freqs_cis=freqs_cis, past_key_value=past_kv, **layer_kwargs) + + next_key_values.append(current_kv if current_kv is not None else ()) + + # Only track the last sliding/global before the sharing boundary + if i < first_kv_shared and shareable_kv is not None: + if is_sliding: + shared_sliding_kv = shareable_kv + else: + shared_global_kv = shareable_kv + + if i == intermediate_output: + intermediate = x.clone() + + if self.norm is not None: + x = self.norm(x) + + if len(next_key_values) > 0: + return x, intermediate, next_key_values + return x, intermediate + + +class Gemma4Base(BaseLlama, BaseGenerate, torch.nn.Module): + """Common base for all Gemma4 variants: text model + vision.""" + def _init_model(self, config, dtype, device, operations): + self.num_layers = config.num_hidden_layers + self.model = Gemma4Transformer(config, device=device, dtype=dtype, ops=operations) + self.dtype = dtype + self.multi_modal_projector = Gemma4MultiModalProjector(config, dtype=dtype, device=device, ops=operations) + self.vision_model = Gemma4VisionEncoder(config.vision_config, dtype=dtype, device=device, ops=operations) + + def logits(self, x): + logits = super().logits(x) + cap = self.model.config.final_logit_softcapping + if cap: + logits = cap * torch.tanh(logits / cap) + return logits + + def init_kv_cache(self, batch, max_cache_len, device, execution_dtype): + past_key_values = [] + for _ in range(self.model.config.num_hidden_layers): + past_key_values.append(()) + return past_key_values + + def preprocess_embed(self, embed, device): + if embed["type"] == "image": + image = embed.pop("data").movedim(-1, 1) # [B, H, W, C] -> [B, C, H, W] + max_soft_tokens = embed.get("max_soft_tokens", None) + vision_out = self.vision_model(image.to(device, dtype=torch.float32), max_soft_tokens=max_soft_tokens) + return self.multi_modal_projector(vision_out), None + return None, None + + +class Gemma4AudioMixin: + """Adds audio support to a Gemma4 model.""" + def _init_audio(self, config, dtype, device, operations): + self.audio_model = Gemma4AudioEncoder(config.audio_config, dtype=dtype, device=device, ops=operations) + self.audio_projector = Gemma4AudioProjector({"audio_output_proj_dims": config.audio_config["output_proj_dims"], "text_hidden_size": config.hidden_size, "rms_norm_eps": config.rms_norm_eps}, dtype=dtype, device=device, ops=operations) + + def preprocess_embed(self, embed, device): + result, extra = super().preprocess_embed(embed, device) + if result is not None: + return result, extra + if embed["type"] == "audio": + audio = embed.pop("data").to(device, dtype=torch.float32) + audio_mask = embed.pop("mask", None) + if audio_mask is not None: + audio_mask = audio_mask.to(device) + audio_out = self.audio_model(audio, audio_mask=audio_mask) + return self.audio_projector(audio_out), None + return None, None + + +# Vision Encoder + +def _compute_vision_2d_rope(head_dim, pixel_position_ids, theta=100.0, device=None): + """Compute 2D RoPE for vision: separate frequencies for x and y dimensions. + + Args: + head_dim: dimension per head (e.g. 64) + pixel_position_ids: [batch, num_patches, 2] with (x, y) coords + theta: RoPE base frequency + Returns: + (cos, sin) each of shape [batch, num_patches, head_dim] + """ + rotary_dim_per_axis = head_dim // 2 + freq_indices = torch.arange(0, rotary_dim_per_axis, 2, device=device).float() + inv_freq = 1.0 / (theta ** (freq_indices / rotary_dim_per_axis)) + + all_cos, all_sin = [], [] + for i in range(2): # x and y + dim_positions = pixel_position_ids[:, :, i].float() # [batch, num_patches] + freqs = torch.einsum('bi,j->bij', dim_positions, inv_freq.to(device)) # [batch, num_patches, rotary_dim/2] + emb = torch.cat([freqs, freqs], dim=-1) # [batch, num_patches, rotary_dim] + all_cos.append(emb.cos()) + all_sin.append(emb.sin()) + + cos = torch.cat(all_cos, dim=-1).to(pixel_position_ids.device) # [batch, num_patches, head_dim] + sin = torch.cat(all_sin, dim=-1).to(pixel_position_ids.device) + return cos, sin + + +def _apply_vision_2d_rope(x, freqs): + """Apply 2D RoPE (multidimensional) to vision query/key states. + + Splits x and cos/sin into ndim=2 parts, applies 1D RoPE to each independently. + + x: [batch, heads, seq, head_dim] + freqs: (cos, sin) each [batch, seq, head_dim] + """ + cos = freqs[0].unsqueeze(1) # [batch, 1, seq, head_dim] + sin = freqs[1].unsqueeze(1) + half = x.shape[-1] // 2 + a = _apply_rotary_pos_emb(x[..., :half], (cos[..., :half], sin[..., :half])) + b = _apply_rotary_pos_emb(x[..., half:], (cos[..., half:], sin[..., half:])) + return torch.cat([a, b], dim=-1) + + +class ClippedLinear(nn.Module): + """Linear layer with activation clipping (from quantization-aware training). + + Stores input_max/min and output_max/min as buffers loaded from checkpoint. + """ + def __init__(self, in_features, out_features, bias=False, device=None, dtype=None, ops=None): + super().__init__() + self.linear = ops.Linear(in_features, out_features, bias=bias, device=device, dtype=dtype) + self.register_buffer('input_max', torch.tensor(float('inf'), device=device, dtype=dtype)) + self.register_buffer('input_min', torch.tensor(float('-inf'), device=device, dtype=dtype)) + self.register_buffer('output_max', torch.tensor(float('inf'), device=device, dtype=dtype)) + self.register_buffer('output_min', torch.tensor(float('-inf'), device=device, dtype=dtype)) + + @property + def weight(self): + return self.linear.weight + + def forward(self, x): + x = x.clamp(min=self.input_min, max=self.input_max) + x = self.linear(x) + return x.clamp_(min=self.output_min, max=self.output_max) + + +class Gemma4VisionMLP(nn.Module): + """SwiGLU MLP matching gate_proj/up_proj/down_proj structure.""" + def __init__(self, config, device=None, dtype=None, ops=None): + super().__init__() + hidden_size = config["hidden_size"] + intermediate_size = config["intermediate_size"] + self.gate_proj = ClippedLinear(hidden_size, intermediate_size, device=device, dtype=dtype, ops=ops) + self.up_proj = ClippedLinear(hidden_size, intermediate_size, device=device, dtype=dtype, ops=ops) + self.down_proj = ClippedLinear(intermediate_size, hidden_size, device=device, dtype=dtype, ops=ops) + + def forward(self, x): + return self.down_proj(torch.nn.functional.gelu(self.gate_proj(x), approximate="tanh") * self.up_proj(x)) + + +class Gemma4VisionAttention(nn.Module): + def __init__(self, config, device=None, dtype=None, ops=None): + super().__init__() + self.hidden_size = config["hidden_size"] + self.num_heads = config["num_attention_heads"] + self.head_dim = config.get("head_dim", self.hidden_size // self.num_heads) + + self.q_proj = ClippedLinear(self.hidden_size, self.num_heads * self.head_dim, device=device, dtype=dtype, ops=ops) + self.k_proj = ClippedLinear(self.hidden_size, self.num_heads * self.head_dim, device=device, dtype=dtype, ops=ops) + self.v_proj = ClippedLinear(self.hidden_size, self.num_heads * self.head_dim, device=device, dtype=dtype, ops=ops) + self.o_proj = ClippedLinear(self.num_heads * self.head_dim, self.hidden_size, device=device, dtype=dtype, ops=ops) + + self.q_norm = RMSNorm(self.head_dim, eps=config["rms_norm_eps"], device=device, dtype=dtype) + self.k_norm = RMSNorm(self.head_dim, eps=config["rms_norm_eps"], device=device, dtype=dtype) + + def forward(self, x, freqs, attention_mask=None): + batch_size, seq_length, _ = x.shape + + xq = self.q_proj(x).view(batch_size, seq_length, self.num_heads, self.head_dim) + xk = self.k_proj(x).view(batch_size, seq_length, self.num_heads, self.head_dim) + xv = self.v_proj(x).view(batch_size, seq_length, self.num_heads, self.head_dim) + + xq = self.q_norm(xq).transpose(1, 2) + xk = self.k_norm(xk).transpose(1, 2) + xv = rms_norm(xv) + + xq = _apply_vision_2d_rope(xq, freqs) + xk = _apply_vision_2d_rope(xk, freqs) + + xv = xv.to(xq.dtype).transpose(1, 2) + + output = optimized_attention_for_device(xq.device, mask=attention_mask is not None, small_input=True)(xq, xk, xv, self.num_heads, mask=attention_mask, skip_reshape=True, scale=1.0) + return self.o_proj(output) + + +class Gemma4VisionLayer(nn.Module): + def __init__(self, config, device=None, dtype=None, ops=None): + super().__init__() + self.self_attn = Gemma4VisionAttention(config, device=device, dtype=dtype, ops=ops) + self.mlp = Gemma4VisionMLP(config, device=device, dtype=dtype, ops=ops) + norm_kwargs = dict(eps=config["rms_norm_eps"], device=device, dtype=dtype) + hidden = config["hidden_size"] + self.input_layernorm = RMSNorm(hidden, **norm_kwargs) + self.post_attention_layernorm = RMSNorm(hidden, **norm_kwargs) + self.pre_feedforward_layernorm = RMSNorm(hidden, **norm_kwargs) + self.post_feedforward_layernorm = RMSNorm(hidden, **norm_kwargs) + + def forward(self, x, freqs, attention_mask=None): + residual = x + x = self.input_layernorm(x) + x = self.self_attn(x, freqs, attention_mask=attention_mask) + x = self.post_attention_layernorm(x) + x = residual + x + + residual = x + x = self.pre_feedforward_layernorm(x) + x = self.mlp(x) + x = self.post_feedforward_layernorm(x) + x = residual + x + return x + + +class Gemma4PatchEmbedder(nn.Module): + """Patch embedding with learned 2D position embeddings via one-hot lookup.""" + def __init__(self, config, device=None, dtype=None, ops=None): + super().__init__() + hidden_size = config["hidden_size"] + patch_size = config["patch_size"] + self.patch_size = patch_size + self.position_embedding_size = config.get("position_embedding_size", 10240) + + self.input_proj = ops.Linear(3 * patch_size * patch_size, hidden_size, bias=False, device=device, dtype=dtype) + self.position_embedding_table = nn.Parameter( + torch.empty(2, self.position_embedding_size, hidden_size, device=device, dtype=dtype) + ) + + def forward(self, patches, pixel_position_ids): + """ + patches: [B, num_patches, 3*patch_size²] in [0,1] range (normalized to [-1,1] inside, matching HF) + pixel_position_ids: [B, num_patches, 2] with (x,y) positions, (-1,-1) for padding + """ + hidden_states = self.input_proj((2.0 * (patches - 0.5)).to(self.input_proj.weight.dtype)) + + clamped_positions = pixel_position_ids.clamp(min=0) + pos_table = comfy.model_management.cast_to_device(self.position_embedding_table, hidden_states.device, hidden_states.dtype) + position_embeddings = pos_table[0][clamped_positions[..., 0]] + pos_table[1][clamped_positions[..., 1]] + + # Zero out position embeddings for padding patches (matching HF) + padding_positions = (pixel_position_ids == -1).all(dim=-1) + position_embeddings = torch.where(padding_positions.unsqueeze(-1), 0.0, position_embeddings) + + return hidden_states + position_embeddings + + +class Gemma4VisionEncoderLayers(nn.Module): + """Wrapper to produce state dict keys as encoder.layers.X.*""" + def __init__(self, config, dtype=None, device=None, ops=None): + super().__init__() + self.layers = nn.ModuleList([ + Gemma4VisionLayer(config, device=device, dtype=dtype, ops=ops) + for _ in range(config["num_hidden_layers"]) + ]) + + +class Gemma4VisionEncoder(nn.Module): + def __init__(self, config, dtype=None, device=None, ops=None): + super().__init__() + self.config = config + self.hidden_size = config["hidden_size"] + self.head_dim = config.get("head_dim", config["hidden_size"] // config["num_attention_heads"]) + self.patch_size = config["patch_size"] + self.pooling_kernel_size = config.get("pooling_kernel_size", 3) + self.root_hidden_size = self.hidden_size ** 0.5 + + self.patch_embedder = Gemma4PatchEmbedder(config, device=device, dtype=dtype, ops=ops) + self.encoder = Gemma4VisionEncoderLayers(config, dtype=dtype, device=device, ops=ops) + + def forward(self, pixel_values, max_soft_tokens=None): + """ + pixel_values: [B, C, H, W] in [0,1] range + max_soft_tokens: if provided, pad to max_soft_tokens * k² total patches + """ + batch_size, _, height, width = pixel_values.shape + ps = self.patch_size + k = self.pooling_kernel_size + patches_h, patches_w = height // ps, width // ps + num_patches = patches_h * patches_w + output_length = max_soft_tokens if max_soft_tokens is not None else num_patches // (k * k) + n_padding = output_length * k * k - num_patches + + # Patchify and build position grid + patches = pixel_values.reshape(batch_size, -1, patches_h, ps, patches_w, ps) + patches = patches.permute(0, 2, 4, 3, 5, 1).reshape(batch_size, num_patches, -1) + grid_y, grid_x = torch.meshgrid(torch.arange(patches_h, device=pixel_values.device), torch.arange(patches_w, device=pixel_values.device), indexing='ij') + position_ids = torch.stack([grid_x.flatten(), grid_y.flatten()], dim=-1).unsqueeze(0).expand(batch_size, -1, -1) + + # Append zero-pixel padding with (-1,-1) positions + if n_padding > 0: + patches = torch.cat([patches, patches.new_zeros(batch_size, n_padding, patches.shape[-1])], dim=1) + position_ids = torch.cat([position_ids, position_ids.new_full((batch_size, n_padding, 2), -1)], dim=1) + + padding = (position_ids == -1).all(dim=-1) + + # Embed, encode, pool + x = self.patch_embedder(patches, position_ids) + freqs = _compute_vision_2d_rope(self.head_dim, position_ids, device=pixel_values.device) + freqs = tuple(t.to(x.dtype) for t in freqs) + if n_padding > 0: + mask = padding.unsqueeze(1).unsqueeze(2).expand(-1, 1, position_ids.shape[1], -1) + mask = torch.zeros_like(mask, dtype=x.dtype).masked_fill_(mask, torch.finfo(x.dtype).min) + else: + mask = None + + for layer in self.encoder.layers: + x = layer(x, freqs, attention_mask=mask) + + if n_padding > 0: + x = x.masked_fill(padding.unsqueeze(-1), 0.0) + + # Average pool by spatial position + clamped = position_ids.clamp(min=0) + max_x = clamped[:, :, 0].max(dim=-1, keepdim=True)[0] + 1 + ki = torch.div(clamped, k, rounding_mode="floor") + ki = ki[:, :, 0] + (max_x // k) * ki[:, :, 1] + weights = torch.nn.functional.one_hot(ki.long(), output_length).float() / (k * k) + x = (weights.transpose(1, 2) @ x.float()).to(x.dtype) + + # Strip empty output tokens + valid_out = ~((weights == 0).all(dim=1)) + if valid_out.any() and not valid_out.all(): + x = x[:, valid_out[0]] if batch_size > 1 else x[valid_out].unsqueeze(0) + + return x * self.root_hidden_size + + +class Gemma4RMSNormProjector(nn.Module): + """Shared projector: parameterless RMSNorm → linear. Used for both vision and audio.""" + def __init__(self, in_dim, out_dim, dtype=None, device=None, ops=None): + super().__init__() + self.embedding_projection = ops.Linear(in_dim, out_dim, bias=False, device=device, dtype=dtype) + + def forward(self, x): + return self.embedding_projection(rms_norm(x)) + + +class Gemma4MultiModalProjector(Gemma4RMSNormProjector): + def __init__(self, config, dtype=None, device=None, ops=None): + super().__init__(config.vision_config["hidden_size"], config.hidden_size, dtype=dtype, device=device, ops=ops) + + +# Audio Encoder + +class Gemma4AudioConvSubsampler(nn.Module): + """2D convolution subsampling for audio features""" + def __init__(self, config, device=None, dtype=None, ops=None): + super().__init__() + eps = config["rms_norm_eps"] + self.layer0 = nn.ModuleDict({ + 'conv': ops.Conv2d(1, 128, kernel_size=3, stride=2, padding=1, bias=False, device=device, dtype=dtype), + 'norm': ops.LayerNorm(128, eps=eps, elementwise_affine=True, bias=False, device=device, dtype=dtype), + }) + self.layer1 = nn.ModuleDict({ + 'conv': ops.Conv2d(128, 32, kernel_size=3, stride=2, padding=1, bias=False, device=device, dtype=dtype), + 'norm': ops.LayerNorm(32, eps=eps, elementwise_affine=True, bias=False, device=device, dtype=dtype), + }) + # proj_input_dim = (128 // 4) * 32 = 1024 + self.input_proj_linear = ops.Linear(1024, config["hidden_size"], bias=False, device=device, dtype=dtype) + + def _conv_layer(self, x, layer, mask): + if mask is not None: + x = x * mask[:, None, :, None].to(x.device) + x = layer['conv'](x.to(layer['conv'].weight.dtype)) + x = torch.relu(layer['norm'](x.permute(0, 2, 3, 1)).permute(0, 3, 1, 2).contiguous()) + if mask is not None: + mask = mask[:, ::2] + return x, mask + + def forward(self, x, mask=None): + x = x.unsqueeze(1) + x, mask = self._conv_layer(x, self.layer0, mask) + x, mask = self._conv_layer(x, self.layer1, mask) + batch_size, _, seq_len, _ = x.shape + x = x.permute(0, 2, 3, 1).contiguous().reshape(batch_size, seq_len, -1) + return self.input_proj_linear(x), mask + + +class Gemma4AudioFeedForward(nn.Module): + """Conformer feed-forward with residual scaling.""" + def __init__(self, config, device=None, dtype=None, ops=None): + super().__init__() + hidden_size = config["hidden_size"] + intermediate_size = config.get("intermediate_size", hidden_size * 4) + self.pre_layer_norm = RMSNorm(hidden_size, eps=config["rms_norm_eps"], device=device, dtype=dtype) + self.ffw_layer_1 = ClippedLinear(hidden_size, intermediate_size, device=device, dtype=dtype, ops=ops) + self.ffw_layer_2 = ClippedLinear(intermediate_size, hidden_size, device=device, dtype=dtype, ops=ops) + self.post_layer_norm = RMSNorm(hidden_size, eps=config["rms_norm_eps"], device=device, dtype=dtype) + self.post_layer_scale = config.get("residual_weight", 0.5) + + def forward(self, x): + residual = x + x = self.pre_layer_norm(x) + x = torch.nn.functional.silu(self.ffw_layer_1(x)) + x = self.ffw_layer_2(x) + x = self.post_layer_norm(x) + x = x * self.post_layer_scale + return x + residual + + +class Gemma4AudioRelPositionalEncoding(nn.Module): + """Sinusoidal relative positional encoding for audio attention.""" + def __init__(self, config, device=None, dtype=None): + super().__init__() + hidden_size = config["hidden_size"] + context_left = config.get("attention_context_left", 13) + context_right = config.get("attention_context_right", 0) + self.chunk_size = config.get("attention_chunk_size", 12) + self.context_size = self.chunk_size + context_left - 1 + context_right + + num_timescales = hidden_size // 2 + log_inc = math.log(10000.0) / max(num_timescales - 1, 1) + inv_timescales = torch.exp(torch.arange(num_timescales) * -log_inc).to(dtype=dtype).unsqueeze(0).unsqueeze(0) + self.register_buffer("inv_timescales", inv_timescales, persistent=False) + + def forward(self, hidden_states): + positions = torch.arange(self.chunk_size, -1, -1, device=hidden_states.device).unsqueeze(-1) + scaled = positions * self.inv_timescales.to(device=hidden_states.device) + return torch.cat([torch.sin(scaled), torch.cos(scaled)], dim=-1).to(dtype=hidden_states.dtype) + + +class Gemma4AudioAttention(nn.Module): + """Chunked block attention with relative position bias and softcap.""" + def __init__(self, config, device=None, dtype=None, ops=None): + super().__init__() + self.hidden_size = config["hidden_size"] + self.num_heads = config["num_attention_heads"] + self.head_dim = self.hidden_size // self.num_heads + self.chunk_size = config.get("attention_chunk_size", 12) + self.max_past_horizon = config.get("attention_context_left", 13) - 1 + self.max_future_horizon = config.get("attention_context_right", 0) + self.context_size = self.chunk_size + self.max_past_horizon + self.max_future_horizon + + self.q_scale = (self.head_dim ** -0.5) / math.log(2) + self.k_scale = math.log(1 + math.e) / math.log(2) + self.register_buffer("softcap", torch.tensor(config.get("attention_logit_cap", 50.0), dtype=dtype), persistent=False) + + self.q_proj = ClippedLinear(self.hidden_size, self.hidden_size, device=device, dtype=dtype, ops=ops) + self.k_proj = ClippedLinear(self.hidden_size, self.hidden_size, device=device, dtype=dtype, ops=ops) + self.v_proj = ClippedLinear(self.hidden_size, self.hidden_size, device=device, dtype=dtype, ops=ops) + self.post = ClippedLinear(self.hidden_size, self.hidden_size, device=device, dtype=dtype, ops=ops) + self.per_dim_scale = nn.Parameter(torch.empty(self.head_dim, device=device, dtype=dtype)) + self.relative_k_proj = ops.Linear(self.hidden_size, self.hidden_size, bias=False, device=device, dtype=dtype) + + def _convert_to_block(self, x): + B, S, H, D = x.shape + num_blocks = (S + self.chunk_size - 1) // self.chunk_size + pad = num_blocks * self.chunk_size - S + x = torch.nn.functional.pad(x, (0, 0, 0, 0, 0, pad)) + return x.reshape(B, num_blocks, self.chunk_size, H, D).contiguous() + + def _extract_block_context(self, x): + x = torch.nn.functional.pad(x, (0, 0, 0, 0, self.max_past_horizon, self.max_future_horizon + self.chunk_size - 1)) + x = x.unfold(1, self.context_size, self.chunk_size) + return torch.movedim(x, -1, 2).contiguous() + + def _rel_shift(self, x): + B, H, NB, BS, PL = x.shape + CS = self.context_size + x = torch.nn.functional.pad(x, (0, CS + 1 - PL)) + x = x.view(B, H, NB, BS * (CS + 1)) + x = x[..., :BS * CS] + return x.view(B, H, NB, BS, CS) + + def _build_blocked_mask(self, seq_len, num_blocks, device, audio_mask=None): + """Build 5D boolean blocked attention mask (True=attend, False=mask)""" + q = torch.arange(seq_len, device=device) + dist = q[:, None] - q[None, :] + mask = (dist >= 0) & (dist < self.max_past_horizon) + if self.max_future_horizon > 0: + mask = mask | ((dist < 0) & ((-dist) < self.max_future_horizon)) + if audio_mask is not None: + mask = mask & audio_mask[0, None, :].bool() + m = mask[None, None] + # Reshape to blocked 5D matching reference code + p = num_blocks * self.chunk_size - seq_len + m = torch.nn.functional.pad(m, (0, p, 0, p), value=False) + m = m.reshape(1, 1, num_blocks, self.chunk_size, -1) + m = torch.nn.functional.pad(m, (self.max_past_horizon, self.max_future_horizon), value=False) + idx = (torch.arange(num_blocks, device=device) * self.chunk_size)[:, None] + torch.arange(self.context_size, device=device)[None, :] + return m.gather(-1, idx[None, None, :, None, :].expand(1, 1, -1, self.chunk_size, -1)) + + def forward(self, x, position_embeddings=None, attn_mask=None): + B, S, _ = x.shape + + q = self.q_proj(x).float().view(B, S, self.num_heads, self.head_dim) + k = self.k_proj(x).float().view(B, S, self.num_heads, self.head_dim) + v = self.v_proj(x).float().view(B, S, self.num_heads, self.head_dim) + + q = q * self.q_scale * torch.nn.functional.softplus(self.per_dim_scale) + k = k * self.k_scale + + q_blocks = self._convert_to_block(q) + k_context = self._extract_block_context(k) + v_context = self._extract_block_context(v) + num_blocks = q_blocks.shape[1] + + rel_k = self.relative_k_proj(position_embeddings).view(-1, self.num_heads, self.head_dim).to(q.dtype) + + queries = q_blocks.permute(0, 3, 1, 2, 4) # [B, H, NB, CS, D] + matrix_ac = queries @ k_context.permute(0, 3, 1, 4, 2) + + queries_flat = queries.reshape(B, self.num_heads, -1, self.head_dim) + matrix_bd = queries_flat @ rel_k.permute(1, 2, 0) + matrix_bd = matrix_bd.reshape(B, self.num_heads, num_blocks, self.chunk_size, -1) + matrix_bd = self._rel_shift(matrix_bd) + + attn_weights = matrix_ac + matrix_bd + attn_weights = torch.tanh(attn_weights / self.softcap) * self.softcap + + # Mask out invalid positions in chunk context (matching reference's masked_fill approach) + if attn_mask is None: + attn_mask = self._build_blocked_mask(S, num_blocks, x.device) + attn_weights = attn_weights.masked_fill(attn_mask.logical_not(), -1e9) + + attn_weights = torch.nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(v.dtype) + out = attn_weights @ v_context.permute(0, 3, 1, 2, 4) + out = out.permute(0, 2, 3, 1, 4).reshape(B, num_blocks * self.chunk_size, -1) + out = out[:, :S].contiguous() + return self.post(out.to(self.post.linear.weight.dtype)) + + +class Gemma4AudioLConv1d(nn.Module): + """Lightweight convolution with standard GLU.""" + def __init__(self, config, device=None, dtype=None, ops=None): + super().__init__() + hidden_size = config["hidden_size"] + conv_kernel_size = config.get("conv_kernel_size", 5) + self.pre_layer_norm = RMSNorm(hidden_size, eps=config["rms_norm_eps"], device=device, dtype=dtype) + self.linear_start = ClippedLinear(hidden_size, hidden_size * 2, device=device, dtype=dtype, ops=ops) + # Causal conv: left-pad only + self.depthwise_conv1d = ops.Conv1d(hidden_size, hidden_size, kernel_size=conv_kernel_size, padding=0, groups=hidden_size, bias=False, device=device, dtype=dtype) + self.conv_left_pad = conv_kernel_size - 1 # causal: pad left by kernel-1 + self.conv_norm = RMSNorm(hidden_size, eps=config["rms_norm_eps"], device=device, dtype=dtype) + self.linear_end = ClippedLinear(hidden_size, hidden_size, device=device, dtype=dtype, ops=ops) + + def forward(self, x): + residual = x + x = self.pre_layer_norm(x) + x = self.linear_start(x) + x = torch.nn.functional.glu(x, dim=-1) + x = x.transpose(1, 2) + x = torch.nn.functional.pad(x, (self.conv_left_pad, 0)) + x = self.depthwise_conv1d(x).transpose(1, 2) + x = self.conv_norm(x) + x = torch.nn.functional.silu(x) + x = self.linear_end(x) + return x + residual + + +class Gemma4AudioLayer(nn.Module): + """Conformer block: FFN1 -> Attention -> LConv -> FFN2.""" + def __init__(self, config, device=None, dtype=None, ops=None): + super().__init__() + self.feed_forward1 = Gemma4AudioFeedForward(config, device=device, dtype=dtype, ops=ops) + self.self_attn = Gemma4AudioAttention(config, device=device, dtype=dtype, ops=ops) + norm_kwargs = dict(eps=config["rms_norm_eps"], device=device, dtype=dtype) + hidden_size = config["hidden_size"] + self.norm_pre_attn = RMSNorm(hidden_size, **norm_kwargs) + self.norm_post_attn = RMSNorm(hidden_size, **norm_kwargs) + self.lconv1d = Gemma4AudioLConv1d(config, device=device, dtype=dtype, ops=ops) + self.feed_forward2 = Gemma4AudioFeedForward(config, device=device, dtype=dtype, ops=ops) + self.norm_out = RMSNorm(hidden_size, **norm_kwargs) + + def forward(self, x, position_embeddings=None, attn_mask=None): + x = self.feed_forward1(x) + + residual = x + x = self.norm_pre_attn(x) + x = self.self_attn(x, position_embeddings=position_embeddings, attn_mask=attn_mask) + x = self.norm_post_attn(x) + x = x + residual + + x = self.lconv1d(x) + x = self.feed_forward2(x) + + x = self.norm_out(x) + return x + + +class Gemma4AudioEncoder(nn.Module): + def __init__(self, config, dtype=None, device=None, ops=None): + super().__init__() + self.hidden_size = config["hidden_size"] + self.output_proj_dims = config.get("output_proj_dims", 1536) + + self.subsample_conv_projection = Gemma4AudioConvSubsampler(config, device=device, dtype=dtype, ops=ops) + self.rel_pos_enc = Gemma4AudioRelPositionalEncoding(config, device=device, dtype=dtype) + + self.layers = nn.ModuleList([ + Gemma4AudioLayer(config, device=device, dtype=dtype, ops=ops) + for _ in range(config["num_hidden_layers"]) + ]) + + self.output_proj = ops.Linear(self.hidden_size, self.output_proj_dims, bias=True, device=device, dtype=dtype) + + def forward(self, audio_features, audio_mask=None): + x, audio_mask = self.subsample_conv_projection(audio_features, audio_mask) + position_embeddings = self.rel_pos_enc(x) + + # Build blocked attention mask once for all layers + attn_mask = self.layers[0].self_attn._build_blocked_mask( + x.shape[1], (x.shape[1] + self.layers[0].self_attn.chunk_size - 1) // self.layers[0].self_attn.chunk_size, + x.device, audio_mask=audio_mask) + + for layer in self.layers: + x = layer(x, position_embeddings=position_embeddings, attn_mask=attn_mask) + + x = self.output_proj(x) + return x + + +class Gemma4AudioProjector(Gemma4RMSNormProjector): + def __init__(self, config, dtype=None, device=None, ops=None): + super().__init__(config.get("audio_output_proj_dims", 1536), config.get("text_hidden_size", 2560), dtype=dtype, device=device, ops=ops) + + +# Tokenizer and Wrappers + +class Gemma4_Tokenizer(): + tokenizer_json_data = None + + def state_dict(self): + if self.tokenizer_json_data is not None: + return {"tokenizer_json": self.tokenizer_json_data} + return {} + + def _extract_mel_spectrogram(self, waveform, sample_rate): + """Extract 128-bin log mel spectrogram. + Uses numpy for FFT/matmul/log to produce bit-identical results with reference code. + """ + # Mix to mono first, then resample to 16kHz + if waveform.dim() > 1 and waveform.shape[0] > 1: + waveform = waveform.mean(dim=0, keepdim=True) + if waveform.dim() == 1: + waveform = waveform.unsqueeze(0) + audio = waveform.squeeze(0).float().numpy() + if sample_rate != 16000: + # Use scipy's resample_poly with a high-quality FIR filter to get as close as possible to librosa's resampling (while still not full match) + from scipy.signal import resample_poly, firwin + from math import gcd + g = gcd(sample_rate, 16000) + up, down = 16000 // g, sample_rate // g + L = max(up, down) + h = firwin(160 * L + 1, 0.96 / L, window=('kaiser', 6.5)) + audio = resample_poly(audio, up, down, window=h).astype(np.float32) + n = len(audio) + + # Pad to multiple of 128, build sample-level mask + if n % 128 != 0: + audio = np.pad(audio, (0, 128 - n % 128)) + mask_raw = np.ones(len(audio), dtype=np.float32) + mask_raw[n:] = 0.0 + + # Semicausal padding: 160 zeros prepended + audio = np.pad(audio, (160, 0)) + mask_raw = np.pad(mask_raw, (160, 0)) + + # Extract 321-sample frames via stride tricks, drop last → 320 + nf = (len(audio) - 321) // 160 + 1 + strides = (audio.strides[0] * 160, audio.strides[0]) + frames = np.lib.stride_tricks.as_strided(audio, (nf, 321), strides)[..., :-1].copy() + + # Periodic Hann window, FFT magnitude, mel filterbank, log + window = (0.5 - 0.5 * np.cos(2 * np.pi * np.arange(320) / 320)).astype(np.float32) + magnitude = np.abs(np.fft.rfft(frames * window, n=512, axis=-1)) + mel_fb = self._build_mel_filterbank() + log_mel = np.log(np.matmul(magnitude, mel_fb) + np.float64(0.001)).astype(np.float32) + + # Frame mask: valid when last sample in window is real audio + mask = mask_raw[np.arange(nf) * 160 + 320].astype(bool) + log_mel = log_mel * mask[:, None] + return torch.from_numpy(log_mel), torch.from_numpy(mask) # [T, 128], [T] + + @staticmethod + def _build_mel_filterbank(): + """Build 128-bin HTK mel filterbank [257, 128] for 512-pt FFT at 16kHz.""" + mel_freqs = np.linspace(0.0, 2595.0 * np.log10(1.0 + 8000.0 / 700.0), 130) + filter_freqs = 700.0 * (10.0 ** (mel_freqs / 2595.0) - 1.0) + fft_freqs = np.linspace(0, 16000 // 2, 257) + filter_diff = np.diff(filter_freqs) + slopes = np.expand_dims(filter_freqs, 0) - np.expand_dims(fft_freqs, 1) + down_slopes = -slopes[:, :-2] / filter_diff[:-1] + up_slopes = slopes[:, 2:] / filter_diff[1:] + return np.maximum(np.zeros(1), np.minimum(down_slopes, up_slopes)) + + def tokenize_with_weights(self, text, return_word_ids=False, image=None, audio=None, video=None, llama_template=None, skip_template=True, thinking=False, **kwargs): + + # Process audio + audio_features = [] + if audio is not None: + waveform = audio["waveform"].squeeze(0) if hasattr(audio, "__getitem__") else audio + sample_rate = audio.get("sample_rate", 16000) if hasattr(audio, "get") else 16000 + mel, mel_mask = self._extract_mel_spectrogram(waveform, sample_rate) + audio_features = [(mel.unsqueeze(0), mel_mask.unsqueeze(0))] # ([1, T, 128], [1, T]) + + # Process image/video frames + is_video = video is not None + source = video if is_video else image + images = [] + if source is not None: + samples = source.movedim(-1, 1) # [B, C, H, W] + num_frames = samples.shape[0] + + # Subsample video to 1fps + if is_video: + fps = kwargs.get("fps", 24) + step = max(1, round(fps)) + indices = list(range(0, num_frames, step)) + if len(indices) == 0: + indices = [0] + samples = samples[indices] + num_frames = len(indices) + + h, w = samples.shape[2], samples.shape[3] + patch_size = 16 + pooling_k = 3 + max_soft_tokens = 70 if is_video else 280 # video uses smaller token budget per frame + max_patches = max_soft_tokens * pooling_k * pooling_k + target_px = max_patches * patch_size * patch_size + factor = (target_px / (h * w)) ** 0.5 + side_mult = pooling_k * patch_size + target_h = max(int(factor * h // side_mult) * side_mult, side_mult) + target_w = max(int(factor * w // side_mult) * side_mult, side_mult) + + import torchvision.transforms.functional as TVF + for i in range(num_frames): + # rescaling to match reference code + s = (samples[i].clamp(0, 1) * 255).to(torch.uint8) # [C, H, W] uint8 + if target_h != h or target_w != w: + s = TVF.resize(s, [target_h, target_w], interpolation=TVF.InterpolationMode.BICUBIC, antialias=True) + s = s.float() * (1.0 / 255.0) + images.append({"pixels": s.unsqueeze(0).movedim(1, -1)[:, :, :, :3], "max_soft_tokens": max_soft_tokens}) + + if text.startswith('<|turn>'): + skip_template = True + + if skip_template: + llama_text = text + else: + if llama_template is not None: + llama_text = llama_template.format(text) + else: + # Build template from modalities present + system = "<|turn>system\n<|think|>\n" if thinking else "" + media = "" + if len(images) > 0: + if is_video: + media += "\n\n" + for i in range(len(images)): + ts = f"{int(i // 60):02d}:{int(i % 60):02d}" + sep = "" if i == 0 else " " + media += f"{sep}{ts} <|image><|video|>" + media += "\n\n" + else: + media += "\n\n" + for i in range(len(images)): + if i > 0: + media += "\n\n\n\n" + media += "<|image><|image|>" + media += "\n\n" + if len(audio_features) > 0: + # Compute audio token count (always at 16kHz) + num_samples = int(waveform.shape[-1] * 16000 / sample_rate) if sample_rate != 16000 else waveform.shape[-1] + _fl = 320 # int(round(16000 * 20.0 / 1000.0)) + _hl = 160 # int(round(16000 * 10.0 / 1000.0)) + _nmel = (num_samples + _fl // 2 - (_fl + 1)) // _hl + 1 + _t = _nmel + for _ in range(2): + _t = (_t + 2 - 3) // 2 + 1 + n_audio_tokens = min(_t, 750) + media += "<|audio>" + "<|audio|>" * n_audio_tokens + "" + llama_text = f"{system}<|turn>user\n{media}{text}\n<|turn>model\n" + + text_tokens = super().tokenize_with_weights(llama_text, return_word_ids) + + def _replace_placeholders(token_list, token_id, embeds): + """Replace first placeholder with embed dict, remove remaining consecutive ones.""" + embed_idx = 0 + i = 0 + while i < len(token_list): + if token_list[i][0] == token_id and embed_idx < len(embeds): + token_list[i] = (embeds[embed_idx],) + token_list[i][1:] + embed_idx += 1 + i += 1 + while i < len(token_list) and token_list[i][0] == token_id: + token_list.pop(i) + else: + i += 1 + + if len(images) > 0: + img_token_id = 258884 if is_video else 258880 + img_embeds = [{"type": "image", "data": img["pixels"], "max_soft_tokens": img["max_soft_tokens"]} for img in images] + for r in text_tokens: + _replace_placeholders(r, img_token_id, img_embeds) + + if len(audio_features) > 0: + aud_embeds = [{"type": "audio", "data": mel, "mask": mask} for mel, mask in audio_features] + for r in text_tokens: + _replace_placeholders(r, 258881, aud_embeds) + + return text_tokens + + +class _Gemma4Tokenizer: + """Tokenizer using the tokenizers (Gemma4 doesn't come with sentencepiece model)""" + def __init__(self, tokenizer_json_bytes=None, **kwargs): + from tokenizers import Tokenizer + if isinstance(tokenizer_json_bytes, torch.Tensor): + tokenizer_json_bytes = bytes(tokenizer_json_bytes.tolist()) + self.tokenizer = Tokenizer.from_str(tokenizer_json_bytes.decode("utf-8")) + + @classmethod + def from_pretrained(cls, tokenizer_data, **kwargs): + return cls(tokenizer_json_bytes=tokenizer_data, **kwargs) + + def __call__(self, text): + return {"input_ids": self.tokenizer.encode(text, add_special_tokens=False).ids} + + def get_vocab(self): + return self.tokenizer.get_vocab() + + def convert_tokens_to_ids(self, tokens): + return [self.tokenizer.token_to_id(t) for t in tokens] + + def decode(self, ids, **kwargs): + return self.tokenizer.decode(ids, skip_special_tokens=kwargs.get("skip_special_tokens", False)) + + +# Tokenizer +class Gemma4SDTokenizer(Gemma4_Tokenizer, sd1_clip.SDTokenizer): + embedding_size = 2560 + def __init__(self, embedding_directory=None, tokenizer_data={}): + tokenizer_json = tokenizer_data.get("tokenizer_json", None) + self.tokenizer_json_data = tokenizer_json + super().__init__(tokenizer_json, pad_with_end=False, embedding_size=self.embedding_size, embedding_key='gemma4', tokenizer_class=_Gemma4Tokenizer, has_start_token=True, has_end_token=False, pad_to_max_length=False, max_length=99999999, min_length=1, pad_left=True, disable_weights=True, start_token=2, tokenizer_data=tokenizer_data) + + def decode(self, token_ids, **kwargs): + text = super().decode(token_ids, skip_special_tokens=False) + # Translate thinking channel markers to standard / tags + text = text.replace("<|channel>thought\n", "\n") + text = text.replace("", "") + # Strip remaining special tokens + text = text.replace("", "").replace("", "").strip() + return text + + +class Gemma4Tokenizer(sd1_clip.SD1Tokenizer): + tokenizer_class = Gemma4SDTokenizer + def __init__(self, embedding_directory=None, tokenizer_data={}): + super().__init__(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data, name="gemma4", tokenizer=self.tokenizer_class) + + +# Model wrappers +class Gemma4Model(sd1_clip.SDClipModel): + model_class = None + def __init__(self, device="cpu", layer="all", layer_idx=None, dtype=None, attention_mask=True, model_options={}): + self.dtypes = set() + self.dtypes.add(dtype) + super().__init__(device=device, layer=layer, layer_idx=layer_idx, textmodel_json_config={}, dtype=dtype, special_tokens={"start": 2, "pad": 0}, layer_norm_hidden_state=False, model_class=self.model_class, enable_attention_masks=attention_mask, return_attention_masks=attention_mask, model_options=model_options) + + def process_tokens(self, tokens, device): + embeds, _, _, _ = super().process_tokens(tokens, device) + return embeds + + def generate(self, tokens, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed, presence_penalty=0.0): + if isinstance(tokens, dict): + tokens = next(iter(tokens.values())) + tokens_only = [[t[0] for t in b] for b in tokens] + embeds, _, _, embeds_info = sd1_clip.SDClipModel.process_tokens(self, tokens_only, self.execution_device) + seq_len = embeds.shape[1] + ids = [0] * seq_len + expanded_idx = 0 + embed_map = {info["index"]: info["size"] for info in embeds_info} + for t in tokens_only[0]: + if expanded_idx in embed_map: + expanded_idx += embed_map[expanded_idx] + elif isinstance(t, int): + if expanded_idx < seq_len: + ids[expanded_idx] = t + expanded_idx += 1 + else: + expanded_idx += 1 + initial_token_ids = [ids] + input_ids = torch.tensor(initial_token_ids, device=self.execution_device) + return self.transformer.generate(embeds, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed, initial_tokens=initial_token_ids[0], presence_penalty=presence_penalty, initial_input_ids=input_ids) + + +def gemma4_te(dtype_llama=None, llama_quantization_metadata=None, model_class=None): + clip_model = type('Gemma4Model_', (Gemma4Model,), {'model_class': model_class}) + class Gemma4TEModel_(sd1_clip.SD1ClipModel): + def __init__(self, device="cpu", dtype=None, model_options={}): + if llama_quantization_metadata is not None: + model_options = model_options.copy() + model_options["quantization_metadata"] = llama_quantization_metadata + if dtype_llama is not None: + dtype = dtype_llama + super().__init__(device=device, dtype=dtype, name="gemma4", clip_model=clip_model, model_options=model_options) + return Gemma4TEModel_ + + +# Variants + +def _make_variant(config_cls): + audio = config_cls.audio_config is not None + bases = (Gemma4AudioMixin, Gemma4Base) if audio else (Gemma4Base,) + class Variant(*bases): + def __init__(self, config_dict, dtype, device, operations): + super().__init__() + self._init_model(config_cls(**config_dict), dtype, device, operations) + if audio: + self._init_audio(self.model.config, dtype, device, operations) + embedding_size = config_cls.hidden_size + if embedding_size != Gemma4SDTokenizer.embedding_size: + tok_cls = type('T', (Gemma4SDTokenizer,), {'embedding_size': embedding_size}) + class Tokenizer(Gemma4Tokenizer): + tokenizer_class = tok_cls + Variant.tokenizer = Tokenizer + else: + Variant.tokenizer = Gemma4Tokenizer + return Variant + +Gemma4_E4B = _make_variant(Gemma4Config) +Gemma4_E2B = _make_variant(Gemma4_E2B_Config) +Gemma4_31B = _make_variant(Gemma4_31B_Config) diff --git a/comfy/text_encoders/hidream_o1.py b/comfy/text_encoders/hidream_o1.py new file mode 100644 index 000000000..5d287b784 --- /dev/null +++ b/comfy/text_encoders/hidream_o1.py @@ -0,0 +1,119 @@ +"""HiDream-O1-Image tokenizer-only text encoder. + +The real Qwen3-VL backbone runs inside diffusion_model.* every step, so this +module just tokenizes the prompt into text_input_ids and emits them as +conditioning. Position ids / token_types / vinput_mask depend on target H/W +and are built later in model_base.HiDreamO1.extra_conds. +""" + +import os + +import torch +from transformers import Qwen2Tokenizer + +from comfy import sd1_clip + + +# Qwen3-VL special tokens +IM_START_ID = 151644 +IM_END_ID = 151645 +ASSISTANT_ID = 77091 +USER_ID = 872 +NEWLINE_ID = 198 +VISION_START_ID = 151652 +VISION_END_ID = 151653 +IMAGE_TOKEN_ID = 151655 +VIDEO_TOKEN_ID = 151656 +# HiDream-O1-specific tokens +BOI_TOKEN_ID = 151669 +BOR_TOKEN_ID = 151670 +EOR_TOKEN_ID = 151671 +BOT_TOKEN_ID = 151672 +TMS_TOKEN_ID = 151673 + + +class HiDreamO1QwenTokenizer(sd1_clip.SDTokenizer): + def __init__(self, embedding_directory=None, tokenizer_data={}): + tokenizer_path = os.path.join( + os.path.dirname(os.path.realpath(__file__)), "qwen25_tokenizer" + ) + super().__init__( + tokenizer_path, + pad_with_end=False, + embedding_size=4096, + embedding_key="hidream_o1", + tokenizer_class=Qwen2Tokenizer, + has_start_token=False, + has_end_token=False, + pad_to_max_length=False, + max_length=99999999, + min_length=1, + pad_token=151643, + tokenizer_data=tokenizer_data, + ) + + +class HiDreamO1Tokenizer(sd1_clip.SD1Tokenizer): + """Wraps prompt in the upstream chat template ending with boi/tms markers. + Image tokens get spliced in at sample time once target H/W is known. + """ + + def __init__(self, embedding_directory=None, tokenizer_data={}): + super().__init__( + embedding_directory=embedding_directory, + tokenizer_data=tokenizer_data, + name="hidream_o1", + tokenizer=HiDreamO1QwenTokenizer, + ) + + def tokenize_with_weights(self, text, return_word_ids=False, **kwargs): + text_tokens_dict = super().tokenize_with_weights( + text, return_word_ids=return_word_ids, disable_weights=True, **kwargs + ) + text_tuples = text_tokens_dict["hidream_o1"][0] + text_tuples = [t for t in text_tuples if int(t[0]) != 151643] # strip pad + + # <|im_start|>user\n{text}<|im_end|>\n<|im_start|>assistant\n<|boi|><|tms|> + def tok(tid): + return (tid, 1.0) if not return_word_ids else (tid, 1.0, 0) + + prefix = [tok(IM_START_ID), tok(USER_ID), tok(NEWLINE_ID)] + suffix = [ + tok(IM_END_ID), tok(NEWLINE_ID), + tok(IM_START_ID), tok(ASSISTANT_ID), tok(NEWLINE_ID), + tok(BOI_TOKEN_ID), tok(TMS_TOKEN_ID), + ] + full = prefix + list(text_tuples) + suffix + return {"hidream_o1": [full]} + + +class HiDreamO1TE(torch.nn.Module): + """Passthrough TE: emits int token ids; the Qwen3-VL backbone in diffusion_model does the actual encoding.""" + + def __init__(self, device="cpu", dtype=None, model_options={}): + super().__init__() + self.dtypes = {torch.float32} + self.disable_offload = True # skips dynamic VRAM management for this zero-parameter module + self.device = torch.device("cpu") if device is None else torch.device(device) + + def encode_token_weights(self, token_weight_pairs): + tok_pairs = token_weight_pairs["hidream_o1"][0] + ids = [int(t[0]) for t in tok_pairs] + input_ids = torch.tensor([ids], dtype=torch.long) + # Surrogate keeps the cross_attn slot non-empty for CONDITIONING + # plumbing; the model reads text_input_ids out of `extra` instead. + cross_attn = input_ids.unsqueeze(-1).to(torch.float32) + extra = {"text_input_ids": input_ids} + return cross_attn, None, extra + + def load_sd(self, sd): + return [] + + def get_sd(self): + return {} + + def reset_clip_options(self): + pass + + def set_clip_options(self, options): + pass diff --git a/comfy/text_encoders/llama.py b/comfy/text_encoders/llama.py index 06f2fbf74..5087228ca 100644 --- a/comfy/text_encoders/llama.py +++ b/comfy/text_encoders/llama.py @@ -60,6 +60,30 @@ class Mistral3Small24BConfig: final_norm: bool = True lm_head: bool = False +@dataclass +class Ministral3_3BConfig: + vocab_size: int = 131072 + hidden_size: int = 3072 + intermediate_size: int = 9216 + num_hidden_layers: int = 26 + num_attention_heads: int = 32 + num_key_value_heads: int = 8 + max_position_embeddings: int = 262144 + rms_norm_eps: float = 1e-5 + rope_theta: float = 1000000.0 + transformer_type: str = "llama" + head_dim = 128 + rms_norm_add = False + mlp_activation = "silu" + qkv_bias = False + rope_dims = None + q_norm = None + k_norm = None + rope_scale = None + final_norm: bool = True + lm_head: bool = False + stop_tokens = [2] + @dataclass class Qwen25_3BConfig: vocab_size: int = 151936 @@ -373,7 +397,7 @@ class RMSNorm(nn.Module): -def precompute_freqs_cis(head_dim, position_ids, theta, rope_scale=None, rope_dims=None, device=None): +def precompute_freqs_cis(head_dim, position_ids, theta, rope_scale=None, rope_dims=None, device=None, interleaved_mrope=False): if not isinstance(theta, list): theta = [theta] @@ -391,16 +415,27 @@ def precompute_freqs_cis(head_dim, position_ids, theta, rope_scale=None, rope_di inv_freq_expanded = inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1) position_ids_expanded = position_ids[:, None, :].float() freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2) - emb = torch.cat((freqs, freqs), dim=-1) - cos = emb.cos() - sin = emb.sin() - if rope_dims is not None and position_ids.shape[0] > 1: - mrope_section = rope_dims * 2 - cos = torch.cat([m[i % 3] for i, m in enumerate(cos.split(mrope_section, dim=-1))], dim=-1).unsqueeze(0) - sin = torch.cat([m[i % 3] for i, m in enumerate(sin.split(mrope_section, dim=-1))], dim=-1).unsqueeze(0) + if rope_dims is not None and position_ids.shape[0] > 1 and interleaved_mrope: + # Qwen3-VL interleaved MRoPE: T-freqs by default, H/W replace every 3rd dim. + freqs_inter = freqs[0].clone() + for axis_idx, offset in ((1, 1), (2, 2)): + length = rope_dims[axis_idx] * 3 + idx = slice(offset, length, 3) + freqs_inter[..., idx] = freqs[axis_idx, ..., idx] + emb = torch.cat((freqs_inter, freqs_inter), dim=-1) + cos = emb.cos().unsqueeze(0) + sin = emb.sin().unsqueeze(0) else: - cos = cos.unsqueeze(1) - sin = sin.unsqueeze(1) + emb = torch.cat((freqs, freqs), dim=-1) + cos = emb.cos() + sin = emb.sin() + if rope_dims is not None and position_ids.shape[0] > 1: + mrope_section = rope_dims * 2 + cos = torch.cat([m[i % 3] for i, m in enumerate(cos.split(mrope_section, dim=-1))], dim=-1).unsqueeze(0) + sin = torch.cat([m[i % 3] for i, m in enumerate(sin.split(mrope_section, dim=-1))], dim=-1).unsqueeze(0) + else: + cos = cos.unsqueeze(1) + sin = sin.unsqueeze(1) sin_split = sin.shape[-1] // 2 out.append((cos, sin[..., : sin_split], -sin[..., sin_split :])) @@ -497,7 +532,7 @@ class Attention(nn.Module): else: present_key_value = (xk, xv, index + num_tokens) - if sliding_window is not None and xk.shape[2] > sliding_window: + if sliding_window is not None and xk.shape[2] > sliding_window and seq_length == 1: xk = xk[:, :, -sliding_window:] xv = xv[:, :, -sliding_window:] attention_mask = attention_mask[..., -sliding_window:] if attention_mask is not None else None @@ -509,12 +544,12 @@ class Attention(nn.Module): return self.o_proj(output), present_key_value class MLP(nn.Module): - def __init__(self, config: Llama2Config, device=None, dtype=None, ops: Any = None): + def __init__(self, config: Llama2Config, device=None, dtype=None, ops: Any = None, intermediate_size=None): super().__init__() - ops = ops or nn - self.gate_proj = ops.Linear(config.hidden_size, config.intermediate_size, bias=False, device=device, dtype=dtype) - self.up_proj = ops.Linear(config.hidden_size, config.intermediate_size, bias=False, device=device, dtype=dtype) - self.down_proj = ops.Linear(config.intermediate_size, config.hidden_size, bias=False, device=device, dtype=dtype) + intermediate_size = intermediate_size or config.intermediate_size + self.gate_proj = ops.Linear(config.hidden_size, intermediate_size, bias=False, device=device, dtype=dtype) + self.up_proj = ops.Linear(config.hidden_size, intermediate_size, bias=False, device=device, dtype=dtype) + self.down_proj = ops.Linear(intermediate_size, config.hidden_size, bias=False, device=device, dtype=dtype) if config.mlp_activation == "silu": self.activation = torch.nn.functional.silu elif config.mlp_activation == "gelu_pytorch_tanh": @@ -623,24 +658,25 @@ class TransformerBlockGemma2(nn.Module): return x, present_key_value +def _make_scaled_embedding(ops, vocab_size, hidden_size, scale, device, dtype): + class ScaledEmbedding(ops.Embedding): + def forward(self, input_ids, out_dtype=None): + return super().forward(input_ids, out_dtype=out_dtype) * scale + return ScaledEmbedding(vocab_size, hidden_size, device=device, dtype=dtype) + + class Llama2_(nn.Module): def __init__(self, config, device=None, dtype=None, ops=None): super().__init__() self.config = config self.vocab_size = config.vocab_size - self.embed_tokens = ops.Embedding( - config.vocab_size, - config.hidden_size, - device=device, - dtype=dtype - ) if self.config.transformer_type == "gemma2" or self.config.transformer_type == "gemma3": transformer = TransformerBlockGemma2 - self.normalize_in = True + self.embed_tokens = _make_scaled_embedding(ops, config.vocab_size, config.hidden_size, config.hidden_size ** 0.5, device, dtype) else: transformer = TransformerBlock - self.normalize_in = False + self.embed_tokens = ops.Embedding(config.vocab_size, config.hidden_size, device=device, dtype=dtype) self.layers = nn.ModuleList([ transformer(config, index=i, device=device, dtype=dtype, ops=ops) @@ -664,17 +700,15 @@ class Llama2_(nn.Module): self.config.rope_theta, self.config.rope_scale, self.config.rope_dims, + interleaved_mrope=getattr(self.config, "interleaved_mrope", False), device=device) - def forward(self, x, attention_mask=None, embeds=None, num_tokens=None, intermediate_output=None, final_layer_norm_intermediate=True, dtype=None, position_ids=None, embeds_info=[], past_key_values=None): + def forward(self, x, attention_mask=None, embeds=None, num_tokens=None, intermediate_output=None, final_layer_norm_intermediate=True, dtype=None, position_ids=None, embeds_info=[], past_key_values=None, input_ids=None): if embeds is not None: x = embeds else: x = self.embed_tokens(x, out_dtype=dtype) - if self.normalize_in: - x *= self.config.hidden_size ** 0.5 - seq_len = x.shape[1] past_len = 0 if past_key_values is not None and len(past_key_values) > 0: @@ -826,7 +860,7 @@ class BaseGenerate: torch.empty([batch, model_config.num_key_value_heads, max_cache_len, model_config.head_dim], device=device, dtype=execution_dtype), 0)) return past_key_values - def generate(self, embeds=None, do_sample=True, max_length=256, temperature=1.0, top_k=50, top_p=0.9, min_p=0.0, repetition_penalty=1.0, seed=42, stop_tokens=None, initial_tokens=[], execution_dtype=None, min_tokens=0, presence_penalty=0.0): + def generate(self, embeds=None, do_sample=True, max_length=256, temperature=1.0, top_k=50, top_p=0.9, min_p=0.0, repetition_penalty=1.0, seed=42, stop_tokens=None, initial_tokens=[], execution_dtype=None, min_tokens=0, presence_penalty=0.0, initial_input_ids=None): device = embeds.device if stop_tokens is None: @@ -851,14 +885,16 @@ class BaseGenerate: pbar = comfy.utils.ProgressBar(max_length) # Generation loop + current_input_ids = initial_input_ids for step in tqdm(range(max_length), desc="Generating tokens"): - x, _, past_key_values = self.model.forward(None, embeds=embeds, attention_mask=None, past_key_values=past_key_values) + x, _, past_key_values = self.model.forward(None, embeds=embeds, attention_mask=None, past_key_values=past_key_values, input_ids=current_input_ids) logits = self.logits(x)[:, -1] next_token = self.sample_token(logits, temperature, top_k, top_p, min_p, repetition_penalty, initial_tokens + generated_token_ids, generator, do_sample=do_sample, presence_penalty=presence_penalty) token_id = next_token[0].item() generated_token_ids.append(token_id) embeds = self.model.embed_tokens(next_token).to(execution_dtype) + current_input_ids = next_token if initial_input_ids is not None else None pbar.update(1) if token_id in stop_tokens: @@ -946,6 +982,15 @@ class Mistral3Small24B(BaseLlama, torch.nn.Module): self.model = Llama2_(config, device=device, dtype=dtype, ops=operations) self.dtype = dtype +class Ministral3_3B(BaseLlama, BaseQwen3, BaseGenerate, torch.nn.Module): + def __init__(self, config_dict, dtype, device, operations): + super().__init__() + config = Ministral3_3BConfig(**config_dict) + self.num_layers = config.num_hidden_layers + + self.model = Llama2_(config, device=device, dtype=dtype, ops=operations) + self.dtype = dtype + class Qwen25_3B(BaseLlama, torch.nn.Module): def __init__(self, config_dict, dtype, device, operations): super().__init__() diff --git a/comfy/text_encoders/lt.py b/comfy/text_encoders/lt.py index 5aee1f4c0..bc5cbae28 100644 --- a/comfy/text_encoders/lt.py +++ b/comfy/text_encoders/lt.py @@ -93,8 +93,7 @@ class Gemma3_12BModel(sd1_clip.SDClipModel): def generate(self, tokens, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed, presence_penalty): tokens_only = [[t[0] for t in b] for b in tokens] - embeds, _, _, embeds_info = self.process_tokens(tokens_only, self.execution_device) - comfy.utils.normalize_image_embeddings(embeds, embeds_info, self.transformer.model.config.hidden_size ** 0.5) + embeds, _, _, _ = self.process_tokens(tokens_only, self.execution_device) return self.transformer.generate(embeds, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed, stop_tokens=[106], presence_penalty=presence_penalty) # 106 is class DualLinearProjection(torch.nn.Module): diff --git a/comfy/text_encoders/lumina2.py b/comfy/text_encoders/lumina2.py index 01ebdfabe..b1f1dbb9f 100644 --- a/comfy/text_encoders/lumina2.py +++ b/comfy/text_encoders/lumina2.py @@ -50,8 +50,7 @@ class Gemma3_4B_Vision_Model(sd1_clip.SDClipModel): super().__init__(device=device, layer=layer, layer_idx=layer_idx, textmodel_json_config={}, dtype=dtype, special_tokens={"start": 2, "pad": 0}, layer_norm_hidden_state=False, model_class=comfy.text_encoders.llama.Gemma3_4B_Vision, enable_attention_masks=attention_mask, return_attention_masks=attention_mask, model_options=model_options) def process_tokens(self, tokens, device): - embeds, _, _, embeds_info = super().process_tokens(tokens, device) - comfy.utils.normalize_image_embeddings(embeds, embeds_info, self.transformer.model.config.hidden_size ** 0.5) + embeds, _, _, _ = super().process_tokens(tokens, device) return embeds class LuminaModel(sd1_clip.SD1ClipModel): diff --git a/comfy/text_encoders/qwen35.py b/comfy/text_encoders/qwen35.py index ce9b07464..b022009b1 100644 --- a/comfy/text_encoders/qwen35.py +++ b/comfy/text_encoders/qwen35.py @@ -408,8 +408,6 @@ class Qwen35Transformer(Llama2_): nn.Module.__init__(self) self.config = config self.vocab_size = config.vocab_size - self.normalize_in = False - self.embed_tokens = ops.Embedding(config.vocab_size, config.hidden_size, device=device, dtype=dtype) self.layers = nn.ModuleList([ Qwen35TransformerBlock(config, index=i, device=device, dtype=dtype, ops=ops) @@ -453,9 +451,8 @@ class Qwen35VisionPatchEmbed(nn.Module): self.proj = ops.Conv3d(self.in_channels, self.embed_dim, kernel_size=kernel_size, stride=kernel_size, bias=True, device=device, dtype=dtype) def forward(self, x): - target_dtype = self.proj.weight.dtype x = x.view(-1, self.in_channels, self.temporal_patch_size, self.patch_size, self.patch_size) - return self.proj(x.to(target_dtype)).view(-1, self.embed_dim) + return self.proj(x).view(-1, self.embed_dim) class Qwen35VisionMLP(nn.Module): @@ -653,7 +650,7 @@ class Qwen35VisionModel(nn.Module): x = self.patch_embed(x) pos_embeds = self.fast_pos_embed_interpolate(grid_thw).to(x.device) x = x + pos_embeds - rotary_pos_emb = self.rot_pos_emb(grid_thw) + rotary_pos_emb = self.rot_pos_emb(grid_thw).to(x.device) seq_len = x.shape[0] x = x.reshape(seq_len, -1) rotary_pos_emb = rotary_pos_emb.reshape(seq_len, -1) diff --git a/comfy/text_encoders/sam3_clip.py b/comfy/text_encoders/sam3_clip.py new file mode 100644 index 000000000..11cb7d9db --- /dev/null +++ b/comfy/text_encoders/sam3_clip.py @@ -0,0 +1,97 @@ +import re +from comfy import sd1_clip + +SAM3_CLIP_CONFIG = { + "architectures": ["CLIPTextModel"], + "hidden_act": "quick_gelu", + "hidden_size": 1024, + "intermediate_size": 4096, + "num_attention_heads": 16, + "num_hidden_layers": 24, + "max_position_embeddings": 32, + "projection_dim": 512, + "vocab_size": 49408, + "layer_norm_eps": 1e-5, + "eos_token_id": 49407, +} + + +class SAM3ClipModel(sd1_clip.SDClipModel): + def __init__(self, device="cpu", dtype=None, model_options={}): + super().__init__(device=device, dtype=dtype, max_length=32, layer="last", textmodel_json_config=SAM3_CLIP_CONFIG, special_tokens={"start": 49406, "end": 49407, "pad": 0}, return_projected_pooled=False, return_attention_masks=True, enable_attention_masks=True, model_options=model_options) + + +class SAM3Tokenizer(sd1_clip.SDTokenizer): + def __init__(self, embedding_directory=None, tokenizer_data={}): + super().__init__(max_length=32, pad_with_end=False, pad_token=0, embedding_directory=embedding_directory, embedding_size=1024, embedding_key="sam3_clip", tokenizer_data=tokenizer_data) + self.disable_weights = True + + +def _parse_prompts(text): + """Split comma-separated prompts with optional :N max detections per category""" + text = text.replace("(", "").replace(")", "") + parts = [p.strip() for p in text.split(",") if p.strip()] + result = [] + for part in parts: + m = re.match(r'^(.+?)\s*:\s*([\d.]+)\s*$', part) + if m: + text_part = m.group(1).strip() + val = m.group(2) + max_det = max(1, round(float(val))) + result.append((text_part, max_det)) + else: + result.append((part, 1)) + return result + + +class SAM3TokenizerWrapper(sd1_clip.SD1Tokenizer): + def __init__(self, embedding_directory=None, tokenizer_data={}): + super().__init__(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data, clip_name="l", tokenizer=SAM3Tokenizer, name="sam3_clip") + + def tokenize_with_weights(self, text: str, return_word_ids=False, **kwargs): + parsed = _parse_prompts(text) + if len(parsed) <= 1 and (not parsed or parsed[0][1] == 1): + return super().tokenize_with_weights(text, return_word_ids, **kwargs) + # Tokenize each prompt part separately, store per-part batches and metadata + inner = getattr(self, self.clip) + per_prompt = [] + for prompt_text, max_det in parsed: + batches = inner.tokenize_with_weights(prompt_text, return_word_ids, **kwargs) + per_prompt.append((batches, max_det)) + # Main output uses first prompt's tokens (for compatibility) + out = {self.clip_name: per_prompt[0][0], "sam3_per_prompt": per_prompt} + return out + + +class SAM3ClipModelWrapper(sd1_clip.SD1ClipModel): + def __init__(self, device="cpu", dtype=None, model_options={}, **kwargs): + super().__init__(device=device, dtype=dtype, model_options=model_options, clip_name="l", clip_model=SAM3ClipModel, name="sam3_clip") + + def encode_token_weights(self, token_weight_pairs): + per_prompt = token_weight_pairs.pop("sam3_per_prompt", None) + if per_prompt is None: + return super().encode_token_weights(token_weight_pairs) + + # Encode each prompt separately, pack into extra dict + inner = getattr(self, self.clip) + multi_cond = [] + first_pooled = None + for batches, max_det in per_prompt: + out = inner.encode_token_weights(batches) + cond, pooled = out[0], out[1] + extra = out[2] if len(out) > 2 else {} + if first_pooled is None: + first_pooled = pooled + multi_cond.append({ + "cond": cond, + "attention_mask": extra.get("attention_mask"), + "max_detections": max_det, + }) + + # Return first prompt as main (for non-SAM3 consumers), all prompts in metadata + main = multi_cond[0] + main_extra = {} + if main["attention_mask"] is not None: + main_extra["attention_mask"] = main["attention_mask"] + main_extra["sam3_multi_cond"] = multi_cond + return (main["cond"], first_pooled, main_extra) diff --git a/comfy/utils.py b/comfy/utils.py index 78c491b98..66682690a 100644 --- a/comfy/utils.py +++ b/comfy/utils.py @@ -1164,12 +1164,18 @@ def tiled_scale_multidim(samples, function, tile=(64, 64), overlap=8, upscale_am o = out o_d = out_div + ps_view = ps + mask_view = mask for d in range(dims): - o = o.narrow(d + 2, upscaled[d], mask.shape[d + 2]) - o_d = o_d.narrow(d + 2, upscaled[d], mask.shape[d + 2]) + l = min(ps_view.shape[d + 2], o.shape[d + 2] - upscaled[d]) + o = o.narrow(d + 2, upscaled[d], l) + o_d = o_d.narrow(d + 2, upscaled[d], l) + if l < ps_view.shape[d + 2]: + ps_view = ps_view.narrow(d + 2, 0, l) + mask_view = mask_view.narrow(d + 2, 0, l) - o.add_(ps * mask) - o_d.add_(mask) + o.add_(ps_view * mask_view) + o_d.add_(mask_view) if pbar is not None: pbar.update(1) @@ -1196,7 +1202,7 @@ def model_trange(*args, **kwargs): pbar.i1_time = time.time() pbar.set_postfix_str(" Model Initialization complete! ") elif pbar._i == 2: - #bring forward the effective start time based the the diff between first and second iteration + #bring forward the effective start time based the diff between first and second iteration #to attempt to remove load overhead from the final step rate estimate. pbar.start_t = pbar.i1_time - (time.time() - pbar.i1_time) pbar.set_postfix_str("") @@ -1390,7 +1396,7 @@ def convert_old_quants(state_dict, model_prefix="", metadata={}): k_out = "{}.weight_scale".format(layer) if layer is not None: - layer_conf = {"format": "float8_e4m3fn"} # TODO: check if anyone did some non e4m3fn scaled checkpoints + layer_conf = {"format": "float8_e4m3fn"} if full_precision_matrix_mult: layer_conf["full_precision_matrix_mult"] = full_precision_matrix_mult layers[layer] = layer_conf @@ -1446,10 +1452,3 @@ def deepcopy_list_dict(obj, memo=None): memo[obj_id] = res return res -def normalize_image_embeddings(embeds, embeds_info, scale_factor): - """Normalize image embeddings to match text embedding scale""" - for info in embeds_info: - if info.get("type") == "image": - start_idx = info["index"] - end_idx = start_idx + info["size"] - embeds[:, start_idx:end_idx, :] /= scale_factor diff --git a/comfy_api/feature_flags.py b/comfy_api/feature_flags.py index 9f6918315..adb5a3144 100644 --- a/comfy_api/feature_flags.py +++ b/comfy_api/feature_flags.py @@ -5,12 +5,95 @@ This module handles capability negotiation between frontend and backend, allowing graceful protocol evolution while maintaining backward compatibility. """ -from typing import Any +import logging +from typing import Any, TypedDict from comfy.cli_args import args + +class FeatureFlagInfo(TypedDict): + type: str + default: Any + description: str + + +# Registry of known CLI-settable feature flags. +# Launchers can query this via --list-feature-flags to discover valid flags. +CLI_FEATURE_FLAG_REGISTRY: dict[str, FeatureFlagInfo] = { + "show_signin_button": { + "type": "bool", + "default": False, + "description": "Show the sign-in button in the frontend even when not signed in", + }, +} + + +def _coerce_bool(v: str) -> bool: + """Strict bool coercion: only 'true'/'false' (case-insensitive). + + Anything else raises ValueError so the caller can warn and drop the flag, + rather than silently treating typos like 'ture' or 'yes' as False. + """ + lower = v.lower() + if lower == "true": + return True + if lower == "false": + return False + raise ValueError(f"expected 'true' or 'false', got {v!r}") + + +_COERCE_FNS: dict[str, Any] = { + "bool": _coerce_bool, + "int": lambda v: int(v), + "float": lambda v: float(v), +} + + +def _coerce_flag_value(key: str, raw_value: str) -> Any: + """Coerce a raw string value using the registry type, or keep as string. + + Returns the raw string if the key is unregistered or the type is unknown. + Raises ValueError/TypeError if the key is registered with a known type but + the value cannot be coerced; callers are expected to warn and drop the flag. + """ + info = CLI_FEATURE_FLAG_REGISTRY.get(key) + if info is None: + return raw_value + coerce = _COERCE_FNS.get(info["type"]) + if coerce is None: + return raw_value + return coerce(raw_value) + + +def _parse_cli_feature_flags() -> dict[str, Any]: + """Parse --feature-flag key=value pairs from CLI args into a dict. + + Items without '=' default to the value 'true' (bare flag form). + Flags whose value cannot be coerced to the registered type are dropped + with a warning, so a typo like '--feature-flag some_bool=ture' does not + silently take effect as the wrong value. + """ + result: dict[str, Any] = {} + for item in getattr(args, "feature_flag", []): + key, sep, raw_value = item.partition("=") + key = key.strip() + if not key: + continue + if not sep: + raw_value = "true" + try: + result[key] = _coerce_flag_value(key, raw_value.strip()) + except (ValueError, TypeError) as e: + info = CLI_FEATURE_FLAG_REGISTRY.get(key, {}) + logging.warning( + "Could not coerce --feature-flag %s=%r to %s (%s); dropping flag.", + key, raw_value.strip(), info.get("type", "?"), e, + ) + return result + + # Default server capabilities -SERVER_FEATURE_FLAGS: dict[str, Any] = { +_CORE_FEATURE_FLAGS: dict[str, Any] = { "supports_preview_metadata": True, "max_upload_size": args.max_upload_size * 1024 * 1024, # Convert MB to bytes "extension": {"manager": {"supports_v4": True}}, @@ -18,6 +101,11 @@ SERVER_FEATURE_FLAGS: dict[str, Any] = { "assets": args.enable_assets, } +# CLI-provided flags cannot overwrite core flags +_cli_flags = {k: v for k, v in _parse_cli_feature_flags().items() if k not in _CORE_FEATURE_FLAGS} + +SERVER_FEATURE_FLAGS: dict[str, Any] = {**_CORE_FEATURE_FLAGS, **_cli_flags} + def get_connection_feature( sockets_metadata: dict[str, dict[str, Any]], diff --git a/comfy_api/input/__init__.py b/comfy_api/input/__init__.py index 16d4acfd1..dc33533cc 100644 --- a/comfy_api/input/__init__.py +++ b/comfy_api/input/__init__.py @@ -9,6 +9,7 @@ from comfy_api.latest._input import ( CurveInput, MonotoneCubicCurve, LinearCurve, + RangeInput, ) __all__ = [ @@ -21,4 +22,5 @@ __all__ = [ "CurveInput", "MonotoneCubicCurve", "LinearCurve", + "RangeInput", ] diff --git a/comfy_api/latest/_input/__init__.py b/comfy_api/latest/_input/__init__.py index 05cd3d40a..f0229717e 100644 --- a/comfy_api/latest/_input/__init__.py +++ b/comfy_api/latest/_input/__init__.py @@ -1,5 +1,6 @@ from .basic_types import ImageInput, AudioInput, MaskInput, LatentInput from .curve_types import CurvePoint, CurveInput, MonotoneCubicCurve, LinearCurve +from .range_types import RangeInput from .video_types import VideoInput __all__ = [ @@ -12,4 +13,5 @@ __all__ = [ "CurveInput", "MonotoneCubicCurve", "LinearCurve", + "RangeInput", ] diff --git a/comfy_api/latest/_input/range_types.py b/comfy_api/latest/_input/range_types.py new file mode 100644 index 000000000..f4c5cb290 --- /dev/null +++ b/comfy_api/latest/_input/range_types.py @@ -0,0 +1,70 @@ +from __future__ import annotations + +import logging +import math +import numpy as np + +logger = logging.getLogger(__name__) + + +class RangeInput: + """Represents a levels/range adjustment: input range [min, max] with + optional midpoint (gamma control). + + Generates a 1D LUT identical to GIMP's levels mapping: + 1. Normalize input to [0, 1] using [min, max] + 2. Apply gamma correction: pow(value, 1/gamma) + 3. Clamp to [0, 1] + + The midpoint field is a position in [0, 1] representing where the + midtone falls within [min, max]. It maps to gamma via: + gamma = -log2(midpoint) + So midpoint=0.5 → gamma=1.0 (linear). + """ + + def __init__(self, min_val: float, max_val: float, midpoint: float | None = None): + self.min_val = min_val + self.max_val = max_val + self.midpoint = midpoint + + @staticmethod + def from_raw(data) -> RangeInput: + if isinstance(data, RangeInput): + return data + if isinstance(data, dict): + return RangeInput( + min_val=float(data.get("min", 0.0)), + max_val=float(data.get("max", 1.0)), + midpoint=float(data["midpoint"]) if data.get("midpoint") is not None else None, + ) + raise TypeError(f"Cannot convert {type(data)} to RangeInput") + + def to_lut(self, size: int = 256) -> np.ndarray: + """Generate a float64 lookup table mapping [0, 1] input through this + levels adjustment. + + The LUT maps normalized input values (0..1) to output values (0..1), + matching the GIMP levels formula. + """ + xs = np.linspace(0.0, 1.0, size, dtype=np.float64) + + in_range = self.max_val - self.min_val + if abs(in_range) < 1e-10: + return np.where(xs >= self.min_val, 1.0, 0.0).astype(np.float64) + + # Normalize: map [min, max] → [0, 1] + result = (xs - self.min_val) / in_range + result = np.clip(result, 0.0, 1.0) + + # Gamma correction from midpoint + if self.midpoint is not None and self.midpoint > 0 and self.midpoint != 0.5: + gamma = max(-math.log2(self.midpoint), 0.001) + inv_gamma = 1.0 / gamma + mask = result > 0 + result[mask] = np.power(result[mask], inv_gamma) + + return result + + def __repr__(self) -> str: + mid = f", midpoint={self.midpoint}" if self.midpoint is not None else "" + return f"RangeInput(min={self.min_val}, max={self.max_val}{mid})" diff --git a/comfy_api/latest/_input_impl/video_types.py b/comfy_api/latest/_input_impl/video_types.py index 1b4993aa7..942278d88 100644 --- a/comfy_api/latest/_input_impl/video_types.py +++ b/comfy_api/latest/_input_impl/video_types.py @@ -12,6 +12,7 @@ import numpy as np import math import torch from .._util import VideoContainer, VideoCodec, VideoComponents +import logging def container_to_output_format(container_format: str | None) -> str | None: @@ -238,64 +239,125 @@ class VideoFromFile(VideoInput): start_time = max(self._get_raw_duration() + self.__start_time, 0) else: start_time = self.__start_time + # Get video frames frames = [] + audio_frames = [] + alphas = None start_pts = int(start_time / video_stream.time_base) end_pts = int((start_time + self.__duration) / video_stream.time_base) - container.seek(start_pts, stream=video_stream) - for frame in container.decode(video_stream): - if frame.pts < start_pts: - continue - if self.__duration and frame.pts >= end_pts: - break - img = frame.to_ndarray(format='rgb24') # shape: (H, W, 3) - img = torch.from_numpy(img) / 255.0 # shape: (H, W, 3) - frames.append(img) - images = torch.stack(frames) if len(frames) > 0 else torch.zeros(0, 3, 0, 0) + if start_pts != 0: + container.seek(start_pts, stream=video_stream) + + image_format = 'gbrpf32le' + process_image_format = lambda a: a + audio = None + + streams = [video_stream] + has_first_audio_frame = False + checked_alpha = False + + # Default to False so we decode until EOF if duration is 0 + video_done = False + audio_done = True + + if len(container.streams.audio): + audio_stream = container.streams.audio[-1] + streams += [audio_stream] + resampler = av.audio.resampler.AudioResampler(format='fltp') + audio_done = False + + for packet in container.demux(*streams): + if video_done and audio_done: + break + + if packet.stream.type == "video": + if video_done: + continue + try: + for frame in packet.decode(): + if frame.pts < start_pts: + continue + if self.__duration and frame.pts >= end_pts: + video_done = True + break + + if not checked_alpha: + alpha_channel = False + for comp in frame.format.components: + if comp.is_alpha or frame.format.name == "pal8": + alphas = [] + alpha_channel = True + break + if frame.format.name in ("yuvj420p", "yuvj422p", "yuvj444p", "rgb24", "rgba", "pal8"): + process_image_format = lambda a: a.float() / 255.0 + if alpha_channel: + image_format = 'rgba' + else: + image_format = 'rgb24' + else: + process_image_format = lambda a: a + if alpha_channel: + image_format = 'gbrapf32le' + else: + image_format = 'gbrpf32le' + + checked_alpha = True + + img = frame.to_ndarray(format=image_format) # shape: (H, W, 4) + if frame.rotation != 0: + k = int(round(frame.rotation // 90)) + img = np.rot90(img, k=k, axes=(0, 1)).copy() + if alphas is None: + frames.append(torch.from_numpy(img)) + else: + frames.append(torch.from_numpy(img[..., :-1])) + alphas.append(torch.from_numpy(img[..., -1:])) + except av.error.InvalidDataError: + logging.info("pyav decode error") + + elif packet.stream.type == "audio": + if audio_done: + continue + + aframes = itertools.chain.from_iterable( + map(resampler.resample, packet.decode()) + ) + for frame in aframes: + if self.__duration and frame.time > start_time + self.__duration: + audio_done = True + break + + if not has_first_audio_frame: + offset_seconds = start_time - frame.pts * audio_stream.time_base + to_skip = max(0, int(offset_seconds * audio_stream.sample_rate)) + if to_skip < frame.samples: + has_first_audio_frame = True + audio_frames.append(frame.to_ndarray()[..., to_skip:]) + else: + audio_frames.append(frame.to_ndarray()) + + images = process_image_format(torch.stack(frames)) if len(frames) > 0 else torch.zeros(0, 0, 0, 3) + if alphas is not None: + alphas = process_image_format(torch.stack(alphas)) if len(alphas) > 0 else torch.zeros(0, 0, 0, 1) # Get frame rate frame_rate = Fraction(video_stream.average_rate) if video_stream.average_rate else Fraction(1) - # Get audio if available - audio = None - container.seek(start_pts, stream=video_stream) - # Use last stream for consistency - if len(container.streams.audio): - audio_stream = container.streams.audio[-1] - audio_frames = [] - resample = av.audio.resampler.AudioResampler(format='fltp').resample - frames = itertools.chain.from_iterable( - map(resample, container.decode(audio_stream)) - ) + if len(audio_frames) > 0: + audio_data = np.concatenate(audio_frames, axis=1) # shape: (channels, total_samples) + if self.__duration: + audio_data = audio_data[..., :int(self.__duration * audio_stream.sample_rate)] - has_first_frame = False - for frame in frames: - offset_seconds = start_time - frame.pts * audio_stream.time_base - to_skip = max(0, int(offset_seconds * audio_stream.sample_rate)) - if to_skip < frame.samples: - has_first_frame = True - break - if has_first_frame: - audio_frames.append(frame.to_ndarray()[..., to_skip:]) - - for frame in frames: - if self.__duration and frame.time > start_time + self.__duration: - break - audio_frames.append(frame.to_ndarray()) # shape: (channels, samples) - if len(audio_frames) > 0: - audio_data = np.concatenate(audio_frames, axis=1) # shape: (channels, total_samples) - if self.__duration: - audio_data = audio_data[..., :int(self.__duration * audio_stream.sample_rate)] - - audio_tensor = torch.from_numpy(audio_data).unsqueeze(0) # shape: (1, channels, total_samples) - audio = AudioInput({ - "waveform": audio_tensor, - "sample_rate": int(audio_stream.sample_rate) if audio_stream.sample_rate else 1, - }) + audio_tensor = torch.from_numpy(audio_data).unsqueeze(0) # shape: (1, channels, total_samples) + audio = AudioInput({ + "waveform": audio_tensor, + "sample_rate": int(audio_stream.sample_rate) if audio_stream.sample_rate else 1, + }) metadata = container.metadata - return VideoComponents(images=images, audio=audio, frame_rate=frame_rate, metadata=metadata) + return VideoComponents(images=images, alpha=alphas, audio=audio, frame_rate=frame_rate, metadata=metadata) def get_components(self) -> VideoComponents: if isinstance(self.__file, io.BytesIO): diff --git a/comfy_api/latest/_io.py b/comfy_api/latest/_io.py index fdeffea2d..5ed968960 100644 --- a/comfy_api/latest/_io.py +++ b/comfy_api/latest/_io.py @@ -17,6 +17,7 @@ if TYPE_CHECKING: from spandrel import ImageModelDescriptor from comfy.clip_vision import ClipVisionModel from comfy.clip_vision import Output as ClipVisionOutput_ + from comfy.bg_removal_model import BackgroundRemovalModel from comfy.controlnet import ControlNet from comfy.hooks import HookGroup, HookKeyframeGroup from comfy.model_patcher import ModelPatcher @@ -395,7 +396,6 @@ class Combo(ComfyTypeIO): @comfytype(io_type="COMBO") class MultiCombo(ComfyTypeI): '''Multiselect Combo input (dropdown for selecting potentially more than one value).''' - # TODO: something is wrong with the serialization, frontend does not recognize it as multiselect Type = list[str] class Input(Combo.Input): def __init__(self, id: str, options: list[str], display_name: str=None, optional=False, tooltip: str=None, lazy: bool=None, @@ -408,12 +408,14 @@ class MultiCombo(ComfyTypeI): self.default: list[str] def as_dict(self): - to_return = super().as_dict() | prune_dict({ - "multi_select": self.multiselect, - "placeholder": self.placeholder, - "chip": self.chip, + # Frontend expects `multi_select` to be an object config (not a boolean). + # Keep top-level `multiselect` from Combo.Input for backwards compatibility. + return super().as_dict() | prune_dict({ + "multi_select": prune_dict({ + "placeholder": self.placeholder, + "chip": self.chip, + }), }) - return to_return @comfytype(io_type="IMAGE") class Image(ComfyTypeIO): @@ -613,6 +615,11 @@ class Model(ComfyTypeIO): if TYPE_CHECKING: Type = ModelPatcher +@comfytype(io_type="BACKGROUND_REMOVAL") +class BackgroundRemoval(ComfyTypeIO): + if TYPE_CHECKING: + Type = BackgroundRemovalModel + @comfytype(io_type="CLIP_VISION") class ClipVision(ComfyTypeIO): if TYPE_CHECKING: @@ -1266,6 +1273,43 @@ class Histogram(ComfyTypeIO): Type = list[int] +@comfytype(io_type="RANGE") +class Range(ComfyTypeIO): + from comfy_api.input import RangeInput + if TYPE_CHECKING: + Type = RangeInput + + class Input(WidgetInput): + def __init__(self, id: str, display_name: str=None, optional=False, tooltip: str=None, + socketless: bool=True, default: dict=None, + display: str=None, + gradient_stops: list=None, + show_midpoint: bool=None, + midpoint_scale: str=None, + value_min: float=None, + value_max: float=None, + advanced: bool=None): + super().__init__(id, display_name, optional, tooltip, None, default, socketless, None, None, None, None, advanced) + if default is None: + self.default = {"min": 0.0, "max": 1.0} + self.display = display + self.gradient_stops = gradient_stops + self.show_midpoint = show_midpoint + self.midpoint_scale = midpoint_scale + self.value_min = value_min + self.value_max = value_max + + def as_dict(self): + return super().as_dict() | prune_dict({ + "display": self.display, + "gradient_stops": self.gradient_stops, + "show_midpoint": self.show_midpoint, + "midpoint_scale": self.midpoint_scale, + "value_min": self.value_min, + "value_max": self.value_max, + }) + + DYNAMIC_INPUT_LOOKUP: dict[str, Callable[[dict[str, Any], dict[str, Any], tuple[str, dict[str, Any]], str, list[str] | None], None]] = {} def register_dynamic_input_func(io_type: str, func: Callable[[dict[str, Any], dict[str, Any], tuple[str, dict[str, Any]], str, list[str] | None], None]): DYNAMIC_INPUT_LOOKUP[io_type] = func @@ -2219,6 +2263,7 @@ __all__ = [ "ModelPatch", "ClipVision", "ClipVisionOutput", + "BackgroundRemoval", "AudioEncoder", "AudioEncoderOutput", "StyleModel", @@ -2276,5 +2321,6 @@ __all__ = [ "BoundingBox", "Curve", "Histogram", + "Range", "NodeReplace", ] diff --git a/comfy_api/latest/_util/geometry_types.py b/comfy_api/latest/_util/geometry_types.py index b586fceb3..cdde60b10 100644 --- a/comfy_api/latest/_util/geometry_types.py +++ b/comfy_api/latest/_util/geometry_types.py @@ -12,9 +12,24 @@ class VOXEL: class MESH: - def __init__(self, vertices: torch.Tensor, faces: torch.Tensor): - self.vertices = vertices - self.faces = faces + def __init__(self, vertices: torch.Tensor, faces: torch.Tensor, + uvs: torch.Tensor | None = None, + vertex_colors: torch.Tensor | None = None, + texture: torch.Tensor | None = None, + vertex_counts: torch.Tensor | None = None, + face_counts: torch.Tensor | None = None): + + assert (vertex_counts is None) == (face_counts is None), \ + "vertex_counts and face_counts must be provided together (both or neither)" + self.vertices = vertices # vertices: (B, N, 3) + self.faces = faces # faces: (B, M, 3) + self.uvs = uvs # uvs: (B, N, 2) + self.vertex_colors = vertex_colors # vertex_colors: (B, N, 3 or 4) + self.texture = texture # texture: (B, H, W, 3) + # When vertices/faces are zero-padded to a common N/M across the batch (variable-size mesh batch), + # these hold the real per-item lengths (B,). None means rows are uniform and no slicing is needed. + self.vertex_counts = vertex_counts + self.face_counts = face_counts class File3D: diff --git a/comfy_api/latest/_util/video_types.py b/comfy_api/latest/_util/video_types.py index fd3b5a510..c92477f08 100644 --- a/comfy_api/latest/_util/video_types.py +++ b/comfy_api/latest/_util/video_types.py @@ -3,7 +3,7 @@ from dataclasses import dataclass from enum import Enum from fractions import Fraction from typing import Optional -from .._input import ImageInput, AudioInput +from .._input import ImageInput, AudioInput, MaskInput class VideoCodec(str, Enum): AUTO = "auto" @@ -48,5 +48,4 @@ class VideoComponents: frame_rate: Fraction audio: Optional[AudioInput] = None metadata: Optional[dict] = None - - + alpha: Optional[MaskInput] = None diff --git a/comfy_api_nodes/apis/anthropic.py b/comfy_api_nodes/apis/anthropic.py new file mode 100644 index 000000000..6cac537ea --- /dev/null +++ b/comfy_api_nodes/apis/anthropic.py @@ -0,0 +1,75 @@ +from enum import Enum +from typing import Literal + +from pydantic import BaseModel, Field + + +class AnthropicRole(str, Enum): + user = "user" + assistant = "assistant" + + +class AnthropicTextContent(BaseModel): + type: Literal["text"] = "text" + text: str = Field(...) + + +class AnthropicImageSourceBase64(BaseModel): + type: Literal["base64"] = "base64" + media_type: str = Field(..., description="MIME type of the image, e.g. image/png, image/jpeg") + data: str = Field(..., description="Base64-encoded image data") + + +class AnthropicImageSourceUrl(BaseModel): + type: Literal["url"] = "url" + url: str = Field(...) + + +class AnthropicImageContent(BaseModel): + type: Literal["image"] = "image" + source: AnthropicImageSourceBase64 | AnthropicImageSourceUrl = Field(...) + + +class AnthropicMessage(BaseModel): + role: AnthropicRole = Field(...) + content: list[AnthropicTextContent | AnthropicImageContent] = Field(...) + + +class AnthropicMessagesRequest(BaseModel): + model: str = Field(...) + messages: list[AnthropicMessage] = Field(...) + max_tokens: int = Field(..., ge=1) + system: str | None = Field(None, description="Top-level system prompt") + temperature: float | None = Field(None, ge=0.0, le=1.0) + top_p: float | None = Field(None, ge=0.0, le=1.0) + top_k: int | None = Field(None, ge=0) + stop_sequences: list[str] | None = Field(None) + + +class AnthropicResponseTextBlock(BaseModel): + type: Literal["text"] = "text" + text: str = Field(...) + + +class AnthropicCacheCreationUsage(BaseModel): + ephemeral_5m_input_tokens: int | None = Field(None) + ephemeral_1h_input_tokens: int | None = Field(None) + + +class AnthropicMessagesUsage(BaseModel): + input_tokens: int | None = Field(None) + output_tokens: int | None = Field(None) + cache_creation_input_tokens: int | None = Field(None) + cache_read_input_tokens: int | None = Field(None) + cache_creation: AnthropicCacheCreationUsage | None = Field(None) + + +class AnthropicMessagesResponse(BaseModel): + id: str | None = Field(None) + type: str | None = Field(None) + role: str | None = Field(None) + model: str | None = Field(None) + content: list[AnthropicResponseTextBlock] | None = Field(None) + stop_reason: str | None = Field(None) + stop_sequence: str | None = Field(None) + usage: AnthropicMessagesUsage | None = Field(None) diff --git a/comfy_api_nodes/apis/bria.py b/comfy_api_nodes/apis/bria.py index 8c496b56c..e08a519a8 100644 --- a/comfy_api_nodes/apis/bria.py +++ b/comfy_api_nodes/apis/bria.py @@ -23,7 +23,7 @@ class BriaEditImageRequest(BaseModel): None, description="Mask image (black and white). Black areas will be preserved, white areas will be edited. " "If omitted, the edit applies to the entire image. " - "The input image and the the input mask must be of the same size.", + "The input image and the input mask must be of the same size.", ) negative_prompt: str | None = Field(None) guidance_scale: float = Field(...) diff --git a/comfy_api_nodes/apis/bytedance.py b/comfy_api_nodes/apis/bytedance.py index 18455396d..03f4c445b 100644 --- a/comfy_api_nodes/apis/bytedance.py +++ b/comfy_api_nodes/apis/bytedance.py @@ -52,6 +52,26 @@ class TaskImageContent(BaseModel): role: Literal["first_frame", "last_frame", "reference_image"] | None = Field(None) +class TaskVideoContentUrl(BaseModel): + url: str = Field(...) + + +class TaskVideoContent(BaseModel): + type: str = Field("video_url") + video_url: TaskVideoContentUrl = Field(...) + role: str = Field("reference_video") + + +class TaskAudioContentUrl(BaseModel): + url: str = Field(...) + + +class TaskAudioContent(BaseModel): + type: str = Field("audio_url") + audio_url: TaskAudioContentUrl = Field(...) + role: str = Field("reference_audio") + + class Text2VideoTaskCreationRequest(BaseModel): model: str = Field(...) content: list[TaskTextContent] = Field(..., min_length=1) @@ -64,6 +84,17 @@ class Image2VideoTaskCreationRequest(BaseModel): generate_audio: bool | None = Field(...) +class Seedance2TaskCreationRequest(BaseModel): + model: str = Field(...) + content: list[TaskTextContent | TaskImageContent | TaskVideoContent | TaskAudioContent] = Field(..., min_length=1) + generate_audio: bool | None = Field(None) + resolution: str | None = Field(None) + ratio: str | None = Field(None) + duration: int | None = Field(None, ge=4, le=15) + seed: int | None = Field(None, ge=0, le=2147483647) + watermark: bool | None = Field(None) + + class TaskCreationResponse(BaseModel): id: str = Field(...) @@ -77,12 +108,67 @@ class TaskStatusResult(BaseModel): video_url: str = Field(...) +class TaskStatusUsage(BaseModel): + completion_tokens: int = Field(0) + total_tokens: int = Field(0) + + class TaskStatusResponse(BaseModel): id: str = Field(...) model: str = Field(...) status: Literal["queued", "running", "cancelled", "succeeded", "failed"] = Field(...) error: TaskStatusError | None = Field(None) content: TaskStatusResult | None = Field(None) + usage: TaskStatusUsage | None = Field(None) + + +class GetAssetResponse(BaseModel): + id: str = Field(...) + name: str | None = Field(None) + url: str | None = Field(None) + asset_type: str = Field(...) + group_id: str = Field(...) + status: str = Field(...) + error: TaskStatusError | None = Field(None) + + +class SeedanceCreateVisualValidateSessionResponse(BaseModel): + session_id: str = Field(...) + h5_link: str = Field(...) + + +class SeedanceGetVisualValidateSessionResponse(BaseModel): + session_id: str = Field(...) + status: str = Field(...) + group_id: str | None = Field(None) + error_code: str | None = Field(None) + error_message: str | None = Field(None) + + +class SeedanceCreateAssetRequest(BaseModel): + group_id: str = Field(...) + url: str = Field(...) + asset_type: str = Field(...) + name: str | None = Field(None, max_length=64) + project_name: str | None = Field(None) + + +class SeedanceCreateAssetResponse(BaseModel): + asset_id: str = Field(...) + + +class SeedanceVirtualLibraryCreateAssetRequest(BaseModel): + url: str = Field(..., description="Publicly accessible URL of the image asset to upload.") + hash: str = Field(..., description="Dedup key. Re-submitting the same hash returns the existing asset id.") + + +# Dollars per 1K tokens, keyed by (model_id, has_video_input). +SEEDANCE2_PRICE_PER_1K_TOKENS = { + ("dreamina-seedance-2-0-260128", False): 0.007, + ("dreamina-seedance-2-0-260128", True): 0.0043, + ("dreamina-seedance-2-0-fast-260128", False): 0.0056, + ("dreamina-seedance-2-0-fast-260128", True): 0.0033, +} RECOMMENDED_PRESETS = [ @@ -112,6 +198,75 @@ RECOMMENDED_PRESETS_SEEDREAM_4 = [ ("Custom", None, None), ] +_PRESETS_SEEDREAM_1K = [ + ("(1K) 1024x1024 (1:1)", 1024, 1024), + ("(1K) 864x1152 (3:4)", 864, 1152), + ("(1K) 1152x864 (4:3)", 1152, 864), + ("(1K) 1312x736 (16:9)", 1312, 736), + ("(1K) 736x1312 (9:16)", 736, 1312), + ("(1K) 832x1248 (2:3)", 832, 1248), + ("(1K) 1248x832 (3:2)", 1248, 832), + ("(1K) 1568x672 (21:9)", 1568, 672), +] + +_PRESETS_SEEDREAM_2K = [ + ("(2K) 2048x2048 (1:1)", 2048, 2048), + ("(2K) 1728x2304 (3:4)", 1728, 2304), + ("(2K) 2304x1728 (4:3)", 2304, 1728), + ("(2K) 2848x1600 (16:9)", 2848, 1600), + ("(2K) 1600x2848 (9:16)", 1600, 2848), + ("(2K) 1664x2496 (2:3)", 1664, 2496), + ("(2K) 2496x1664 (3:2)", 2496, 1664), + ("(2K) 3136x1344 (21:9)", 3136, 1344), +] + +_PRESETS_SEEDREAM_3K = [ + ("(3K) 3072x3072 (1:1)", 3072, 3072), + ("(3K) 2592x3456 (3:4)", 2592, 3456), + ("(3K) 3456x2592 (4:3)", 3456, 2592), + ("(3K) 4096x2304 (16:9)", 4096, 2304), + ("(3K) 2304x4096 (9:16)", 2304, 4096), + ("(3K) 2496x3744 (2:3)", 2496, 3744), + ("(3K) 3744x2496 (3:2)", 3744, 2496), + ("(3K) 4704x2016 (21:9)", 4704, 2016), +] + +_PRESETS_SEEDREAM_4K = [ + ("(4K) 4096x4096 (1:1)", 4096, 4096), + ("(4K) 3520x4704 (3:4)", 3520, 4704), + ("(4K) 4704x3520 (4:3)", 4704, 3520), + ("(4K) 5504x3040 (16:9)", 5504, 3040), + ("(4K) 3040x5504 (9:16)", 3040, 5504), + ("(4K) 3328x4992 (2:3)", 3328, 4992), + ("(4K) 4992x3328 (3:2)", 4992, 3328), + ("(4K) 6240x2656 (21:9)", 6240, 2656), +] + +_CUSTOM_PRESET = [("Custom", None, None)] + +RECOMMENDED_PRESETS_SEEDREAM_5_LITE = ( + _PRESETS_SEEDREAM_2K + _PRESETS_SEEDREAM_3K + _PRESETS_SEEDREAM_4K + _CUSTOM_PRESET +) +RECOMMENDED_PRESETS_SEEDREAM_4_5 = ( + _PRESETS_SEEDREAM_2K + _PRESETS_SEEDREAM_4K + _CUSTOM_PRESET +) +RECOMMENDED_PRESETS_SEEDREAM_4_0 = ( + _PRESETS_SEEDREAM_1K + _PRESETS_SEEDREAM_2K + _PRESETS_SEEDREAM_4K + _CUSTOM_PRESET +) + +# Seedance 2.0 reference video pixel count limits per model and output resolution. +SEEDANCE2_REF_VIDEO_PIXEL_LIMITS = { + "dreamina-seedance-2-0-260128": { + "480p": {"min": 409_600, "max": 927_408}, + "720p": {"min": 409_600, "max": 927_408}, + "1080p": {"min": 409_600, "max": 2_073_600}, + }, + "dreamina-seedance-2-0-fast-260128": { + "480p": {"min": 409_600, "max": 927_408}, + "720p": {"min": 409_600, "max": 927_408}, + }, +} + # The time in this dictionary are given for 10 seconds duration. VIDEO_TASKS_EXECUTION_TIME = { "seedance-1-0-lite-t2v-250428": { diff --git a/comfy_api_nodes/apis/luma.py b/comfy_api_nodes/apis/luma.py index 632c4ab96..8c6db2022 100644 --- a/comfy_api_nodes/apis/luma.py +++ b/comfy_api_nodes/apis/luma.py @@ -1,15 +1,12 @@ from __future__ import annotations - -import torch - from enum import Enum from typing import Optional, Union +import torch from pydantic import BaseModel, Field, confloat - class LumaIO: LUMA_REF = "LUMA_REF" LUMA_CONCEPTS = "LUMA_CONCEPTS" @@ -183,13 +180,13 @@ class LumaAssets(BaseModel): class LumaImageRef(BaseModel): - '''Used for image gen''' + """Used for image gen""" url: str = Field(..., description='The URL of the image reference') weight: confloat(ge=0.0, le=1.0) = Field(..., description='The weight of the image reference') class LumaImageReference(BaseModel): - '''Used for video gen''' + """Used for video gen""" type: Optional[str] = Field('image', description='Input type, defaults to image') url: str = Field(..., description='The URL of the image') @@ -251,3 +248,32 @@ class LumaGeneration(BaseModel): assets: Optional[LumaAssets] = Field(None, description='The assets of the generation') model: str = Field(..., description='The model used for the generation') request: Union[LumaGenerationRequest, LumaImageGenerationRequest] = Field(..., description="The request used for the generation") + + +class Luma2ImageRef(BaseModel): + url: str | None = None + data: str | None = None + media_type: str | None = None + + +class Luma2GenerationRequest(BaseModel): + prompt: str = Field(..., min_length=1, max_length=6000) + model: str | None = None + type: str | None = None + aspect_ratio: str | None = None + style: str | None = None + output_format: str | None = None + web_search: bool | None = None + image_ref: list[Luma2ImageRef] | None = None + source: Luma2ImageRef | None = None + + +class Luma2Generation(BaseModel): + id: str | None = None + type: str | None = None + state: str | None = None + model: str | None = None + created_at: str | None = None + output: list[LumaImageReference] | None = None + failure_reason: str | None = None + failure_code: str | None = None diff --git a/comfy_api_nodes/apis/moonvalley.py b/comfy_api_nodes/apis/moonvalley.py deleted file mode 100644 index 7ec7a4ade..000000000 --- a/comfy_api_nodes/apis/moonvalley.py +++ /dev/null @@ -1,152 +0,0 @@ -from enum import Enum -from typing import Optional, Dict, Any - -from pydantic import BaseModel, Field, StrictBytes - - -class MoonvalleyPromptResponse(BaseModel): - error: Optional[Dict[str, Any]] = None - frame_conditioning: Optional[Dict[str, Any]] = None - id: Optional[str] = None - inference_params: Optional[Dict[str, Any]] = None - meta: Optional[Dict[str, Any]] = None - model_params: Optional[Dict[str, Any]] = None - output_url: Optional[str] = None - prompt_text: Optional[str] = None - status: Optional[str] = None - - -class MoonvalleyTextToVideoInferenceParams(BaseModel): - add_quality_guidance: Optional[bool] = Field( - True, description='Whether to add quality guidance' - ) - caching_coefficient: Optional[float] = Field( - 0.3, description='Caching coefficient for optimization' - ) - caching_cooldown: Optional[int] = Field( - 3, description='Number of caching cooldown steps' - ) - caching_warmup: Optional[int] = Field( - 3, description='Number of caching warmup steps' - ) - clip_value: Optional[float] = Field( - 3, description='CLIP value for generation control' - ) - conditioning_frame_index: Optional[int] = Field( - 0, description='Index of the conditioning frame' - ) - cooldown_steps: Optional[int] = Field( - 75, description='Number of cooldown steps (calculated based on num_frames)' - ) - fps: Optional[int] = Field( - 24, description='Frames per second of the generated video' - ) - guidance_scale: Optional[float] = Field( - 10, description='Guidance scale for generation control' - ) - height: Optional[int] = Field( - 1080, description='Height of the generated video in pixels' - ) - negative_prompt: Optional[str] = Field(None, description='Negative prompt text') - num_frames: Optional[int] = Field(64, description='Number of frames to generate') - seed: Optional[int] = Field( - None, description='Random seed for generation (default: random)' - ) - shift_value: Optional[float] = Field( - 3, description='Shift value for generation control' - ) - steps: Optional[int] = Field(80, description='Number of denoising steps') - use_guidance_schedule: Optional[bool] = Field( - True, description='Whether to use guidance scheduling' - ) - use_negative_prompts: Optional[bool] = Field( - False, description='Whether to use negative prompts' - ) - use_timestep_transform: Optional[bool] = Field( - True, description='Whether to use timestep transformation' - ) - warmup_steps: Optional[int] = Field( - 0, description='Number of warmup steps (calculated based on num_frames)' - ) - width: Optional[int] = Field( - 1920, description='Width of the generated video in pixels' - ) - - -class MoonvalleyTextToVideoRequest(BaseModel): - image_url: Optional[str] = None - inference_params: Optional[MoonvalleyTextToVideoInferenceParams] = None - prompt_text: Optional[str] = None - webhook_url: Optional[str] = None - - -class MoonvalleyUploadFileRequest(BaseModel): - file: Optional[StrictBytes] = None - - -class MoonvalleyUploadFileResponse(BaseModel): - access_url: Optional[str] = None - - -class MoonvalleyVideoToVideoInferenceParams(BaseModel): - add_quality_guidance: Optional[bool] = Field( - True, description='Whether to add quality guidance' - ) - caching_coefficient: Optional[float] = Field( - 0.3, description='Caching coefficient for optimization' - ) - caching_cooldown: Optional[int] = Field( - 3, description='Number of caching cooldown steps' - ) - caching_warmup: Optional[int] = Field( - 3, description='Number of caching warmup steps' - ) - clip_value: Optional[float] = Field( - 3, description='CLIP value for generation control' - ) - conditioning_frame_index: Optional[int] = Field( - 0, description='Index of the conditioning frame' - ) - cooldown_steps: Optional[int] = Field( - 36, description='Number of cooldown steps (calculated based on num_frames)' - ) - guidance_scale: Optional[float] = Field( - 15, description='Guidance scale for generation control' - ) - negative_prompt: Optional[str] = Field(None, description='Negative prompt text') - seed: Optional[int] = Field( - None, description='Random seed for generation (default: random)' - ) - shift_value: Optional[float] = Field( - 3, description='Shift value for generation control' - ) - steps: Optional[int] = Field(80, description='Number of denoising steps') - use_guidance_schedule: Optional[bool] = Field( - True, description='Whether to use guidance scheduling' - ) - use_negative_prompts: Optional[bool] = Field( - False, description='Whether to use negative prompts' - ) - use_timestep_transform: Optional[bool] = Field( - True, description='Whether to use timestep transformation' - ) - warmup_steps: Optional[int] = Field( - 24, description='Number of warmup steps (calculated based on num_frames)' - ) - - -class ControlType(str, Enum): - motion_control = 'motion_control' - pose_control = 'pose_control' - - -class MoonvalleyVideoToVideoRequest(BaseModel): - control_type: ControlType = Field( - ..., description='Supported types for video control' - ) - inference_params: Optional[MoonvalleyVideoToVideoInferenceParams] = None - prompt_text: str = Field(..., description='Describes the video to generate') - video_url: str = Field(..., description='Url to control video') - webhook_url: Optional[str] = Field( - None, description='Optional webhook URL for notifications' - ) diff --git a/comfy_api_nodes/apis/openai.py b/comfy_api_nodes/apis/openai.py index b85ef252b..bee75d639 100644 --- a/comfy_api_nodes/apis/openai.py +++ b/comfy_api_nodes/apis/openai.py @@ -56,14 +56,14 @@ class ModelResponseProperties(BaseModel): instructions: str | None = Field(None) max_output_tokens: int | None = Field(None) model: str | None = Field(None) - temperature: float | None = Field(1, description="Controls randomness in the response", ge=0.0, le=2.0) + temperature: float | None = Field(None, description="Controls randomness in the response", ge=0.0, le=2.0) top_p: float | None = Field( - 1, + None, description="Controls diversity of the response via nucleus sampling", ge=0.0, le=1.0, ) - truncation: str | None = Field("disabled", description="Allowed values: 'auto' or 'disabled'") + truncation: str | None = Field(None, description="Allowed values: 'auto' or 'disabled'") class ResponseProperties(BaseModel): diff --git a/comfy_api_nodes/apis/topaz.py b/comfy_api_nodes/apis/topaz.py index a9e6235a7..f91980e3d 100644 --- a/comfy_api_nodes/apis/topaz.py +++ b/comfy_api_nodes/apis/topaz.py @@ -1,4 +1,4 @@ -from typing import Optional, Union +from typing import Optional from pydantic import BaseModel, Field @@ -72,8 +72,11 @@ class VideoEnhancementFilter(BaseModel): grain: Optional[float] = Field(None, description="Grain after AI model processing") grainSize: Optional[float] = Field(None, description="Size of generated grain") recoverOriginalDetailValue: Optional[float] = Field(None, description="Source details into the output video") - creativity: Optional[str] = Field(None, description="Creativity level(high, low) for slc-1 only") + creativity: float | str | None = Field(None, description="slc-1/slp-2.5: enum (low/middle/high). ast-2: decimal 0.0-1.0.") isOptimizedMode: Optional[bool] = Field(None, description="Set to true for Starlight Creative (slc-1) only") + prompt: str | None = Field(None, description="Descriptive scene prompt (ast-2 only)") + sharp: float | None = Field(None, description="ast-2 pre-enhance sharpness") + realism: float | None = Field(None, description="ast-2 realism control") class OutputInformationVideo(BaseModel): @@ -90,7 +93,7 @@ class Overrides(BaseModel): class CreateVideoRequest(BaseModel): source: CreateVideoRequestSource = Field(...) - filters: list[Union[VideoFrameInterpolationFilter, VideoEnhancementFilter]] = Field(...) + filters: list[VideoFrameInterpolationFilter | VideoEnhancementFilter] = Field(...) output: OutputInformationVideo = Field(...) overrides: Overrides = Field(Overrides(isPaidDiffusion=True)) diff --git a/comfy_api_nodes/apis/tripo.py b/comfy_api_nodes/apis/tripo.py index ffaaa7dc1..bce6b0e89 100644 --- a/comfy_api_nodes/apis/tripo.py +++ b/comfy_api_nodes/apis/tripo.py @@ -1,10 +1,11 @@ -from __future__ import annotations from enum import Enum -from typing import Optional, List, Dict, Any, Union +from typing import Optional, Any from pydantic import BaseModel, Field, RootModel + class TripoModelVersion(str, Enum): + v3_1_20260211 = 'v3.1-20260211' v3_0_20250812 = 'v3.0-20250812' v2_5_20250123 = 'v2.5-20250123' v2_0_20240919 = 'v2.0-20240919' @@ -142,7 +143,7 @@ class TripoFileEmptyReference(BaseModel): pass class TripoFileReference(RootModel): - root: Union[TripoFileTokenReference, TripoUrlReference, TripoObjectReference, TripoFileEmptyReference] + root: TripoFileTokenReference | TripoUrlReference | TripoObjectReference | TripoFileEmptyReference class TripoGetStsTokenRequest(BaseModel): format: str = Field(..., description='The format of the image') @@ -183,7 +184,7 @@ class TripoImageToModelRequest(BaseModel): class TripoMultiviewToModelRequest(BaseModel): type: TripoTaskType = TripoTaskType.MULTIVIEW_TO_MODEL - files: List[TripoFileReference] = Field(..., description='The file references to convert to a model') + files: list[TripoFileReference] = Field(..., description='The file references to convert to a model') model_version: Optional[TripoModelVersion] = Field(None, description='The model version to use for generation') orthographic_projection: Optional[bool] = Field(False, description='Whether to use orthographic projection') face_limit: Optional[int] = Field(None, description='The number of faces to limit the generation to') @@ -251,27 +252,13 @@ class TripoConvertModelRequest(BaseModel): with_animation: Optional[bool] = Field(None, description='Whether to include animations') pack_uv: Optional[bool] = Field(None, description='Whether to pack the UVs') bake: Optional[bool] = Field(None, description='Whether to bake the model') - part_names: Optional[List[str]] = Field(None, description='The names of the parts to include') + part_names: Optional[list[str]] = Field(None, description='The names of the parts to include') fbx_preset: Optional[TripoFbxPreset] = Field(None, description='The preset for the FBX export') export_vertex_colors: Optional[bool] = Field(None, description='Whether to export the vertex colors') export_orientation: Optional[TripoOrientation] = Field(None, description='The orientation for the export') animate_in_place: Optional[bool] = Field(None, description='Whether to animate in place') -class TripoTaskRequest(RootModel): - root: Union[ - TripoTextToModelRequest, - TripoImageToModelRequest, - TripoMultiviewToModelRequest, - TripoTextureModelRequest, - TripoRefineModelRequest, - TripoAnimatePrerigcheckRequest, - TripoAnimateRigRequest, - TripoAnimateRetargetRequest, - TripoStylizeModelRequest, - TripoConvertModelRequest - ] - class TripoTaskOutput(BaseModel): model: Optional[str] = Field(None, description='URL to the model') base_model: Optional[str] = Field(None, description='URL to the base model') @@ -283,12 +270,13 @@ class TripoTask(BaseModel): task_id: str = Field(..., description='The task ID') type: Optional[str] = Field(None, description='The type of task') status: Optional[TripoTaskStatus] = Field(None, description='The status of the task') - input: Optional[Dict[str, Any]] = Field(None, description='The input parameters for the task') + input: Optional[dict[str, Any]] = Field(None, description='The input parameters for the task') output: Optional[TripoTaskOutput] = Field(None, description='The output of the task') progress: Optional[int] = Field(None, description='The progress of the task', ge=0, le=100) create_time: Optional[int] = Field(None, description='The creation time of the task') running_left_time: Optional[int] = Field(None, description='The estimated time left for the task') queue_position: Optional[int] = Field(None, description='The position in the queue') + consumed_credit: int | None = Field(None) class TripoTaskResponse(BaseModel): code: int = Field(0, description='The response code') @@ -296,7 +284,7 @@ class TripoTaskResponse(BaseModel): class TripoGeneralResponse(BaseModel): code: int = Field(0, description='The response code') - data: Dict[str, str] = Field(..., description='The task ID data') + data: dict[str, str] = Field(..., description='The task ID data') class TripoBalanceData(BaseModel): balance: float = Field(..., description='The account balance') diff --git a/comfy_api_nodes/apis/wan.py b/comfy_api_nodes/apis/wan.py index 44b65e4f6..c64acae97 100644 --- a/comfy_api_nodes/apis/wan.py +++ b/comfy_api_nodes/apis/wan.py @@ -118,7 +118,7 @@ class Wan27ReferenceVideoInputField(BaseModel): class Wan27ReferenceVideoParametersField(BaseModel): resolution: str = Field(...) ratio: str | None = Field(None) - duration: int = Field(5, ge=2, le=10) + duration: int = Field(5, ge=2, le=15) watermark: bool = Field(False) seed: int = Field(..., ge=0, le=2147483647) @@ -157,7 +157,7 @@ class Wan27VideoEditInputField(BaseModel): class Wan27VideoEditParametersField(BaseModel): resolution: str = Field(...) ratio: str | None = Field(None) - duration: int = Field(0) + duration: int | None = Field(0) audio_setting: str = Field("auto") watermark: bool = Field(False) seed: int = Field(..., ge=0, le=2147483647) diff --git a/comfy_api_nodes/nodes_anthropic.py b/comfy_api_nodes/nodes_anthropic.py new file mode 100644 index 000000000..60e1624f7 --- /dev/null +++ b/comfy_api_nodes/nodes_anthropic.py @@ -0,0 +1,245 @@ +"""API Nodes for Anthropic Claude (Messages API). See: https://docs.anthropic.com/en/api/messages""" + +from typing_extensions import override + +from comfy_api.latest import IO, ComfyExtension, Input +from comfy_api_nodes.apis.anthropic import ( + AnthropicImageContent, + AnthropicImageSourceUrl, + AnthropicMessage, + AnthropicMessagesRequest, + AnthropicMessagesResponse, + AnthropicRole, + AnthropicTextContent, +) +from comfy_api_nodes.util import ( + ApiEndpoint, + get_number_of_images, + sync_op, + upload_images_to_comfyapi, + validate_string, +) + +ANTHROPIC_MESSAGES_ENDPOINT = "/proxy/anthropic/v1/messages" +ANTHROPIC_IMAGE_MAX_PIXELS = 1568 * 1568 +CLAUDE_MAX_IMAGES = 20 + +CLAUDE_MODELS: dict[str, str] = { + "Opus 4.7": "claude-opus-4-7", + "Opus 4.6": "claude-opus-4-6", + "Sonnet 4.6": "claude-sonnet-4-6", + "Sonnet 4.5": "claude-sonnet-4-5-20250929", + "Haiku 4.5": "claude-haiku-4-5-20251001", +} + + +def _claude_model_inputs(): + return [ + IO.Int.Input( + "max_tokens", + default=16000, + min=32, + max=32000, + tooltip="Maximum number of tokens to generate before stopping.", + advanced=True, + ), + IO.Float.Input( + "temperature", + default=1.0, + min=0.0, + max=1.0, + step=0.01, + tooltip="Controls randomness. 0.0 is deterministic, 1.0 is most random.", + advanced=True, + ), + ] + + +def _model_price_per_million(model: str) -> tuple[float, float] | None: + """Return (input_per_1M, output_per_1M) USD for a Claude model, or None if unknown.""" + if "opus-4-7" in model or "opus-4-6" in model or "opus-4-5" in model: + return 5.0, 25.0 + if "sonnet-4" in model: + return 3.0, 15.0 + if "haiku-4-5" in model: + return 1.0, 5.0 + return None + + +def calculate_tokens_price(response: AnthropicMessagesResponse) -> float | None: + """Compute approximate USD price from response usage. Server-side billing is authoritative.""" + if not response.usage or not response.model: + return None + rates = _model_price_per_million(response.model) + if rates is None: + return None + input_rate, output_rate = rates + input_tokens = response.usage.input_tokens or 0 + output_tokens = response.usage.output_tokens or 0 + cache_read = response.usage.cache_read_input_tokens or 0 + cache_5m = 0 + cache_1h = 0 + if response.usage.cache_creation: + cache_5m = response.usage.cache_creation.ephemeral_5m_input_tokens or 0 + cache_1h = response.usage.cache_creation.ephemeral_1h_input_tokens or 0 + total = ( + input_tokens * input_rate + + output_tokens * output_rate + + cache_read * input_rate * 0.1 + + cache_5m * input_rate * 1.25 + + cache_1h * input_rate * 2.0 + ) + return total / 1_000_000.0 + + +def _get_text_from_response(response: AnthropicMessagesResponse) -> str: + if not response.content: + return "" + return "\n".join(block.text for block in response.content if block.text) + + +async def _build_image_content_blocks( + cls: type[IO.ComfyNode], + image_tensors: list[Input.Image], +) -> list[AnthropicImageContent]: + urls = await upload_images_to_comfyapi( + cls, + image_tensors, + max_images=CLAUDE_MAX_IMAGES, + total_pixels=ANTHROPIC_IMAGE_MAX_PIXELS, + wait_label="Uploading reference images", + ) + return [AnthropicImageContent(source=AnthropicImageSourceUrl(url=url)) for url in urls] + + +class ClaudeNode(IO.ComfyNode): + """Generate text responses from an Anthropic Claude model.""" + + @classmethod + def define_schema(cls): + return IO.Schema( + node_id="ClaudeNode", + display_name="Anthropic Claude", + category="api node/text/Anthropic", + essentials_category="Text Generation", + description="Generate text responses with Anthropic's Claude models. " + "Provide a text prompt and optionally one or more images for multimodal context.", + inputs=[ + IO.String.Input( + "prompt", + multiline=True, + default="", + tooltip="Text input to the model.", + ), + IO.DynamicCombo.Input( + "model", + options=[IO.DynamicCombo.Option(label, _claude_model_inputs()) for label in CLAUDE_MODELS], + tooltip="The Claude model used to generate the response.", + ), + IO.Int.Input( + "seed", + default=0, + min=0, + max=2147483647, + control_after_generate=True, + tooltip="Seed controls whether the node should re-run; " + "results are non-deterministic regardless of seed.", + ), + IO.Autogrow.Input( + "images", + template=IO.Autogrow.TemplateNames( + IO.Image.Input("image"), + names=[f"image_{i}" for i in range(1, CLAUDE_MAX_IMAGES + 1)], + min=0, + ), + tooltip=f"Optional image(s) to use as context for the model. Up to {CLAUDE_MAX_IMAGES} images.", + ), + IO.String.Input( + "system_prompt", + multiline=True, + default="", + optional=True, + advanced=True, + tooltip="Foundational instructions that dictate the model's behavior.", + ), + ], + outputs=[IO.String.Output()], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + depends_on=IO.PriceBadgeDepends(widgets=["model"]), + expr=""" + ( + $m := widgets.model; + $contains($m, "opus") ? { + "type": "list_usd", + "usd": [0.005, 0.025], + "format": { "approximate": true, "separator": "-", "suffix": " per 1K tokens" } + } + : $contains($m, "sonnet") ? { + "type": "list_usd", + "usd": [0.003, 0.015], + "format": { "approximate": true, "separator": "-", "suffix": " per 1K tokens" } + } + : $contains($m, "haiku") ? { + "type": "list_usd", + "usd": [0.001, 0.005], + "format": { "approximate": true, "separator": "-", "suffix": " per 1K tokens" } + } + : {"type":"text", "text":"Token-based"} + ) + """, + ), + ) + + @classmethod + async def execute( + cls, + prompt: str, + model: dict, + seed: int, + images: dict | None = None, + system_prompt: str = "", + ) -> IO.NodeOutput: + validate_string(prompt, strip_whitespace=True, min_length=1) + model_label = model["model"] + max_tokens = model["max_tokens"] + temperature = model["temperature"] + + image_tensors: list[Input.Image] = [t for t in (images or {}).values() if t is not None] + if sum(get_number_of_images(t) for t in image_tensors) > CLAUDE_MAX_IMAGES: + raise ValueError(f"Up to {CLAUDE_MAX_IMAGES} images are supported per request.") + + content: list[AnthropicTextContent | AnthropicImageContent] = [] + if image_tensors: + content.extend(await _build_image_content_blocks(cls, image_tensors)) + content.append(AnthropicTextContent(text=prompt)) + + response = await sync_op( + cls, + ApiEndpoint(path=ANTHROPIC_MESSAGES_ENDPOINT, method="POST"), + response_model=AnthropicMessagesResponse, + data=AnthropicMessagesRequest( + model=CLAUDE_MODELS[model_label], + max_tokens=max_tokens, + messages=[AnthropicMessage(role=AnthropicRole.user, content=content)], + system=system_prompt or None, + temperature=temperature, + ), + price_extractor=calculate_tokens_price, + ) + return IO.NodeOutput(_get_text_from_response(response) or "Empty response from Claude model.") + + +class AnthropicExtension(ComfyExtension): + @override + async def get_node_list(self) -> list[type[IO.ComfyNode]]: + return [ClaudeNode] + + +async def comfy_entrypoint() -> AnthropicExtension: + return AnthropicExtension() diff --git a/comfy_api_nodes/nodes_bfl.py b/comfy_api_nodes/nodes_bfl.py index 23590bf24..3f0ce29d8 100644 --- a/comfy_api_nodes/nodes_bfl.py +++ b/comfy_api_nodes/nodes_bfl.py @@ -596,6 +596,7 @@ class Flux2ProImageNode(IO.ComfyNode): depends_on=IO.PriceBadgeDepends(widgets=["width", "height"], inputs=["images"]), expr=cls.PRICE_BADGE_EXPR, ), + is_deprecated=True, ) @classmethod @@ -674,6 +675,175 @@ class Flux2MaxImageNode(Flux2ProImageNode): """ +_FLUX2_MODEL_ENDPOINTS = { + "Flux.2 [pro]": "/proxy/bfl/flux-2-pro/generate", + "Flux.2 [max]": "/proxy/bfl/flux-2-max/generate", +} + + +def _flux2_model_inputs(): + return [ + IO.Int.Input( + "width", + default=1024, + min=256, + max=2048, + step=32, + ), + IO.Int.Input( + "height", + default=768, + min=256, + max=2048, + step=32, + ), + IO.Autogrow.Input( + "images", + template=IO.Autogrow.TemplateNames( + IO.Image.Input("image"), + names=[f"image_{i}" for i in range(1, 9)], + min=0, + ), + tooltip="Optional reference image(s) for image-to-image generation. Up to 8 images.", + ), + ] + + +class Flux2ImageNode(IO.ComfyNode): + + @classmethod + def define_schema(cls) -> IO.Schema: + return IO.Schema( + node_id="Flux2ImageNode", + display_name="Flux.2 Image", + category="api node/image/BFL", + description="Generate images via Flux.2 [pro] or Flux.2 [max] from a prompt and optional reference images.", + inputs=[ + IO.String.Input( + "prompt", + multiline=True, + default="", + tooltip="Prompt for the image generation or edit", + ), + IO.DynamicCombo.Input( + "model", + options=[ + IO.DynamicCombo.Option("Flux.2 [pro]", _flux2_model_inputs()), + IO.DynamicCombo.Option("Flux.2 [max]", _flux2_model_inputs()), + ], + ), + IO.Int.Input( + "seed", + default=0, + min=0, + max=0xFFFFFFFFFFFFFFFF, + control_after_generate=True, + tooltip="The random seed used for creating the noise.", + ), + ], + outputs=[IO.Image.Output()], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + depends_on=IO.PriceBadgeDepends( + widgets=["model", "model.width", "model.height"], + input_groups=["model.images"], + ), + expr=""" + ( + $isMax := widgets.model = "flux.2 [max]"; + $MP := 1024 * 1024; + $w := $lookup(widgets, "model.width"); + $h := $lookup(widgets, "model.height"); + $outMP := $max([1, $floor((($w * $h) + $MP - 1) / $MP)]); + $outputCost := $isMax + ? (0.07 + 0.03 * ($outMP - 1)) + : (0.03 + 0.015 * ($outMP - 1)); + $refMin := $isMax ? 0.03 : 0.015; + $refMax := $isMax ? 0.24 : 0.12; + $hasRefs := $lookup(inputGroups, "model.images") > 0; + $hasRefs + ? { + "type": "range_usd", + "min_usd": $outputCost + $refMin, + "max_usd": $outputCost + $refMax, + "format": { "approximate": true } + } + : {"type": "usd", "usd": $outputCost} + ) + """, + ), + ) + + @classmethod + async def execute( + cls, + prompt: str, + model: dict, + seed: int, + ) -> IO.NodeOutput: + model_choice = model["model"] + endpoint = _FLUX2_MODEL_ENDPOINTS[model_choice] + width = model["width"] + height = model["height"] + images_dict = model.get("images") or {} + + image_tensors: list[Input.Image] = [t for t in images_dict.values() if t is not None] + n_images = sum(get_number_of_images(t) for t in image_tensors) + if n_images > 8: + raise ValueError("The current maximum number of supported images is 8.") + + flat_tensors: list[torch.Tensor] = [] + for tensor in image_tensors: + if len(tensor.shape) == 4: + flat_tensors.extend(tensor[i] for i in range(tensor.shape[0])) + else: + flat_tensors.append(tensor) + + reference_images: dict[str, str] = {} + for idx, tensor in enumerate(flat_tensors): + key_name = f"input_image_{idx + 1}" if idx else "input_image" + reference_images[key_name] = tensor_to_base64_string(tensor, total_pixels=2048 * 2048) + + initial_response = await sync_op( + cls, + ApiEndpoint(path=endpoint, method="POST"), + response_model=BFLFluxProGenerateResponse, + data=Flux2ProGenerateRequest( + prompt=prompt, + width=width, + height=height, + seed=seed, + **reference_images, + ), + ) + + def price_extractor(_r: BaseModel) -> float | None: + return None if initial_response.cost is None else initial_response.cost / 100 + + response = await poll_op( + cls, + ApiEndpoint(initial_response.polling_url), + response_model=BFLFluxStatusResponse, + status_extractor=lambda r: r.status, + progress_extractor=lambda r: r.progress, + price_extractor=price_extractor, + completed_statuses=[BFLStatus.ready], + failed_statuses=[ + BFLStatus.request_moderated, + BFLStatus.content_moderated, + BFLStatus.error, + BFLStatus.task_not_found, + ], + queued_statuses=[], + ) + return IO.NodeOutput(await download_url_to_image_tensor(response.result["sample"])) + + class BFLExtension(ComfyExtension): @override async def get_node_list(self) -> list[type[IO.ComfyNode]]: @@ -685,6 +855,7 @@ class BFLExtension(ComfyExtension): FluxProFillNode, Flux2ProImageNode, Flux2MaxImageNode, + Flux2ImageNode, ] diff --git a/comfy_api_nodes/nodes_bytedance.py b/comfy_api_nodes/nodes_bytedance.py index de0c22e70..d6b479336 100644 --- a/comfy_api_nodes/nodes_bytedance.py +++ b/comfy_api_nodes/nodes_bytedance.py @@ -1,5 +1,7 @@ +import hashlib import logging import math +import re import torch from typing_extensions import override @@ -8,16 +10,32 @@ from comfy_api.latest import IO, ComfyExtension, Input from comfy_api_nodes.apis.bytedance import ( RECOMMENDED_PRESETS, RECOMMENDED_PRESETS_SEEDREAM_4, + RECOMMENDED_PRESETS_SEEDREAM_4_0, + RECOMMENDED_PRESETS_SEEDREAM_4_5, + RECOMMENDED_PRESETS_SEEDREAM_5_LITE, + SEEDANCE2_PRICE_PER_1K_TOKENS, + SEEDANCE2_REF_VIDEO_PIXEL_LIMITS, VIDEO_TASKS_EXECUTION_TIME, + GetAssetResponse, Image2VideoTaskCreationRequest, ImageTaskCreationResponse, + Seedance2TaskCreationRequest, + SeedanceCreateAssetRequest, + SeedanceCreateAssetResponse, + SeedanceCreateVisualValidateSessionResponse, + SeedanceGetVisualValidateSessionResponse, + SeedanceVirtualLibraryCreateAssetRequest, Seedream4Options, Seedream4TaskCreationRequest, + TaskAudioContent, + TaskAudioContentUrl, TaskCreationResponse, TaskImageContent, TaskImageContentUrl, TaskStatusResponse, TaskTextContent, + TaskVideoContent, + TaskVideoContentUrl, Text2ImageTaskCreationRequest, Text2VideoTaskCreationRequest, ) @@ -28,30 +46,280 @@ from comfy_api_nodes.util import ( get_number_of_images, image_tensor_pair_to_batch, poll_op, + resize_video_to_pixel_budget, sync_op, + upload_audio_to_comfyapi, + upload_image_to_comfyapi, upload_images_to_comfyapi, + upload_video_to_comfyapi, validate_image_aspect_ratio, validate_image_dimensions, validate_string, + validate_video_dimensions, + validate_video_duration, ) +from server import PromptServer BYTEPLUS_IMAGE_ENDPOINT = "/proxy/byteplus/api/v3/images/generations" +_VERIFICATION_POLL_TIMEOUT_SEC = 120 +_VERIFICATION_POLL_INTERVAL_SEC = 3 + SEEDREAM_MODELS = { "seedream 5.0 lite": "seedream-5-0-260128", "seedream-4-5-251128": "seedream-4-5-251128", "seedream-4-0-250828": "seedream-4-0-250828", } +SEEDREAM_PRESETS = { + "seedream-5-0-260128": RECOMMENDED_PRESETS_SEEDREAM_5_LITE, + "seedream-4-5-251128": RECOMMENDED_PRESETS_SEEDREAM_4_5, + "seedream-4-0-250828": RECOMMENDED_PRESETS_SEEDREAM_4_0, +} + # Long-running tasks endpoints(e.g., video) BYTEPLUS_TASK_ENDPOINT = "/proxy/byteplus/api/v3/contents/generations/tasks" BYTEPLUS_TASK_STATUS_ENDPOINT = "/proxy/byteplus/api/v3/contents/generations/tasks" # + /{task_id} +BYTEPLUS_SEEDANCE2_TASK_STATUS_ENDPOINT = "/proxy/byteplus-seedance2/api/v3/contents/generations/tasks" # + /{task_id} + +SEEDANCE_MODELS = { + "Seedance 2.0": "dreamina-seedance-2-0-260128", + "Seedance 2.0 Fast": "dreamina-seedance-2-0-fast-260128", +} DEPRECATED_MODELS = {"seedance-1-0-lite-t2v-250428", "seedance-1-0-lite-i2v-250428"} + logger = logging.getLogger(__name__) +def _validate_ref_video_pixels(video: Input.Video, model_id: str, resolution: str, index: int) -> None: + """Validate reference video pixel count against Seedance 2.0 model limits for the selected resolution.""" + model_limits = SEEDANCE2_REF_VIDEO_PIXEL_LIMITS.get(model_id) + if not model_limits: + return + limits = model_limits.get(resolution) + if not limits: + return + try: + w, h = video.get_dimensions() + except Exception: + return + pixels = w * h + min_px = limits.get("min") + max_px = limits.get("max") + if min_px and pixels < min_px: + raise ValueError( + f"Reference video {index} is too small: {w}x{h} = {pixels:,}px. " f"Minimum is {min_px:,}px for this model." + ) + if max_px and pixels > max_px: + raise ValueError( + f"Reference video {index} is too large: {w}x{h} = {pixels:,}px. " + f"Maximum is {max_px:,}px for this model. Try downscaling the video." + ) + + +async def _resolve_reference_assets( + cls: type[IO.ComfyNode], + asset_ids: list[str], +) -> tuple[dict[str, str], dict[str, str], dict[str, str]]: + """Look up each asset, validate Active status, group by asset_type. + + Returns (image_assets, video_assets, audio_assets), each mapping asset_id -> "asset://". + """ + image_assets: dict[str, str] = {} + video_assets: dict[str, str] = {} + audio_assets: dict[str, str] = {} + for i, raw_id in enumerate(asset_ids, 1): + asset_id = (raw_id or "").strip() + if not asset_id: + continue + result = await sync_op( + cls, + ApiEndpoint(path=f"/proxy/seedance/assets/{asset_id}"), + response_model=GetAssetResponse, + ) + if result.status != "Active": + extra = f" {result.error.code}: {result.error.message}" if result.error else "" + raise ValueError(f"Reference asset {i} (Id={asset_id}) is not Active (Status={result.status}).{extra}") + asset_uri = f"asset://{asset_id}" + if result.asset_type == "Image": + image_assets[asset_id] = asset_uri + elif result.asset_type == "Video": + video_assets[asset_id] = asset_uri + elif result.asset_type == "Audio": + audio_assets[asset_id] = asset_uri + return image_assets, video_assets, audio_assets + + +_ASSET_REF_RE = re.compile(r"\basset ?(\d{1,2})\b", re.IGNORECASE) + + +def _build_asset_labels( + reference_assets: dict[str, str], + image_asset_uris: dict[str, str], + video_asset_uris: dict[str, str], + audio_asset_uris: dict[str, str], + n_reference_images: int, + n_reference_videos: int, + n_reference_audios: int, +) -> dict[int, str]: + """Map asset slot number (from 'asset_N' keys) to its positional label. + + Asset entries are appended to `content` after the reference_images/videos/audios, + so their 1-indexed labels continue from the count of existing same-type refs: + one reference_images entry + one Image-type asset -> asset labelled "Image 2". + """ + image_n = n_reference_images + video_n = n_reference_videos + audio_n = n_reference_audios + labels: dict[int, str] = {} + for slot_key, raw_id in reference_assets.items(): + asset_id = (raw_id or "").strip() + if not asset_id: + continue + try: + slot_num = int(slot_key.rsplit("_", 1)[-1]) + except ValueError: + continue + if asset_id in image_asset_uris: + image_n += 1 + labels[slot_num] = f"Image {image_n}" + elif asset_id in video_asset_uris: + video_n += 1 + labels[slot_num] = f"Video {video_n}" + elif asset_id in audio_asset_uris: + audio_n += 1 + labels[slot_num] = f"Audio {audio_n}" + return labels + + +def _rewrite_asset_refs(prompt: str, labels: dict[int, str]) -> str: + """Case-insensitively replace 'assetNN' (1-2 digit) tokens with their labels.""" + if not labels: + return prompt + + def _sub(m: "re.Match[str]") -> str: + return labels.get(int(m.group(1)), m.group(0)) + + return _ASSET_REF_RE.sub(_sub, prompt) + + +async def _obtain_group_id_via_h5_auth(cls: type[IO.ComfyNode]) -> str: + session = await sync_op( + cls, + ApiEndpoint(path="/proxy/seedance/visual-validate/sessions", method="POST"), + response_model=SeedanceCreateVisualValidateSessionResponse, + ) + logger.warning("Seedance authentication required. Open link: %s", session.h5_link) + + h5_text = f"Open this link in your browser and complete face verification:\n\n{session.h5_link}" + + result = await poll_op( + cls, + ApiEndpoint(path=f"/proxy/seedance/visual-validate/sessions/{session.session_id}"), + response_model=SeedanceGetVisualValidateSessionResponse, + status_extractor=lambda r: r.status, + completed_statuses=["completed"], + failed_statuses=["failed"], + poll_interval=_VERIFICATION_POLL_INTERVAL_SEC, + max_poll_attempts=(_VERIFICATION_POLL_TIMEOUT_SEC // _VERIFICATION_POLL_INTERVAL_SEC) - 1, + estimated_duration=_VERIFICATION_POLL_TIMEOUT_SEC - 1, + extra_text=h5_text, + ) + + if not result.group_id: + raise RuntimeError(f"Seedance session {session.session_id} completed without a group_id") + + logger.warning("Seedance authentication complete. New GroupId: %s", result.group_id) + PromptServer.instance.send_progress_text( + f"Authentication complete. New GroupId: {result.group_id}", cls.hidden.unique_id + ) + return result.group_id + + +async def _resolve_group_id(cls: type[IO.ComfyNode], group_id: str) -> str: + if group_id and group_id.strip(): + return group_id.strip() + return await _obtain_group_id_via_h5_auth(cls) + + +async def _create_seedance_asset( + cls: type[IO.ComfyNode], + *, + group_id: str, + url: str, + name: str, + asset_type: str, +) -> str: + req = SeedanceCreateAssetRequest( + group_id=group_id, + url=url, + asset_type=asset_type, + name=name or None, + ) + result = await sync_op( + cls, + ApiEndpoint(path="/proxy/seedance/assets", method="POST"), + response_model=SeedanceCreateAssetResponse, + data=req, + ) + return result.asset_id + + +async def _wait_for_asset_active(cls: type[IO.ComfyNode], asset_id: str, group_id: str) -> GetAssetResponse: + """Poll the newly created asset until its status becomes Active.""" + return await poll_op( + cls, + ApiEndpoint(path=f"/proxy/seedance/assets/{asset_id}"), + response_model=GetAssetResponse, + status_extractor=lambda r: r.status, + completed_statuses=["Active"], + failed_statuses=["Failed"], + poll_interval=5, + max_poll_attempts=1200, + extra_text=f"Waiting for asset pre-processing...\n\nasset_id: {asset_id}\n\ngroup_id: {group_id}", + ) + + +async def _seedance_virtual_library_upload_image_asset( + cls: type[IO.ComfyNode], + image: torch.Tensor, + *, + wait_label: str = "Uploading image", +) -> str: + """Upload an image into the caller's per-customer Seedance virtual library.""" + public_url = await upload_image_to_comfyapi(cls, image, wait_label=wait_label) + normalized = image.detach().cpu().contiguous().to(torch.float32) + digest = hashlib.sha256() + digest.update(str(tuple(normalized.shape)).encode("utf-8")) + digest.update(b"\0") + digest.update(normalized.numpy().tobytes()) + image_hash = digest.hexdigest() + create_resp = await sync_op( + cls, + ApiEndpoint(path="/proxy/seedance/virtual-library/assets", method="POST"), + response_model=SeedanceCreateAssetResponse, + data=SeedanceVirtualLibraryCreateAssetRequest(url=public_url, hash=image_hash), + ) + await _wait_for_asset_active(cls, create_resp.asset_id, group_id="virtual-library") + return f"asset://{create_resp.asset_id}" + + +def _seedance2_price_extractor(model_id: str, has_video_input: bool): + """Returns a price_extractor closure for Seedance 2.0 poll_op.""" + rate = SEEDANCE2_PRICE_PER_1K_TOKENS.get((model_id, has_video_input)) + if rate is None: + return None + + def extractor(response: TaskStatusResponse) -> float | None: + if response.usage is None: + return None + return response.usage.total_tokens * 1.43 * rate / 1_000.0 + + return extractor + + def get_image_url_from_response(response: ImageTaskCreationResponse) -> str: if response.error: error_msg = f"ByteDance request failed. Code: {response.error['code']}, message: {response.error['message']}" @@ -303,6 +571,7 @@ class ByteDanceSeedreamNode(IO.ComfyNode): ) """, ), + is_deprecated=True, ) @classmethod @@ -335,8 +604,7 @@ class ByteDanceSeedreamNode(IO.ComfyNode): mp_provided = out_num_pixels / 1_000_000.0 if ("seedream-4-5" in model or "seedream-5-0" in model) and out_num_pixels < 3686400: raise ValueError( - f"Minimum image resolution for the selected model is 3.68MP, " - f"but {mp_provided:.2f}MP provided." + f"Minimum image resolution for the selected model is 3.68MP, " f"but {mp_provided:.2f}MP provided." ) if "seedream-4-0" in model and out_num_pixels < 921600: raise ValueError( @@ -393,6 +661,226 @@ class ByteDanceSeedreamNode(IO.ComfyNode): return IO.NodeOutput(torch.cat([await download_url_to_image_tensor(i) for i in urls])) +def _seedream_model_inputs(*, max_ref_images: int, presets: list): + return [ + IO.Combo.Input( + "size_preset", + options=[label for label, _, _ in presets], + tooltip="Pick a recommended size. Select Custom to use the width and height below.", + ), + IO.Int.Input( + "width", + default=2048, + min=1024, + max=6240, + step=2, + tooltip="Custom width for image. Value is working only if `size_preset` is set to `Custom`", + ), + IO.Int.Input( + "height", + default=2048, + min=1024, + max=4992, + step=2, + tooltip="Custom height for image. Value is working only if `size_preset` is set to `Custom`", + ), + IO.Int.Input( + "max_images", + default=1, + min=1, + max=max_ref_images, + step=1, + display_mode=IO.NumberDisplay.number, + tooltip="Maximum number of images to generate. With 1, exactly one image is produced. " + "With >1, the model generates between 1 and max_images related images " + "(e.g., story scenes, character variations). " + "Total images (input + generated) cannot exceed 15.", + ), + IO.Autogrow.Input( + "images", + template=IO.Autogrow.TemplateNames( + IO.Image.Input("image"), + names=[f"image_{i}" for i in range(1, max_ref_images + 1)], + min=0, + ), + tooltip=f"Optional reference image(s) for image-to-image or multi-reference generation. " + f"Up to {max_ref_images} images.", + ), + IO.Boolean.Input( + "fail_on_partial", + default=False, + tooltip="If enabled, abort execution if any requested images are missing or return an error.", + advanced=True, + ), + ] + + +class ByteDanceSeedreamNodeV2(IO.ComfyNode): + + @classmethod + def define_schema(cls): + return IO.Schema( + node_id="ByteDanceSeedreamNodeV2", + display_name="ByteDance Seedream 4.5 & 5.0", + category="api node/image/ByteDance", + description="Unified text-to-image generation and precise single-sentence editing at up to 4K resolution.", + inputs=[ + IO.String.Input( + "prompt", + multiline=True, + default="", + tooltip="Text prompt for creating or editing an image.", + ), + IO.DynamicCombo.Input( + "model", + options=[ + IO.DynamicCombo.Option( + "seedream 5.0 lite", + _seedream_model_inputs(max_ref_images=14, presets=RECOMMENDED_PRESETS_SEEDREAM_5_LITE), + ), + IO.DynamicCombo.Option( + "seedream-4-5-251128", + _seedream_model_inputs(max_ref_images=10, presets=RECOMMENDED_PRESETS_SEEDREAM_4_5), + ), + IO.DynamicCombo.Option( + "seedream-4-0-250828", + _seedream_model_inputs(max_ref_images=10, presets=RECOMMENDED_PRESETS_SEEDREAM_4_0), + ), + ], + ), + IO.Int.Input( + "seed", + default=0, + min=0, + max=2147483647, + step=1, + display_mode=IO.NumberDisplay.number, + control_after_generate=True, + tooltip="Seed to use for generation.", + ), + IO.Boolean.Input( + "watermark", + default=False, + tooltip='Whether to add an "AI generated" watermark to the image.', + advanced=True, + ), + ], + outputs=[ + IO.Image.Output(), + ], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + depends_on=IO.PriceBadgeDepends(widgets=["model"]), + expr=""" + ( + $price := $contains(widgets.model, "5.0 lite") ? 0.035 : + $contains(widgets.model, "4-5") ? 0.04 : 0.03; + { + "type":"usd", + "usd": $price, + "format": { "suffix":" x images/Run", "approximate": true } + } + ) + """, + ), + ) + + @classmethod + async def execute( + cls, + prompt: str, + model: dict, + seed: int = 0, + watermark: bool = False, + ) -> IO.NodeOutput: + validate_string(prompt, strip_whitespace=True, min_length=1) + model_id = SEEDREAM_MODELS[model["model"]] + presets = SEEDREAM_PRESETS[model_id] + + size_preset = model.get("size_preset", presets[0][0]) + width = model.get("width", 2048) + height = model.get("height", 2048) + max_images = model.get("max_images", 1) + sequential_image_generation = "disabled" if max_images == 1 else "auto" + images_dict = model.get("images") or {} + fail_on_partial = model.get("fail_on_partial", False) + + w = h = None + for label, tw, th in presets: + if label == size_preset: + w, h = tw, th + break + if w is None or h is None: + w, h = width, height + + out_num_pixels = w * h + mp_provided = out_num_pixels / 1_000_000.0 + if ("seedream-4-5" in model_id or "seedream-5-0" in model_id) and out_num_pixels < 3686400: + raise ValueError( + f"Minimum image resolution for the selected model is 3.68MP, but {mp_provided:.2f}MP provided." + ) + if "seedream-4-0" in model_id and out_num_pixels < 921600: + raise ValueError( + f"Minimum image resolution that the selected model can generate is 0.92MP, " + f"but {mp_provided:.2f}MP provided." + ) + if out_num_pixels > 16_777_216: + raise ValueError( + f"Maximum image resolution for the selected model is 16.78MP, but {mp_provided:.2f}MP provided." + ) + + image_tensors: list[Input.Image] = [t for t in images_dict.values() if t is not None] + n_input_images = sum(get_number_of_images(t) for t in image_tensors) + max_num_of_images = 14 if model_id == "seedream-5-0-260128" else 10 + if n_input_images > max_num_of_images: + raise ValueError( + f"Maximum of {max_num_of_images} reference images are supported, but {n_input_images} received." + ) + if sequential_image_generation == "auto" and n_input_images + max_images > 15: + raise ValueError( + "The maximum number of generated images plus the number of reference images cannot exceed 15." + ) + + reference_images_urls: list[str] = [] + if image_tensors: + for tensor in image_tensors: + validate_image_aspect_ratio(tensor, (1, 3), (3, 1)) + reference_images_urls = await upload_images_to_comfyapi( + cls, + image_tensors, + max_images=n_input_images, + mime_type="image/png", + wait_label="Uploading reference images", + ) + + response = await sync_op( + cls, + ApiEndpoint(path=BYTEPLUS_IMAGE_ENDPOINT, method="POST"), + response_model=ImageTaskCreationResponse, + data=Seedream4TaskCreationRequest( + model=model_id, + prompt=prompt, + image=reference_images_urls, + size=f"{w}x{h}", + seed=seed, + sequential_image_generation=sequential_image_generation, + sequential_image_generation_options=Seedream4Options(max_images=max_images), + watermark=watermark, + ), + ) + if len(response.data) == 1: + return IO.NodeOutput(await download_url_to_image_tensor(get_image_url_from_response(response))) + urls = [str(d["url"]) for d in response.data if isinstance(d, dict) and "url" in d] + if fail_on_partial and len(urls) < len(response.data): + raise RuntimeError(f"Only {len(urls)} of {len(response.data)} images were generated before error.") + return IO.NodeOutput(torch.cat([await download_url_to_image_tensor(i) for i in urls])) + + class ByteDanceTextToVideoNode(IO.ComfyNode): @classmethod @@ -952,33 +1440,6 @@ class ByteDanceImageReferenceNode(IO.ComfyNode): ) -async def process_video_task( - cls: type[IO.ComfyNode], - payload: Text2VideoTaskCreationRequest | Image2VideoTaskCreationRequest, - estimated_duration: int | None, -) -> IO.NodeOutput: - if payload.model in DEPRECATED_MODELS: - logger.warning( - "Model '%s' is deprecated and will be deactivated on May 13, 2026. " - "Please switch to a newer model. Recommended: seedance-1-0-pro-fast-251015.", - payload.model, - ) - initial_response = await sync_op( - cls, - ApiEndpoint(path=BYTEPLUS_TASK_ENDPOINT, method="POST"), - data=payload, - response_model=TaskCreationResponse, - ) - response = await poll_op( - cls, - ApiEndpoint(path=f"{BYTEPLUS_TASK_STATUS_ENDPOINT}/{initial_response.id}"), - status_extractor=lambda r: r.status, - estimated_duration=estimated_duration, - response_model=TaskStatusResponse, - ) - return IO.NodeOutput(await download_url_to_video_output(response.content.video_url)) - - def raise_if_text_params(prompt: str, text_params: list[str]) -> None: for i in text_params: if f"--{i} " in prompt: @@ -1040,16 +1501,850 @@ PRICE_BADGE_VIDEO = IO.PriceBadge( ) +def _seedance2_text_inputs(resolutions: list[str], default_ratio: str = "16:9"): + return [ + IO.String.Input( + "prompt", + multiline=True, + default="", + tooltip="Text prompt for video generation.", + ), + IO.Combo.Input( + "resolution", + options=resolutions, + tooltip="Resolution of the output video.", + ), + IO.Combo.Input( + "ratio", + options=["16:9", "4:3", "1:1", "3:4", "9:16", "21:9", "adaptive"], + default=default_ratio, + tooltip="Aspect ratio of the output video.", + ), + IO.Int.Input( + "duration", + default=7, + min=4, + max=15, + step=1, + tooltip="Duration of the output video in seconds (4-15).", + display_mode=IO.NumberDisplay.slider, + ), + IO.Boolean.Input( + "generate_audio", + default=True, + tooltip="Enable audio generation for the output video.", + ), + ] + + +class ByteDance2TextToVideoNode(IO.ComfyNode): + + @classmethod + def define_schema(cls): + return IO.Schema( + node_id="ByteDance2TextToVideoNode", + display_name="ByteDance Seedance 2.0 Text to Video", + category="api node/video/ByteDance", + description="Generate video using Seedance 2.0 models based on a text prompt.", + inputs=[ + IO.DynamicCombo.Input( + "model", + options=[ + IO.DynamicCombo.Option("Seedance 2.0", _seedance2_text_inputs(["480p", "720p", "1080p"])), + IO.DynamicCombo.Option("Seedance 2.0 Fast", _seedance2_text_inputs(["480p", "720p"])), + ], + tooltip="Seedance 2.0 for maximum quality; Seedance 2.0 Fast for speed optimization.", + ), + IO.Int.Input( + "seed", + default=0, + min=0, + max=2147483647, + step=1, + display_mode=IO.NumberDisplay.number, + control_after_generate=True, + tooltip="Seed controls whether the node should re-run; " + "results are non-deterministic regardless of seed.", + ), + IO.Boolean.Input( + "watermark", + default=False, + tooltip="Whether to add a watermark to the video.", + advanced=True, + ), + ], + outputs=[ + IO.Video.Output(), + ], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + depends_on=IO.PriceBadgeDepends(widgets=["model", "model.resolution", "model.duration"]), + expr=""" + ( + $rate480 := 10044; + $rate720 := 21600; + $rate1080 := 48800; + $m := widgets.model; + $pricePer1K := $contains($m, "fast") ? 0.008008 : 0.01001; + $res := $lookup(widgets, "model.resolution"); + $dur := $lookup(widgets, "model.duration"); + $rate := $res = "1080p" ? $rate1080 : + $res = "720p" ? $rate720 : + $rate480; + $cost := $dur * $rate * $pricePer1K / 1000; + {"type": "usd", "usd": $cost, "format": {"approximate": true}} + ) + """, + ), + ) + + @classmethod + async def execute( + cls, + model: dict, + seed: int, + watermark: bool, + ) -> IO.NodeOutput: + validate_string(model["prompt"], strip_whitespace=True, min_length=1) + model_id = SEEDANCE_MODELS[model["model"]] + initial_response = await sync_op( + cls, + ApiEndpoint(path=BYTEPLUS_TASK_ENDPOINT, method="POST"), + data=Seedance2TaskCreationRequest( + model=model_id, + content=[TaskTextContent(text=model["prompt"])], + generate_audio=model["generate_audio"], + resolution=model["resolution"], + ratio=model["ratio"], + duration=model["duration"], + seed=seed, + watermark=watermark, + ), + response_model=TaskCreationResponse, + ) + response = await poll_op( + cls, + ApiEndpoint(path=f"{BYTEPLUS_SEEDANCE2_TASK_STATUS_ENDPOINT}/{initial_response.id}"), + response_model=TaskStatusResponse, + status_extractor=lambda r: r.status, + price_extractor=_seedance2_price_extractor(model_id, has_video_input=False), + poll_interval=9, + ) + return IO.NodeOutput(await download_url_to_video_output(response.content.video_url)) + + +class ByteDance2FirstLastFrameNode(IO.ComfyNode): + + @classmethod + def define_schema(cls): + return IO.Schema( + node_id="ByteDance2FirstLastFrameNode", + display_name="ByteDance Seedance 2.0 First-Last-Frame to Video", + category="api node/video/ByteDance", + description="Generate video using Seedance 2.0 from a first frame image and optional last frame image.", + inputs=[ + IO.DynamicCombo.Input( + "model", + options=[ + IO.DynamicCombo.Option( + "Seedance 2.0", + _seedance2_text_inputs(["480p", "720p", "1080p"], default_ratio="adaptive"), + ), + IO.DynamicCombo.Option( + "Seedance 2.0 Fast", + _seedance2_text_inputs(["480p", "720p"], default_ratio="adaptive"), + ), + ], + tooltip="Seedance 2.0 for maximum quality; Seedance 2.0 Fast for speed optimization.", + ), + IO.Image.Input( + "first_frame", + tooltip="First frame image for the video.", + optional=True, + ), + IO.Image.Input( + "last_frame", + tooltip="Last frame image for the video.", + optional=True, + ), + IO.String.Input( + "first_frame_asset_id", + default="", + tooltip="Seedance asset_id to use as the first frame. " + "Mutually exclusive with the first_frame image input.", + optional=True, + ), + IO.String.Input( + "last_frame_asset_id", + default="", + tooltip="Seedance asset_id to use as the last frame. " + "Mutually exclusive with the last_frame image input.", + optional=True, + ), + IO.Int.Input( + "seed", + default=0, + min=0, + max=2147483647, + step=1, + display_mode=IO.NumberDisplay.number, + control_after_generate=True, + tooltip="Seed controls whether the node should re-run; " + "results are non-deterministic regardless of seed.", + ), + IO.Boolean.Input( + "watermark", + default=False, + tooltip="Whether to add a watermark to the video.", + advanced=True, + ), + ], + outputs=[ + IO.Video.Output(), + ], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + depends_on=IO.PriceBadgeDepends(widgets=["model", "model.resolution", "model.duration"]), + expr=""" + ( + $rate480 := 10044; + $rate720 := 21600; + $rate1080 := 48800; + $m := widgets.model; + $pricePer1K := $contains($m, "fast") ? 0.008008 : 0.01001; + $res := $lookup(widgets, "model.resolution"); + $dur := $lookup(widgets, "model.duration"); + $rate := $res = "1080p" ? $rate1080 : + $res = "720p" ? $rate720 : + $rate480; + $cost := $dur * $rate * $pricePer1K / 1000; + {"type": "usd", "usd": $cost, "format": {"approximate": true}} + ) + """, + ), + ) + + @classmethod + async def execute( + cls, + model: dict, + seed: int, + watermark: bool, + first_frame: Input.Image | None = None, + last_frame: Input.Image | None = None, + first_frame_asset_id: str = "", + last_frame_asset_id: str = "", + ) -> IO.NodeOutput: + validate_string(model["prompt"], strip_whitespace=True, min_length=1) + model_id = SEEDANCE_MODELS[model["model"]] + + first_frame_asset_id = first_frame_asset_id.strip() + last_frame_asset_id = last_frame_asset_id.strip() + + if first_frame is not None and first_frame_asset_id: + raise ValueError("Provide only one of first_frame or first_frame_asset_id, not both.") + if first_frame is None and not first_frame_asset_id: + raise ValueError("Either first_frame or first_frame_asset_id is required.") + if last_frame is not None and last_frame_asset_id: + raise ValueError("Provide only one of last_frame or last_frame_asset_id, not both.") + + asset_ids_to_resolve = [a for a in (first_frame_asset_id, last_frame_asset_id) if a] + image_assets: dict[str, str] = {} + if asset_ids_to_resolve: + image_assets, _, _ = await _resolve_reference_assets(cls, asset_ids_to_resolve) + for aid in asset_ids_to_resolve: + if aid not in image_assets: + raise ValueError(f"Asset {aid} is not an Image asset.") + + if first_frame_asset_id: + first_frame_url = image_assets[first_frame_asset_id] + else: + first_frame_url = await _seedance_virtual_library_upload_image_asset( + cls, first_frame, wait_label="Uploading first frame." + ) + + content: list[TaskTextContent | TaskImageContent] = [ + TaskTextContent(text=model["prompt"]), + TaskImageContent( + image_url=TaskImageContentUrl(url=first_frame_url), + role="first_frame", + ), + ] + if last_frame_asset_id: + content.append( + TaskImageContent( + image_url=TaskImageContentUrl(url=image_assets[last_frame_asset_id]), + role="last_frame", + ), + ) + elif last_frame is not None: + content.append( + TaskImageContent( + image_url=TaskImageContentUrl( + url=await _seedance_virtual_library_upload_image_asset( + cls, last_frame, wait_label="Uploading last frame." + ) + ), + role="last_frame", + ), + ) + + initial_response = await sync_op( + cls, + ApiEndpoint(path=BYTEPLUS_TASK_ENDPOINT, method="POST"), + data=Seedance2TaskCreationRequest( + model=model_id, + content=content, + generate_audio=model["generate_audio"], + resolution=model["resolution"], + ratio=model["ratio"], + duration=model["duration"], + seed=seed, + watermark=watermark, + ), + response_model=TaskCreationResponse, + ) + response = await poll_op( + cls, + ApiEndpoint(path=f"{BYTEPLUS_SEEDANCE2_TASK_STATUS_ENDPOINT}/{initial_response.id}"), + response_model=TaskStatusResponse, + status_extractor=lambda r: r.status, + price_extractor=_seedance2_price_extractor(model_id, has_video_input=False), + poll_interval=9, + ) + return IO.NodeOutput(await download_url_to_video_output(response.content.video_url)) + + +def _seedance2_reference_inputs(resolutions: list[str], default_ratio: str = "16:9"): + return [ + *_seedance2_text_inputs(resolutions, default_ratio=default_ratio), + IO.Autogrow.Input( + "reference_images", + template=IO.Autogrow.TemplateNames( + IO.Image.Input("reference_image"), + names=[ + "image_1", + "image_2", + "image_3", + "image_4", + "image_5", + "image_6", + "image_7", + "image_8", + "image_9", + ], + min=0, + ), + ), + IO.Autogrow.Input( + "reference_videos", + template=IO.Autogrow.TemplateNames( + IO.Video.Input("reference_video"), + names=["video_1", "video_2", "video_3"], + min=0, + ), + ), + IO.Autogrow.Input( + "reference_audios", + template=IO.Autogrow.TemplateNames( + IO.Audio.Input("reference_audio"), + names=["audio_1", "audio_2", "audio_3"], + min=0, + ), + ), + IO.Boolean.Input( + "auto_downscale", + default=False, + advanced=True, + optional=True, + tooltip="Automatically downscale reference videos that exceed the model's pixel budget " + "for the selected resolution. Aspect ratio is preserved; videos already within limits are untouched.", + ), + IO.Autogrow.Input( + "reference_assets", + template=IO.Autogrow.TemplateNames( + IO.String.Input("reference_asset"), + names=[ + "asset_1", + "asset_2", + "asset_3", + "asset_4", + "asset_5", + "asset_6", + "asset_7", + "asset_8", + "asset_9", + ], + min=0, + ), + ), + ] + + +class ByteDance2ReferenceNode(IO.ComfyNode): + + @classmethod + def define_schema(cls): + return IO.Schema( + node_id="ByteDance2ReferenceNode", + display_name="ByteDance Seedance 2.0 Reference to Video", + category="api node/video/ByteDance", + description="Generate, edit, or extend video using Seedance 2.0 with reference images, " + "videos, and audio. Supports multimodal reference, video editing, and video extension.", + inputs=[ + IO.DynamicCombo.Input( + "model", + options=[ + IO.DynamicCombo.Option( + "Seedance 2.0", + _seedance2_reference_inputs(["480p", "720p", "1080p"], default_ratio="adaptive"), + ), + IO.DynamicCombo.Option( + "Seedance 2.0 Fast", + _seedance2_reference_inputs(["480p", "720p"], default_ratio="adaptive"), + ), + ], + tooltip="Seedance 2.0 for maximum quality; Seedance 2.0 Fast for speed optimization.", + ), + IO.Int.Input( + "seed", + default=0, + min=0, + max=2147483647, + step=1, + display_mode=IO.NumberDisplay.number, + control_after_generate=True, + tooltip="Seed controls whether the node should re-run; " + "results are non-deterministic regardless of seed.", + ), + IO.Boolean.Input( + "watermark", + default=False, + tooltip="Whether to add a watermark to the video.", + advanced=True, + ), + ], + outputs=[ + IO.Video.Output(), + ], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + depends_on=IO.PriceBadgeDepends( + widgets=["model", "model.resolution", "model.duration"], + input_groups=["model.reference_videos"], + ), + expr=""" + ( + $rate480 := 10044; + $rate720 := 21600; + $rate1080 := 48800; + $m := widgets.model; + $hasVideo := $lookup(inputGroups, "model.reference_videos") > 0; + $noVideoPricePer1K := $contains($m, "fast") ? 0.008008 : 0.01001; + $videoPricePer1K := $contains($m, "fast") ? 0.004719 : 0.006149; + $res := $lookup(widgets, "model.resolution"); + $dur := $lookup(widgets, "model.duration"); + $rate := $res = "1080p" ? $rate1080 : + $res = "720p" ? $rate720 : + $rate480; + $noVideoCost := $dur * $rate * $noVideoPricePer1K / 1000; + $minVideoFactor := $ceil($dur * 5 / 3); + $minVideoCost := $minVideoFactor * $rate * $videoPricePer1K / 1000; + $maxVideoCost := (15 + $dur) * $rate * $videoPricePer1K / 1000; + $hasVideo + ? { + "type": "range_usd", + "min_usd": $minVideoCost, + "max_usd": $maxVideoCost, + "format": {"approximate": true} + } + : { + "type": "usd", + "usd": $noVideoCost, + "format": {"approximate": true} + } + ) + """, + ), + ) + + @classmethod + async def execute( + cls, + model: dict, + seed: int, + watermark: bool, + ) -> IO.NodeOutput: + validate_string(model["prompt"], strip_whitespace=True, min_length=1) + + reference_images = model.get("reference_images", {}) + reference_videos = model.get("reference_videos", {}) + reference_audios = model.get("reference_audios", {}) + reference_assets = model.get("reference_assets", {}) + + reference_image_assets, reference_video_assets, reference_audio_assets = await _resolve_reference_assets( + cls, list(reference_assets.values()) + ) + + if not reference_images and not reference_videos and not reference_image_assets and not reference_video_assets: + raise ValueError("At least one reference image or video or asset is required.") + + total_images = len(reference_images) + len(reference_image_assets) + if total_images > 9: + raise ValueError( + f"Too many reference images: {total_images} " + f"(images={len(reference_images)}, image assets={len(reference_image_assets)}). Maximum is 9." + ) + total_videos = len(reference_videos) + len(reference_video_assets) + if total_videos > 3: + raise ValueError( + f"Too many reference videos: {total_videos} " + f"(videos={len(reference_videos)}, video assets={len(reference_video_assets)}). Maximum is 3." + ) + total_audios = len(reference_audios) + len(reference_audio_assets) + if total_audios > 3: + raise ValueError( + f"Too many reference audios: {total_audios} " + f"(audios={len(reference_audios)}, audio assets={len(reference_audio_assets)}). Maximum is 3." + ) + + model_id = SEEDANCE_MODELS[model["model"]] + has_video_input = total_videos > 0 + + if model.get("auto_downscale") and reference_videos: + max_px = SEEDANCE2_REF_VIDEO_PIXEL_LIMITS.get(model_id, {}).get(model["resolution"], {}).get("max") + if max_px: + for key in reference_videos: + reference_videos[key] = resize_video_to_pixel_budget(reference_videos[key], max_px) + + total_video_duration = 0.0 + for i, key in enumerate(reference_videos, 1): + video = reference_videos[key] + _validate_ref_video_pixels(video, model_id, model["resolution"], i) + try: + dur = video.get_duration() + if dur < 1.8: + raise ValueError(f"Reference video {i} is too short: {dur:.1f}s. Minimum duration is 1.8 seconds.") + total_video_duration += dur + except ValueError: + raise + except Exception: + pass + if total_video_duration > 15.1: + raise ValueError(f"Total reference video duration is {total_video_duration:.1f}s. Maximum is 15.1 seconds.") + + total_audio_duration = 0.0 + for i, key in enumerate(reference_audios, 1): + audio = reference_audios[key] + dur = int(audio["waveform"].shape[-1]) / int(audio["sample_rate"]) + if dur < 1.8: + raise ValueError(f"Reference audio {i} is too short: {dur:.1f}s. Minimum duration is 1.8 seconds.") + total_audio_duration += dur + if total_audio_duration > 15.1: + raise ValueError(f"Total reference audio duration is {total_audio_duration:.1f}s. Maximum is 15.1 seconds.") + + asset_labels = _build_asset_labels( + reference_assets, + reference_image_assets, + reference_video_assets, + reference_audio_assets, + len(reference_images), + len(reference_videos), + len(reference_audios), + ) + prompt_text = _rewrite_asset_refs(model["prompt"], asset_labels) + + content: list[TaskTextContent | TaskImageContent | TaskVideoContent | TaskAudioContent] = [ + TaskTextContent(text=prompt_text), + ] + for i, key in enumerate(reference_images, 1): + content.append( + TaskImageContent( + image_url=TaskImageContentUrl( + url=await _seedance_virtual_library_upload_image_asset( + cls, + reference_images[key], + wait_label=f"Uploading image {i}", + ), + ), + role="reference_image", + ), + ) + for i, key in enumerate(reference_videos, 1): + content.append( + TaskVideoContent( + video_url=TaskVideoContentUrl( + url=await upload_video_to_comfyapi( + cls, + reference_videos[key], + wait_label=f"Uploading video {i}", + ), + ), + ), + ) + for key in reference_audios: + content.append( + TaskAudioContent( + audio_url=TaskAudioContentUrl( + url=await upload_audio_to_comfyapi( + cls, + reference_audios[key], + container_format="mp3", + codec_name="libmp3lame", + mime_type="audio/mpeg", + ), + ), + ), + ) + for url in reference_image_assets.values(): + content.append( + TaskImageContent( + image_url=TaskImageContentUrl(url=url), + role="reference_image", + ), + ) + for url in reference_video_assets.values(): + content.append( + TaskVideoContent(video_url=TaskVideoContentUrl(url=url)), + ) + for url in reference_audio_assets.values(): + content.append( + TaskAudioContent(audio_url=TaskAudioContentUrl(url=url)), + ) + initial_response = await sync_op( + cls, + ApiEndpoint(path=BYTEPLUS_TASK_ENDPOINT, method="POST"), + data=Seedance2TaskCreationRequest( + model=model_id, + content=content, + generate_audio=model["generate_audio"], + resolution=model["resolution"], + ratio=model["ratio"], + duration=model["duration"], + seed=seed, + watermark=watermark, + ), + response_model=TaskCreationResponse, + ) + response = await poll_op( + cls, + ApiEndpoint(path=f"{BYTEPLUS_SEEDANCE2_TASK_STATUS_ENDPOINT}/{initial_response.id}"), + response_model=TaskStatusResponse, + status_extractor=lambda r: r.status, + price_extractor=_seedance2_price_extractor(model_id, has_video_input=has_video_input), + poll_interval=9, + ) + return IO.NodeOutput(await download_url_to_video_output(response.content.video_url)) + + +async def process_video_task( + cls: type[IO.ComfyNode], + payload: Text2VideoTaskCreationRequest | Image2VideoTaskCreationRequest, + estimated_duration: int | None, +) -> IO.NodeOutput: + if payload.model in DEPRECATED_MODELS: + logger.warning( + "Model '%s' is deprecated and will be deactivated on May 13, 2026. " + "Please switch to a newer model. Recommended: seedance-1-0-pro-fast-251015.", + payload.model, + ) + initial_response = await sync_op( + cls, + ApiEndpoint(path=BYTEPLUS_TASK_ENDPOINT, method="POST"), + data=payload, + response_model=TaskCreationResponse, + ) + response = await poll_op( + cls, + ApiEndpoint(path=f"{BYTEPLUS_TASK_STATUS_ENDPOINT}/{initial_response.id}"), + status_extractor=lambda r: r.status, + estimated_duration=estimated_duration, + response_model=TaskStatusResponse, + ) + return IO.NodeOutput(await download_url_to_video_output(response.content.video_url)) + + +class ByteDanceCreateImageAsset(IO.ComfyNode): + + @classmethod + def define_schema(cls) -> IO.Schema: + return IO.Schema( + node_id="ByteDanceCreateImageAsset", + display_name="ByteDance Create Image Asset", + category="api node/image/ByteDance", + description=( + "Create a Seedance 2.0 personal image asset. Uploads the input image and " + "registers it in the given asset group. If group_id is empty, runs a real-person " + "H5 authentication flow to create a new group before adding the asset." + ), + inputs=[ + IO.Image.Input("image", tooltip="Image to register as a personal asset."), + IO.String.Input( + "group_id", + default="", + tooltip="Reuse an existing Seedance asset group ID to skip repeated human verification for the " + "same person. Leave empty to run real-person authentication in the browser and create a new group.", + ), + # IO.String.Input( + # "name", + # default="", + # tooltip="Asset name (up to 64 characters).", + # ), + ], + outputs=[ + IO.String.Output(display_name="asset_id"), + IO.String.Output(display_name="group_id"), + ], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + # is_api_node=True, + ) + + @classmethod + async def execute( + cls, + image: Input.Image, + group_id: str = "", + # name: str = "", + ) -> IO.NodeOutput: + # if len(name) > 64: + # raise ValueError("Name of asset can not be greater then 64 symbols") + validate_image_dimensions(image, min_width=300, max_width=6000, min_height=300, max_height=6000) + validate_image_aspect_ratio(image, min_ratio=(0.4, 1), max_ratio=(2.5, 1)) + resolved_group = await _resolve_group_id(cls, group_id) + asset_id = await _create_seedance_asset( + cls, + group_id=resolved_group, + url=await upload_image_to_comfyapi(cls, image), + name="", + asset_type="Image", + ) + await _wait_for_asset_active(cls, asset_id, resolved_group) + PromptServer.instance.send_progress_text( + f"Please save the asset_id and group_id for reuse.\n\nasset_id: {asset_id}\n\n" + f"group_id: {resolved_group}", + cls.hidden.unique_id, + ) + return IO.NodeOutput(asset_id, resolved_group) + + +class ByteDanceCreateVideoAsset(IO.ComfyNode): + + @classmethod + def define_schema(cls) -> IO.Schema: + return IO.Schema( + node_id="ByteDanceCreateVideoAsset", + display_name="ByteDance Create Video Asset", + category="api node/video/ByteDance", + description=( + "Create a Seedance 2.0 personal video asset. Uploads the input video and " + "registers it in the given asset group. If group_id is empty, runs a real-person " + "H5 authentication flow to create a new group before adding the asset." + ), + inputs=[ + IO.Video.Input("video", tooltip="Video to register as a personal asset."), + IO.String.Input( + "group_id", + default="", + tooltip="Reuse an existing Seedance asset group ID to skip repeated human verification for the " + "same person. Leave empty to run real-person authentication in the browser and create a new group.", + ), + # IO.String.Input( + # "name", + # default="", + # tooltip="Asset name (up to 64 characters).", + # ), + ], + outputs=[ + IO.String.Output(display_name="asset_id"), + IO.String.Output(display_name="group_id"), + ], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + # is_api_node=True, + ) + + @classmethod + async def execute( + cls, + video: Input.Video, + group_id: str = "", + # name: str = "", + ) -> IO.NodeOutput: + # if len(name) > 64: + # raise ValueError("Name of asset can not be greater then 64 symbols") + validate_video_duration(video, min_duration=2, max_duration=15) + validate_video_dimensions(video, min_width=300, max_width=6000, min_height=300, max_height=6000) + + w, h = video.get_dimensions() + if h > 0: + ratio = w / h + if not (0.4 <= ratio <= 2.5): + raise ValueError(f"Asset video aspect ratio (W/H) must be in [0.4, 2.5], got {ratio:.3f} ({w}x{h}).") + pixels = w * h + if not (409_600 <= pixels <= 927_408): + raise ValueError( + f"Asset video total pixels (W×H) must be in [409600, 927408], " f"got {pixels:,} ({w}x{h})." + ) + + fps = float(video.get_frame_rate()) + if not (24 <= fps <= 60): + raise ValueError(f"Asset video FPS must be in [24, 60], got {fps:.2f}.") + + resolved_group = await _resolve_group_id(cls, group_id) + asset_id = await _create_seedance_asset( + cls, + group_id=resolved_group, + url=await upload_video_to_comfyapi(cls, video), + name="", + asset_type="Video", + ) + await _wait_for_asset_active(cls, asset_id, resolved_group) + PromptServer.instance.send_progress_text( + f"Please save the asset_id and group_id for reuse.\n\nasset_id: {asset_id}\n\n" + f"group_id: {resolved_group}", + cls.hidden.unique_id, + ) + return IO.NodeOutput(asset_id, resolved_group) + + class ByteDanceExtension(ComfyExtension): @override async def get_node_list(self) -> list[type[IO.ComfyNode]]: return [ ByteDanceImageNode, ByteDanceSeedreamNode, + ByteDanceSeedreamNodeV2, ByteDanceTextToVideoNode, ByteDanceImageToVideoNode, ByteDanceFirstLastFrameNode, ByteDanceImageReferenceNode, + ByteDance2TextToVideoNode, + ByteDance2FirstLastFrameNode, + ByteDance2ReferenceNode, + ByteDanceCreateImageAsset, + ByteDanceCreateVideoAsset, ] diff --git a/comfy_api_nodes/nodes_gemini.py b/comfy_api_nodes/nodes_gemini.py index 2b77a022e..d18c958a8 100644 --- a/comfy_api_nodes/nodes_gemini.py +++ b/comfy_api_nodes/nodes_gemini.py @@ -83,13 +83,16 @@ class GeminiImageModel(str, Enum): async def create_image_parts( cls: type[IO.ComfyNode], - images: Input.Image, + images: Input.Image | list[Input.Image], image_limit: int = 0, ) -> list[GeminiPart]: image_parts: list[GeminiPart] = [] if image_limit < 0: raise ValueError("image_limit must be greater than or equal to 0 when creating Gemini image parts.") - total_images = get_number_of_images(images) + + # Accept either a single (possibly-batched) tensor or a list of them; share URL budget across all. + images_list: list[Input.Image] = images if isinstance(images, list) else [images] + total_images = sum(get_number_of_images(img) for img in images_list) if total_images <= 0: raise ValueError("No images provided to create_image_parts; at least one image is required.") @@ -98,10 +101,18 @@ async def create_image_parts( # Number of images we'll send as URLs (fileData) num_url_images = min(effective_max, 10) # Vertex API max number of image links + upload_kwargs: dict = {"wait_label": "Uploading reference images"} + if effective_max > num_url_images: + # Split path (e.g. 11+ images): suppress per-image counter to avoid a confusing dual-fraction label. + upload_kwargs = { + "wait_label": f"Uploading reference images ({num_url_images}+)", + "show_batch_index": False, + } reference_images_urls = await upload_images_to_comfyapi( cls, - images, + images_list, max_images=num_url_images, + **upload_kwargs, ) for reference_image_url in reference_images_urls: image_parts.append( @@ -112,15 +123,22 @@ async def create_image_parts( ) ) ) - for idx in range(num_url_images, effective_max): - image_parts.append( - GeminiPart( - inlineData=GeminiInlineData( - mimeType=GeminiMimeType.image_png, - data=tensor_to_base64_string(images[idx]), + if effective_max > num_url_images: + flat: list[torch.Tensor] = [] + for tensor in images_list: + if len(tensor.shape) == 4: + flat.extend(tensor[i] for i in range(tensor.shape[0])) + else: + flat.append(tensor) + for idx in range(num_url_images, effective_max): + image_parts.append( + GeminiPart( + inlineData=GeminiInlineData( + mimeType=GeminiMimeType.image_png, + data=tensor_to_base64_string(flat[idx]), + ) ) ) - ) return image_parts @@ -891,10 +909,6 @@ class GeminiNanoBanana2(IO.ComfyNode): "9:16", "16:9", "21:9", - # "1:4", - # "4:1", - # "8:1", - # "1:8", ], default="auto", tooltip="If set to 'auto', matches your input image's aspect ratio; " @@ -902,12 +916,7 @@ class GeminiNanoBanana2(IO.ComfyNode): ), IO.Combo.Input( "resolution", - options=[ - # "512px", - "1K", - "2K", - "4K", - ], + options=["1K", "2K", "4K"], tooltip="Target output resolution. For 2K/4K the native Gemini upscaler is used.", ), IO.Combo.Input( @@ -956,6 +965,7 @@ class GeminiNanoBanana2(IO.ComfyNode): ], is_api_node=True, price_badge=GEMINI_IMAGE_2_PRICE_BADGE, + is_deprecated=True, ) @classmethod @@ -1016,6 +1026,197 @@ class GeminiNanoBanana2(IO.ComfyNode): ) +def _nano_banana_2_v2_model_inputs(): + return [ + IO.Combo.Input( + "aspect_ratio", + options=[ + "auto", + "1:1", + "2:3", + "3:2", + "3:4", + "4:3", + "4:5", + "5:4", + "9:16", + "16:9", + "21:9", + "1:4", + "4:1", + "8:1", + "1:8", + ], + default="auto", + tooltip="If set to 'auto', matches your input image's aspect ratio; " + "if no image is provided, a 16:9 square is usually generated.", + ), + IO.Combo.Input( + "resolution", + options=["1K", "2K", "4K"], + tooltip="Target output resolution. For 2K/4K the native Gemini upscaler is used.", + ), + IO.Combo.Input( + "thinking_level", + options=["MINIMAL", "HIGH"], + ), + IO.Autogrow.Input( + "images", + template=IO.Autogrow.TemplateNames( + IO.Image.Input("image"), + names=[f"image_{i}" for i in range(1, 15)], + min=0, + ), + tooltip="Optional reference image(s). Up to 14 images total.", + ), + IO.Custom("GEMINI_INPUT_FILES").Input( + "files", + optional=True, + tooltip="Optional file(s) to use as context for the model. " + "Accepts inputs from the Gemini Generate Content Input Files node.", + ), + ] + + +class GeminiNanoBanana2V2(IO.ComfyNode): + + @classmethod + def define_schema(cls): + return IO.Schema( + node_id="GeminiNanoBanana2V2", + display_name="Nano Banana 2", + category="api node/image/Gemini", + description="Generate or edit images synchronously via Google Vertex API.", + inputs=[ + IO.String.Input( + "prompt", + multiline=True, + tooltip="Text prompt describing the image to generate or the edits to apply. " + "Include any constraints, styles, or details the model should follow.", + default="", + ), + IO.DynamicCombo.Input( + "model", + options=[ + IO.DynamicCombo.Option( + "Nano Banana 2 (Gemini 3.1 Flash Image)", + _nano_banana_2_v2_model_inputs(), + ), + ], + ), + IO.Int.Input( + "seed", + default=42, + min=0, + max=0xFFFFFFFFFFFFFFFF, + control_after_generate=True, + tooltip="When the seed is fixed to a specific value, the model makes a best effort to provide " + "the same response for repeated requests. Deterministic output isn't guaranteed. " + "Also, changing the model or parameter settings, such as the temperature, " + "can cause variations in the response even when you use the same seed value. " + "By default, a random seed value is used.", + ), + IO.Combo.Input( + "response_modalities", + options=["IMAGE", "IMAGE+TEXT"], + advanced=True, + ), + IO.String.Input( + "system_prompt", + multiline=True, + default=GEMINI_IMAGE_SYS_PROMPT, + optional=True, + tooltip="Foundational instructions that dictate an AI's behavior.", + advanced=True, + ), + ], + outputs=[ + IO.Image.Output(), + IO.String.Output(), + IO.Image.Output( + display_name="thought_image", + tooltip="First image from the model's thinking process. " + "Only available with thinking_level HIGH and IMAGE+TEXT modality.", + ), + ], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + depends_on=IO.PriceBadgeDepends(widgets=["model", "model.resolution"]), + expr=""" + ( + $r := $lookup(widgets, "model.resolution"); + $prices := {"1k": 0.0696, "2k": 0.1014, "4k": 0.154}; + {"type":"usd","usd": $lookup($prices, $r), "format":{"suffix":"/Image","approximate":true}} + ) + """, + ), + ) + + @classmethod + async def execute( + cls, + prompt: str, + model: dict, + seed: int, + response_modalities: str, + system_prompt: str = "", + ) -> IO.NodeOutput: + validate_string(prompt, strip_whitespace=True, min_length=1) + model_choice = model["model"] + if model_choice == "Nano Banana 2 (Gemini 3.1 Flash Image)": + model_id = "gemini-3.1-flash-image-preview" + else: + model_id = model_choice + + images = model.get("images") or {} + parts: list[GeminiPart] = [GeminiPart(text=prompt)] + if images: + image_tensors: list[Input.Image] = [t for t in images.values() if t is not None] + if image_tensors: + if sum(get_number_of_images(t) for t in image_tensors) > 14: + raise ValueError("The current maximum number of supported images is 14.") + parts.extend(await create_image_parts(cls, image_tensors)) + files = model.get("files") + if files is not None: + parts.extend(files) + + image_config = GeminiImageConfig(imageSize=model["resolution"]) + if model["aspect_ratio"] != "auto": + image_config.aspectRatio = model["aspect_ratio"] + + gemini_system_prompt = None + if system_prompt: + gemini_system_prompt = GeminiSystemInstructionContent(parts=[GeminiTextPart(text=system_prompt)], role=None) + + response = await sync_op( + cls, + ApiEndpoint(path=f"/proxy/vertexai/gemini/{model_id}", method="POST"), + data=GeminiImageGenerateContentRequest( + contents=[ + GeminiContent(role=GeminiRole.user, parts=parts), + ], + generationConfig=GeminiImageGenerationConfig( + responseModalities=(["IMAGE"] if response_modalities == "IMAGE" else ["TEXT", "IMAGE"]), + imageConfig=image_config, + thinkingConfig=GeminiThinkingConfig(thinkingLevel=model["thinking_level"]), + ), + systemInstruction=gemini_system_prompt, + ), + response_model=GeminiGenerateContentResponse, + price_extractor=calculate_tokens_price, + ) + return IO.NodeOutput( + await get_image_from_response(response), + get_text_from_response(response), + await get_image_from_response(response, thought=True), + ) + + class GeminiExtension(ComfyExtension): @override async def get_node_list(self) -> list[type[IO.ComfyNode]]: @@ -1024,6 +1225,7 @@ class GeminiExtension(ComfyExtension): GeminiImage, GeminiImage2, GeminiNanoBanana2, + GeminiNanoBanana2V2, GeminiInputFiles, ] diff --git a/comfy_api_nodes/nodes_grok.py b/comfy_api_nodes/nodes_grok.py index f42d84616..a103f24ee 100644 --- a/comfy_api_nodes/nodes_grok.py +++ b/comfy_api_nodes/nodes_grok.py @@ -54,7 +54,12 @@ class GrokImageNode(IO.ComfyNode): inputs=[ IO.Combo.Input( "model", - options=["grok-imagine-image-pro", "grok-imagine-image", "grok-imagine-image-beta"], + options=[ + "grok-imagine-image-quality", + "grok-imagine-image-pro", + "grok-imagine-image", + "grok-imagine-image-beta", + ], ), IO.String.Input( "prompt", @@ -111,10 +116,12 @@ class GrokImageNode(IO.ComfyNode): ], is_api_node=True, price_badge=IO.PriceBadge( - depends_on=IO.PriceBadgeDepends(widgets=["model", "number_of_images"]), + depends_on=IO.PriceBadgeDepends(widgets=["model", "number_of_images", "resolution"]), expr=""" ( - $rate := $contains(widgets.model, "pro") ? 0.07 : 0.02; + $rate := widgets.model = "grok-imagine-image-quality" + ? (widgets.resolution = "1k" ? 0.05 : 0.07) + : ($contains(widgets.model, "pro") ? 0.07 : 0.02); {"type":"usd","usd": $rate * widgets.number_of_images} ) """, @@ -155,6 +162,61 @@ class GrokImageNode(IO.ComfyNode): ) +_GROK_IMAGE_EDIT_ASPECT_RATIO_OPTIONS = [ + "auto", + "1:1", + "2:3", + "3:2", + "3:4", + "4:3", + "9:16", + "16:9", + "9:19.5", + "19.5:9", + "9:20", + "20:9", + "1:2", + "2:1", +] + + +def _grok_image_edit_model_inputs(*, max_ref_images: int, with_aspect_ratio: bool): + inputs = [ + IO.Autogrow.Input( + "images", + template=IO.Autogrow.TemplateNames( + IO.Image.Input("image"), + names=[f"image_{i}" for i in range(1, max_ref_images + 1)], + min=1, + ), + tooltip=( + "Reference image to edit." + if max_ref_images == 1 + else f"Reference image(s) to edit. Up to {max_ref_images} images." + ), + ), + IO.Combo.Input("resolution", options=["1K", "2K"]), + IO.Int.Input( + "number_of_images", + default=1, + min=1, + max=10, + step=1, + tooltip="Number of edited images to generate", + display_mode=IO.NumberDisplay.number, + ), + ] + if with_aspect_ratio: + inputs.append( + IO.Combo.Input( + "aspect_ratio", + options=_GROK_IMAGE_EDIT_ASPECT_RATIO_OPTIONS, + tooltip="Only allowed when multiple images are connected.", + ) + ) + return inputs + + class GrokImageEditNode(IO.ComfyNode): @classmethod @@ -167,7 +229,12 @@ class GrokImageEditNode(IO.ComfyNode): inputs=[ IO.Combo.Input( "model", - options=["grok-imagine-image-pro", "grok-imagine-image", "grok-imagine-image-beta"], + options=[ + "grok-imagine-image-quality", + "grok-imagine-image-pro", + "grok-imagine-image", + "grok-imagine-image-beta", + ], ), IO.Image.Input("image", display_name="images"), IO.String.Input( @@ -228,14 +295,23 @@ class GrokImageEditNode(IO.ComfyNode): ], is_api_node=True, price_badge=IO.PriceBadge( - depends_on=IO.PriceBadgeDepends(widgets=["model", "number_of_images"]), + depends_on=IO.PriceBadgeDepends(widgets=["model", "number_of_images", "resolution"]), expr=""" ( - $rate := $contains(widgets.model, "pro") ? 0.07 : 0.02; - {"type":"usd","usd": 0.002 + $rate * widgets.number_of_images} + $isQualityModel := widgets.model = "grok-imagine-image-quality"; + $isPro := $contains(widgets.model, "pro"); + $rate := $isQualityModel + ? (widgets.resolution = "1k" ? 0.05 : 0.07) + : ($isPro ? 0.07 : 0.02); + $base := $isQualityModel ? 0.01 : 0.002; + $output := $rate * widgets.number_of_images; + $isPro + ? {"type":"usd","usd": $base + $output} + : {"type":"range_usd","min_usd": $base + $output, "max_usd": 3 * $base + $output} ) """, ), + is_deprecated=True, ) @classmethod @@ -283,6 +359,143 @@ class GrokImageEditNode(IO.ComfyNode): ) +class GrokImageEditNodeV2(IO.ComfyNode): + + @classmethod + def define_schema(cls): + return IO.Schema( + node_id="GrokImageEditNodeV2", + display_name="Grok Image Edit", + category="api node/image/Grok", + description="Modify an existing image based on a text prompt", + inputs=[ + IO.String.Input( + "prompt", + multiline=True, + default="", + tooltip="The text prompt used to generate the image", + ), + IO.DynamicCombo.Input( + "model", + options=[ + IO.DynamicCombo.Option( + "grok-imagine-image-quality", + _grok_image_edit_model_inputs(max_ref_images=3, with_aspect_ratio=True), + ), + IO.DynamicCombo.Option( + "grok-imagine-image-pro", + _grok_image_edit_model_inputs(max_ref_images=1, with_aspect_ratio=False), + ), + IO.DynamicCombo.Option( + "grok-imagine-image", + _grok_image_edit_model_inputs(max_ref_images=3, with_aspect_ratio=True), + ), + ], + ), + IO.Int.Input( + "seed", + default=0, + min=0, + max=2147483647, + step=1, + display_mode=IO.NumberDisplay.number, + control_after_generate=True, + tooltip="Seed to determine if node should re-run; " + "actual results are nondeterministic regardless of seed.", + ), + ], + outputs=[ + IO.Image.Output(), + ], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + depends_on=IO.PriceBadgeDepends( + widgets=["model", "model.resolution", "model.number_of_images"], + ), + expr=""" + ( + $isQualityModel := widgets.model = "grok-imagine-image-quality"; + $isPro := $contains(widgets.model, "pro"); + $res := $lookup(widgets, "model.resolution"); + $n := $lookup(widgets, "model.number_of_images"); + $rate := $isQualityModel + ? ($res = "1k" ? 0.05 : 0.07) + : ($isPro ? 0.07 : 0.02); + $base := $isQualityModel ? 0.01 : 0.002; + $output := $rate * $n; + $isPro + ? {"type":"usd","usd": $base + $output} + : {"type":"range_usd","min_usd": $base + $output, "max_usd": 3 * $base + $output} + ) + """, + ), + ) + + @classmethod + async def execute( + cls, + prompt: str, + model: dict, + seed: int, + ) -> IO.NodeOutput: + validate_string(prompt, strip_whitespace=True, min_length=1) + model_id = model["model"] + resolution = model["resolution"] + number_of_images = model["number_of_images"] + images_dict = model.get("images") or {} + aspect_ratio = model.get("aspect_ratio", "auto") + + image_tensors: list[Input.Image] = [t for t in images_dict.values() if t is not None] + n_images = sum(get_number_of_images(t) for t in image_tensors) + if n_images < 1: + raise ValueError("At least one image is required for editing.") + if model_id == "grok-imagine-image-pro" and n_images > 1: + raise ValueError("The pro model supports only 1 input image.") + if model_id != "grok-imagine-image-pro" and n_images > 3: + raise ValueError("A maximum of 3 input images is supported.") + if aspect_ratio != "auto" and n_images == 1: + raise ValueError( + "Custom aspect ratio is only allowed when multiple images are connected to the image input." + ) + + flat_tensors: list[torch.Tensor] = [] + for tensor in image_tensors: + if len(tensor.shape) == 4: + flat_tensors.extend(tensor[i] for i in range(tensor.shape[0])) + else: + flat_tensors.append(tensor) + + response = await sync_op( + cls, + ApiEndpoint(path="/proxy/xai/v1/images/edits", method="POST"), + data=ImageEditRequest( + model=model_id, + images=[ + InputUrlObject(url=f"data:image/png;base64,{tensor_to_base64_string(i)}") for i in flat_tensors + ], + prompt=prompt, + resolution=resolution.lower(), + n=number_of_images, + seed=seed, + aspect_ratio=None if aspect_ratio == "auto" else aspect_ratio, + ), + response_model=ImageGenerationResponse, + price_extractor=_extract_grok_price, + ) + if len(response.data) == 1: + return IO.NodeOutput(await download_url_to_image_tensor(response.data[0].url)) + return IO.NodeOutput( + torch.cat( + [await download_url_to_image_tensor(i) for i in [str(d.url) for d in response.data if d.url]], + ) + ) + + class GrokVideoNode(IO.ComfyNode): @classmethod @@ -717,6 +930,7 @@ class GrokExtension(ComfyExtension): return [ GrokImageNode, GrokImageEditNode, + GrokImageEditNodeV2, GrokVideoNode, GrokVideoReferenceNode, GrokVideoEditNode, diff --git a/comfy_api_nodes/nodes_hitpaw.py b/comfy_api_nodes/nodes_hitpaw.py index 488080a74..bca5170e4 100644 --- a/comfy_api_nodes/nodes_hitpaw.py +++ b/comfy_api_nodes/nodes_hitpaw.py @@ -178,7 +178,6 @@ class HitPawGeneralImageEnhance(IO.ComfyNode): status_extractor=lambda x: x.data.status, price_extractor=lambda x: request_price, poll_interval=10.0, - max_poll_attempts=480, ) return IO.NodeOutput(await download_url_to_image_tensor(final_response.data.res_url)) @@ -324,7 +323,6 @@ class HitPawVideoEnhance(IO.ComfyNode): status_extractor=lambda x: x.data.status, price_extractor=lambda x: request_price, poll_interval=10.0, - max_poll_attempts=320, ) return IO.NodeOutput(await download_url_to_video_output(final_response.data.res_url)) diff --git a/comfy_api_nodes/nodes_hunyuan3d.py b/comfy_api_nodes/nodes_hunyuan3d.py index 44c94a98e..5fc31bccd 100644 --- a/comfy_api_nodes/nodes_hunyuan3d.py +++ b/comfy_api_nodes/nodes_hunyuan3d.py @@ -221,14 +221,17 @@ class TencentTextToModelNode(IO.ComfyNode): response_model=To3DProTaskResultResponse, status_extractor=lambda r: r.Status, ) - obj_result = await download_and_extract_obj_zip(get_file_from_response(result.ResultFile3Ds, "obj").Url) + obj_file_response = get_file_from_response(result.ResultFile3Ds, "obj", raise_if_not_found=False) + obj_result = None + if obj_file_response: + obj_result = await download_and_extract_obj_zip(obj_file_response.Url) return IO.NodeOutput( f"{task_id}.glb", await download_url_to_file_3d( get_file_from_response(result.ResultFile3Ds, "glb").Url, "glb", task_id=task_id ), - obj_result.obj, - obj_result.texture, + obj_result.obj if obj_result else None, + obj_result.texture if obj_result else None, ) @@ -378,17 +381,30 @@ class TencentImageToModelNode(IO.ComfyNode): response_model=To3DProTaskResultResponse, status_extractor=lambda r: r.Status, ) - obj_result = await download_and_extract_obj_zip(get_file_from_response(result.ResultFile3Ds, "obj").Url) + obj_file_response = get_file_from_response(result.ResultFile3Ds, "obj", raise_if_not_found=False) + if obj_file_response: + obj_result = await download_and_extract_obj_zip(obj_file_response.Url) + return IO.NodeOutput( + f"{task_id}.glb", + await download_url_to_file_3d( + get_file_from_response(result.ResultFile3Ds, "glb").Url, "glb", task_id=task_id + ), + obj_result.obj, + obj_result.texture, + obj_result.metallic if obj_result.metallic is not None else torch.zeros(1, 1, 1, 3), + obj_result.normal if obj_result.normal is not None else torch.zeros(1, 1, 1, 3), + obj_result.roughness if obj_result.roughness is not None else torch.zeros(1, 1, 1, 3), + ) return IO.NodeOutput( f"{task_id}.glb", await download_url_to_file_3d( get_file_from_response(result.ResultFile3Ds, "glb").Url, "glb", task_id=task_id ), - obj_result.obj, - obj_result.texture, - obj_result.metallic if obj_result.metallic is not None else torch.zeros(1, 1, 1, 3), - obj_result.normal if obj_result.normal is not None else torch.zeros(1, 1, 1, 3), - obj_result.roughness if obj_result.roughness is not None else torch.zeros(1, 1, 1, 3), + None, + None, + None, + None, + None, ) diff --git a/comfy_api_nodes/nodes_kling.py b/comfy_api_nodes/nodes_kling.py index 9a37ccc53..7586f1816 100644 --- a/comfy_api_nodes/nodes_kling.py +++ b/comfy_api_nodes/nodes_kling.py @@ -862,7 +862,7 @@ class OmniProTextToVideoNode(IO.ComfyNode): ), IO.Combo.Input("aspect_ratio", options=["16:9", "9:16", "1:1"]), IO.Int.Input("duration", default=5, min=3, max=15, display_mode=IO.NumberDisplay.slider), - IO.Combo.Input("resolution", options=["1080p", "720p"], optional=True), + IO.Combo.Input("resolution", options=["4k", "1080p", "720p"], default="1080p", optional=True), IO.DynamicCombo.Input( "storyboards", options=[ @@ -904,12 +904,13 @@ class OmniProTextToVideoNode(IO.ComfyNode): depends_on=IO.PriceBadgeDepends(widgets=["duration", "resolution", "model_name", "generate_audio"]), expr=""" ( - $mode := (widgets.resolution = "720p") ? "std" : "pro"; + $res := widgets.resolution; + $mode := $res = "4k" ? "4k" : ($res = "720p" ? "std" : "pro"); $isV3 := $contains(widgets.model_name, "v3"); $audio := $isV3 and widgets.generate_audio; $rates := $audio - ? {"std": 0.112, "pro": 0.14} - : {"std": 0.084, "pro": 0.112}; + ? {"std": 0.112, "pro": 0.14, "4k": 0.42} + : {"std": 0.084, "pro": 0.112, "4k": 0.42}; {"type":"usd","usd": $lookup($rates, $mode) * widgets.duration} ) """, @@ -934,6 +935,8 @@ class OmniProTextToVideoNode(IO.ComfyNode): raise ValueError("kling-video-o1 only supports durations of 5 or 10 seconds.") if generate_audio: raise ValueError("kling-video-o1 does not support audio generation.") + if resolution == "4k": + raise ValueError("kling-video-o1 does not support 4k resolution.") stories_enabled = storyboards is not None and storyboards["storyboards"] != "disabled" if stories_enabled and model_name == "kling-video-o1": raise ValueError("kling-video-o1 does not support storyboards.") @@ -963,6 +966,12 @@ class OmniProTextToVideoNode(IO.ComfyNode): f"must equal the global duration ({duration}s)." ) + if resolution == "4k": + mode = "4k" + elif resolution == "1080p": + mode = "pro" + else: + mode = "std" response = await sync_op( cls, ApiEndpoint(path="/proxy/kling/v1/videos/omni-video", method="POST"), @@ -972,7 +981,7 @@ class OmniProTextToVideoNode(IO.ComfyNode): prompt=prompt, aspect_ratio=aspect_ratio, duration=str(duration), - mode="pro" if resolution == "1080p" else "std", + mode=mode, multi_shot=multi_shot, multi_prompt=multi_prompt_list, shot_type="customize" if multi_shot else None, @@ -1014,7 +1023,7 @@ class OmniProFirstLastFrameNode(IO.ComfyNode): optional=True, tooltip="Up to 6 additional reference images.", ), - IO.Combo.Input("resolution", options=["1080p", "720p"], optional=True), + IO.Combo.Input("resolution", options=["4k", "1080p", "720p"], default="1080p", optional=True), IO.DynamicCombo.Input( "storyboards", options=[ @@ -1061,12 +1070,13 @@ class OmniProFirstLastFrameNode(IO.ComfyNode): depends_on=IO.PriceBadgeDepends(widgets=["duration", "resolution", "model_name", "generate_audio"]), expr=""" ( - $mode := (widgets.resolution = "720p") ? "std" : "pro"; + $res := widgets.resolution; + $mode := $res = "4k" ? "4k" : ($res = "720p" ? "std" : "pro"); $isV3 := $contains(widgets.model_name, "v3"); $audio := $isV3 and widgets.generate_audio; $rates := $audio - ? {"std": 0.112, "pro": 0.14} - : {"std": 0.084, "pro": 0.112}; + ? {"std": 0.112, "pro": 0.14, "4k": 0.42} + : {"std": 0.084, "pro": 0.112, "4k": 0.42}; {"type":"usd","usd": $lookup($rates, $mode) * widgets.duration} ) """, @@ -1093,6 +1103,8 @@ class OmniProFirstLastFrameNode(IO.ComfyNode): raise ValueError("kling-video-o1 does not support durations greater than 10 seconds.") if generate_audio: raise ValueError("kling-video-o1 does not support audio generation.") + if resolution == "4k": + raise ValueError("kling-video-o1 does not support 4k resolution.") stories_enabled = storyboards is not None and storyboards["storyboards"] != "disabled" if stories_enabled and model_name == "kling-video-o1": raise ValueError("kling-video-o1 does not support storyboards.") @@ -1161,6 +1173,12 @@ class OmniProFirstLastFrameNode(IO.ComfyNode): validate_image_aspect_ratio(i, (1, 2.5), (2.5, 1)) for i in await upload_images_to_comfyapi(cls, reference_images, wait_label="Uploading reference frame(s)"): image_list.append(OmniParamImage(image_url=i)) + if resolution == "4k": + mode = "4k" + elif resolution == "1080p": + mode = "pro" + else: + mode = "std" response = await sync_op( cls, ApiEndpoint(path="/proxy/kling/v1/videos/omni-video", method="POST"), @@ -1170,7 +1188,7 @@ class OmniProFirstLastFrameNode(IO.ComfyNode): prompt=prompt, duration=str(duration), image_list=image_list, - mode="pro" if resolution == "1080p" else "std", + mode=mode, sound="on" if generate_audio else "off", multi_shot=multi_shot, multi_prompt=multi_prompt_list, @@ -1204,7 +1222,7 @@ class OmniProImageToVideoNode(IO.ComfyNode): "reference_images", tooltip="Up to 7 reference images.", ), - IO.Combo.Input("resolution", options=["1080p", "720p"], optional=True), + IO.Combo.Input("resolution", options=["4k", "1080p", "720p"], default="1080p", optional=True), IO.DynamicCombo.Input( "storyboards", options=[ @@ -1251,12 +1269,13 @@ class OmniProImageToVideoNode(IO.ComfyNode): depends_on=IO.PriceBadgeDepends(widgets=["duration", "resolution", "model_name", "generate_audio"]), expr=""" ( - $mode := (widgets.resolution = "720p") ? "std" : "pro"; + $res := widgets.resolution; + $mode := $res = "4k" ? "4k" : ($res = "720p" ? "std" : "pro"); $isV3 := $contains(widgets.model_name, "v3"); $audio := $isV3 and widgets.generate_audio; $rates := $audio - ? {"std": 0.112, "pro": 0.14} - : {"std": 0.084, "pro": 0.112}; + ? {"std": 0.112, "pro": 0.14, "4k": 0.42} + : {"std": 0.084, "pro": 0.112, "4k": 0.42}; {"type":"usd","usd": $lookup($rates, $mode) * widgets.duration} ) """, @@ -1282,6 +1301,8 @@ class OmniProImageToVideoNode(IO.ComfyNode): raise ValueError("kling-video-o1 does not support durations greater than 10 seconds.") if generate_audio: raise ValueError("kling-video-o1 does not support audio generation.") + if resolution == "4k": + raise ValueError("kling-video-o1 does not support 4k resolution.") stories_enabled = storyboards is not None and storyboards["storyboards"] != "disabled" if stories_enabled and model_name == "kling-video-o1": raise ValueError("kling-video-o1 does not support storyboards.") @@ -1320,6 +1341,12 @@ class OmniProImageToVideoNode(IO.ComfyNode): image_list: list[OmniParamImage] = [] for i in await upload_images_to_comfyapi(cls, reference_images, wait_label="Uploading reference image"): image_list.append(OmniParamImage(image_url=i)) + if resolution == "4k": + mode = "4k" + elif resolution == "1080p": + mode = "pro" + else: + mode = "std" response = await sync_op( cls, ApiEndpoint(path="/proxy/kling/v1/videos/omni-video", method="POST"), @@ -1330,7 +1357,7 @@ class OmniProImageToVideoNode(IO.ComfyNode): aspect_ratio=aspect_ratio, duration=str(duration), image_list=image_list, - mode="pro" if resolution == "1080p" else "std", + mode=mode, sound="on" if generate_audio else "off", multi_shot=multi_shot, multi_prompt=multi_prompt_list, @@ -2760,11 +2787,15 @@ class MotionControl(IO.ComfyNode): ], is_api_node=True, price_badge=IO.PriceBadge( - depends_on=IO.PriceBadgeDepends(widgets=["mode"]), + depends_on=IO.PriceBadgeDepends(widgets=["mode", "model"]), expr=""" ( - $prices := {"std": 0.07, "pro": 0.112}; - {"type":"usd","usd": $lookup($prices, widgets.mode), "format":{"suffix":"/second"}} + $prices := { + "kling-v3": {"std": 0.126, "pro": 0.168}, + "kling-v2-6": {"std": 0.07, "pro": 0.112} + }; + $modelPrices := $lookup($prices, widgets.model); + {"type":"usd","usd": $lookup($modelPrices, widgets.mode), "format":{"suffix":"/second"}} ) """, ), @@ -2860,7 +2891,7 @@ class KlingVideoNode(IO.ComfyNode): IO.DynamicCombo.Option( "kling-v3", [ - IO.Combo.Input("resolution", options=["1080p", "720p"]), + IO.Combo.Input("resolution", options=["4k", "1080p", "720p"], default="1080p"), IO.Combo.Input( "aspect_ratio", options=["16:9", "9:16", "1:1"], @@ -2913,7 +2944,11 @@ class KlingVideoNode(IO.ComfyNode): ), expr=""" ( - $rates := {"1080p": {"off": 0.112, "on": 0.168}, "720p": {"off": 0.084, "on": 0.126}}; + $rates := { + "4k": {"off": 0.42, "on": 0.42}, + "1080p": {"off": 0.112, "on": 0.168}, + "720p": {"off": 0.084, "on": 0.126} + }; $res := $lookup(widgets, "model.resolution"); $audio := widgets.generate_audio ? "on" : "off"; $rate := $lookup($lookup($rates, $res), $audio); @@ -2943,7 +2978,12 @@ class KlingVideoNode(IO.ComfyNode): start_frame: Input.Image | None = None, ) -> IO.NodeOutput: _ = seed - mode = "pro" if model["resolution"] == "1080p" else "std" + if model["resolution"] == "4k": + mode = "4k" + elif model["resolution"] == "1080p": + mode = "pro" + else: + mode = "std" custom_multi_shot = False if multi_shot["multi_shot"] == "disabled": shot_type = None @@ -3057,7 +3097,7 @@ class KlingFirstLastFrameNode(IO.ComfyNode): IO.DynamicCombo.Option( "kling-v3", [ - IO.Combo.Input("resolution", options=["1080p", "720p"]), + IO.Combo.Input("resolution", options=["4k", "1080p", "720p"], default="1080p"), ], ), ], @@ -3089,7 +3129,11 @@ class KlingFirstLastFrameNode(IO.ComfyNode): ), expr=""" ( - $rates := {"1080p": {"off": 0.112, "on": 0.168}, "720p": {"off": 0.084, "on": 0.126}}; + $rates := { + "4k": {"off": 0.42, "on": 0.42}, + "1080p": {"off": 0.112, "on": 0.168}, + "720p": {"off": 0.084, "on": 0.126} + }; $res := $lookup(widgets, "model.resolution"); $audio := widgets.generate_audio ? "on" : "off"; $rate := $lookup($lookup($rates, $res), $audio); @@ -3118,6 +3162,12 @@ class KlingFirstLastFrameNode(IO.ComfyNode): validate_image_aspect_ratio(end_frame, (1, 2.5), (2.5, 1)) image_url = await upload_image_to_comfyapi(cls, first_frame, wait_label="Uploading first frame") image_tail_url = await upload_image_to_comfyapi(cls, end_frame, wait_label="Uploading end frame") + if model["resolution"] == "4k": + mode = "4k" + elif model["resolution"] == "1080p": + mode = "pro" + else: + mode = "std" response = await sync_op( cls, ApiEndpoint(path="/proxy/kling/v1/videos/image2video", method="POST"), @@ -3127,7 +3177,7 @@ class KlingFirstLastFrameNode(IO.ComfyNode): image=image_url, image_tail=image_tail_url, prompt=prompt, - mode="pro" if model["resolution"] == "1080p" else "std", + mode=mode, duration=str(duration), sound="on" if generate_audio else "off", ), diff --git a/comfy_api_nodes/nodes_luma.py b/comfy_api_nodes/nodes_luma.py index 9ed6cd299..d92a7c382 100644 --- a/comfy_api_nodes/nodes_luma.py +++ b/comfy_api_nodes/nodes_luma.py @@ -1,10 +1,11 @@ -from typing import Optional - import torch from typing_extensions import override -from comfy_api.latest import IO, ComfyExtension +from comfy_api.latest import IO, ComfyExtension, Input from comfy_api_nodes.apis.luma import ( + Luma2Generation, + Luma2GenerationRequest, + Luma2ImageRef, LumaAspectRatio, LumaCharacterRef, LumaConceptChain, @@ -30,6 +31,7 @@ from comfy_api_nodes.util import ( download_url_to_video_output, poll_op, sync_op, + upload_image_to_comfyapi, upload_images_to_comfyapi, validate_string, ) @@ -212,9 +214,9 @@ class LumaImageGenerationNode(IO.ComfyNode): aspect_ratio: str, seed, style_image_weight: float, - image_luma_ref: Optional[LumaReferenceChain] = None, - style_image: Optional[torch.Tensor] = None, - character_image: Optional[torch.Tensor] = None, + image_luma_ref: LumaReferenceChain | None = None, + style_image: torch.Tensor | None = None, + character_image: torch.Tensor | None = None, ) -> IO.NodeOutput: validate_string(prompt, strip_whitespace=True, min_length=3) # handle image_luma_ref @@ -434,7 +436,7 @@ class LumaTextToVideoGenerationNode(IO.ComfyNode): duration: str, loop: bool, seed, - luma_concepts: Optional[LumaConceptChain] = None, + luma_concepts: LumaConceptChain | None = None, ) -> IO.NodeOutput: validate_string(prompt, strip_whitespace=False, min_length=3) duration = duration if model != LumaVideoModel.ray_1_6 else None @@ -533,7 +535,6 @@ class LumaImageToVideoGenerationNode(IO.ComfyNode): ], is_api_node=True, price_badge=PRICE_BADGE_VIDEO, - ) @classmethod @@ -644,6 +645,293 @@ PRICE_BADGE_VIDEO = IO.PriceBadge( ) +def _luma2_uni1_common_inputs(max_image_refs: int) -> list: + return [ + IO.Combo.Input( + "style", + options=["auto", "manga"], + default="auto", + tooltip="Style preset. 'auto' picks based on the prompt; " + "'manga' applies a manga/anime aesthetic and requires a portrait " + "aspect ratio (2:3, 9:16, 1:2, 1:3).", + ), + IO.Boolean.Input( + "web_search", + default=False, + tooltip="Search the web for visual references before generating.", + ), + IO.Autogrow.Input( + "image_ref", + template=IO.Autogrow.TemplateNames( + IO.Image.Input("image"), + names=[f"image_{i}" for i in range(1, max_image_refs + 1)], + min=0, + ), + optional=True, + tooltip=f"Up to {max_image_refs} reference images for style/content guidance.", + ), + ] + + +async def _luma2_upload_image_refs( + cls: type[IO.ComfyNode], + refs: dict | None, + max_count: int, +) -> list[Luma2ImageRef] | None: + if not refs: + return None + out: list[Luma2ImageRef] = [] + for key in refs: + url = await upload_image_to_comfyapi(cls, refs[key]) + out.append(Luma2ImageRef(url=url)) + if len(out) > max_count: + raise ValueError(f"Maximum {max_count} reference images are allowed.") + return out or None + + +async def _luma2_submit_and_poll( + cls: type[IO.ComfyNode], + request: Luma2GenerationRequest, +) -> Input.Image: + initial = await sync_op( + cls, + ApiEndpoint(path="/proxy/luma_2/generations", method="POST"), + response_model=Luma2Generation, + data=request, + ) + if not initial.id: + raise RuntimeError("Luma 2 API did not return a generation id.") + final = await poll_op( + cls, + ApiEndpoint(path=f"/proxy/luma_2/generations/{initial.id}", method="GET"), + response_model=Luma2Generation, + status_extractor=lambda r: r.state, + progress_extractor=lambda r: None, + ) + if not final.output: + msg = final.failure_reason or "no output returned" + raise RuntimeError(f"Luma 2 generation failed: {msg}") + url = final.output[0].url + if not url: + raise RuntimeError("Luma 2 generation completed without an output URL.") + return await download_url_to_image_tensor(url) + + +class LumaImageNode(IO.ComfyNode): + + @classmethod + def define_schema(cls) -> IO.Schema: + return IO.Schema( + node_id="LumaImageNode2", + display_name="Luma UNI-1 Image", + category="api node/image/Luma", + description="Generate images from text using the Luma UNI-1 model.", + inputs=[ + IO.String.Input( + "prompt", + multiline=True, + default="", + tooltip="Text description of the desired image. 1–6000 characters.", + ), + IO.DynamicCombo.Input( + "model", + options=[ + IO.DynamicCombo.Option( + "uni-1", + [ + IO.Combo.Input( + "aspect_ratio", + options=[ + "auto", + "3:1", + "2:1", + "16:9", + "3:2", + "1:1", + "2:3", + "9:16", + "1:2", + "1:3", + ], + default="auto", + tooltip="Output image aspect ratio. 'auto' lets " + "the model pick based on the prompt.", + ), + *_luma2_uni1_common_inputs(max_image_refs=9), + ], + ), + IO.DynamicCombo.Option( + "uni-1-max", + [ + IO.Combo.Input( + "aspect_ratio", + options=[ + "auto", + "3:1", + "2:1", + "16:9", + "3:2", + "1:1", + "2:3", + "9:16", + "1:2", + "1:3", + ], + default="auto", + tooltip="Output image aspect ratio. 'auto' lets " + "the model pick based on the prompt.", + ), + *_luma2_uni1_common_inputs(max_image_refs=9), + ], + ), + ], + tooltip="Model to use for generation.", + ), + IO.Int.Input( + "seed", + default=0, + min=0, + max=2147483647, + control_after_generate=True, + tooltip="Seed controls whether the node should re-run; " + "results are non-deterministic regardless of seed.", + ), + ], + outputs=[IO.Image.Output()], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + depends_on=IO.PriceBadgeDepends(widgets=["model"], input_groups=["model.image_ref"]), + expr=""" + ( + $m := widgets.model; + $refs := $lookup(inputGroups, "model.image_ref"); + $base := $m = "uni-1-max" ? 0.1 : 0.0404; + {"type":"usd","usd": $round($base + 0.003 * $refs, 4)} + ) + """, + ), + ) + + @classmethod + async def execute( + cls, + prompt: str, + model: dict, + seed: int, + ) -> IO.NodeOutput: + validate_string(prompt, min_length=1, max_length=6000) + aspect_ratio = model["aspect_ratio"] + style = model["style"] + allowed_manga_ratios = {"2:3", "9:16", "1:2", "1:3"} + if style == "manga" and aspect_ratio != "auto" and aspect_ratio not in allowed_manga_ratios: + raise ValueError( + f"'manga' style requires a portrait aspect ratio " + f"({', '.join(sorted(allowed_manga_ratios))}) or 'auto'; got '{aspect_ratio}'." + ) + request = Luma2GenerationRequest( + prompt=prompt, + model=model["model"], + type="image", + aspect_ratio=aspect_ratio if aspect_ratio != "auto" else None, + style=style if style != "auto" else None, + output_format="png", + web_search=model["web_search"], + image_ref=await _luma2_upload_image_refs(cls, model.get("image_ref"), max_count=9), + ) + return IO.NodeOutput(await _luma2_submit_and_poll(cls, request)) + + +class LumaImageEditNode(IO.ComfyNode): + + @classmethod + def define_schema(cls) -> IO.Schema: + return IO.Schema( + node_id="LumaImageEditNode2", + display_name="Luma UNI-1 Image Edit", + category="api node/image/Luma", + description="Edit an existing image with a text prompt using the Luma UNI-1 model.", + inputs=[ + IO.Image.Input( + "source", + tooltip="Source image to edit.", + ), + IO.String.Input( + "prompt", + multiline=True, + default="", + tooltip="Description of the desired edit. 1–6000 characters.", + ), + IO.DynamicCombo.Input( + "model", + options=[ + IO.DynamicCombo.Option( + "uni-1", + _luma2_uni1_common_inputs(max_image_refs=8), + ), + IO.DynamicCombo.Option( + "uni-1-max", + _luma2_uni1_common_inputs(max_image_refs=8), + ), + ], + tooltip="Model to use for editing.", + ), + IO.Int.Input( + "seed", + default=0, + min=0, + max=2147483647, + control_after_generate=True, + tooltip="Seed controls whether the node should re-run; " + "results are non-deterministic regardless of seed.", + ), + ], + outputs=[IO.Image.Output()], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + depends_on=IO.PriceBadgeDepends(widgets=["model"], input_groups=["model.image_ref"]), + expr=""" + ( + $m := widgets.model; + $refs := $lookup(inputGroups, "model.image_ref"); + $base := $m = "uni-1-max" ? 0.103 : 0.0434; + {"type":"usd","usd": $round($base + 0.003 * $refs, 4)} + ) + """, + ), + ) + + @classmethod + async def execute( + cls, + source: Input.Image, + prompt: str, + model: dict, + seed: int, + ) -> IO.NodeOutput: + validate_string(prompt, min_length=1, max_length=6000) + request = Luma2GenerationRequest( + prompt=prompt, + model=model["model"], + type="image_edit", + source=Luma2ImageRef(url=await upload_image_to_comfyapi(cls, source)), + style=model["style"] if model["style"] != "auto" else None, + output_format="png", + web_search=model["web_search"], + image_ref=await _luma2_upload_image_refs(cls, model.get("image_ref"), max_count=8), + ) + return IO.NodeOutput(await _luma2_submit_and_poll(cls, request)) + + class LumaExtension(ComfyExtension): @override async def get_node_list(self) -> list[type[IO.ComfyNode]]: @@ -654,6 +942,8 @@ class LumaExtension(ComfyExtension): LumaImageToVideoGenerationNode, LumaReferenceNode, LumaConceptsNode, + LumaImageNode, + LumaImageEditNode, ] diff --git a/comfy_api_nodes/nodes_magnific.py b/comfy_api_nodes/nodes_magnific.py index 0f53208d4..38b881fea 100644 --- a/comfy_api_nodes/nodes_magnific.py +++ b/comfy_api_nodes/nodes_magnific.py @@ -230,7 +230,6 @@ class MagnificImageUpscalerCreativeNode(IO.ComfyNode): status_extractor=lambda x: x.status, price_extractor=lambda _: price_usd, poll_interval=10.0, - max_poll_attempts=480, ) return IO.NodeOutput(await download_url_to_image_tensor(final_response.generated[0])) @@ -391,7 +390,6 @@ class MagnificImageUpscalerPreciseV2Node(IO.ComfyNode): status_extractor=lambda x: x.status, price_extractor=lambda _: price_usd, poll_interval=10.0, - max_poll_attempts=480, ) return IO.NodeOutput(await download_url_to_image_tensor(final_response.generated[0])) @@ -541,7 +539,6 @@ class MagnificImageStyleTransferNode(IO.ComfyNode): response_model=TaskResponse, status_extractor=lambda x: x.status, poll_interval=10.0, - max_poll_attempts=480, ) return IO.NodeOutput(await download_url_to_image_tensor(final_response.generated[0])) @@ -782,7 +779,6 @@ class MagnificImageRelightNode(IO.ComfyNode): response_model=TaskResponse, status_extractor=lambda x: x.status, poll_interval=10.0, - max_poll_attempts=480, ) return IO.NodeOutput(await download_url_to_image_tensor(final_response.generated[0])) @@ -924,7 +920,6 @@ class MagnificImageSkinEnhancerNode(IO.ComfyNode): response_model=TaskResponse, status_extractor=lambda x: x.status, poll_interval=10.0, - max_poll_attempts=480, ) return IO.NodeOutput(await download_url_to_image_tensor(final_response.generated[0])) diff --git a/comfy_api_nodes/nodes_moonvalley.py b/comfy_api_nodes/nodes_moonvalley.py deleted file mode 100644 index 78a230529..000000000 --- a/comfy_api_nodes/nodes_moonvalley.py +++ /dev/null @@ -1,534 +0,0 @@ -import logging - -from typing_extensions import override - -from comfy_api.latest import IO, ComfyExtension, Input -from comfy_api_nodes.apis.moonvalley import ( - MoonvalleyPromptResponse, - MoonvalleyTextToVideoInferenceParams, - MoonvalleyTextToVideoRequest, - MoonvalleyVideoToVideoInferenceParams, - MoonvalleyVideoToVideoRequest, -) -from comfy_api_nodes.util import ( - ApiEndpoint, - download_url_to_video_output, - poll_op, - sync_op, - trim_video, - upload_images_to_comfyapi, - upload_video_to_comfyapi, - validate_container_format_is_mp4, - validate_image_dimensions, - validate_string, -) - -API_UPLOADS_ENDPOINT = "/proxy/moonvalley/uploads" -API_PROMPTS_ENDPOINT = "/proxy/moonvalley/prompts" -API_VIDEO2VIDEO_ENDPOINT = "/proxy/moonvalley/prompts/video-to-video" -API_TXT2VIDEO_ENDPOINT = "/proxy/moonvalley/prompts/text-to-video" -API_IMG2VIDEO_ENDPOINT = "/proxy/moonvalley/prompts/image-to-video" - -MIN_WIDTH = 300 -MIN_HEIGHT = 300 - -MAX_WIDTH = 10000 -MAX_HEIGHT = 10000 - -MIN_VID_WIDTH = 300 -MIN_VID_HEIGHT = 300 - -MAX_VID_WIDTH = 10000 -MAX_VID_HEIGHT = 10000 - -MAX_VIDEO_SIZE = 1024 * 1024 * 1024 # 1 GB max for in-memory video processing - -MOONVALLEY_MAREY_MAX_PROMPT_LENGTH = 5000 - - -def is_valid_task_creation_response(response: MoonvalleyPromptResponse) -> bool: - """Verifies that the initial response contains a task ID.""" - return bool(response.id) - - -def validate_task_creation_response(response) -> None: - if not is_valid_task_creation_response(response): - error_msg = f"Moonvalley Marey API: Initial request failed. Code: {response.code}, Message: {response.message}, Data: {response}" - logging.error(error_msg) - raise RuntimeError(error_msg) - - -def validate_video_to_video_input(video: Input.Video) -> Input.Video: - """ - Validates and processes video input for Moonvalley Video-to-Video generation. - - Args: - video: Input video to validate - - Returns: - Validated and potentially trimmed video - - Raises: - ValueError: If video doesn't meet requirements - MoonvalleyApiError: If video duration is too short - """ - width, height = _get_video_dimensions(video) - _validate_video_dimensions(width, height) - validate_container_format_is_mp4(video) - - return _validate_and_trim_duration(video) - - -def _get_video_dimensions(video: Input.Video) -> tuple[int, int]: - """Extracts video dimensions with error handling.""" - try: - return video.get_dimensions() - except Exception as e: - logging.error("Error getting dimensions of video: %s", e) - raise ValueError(f"Cannot get video dimensions: {e}") from e - - -def _validate_video_dimensions(width: int, height: int) -> None: - """Validates video dimensions meet Moonvalley V2V requirements.""" - supported_resolutions = { - (1920, 1080), - (1080, 1920), - (1152, 1152), - (1536, 1152), - (1152, 1536), - } - - if (width, height) not in supported_resolutions: - supported_list = ", ".join([f"{w}x{h}" for w, h in sorted(supported_resolutions)]) - raise ValueError(f"Resolution {width}x{height} not supported. Supported: {supported_list}") - - -def _validate_and_trim_duration(video: Input.Video) -> Input.Video: - """Validates video duration and trims to 5 seconds if needed.""" - duration = video.get_duration() - _validate_minimum_duration(duration) - return _trim_if_too_long(video, duration) - - -def _validate_minimum_duration(duration: float) -> None: - """Ensures video is at least 5 seconds long.""" - if duration < 5: - raise ValueError("Input video must be at least 5 seconds long.") - - -def _trim_if_too_long(video: Input.Video, duration: float) -> Input.Video: - """Trims video to 5 seconds if longer.""" - if duration > 5: - return trim_video(video, 5) - return video - - -def parse_width_height_from_res(resolution: str): - # Accepts a string like "16:9 (1920 x 1080)" and returns width, height as a dict - res_map = { - "16:9 (1920 x 1080)": {"width": 1920, "height": 1080}, - "9:16 (1080 x 1920)": {"width": 1080, "height": 1920}, - "1:1 (1152 x 1152)": {"width": 1152, "height": 1152}, - "4:3 (1536 x 1152)": {"width": 1536, "height": 1152}, - "3:4 (1152 x 1536)": {"width": 1152, "height": 1536}, - # "21:9 (2560 x 1080)": {"width": 2560, "height": 1080}, - } - return res_map.get(resolution, {"width": 1920, "height": 1080}) - - -def parse_control_parameter(value): - control_map = { - "Motion Transfer": "motion_control", - "Canny": "canny_control", - "Pose Transfer": "pose_control", - "Depth": "depth_control", - } - return control_map.get(value, control_map["Motion Transfer"]) - - -async def get_response(cls: type[IO.ComfyNode], task_id: str) -> MoonvalleyPromptResponse: - return await poll_op( - cls, - ApiEndpoint(path=f"{API_PROMPTS_ENDPOINT}/{task_id}"), - response_model=MoonvalleyPromptResponse, - status_extractor=lambda r: (r.status if r and r.status else None), - poll_interval=16.0, - max_poll_attempts=240, - ) - - -class MoonvalleyImg2VideoNode(IO.ComfyNode): - - @classmethod - def define_schema(cls) -> IO.Schema: - return IO.Schema( - node_id="MoonvalleyImg2VideoNode", - display_name="Moonvalley Marey Image to Video", - category="api node/video/Moonvalley Marey", - description="Moonvalley Marey Image to Video Node", - inputs=[ - IO.Image.Input( - "image", - tooltip="The reference image used to generate the video", - ), - IO.String.Input( - "prompt", - multiline=True, - ), - IO.String.Input( - "negative_prompt", - multiline=True, - default=" gopro, bright, contrast, static, overexposed, vignette, " - "artifacts, still, noise, texture, scanlines, videogame, 360 camera, VR, transition, " - "flare, saturation, distorted, warped, wide angle, saturated, vibrant, glowing, " - "cross dissolve, cheesy, ugly hands, mutated hands, mutant, disfigured, extra fingers, " - "blown out, horrible, blurry, worst quality, bad, dissolve, melt, fade in, fade out, " - "wobbly, weird, low quality, plastic, stock footage, video camera, boring", - tooltip="Negative prompt text", - ), - IO.Combo.Input( - "resolution", - options=[ - "16:9 (1920 x 1080)", - "9:16 (1080 x 1920)", - "1:1 (1152 x 1152)", - "4:3 (1536 x 1152)", - "3:4 (1152 x 1536)", - # "21:9 (2560 x 1080)", - ], - default="16:9 (1920 x 1080)", - tooltip="Resolution of the output video", - ), - IO.Float.Input( - "prompt_adherence", - default=4.5, - min=1.0, - max=20.0, - step=1.0, - tooltip="Guidance scale for generation control", - ), - IO.Int.Input( - "seed", - default=9, - min=0, - max=4294967295, - step=1, - display_mode=IO.NumberDisplay.number, - tooltip="Random seed value", - control_after_generate=True, - ), - IO.Int.Input( - "steps", - default=80, - min=75, # steps should be greater or equal to cooldown_steps(75) + warmup_steps(0) - max=100, - step=1, - tooltip="Number of denoising steps", - ), - ], - outputs=[IO.Video.Output()], - hidden=[ - IO.Hidden.auth_token_comfy_org, - IO.Hidden.api_key_comfy_org, - IO.Hidden.unique_id, - ], - is_api_node=True, - price_badge=IO.PriceBadge( - depends_on=IO.PriceBadgeDepends(), - expr="""{"type":"usd","usd": 1.5}""", - ), - ) - - @classmethod - async def execute( - cls, - image: Input.Image, - prompt: str, - negative_prompt: str, - resolution: str, - prompt_adherence: float, - seed: int, - steps: int, - ) -> IO.NodeOutput: - validate_image_dimensions(image, min_width=300, min_height=300, max_height=MAX_HEIGHT, max_width=MAX_WIDTH) - validate_string(prompt, min_length=1, max_length=MOONVALLEY_MAREY_MAX_PROMPT_LENGTH) - validate_string(negative_prompt, field_name="negative_prompt", max_length=MOONVALLEY_MAREY_MAX_PROMPT_LENGTH) - width_height = parse_width_height_from_res(resolution) - - inference_params = MoonvalleyTextToVideoInferenceParams( - negative_prompt=negative_prompt, - steps=steps, - seed=seed, - guidance_scale=prompt_adherence, - width=width_height["width"], - height=width_height["height"], - use_negative_prompts=True, - ) - - # Get MIME type from tensor - assuming PNG format for image tensors - mime_type = "image/png" - image_url = (await upload_images_to_comfyapi(cls, image, max_images=1, mime_type=mime_type))[0] - task_creation_response = await sync_op( - cls, - endpoint=ApiEndpoint(path=API_IMG2VIDEO_ENDPOINT, method="POST"), - response_model=MoonvalleyPromptResponse, - data=MoonvalleyTextToVideoRequest( - image_url=image_url, prompt_text=prompt, inference_params=inference_params - ), - ) - validate_task_creation_response(task_creation_response) - final_response = await get_response(cls, task_creation_response.id) - video = await download_url_to_video_output(final_response.output_url) - return IO.NodeOutput(video) - - -class MoonvalleyVideo2VideoNode(IO.ComfyNode): - - @classmethod - def define_schema(cls) -> IO.Schema: - return IO.Schema( - node_id="MoonvalleyVideo2VideoNode", - display_name="Moonvalley Marey Video to Video", - category="api node/video/Moonvalley Marey", - description="", - inputs=[ - IO.String.Input( - "prompt", - multiline=True, - tooltip="Describes the video to generate", - ), - IO.String.Input( - "negative_prompt", - multiline=True, - default=" gopro, bright, contrast, static, overexposed, vignette, " - "artifacts, still, noise, texture, scanlines, videogame, 360 camera, VR, transition, " - "flare, saturation, distorted, warped, wide angle, saturated, vibrant, glowing, " - "cross dissolve, cheesy, ugly hands, mutated hands, mutant, disfigured, extra fingers, " - "blown out, horrible, blurry, worst quality, bad, dissolve, melt, fade in, fade out, " - "wobbly, weird, low quality, plastic, stock footage, video camera, boring", - tooltip="Negative prompt text", - ), - IO.Int.Input( - "seed", - default=9, - min=0, - max=4294967295, - step=1, - display_mode=IO.NumberDisplay.number, - tooltip="Random seed value", - control_after_generate=False, - ), - IO.Video.Input( - "video", - tooltip="The reference video used to generate the output video. Must be at least 5 seconds long. " - "Videos longer than 5s will be automatically trimmed. Only MP4 format supported.", - ), - IO.Combo.Input( - "control_type", - options=["Motion Transfer", "Pose Transfer"], - default="Motion Transfer", - optional=True, - ), - IO.Int.Input( - "motion_intensity", - default=100, - min=0, - max=100, - step=1, - tooltip="Only used if control_type is 'Motion Transfer'", - optional=True, - ), - IO.Int.Input( - "steps", - default=60, - min=60, # steps should be greater or equal to cooldown_steps(36) + warmup_steps(24) - max=100, - step=1, - display_mode=IO.NumberDisplay.number, - tooltip="Number of inference steps", - ), - ], - outputs=[IO.Video.Output()], - hidden=[ - IO.Hidden.auth_token_comfy_org, - IO.Hidden.api_key_comfy_org, - IO.Hidden.unique_id, - ], - is_api_node=True, - price_badge=IO.PriceBadge( - depends_on=IO.PriceBadgeDepends(), - expr="""{"type":"usd","usd": 2.25}""", - ), - ) - - @classmethod - async def execute( - cls, - prompt: str, - negative_prompt: str, - seed: int, - video: Input.Video | None = None, - control_type: str = "Motion Transfer", - motion_intensity: int | None = 100, - steps=60, - prompt_adherence=4.5, - ) -> IO.NodeOutput: - validated_video = validate_video_to_video_input(video) - video_url = await upload_video_to_comfyapi(cls, validated_video) - validate_string(prompt, min_length=1, max_length=MOONVALLEY_MAREY_MAX_PROMPT_LENGTH) - validate_string(negative_prompt, field_name="negative_prompt", max_length=MOONVALLEY_MAREY_MAX_PROMPT_LENGTH) - - # Only include motion_intensity for Motion Transfer - control_params = {} - if control_type == "Motion Transfer" and motion_intensity is not None: - control_params["motion_intensity"] = motion_intensity - - inference_params = MoonvalleyVideoToVideoInferenceParams( - negative_prompt=negative_prompt, - seed=seed, - control_params=control_params, - steps=steps, - guidance_scale=prompt_adherence, - ) - - task_creation_response = await sync_op( - cls, - endpoint=ApiEndpoint(path=API_VIDEO2VIDEO_ENDPOINT, method="POST"), - response_model=MoonvalleyPromptResponse, - data=MoonvalleyVideoToVideoRequest( - control_type=parse_control_parameter(control_type), - video_url=video_url, - prompt_text=prompt, - inference_params=inference_params, - ), - ) - validate_task_creation_response(task_creation_response) - final_response = await get_response(cls, task_creation_response.id) - return IO.NodeOutput(await download_url_to_video_output(final_response.output_url)) - - -class MoonvalleyTxt2VideoNode(IO.ComfyNode): - - @classmethod - def define_schema(cls) -> IO.Schema: - return IO.Schema( - node_id="MoonvalleyTxt2VideoNode", - display_name="Moonvalley Marey Text to Video", - category="api node/video/Moonvalley Marey", - description="", - inputs=[ - IO.String.Input( - "prompt", - multiline=True, - ), - IO.String.Input( - "negative_prompt", - multiline=True, - default=" gopro, bright, contrast, static, overexposed, vignette, " - "artifacts, still, noise, texture, scanlines, videogame, 360 camera, VR, transition, " - "flare, saturation, distorted, warped, wide angle, saturated, vibrant, glowing, " - "cross dissolve, cheesy, ugly hands, mutated hands, mutant, disfigured, extra fingers, " - "blown out, horrible, blurry, worst quality, bad, dissolve, melt, fade in, fade out, " - "wobbly, weird, low quality, plastic, stock footage, video camera, boring", - tooltip="Negative prompt text", - ), - IO.Combo.Input( - "resolution", - options=[ - "16:9 (1920 x 1080)", - "9:16 (1080 x 1920)", - "1:1 (1152 x 1152)", - "4:3 (1536 x 1152)", - "3:4 (1152 x 1536)", - "21:9 (2560 x 1080)", - ], - default="16:9 (1920 x 1080)", - tooltip="Resolution of the output video", - ), - IO.Float.Input( - "prompt_adherence", - default=4.0, - min=1.0, - max=20.0, - step=1.0, - tooltip="Guidance scale for generation control", - ), - IO.Int.Input( - "seed", - default=9, - min=0, - max=4294967295, - step=1, - display_mode=IO.NumberDisplay.number, - control_after_generate=True, - tooltip="Random seed value", - ), - IO.Int.Input( - "steps", - default=80, - min=75, # steps should be greater or equal to cooldown_steps(75) + warmup_steps(0) - max=100, - step=1, - tooltip="Inference steps", - ), - ], - outputs=[IO.Video.Output()], - hidden=[ - IO.Hidden.auth_token_comfy_org, - IO.Hidden.api_key_comfy_org, - IO.Hidden.unique_id, - ], - is_api_node=True, - price_badge=IO.PriceBadge( - depends_on=IO.PriceBadgeDepends(), - expr="""{"type":"usd","usd": 1.5}""", - ), - ) - - @classmethod - async def execute( - cls, - prompt: str, - negative_prompt: str, - resolution: str, - prompt_adherence: float, - seed: int, - steps: int, - ) -> IO.NodeOutput: - validate_string(prompt, min_length=1, max_length=MOONVALLEY_MAREY_MAX_PROMPT_LENGTH) - validate_string(negative_prompt, field_name="negative_prompt", max_length=MOONVALLEY_MAREY_MAX_PROMPT_LENGTH) - width_height = parse_width_height_from_res(resolution) - - inference_params = MoonvalleyTextToVideoInferenceParams( - negative_prompt=negative_prompt, - steps=steps, - seed=seed, - guidance_scale=prompt_adherence, - num_frames=128, - width=width_height["width"], - height=width_height["height"], - ) - - task_creation_response = await sync_op( - cls, - endpoint=ApiEndpoint(path=API_TXT2VIDEO_ENDPOINT, method="POST"), - response_model=MoonvalleyPromptResponse, - data=MoonvalleyTextToVideoRequest(prompt_text=prompt, inference_params=inference_params), - ) - validate_task_creation_response(task_creation_response) - final_response = await get_response(cls, task_creation_response.id) - return IO.NodeOutput(await download_url_to_video_output(final_response.output_url)) - - -class MoonvalleyExtension(ComfyExtension): - @override - async def get_node_list(self) -> list[type[IO.ComfyNode]]: - return [ - MoonvalleyImg2VideoNode, - MoonvalleyTxt2VideoNode, - MoonvalleyVideo2VideoNode, - ] - - -async def comfy_entrypoint() -> MoonvalleyExtension: - return MoonvalleyExtension() diff --git a/comfy_api_nodes/nodes_openai.py b/comfy_api_nodes/nodes_openai.py index 4ee896fa8..a5a188634 100644 --- a/comfy_api_nodes/nodes_openai.py +++ b/comfy_api_nodes/nodes_openai.py @@ -27,6 +27,7 @@ from comfy_api_nodes.util import ( ApiEndpoint, download_url_to_bytesio, downscale_image_tensor, + get_number_of_images, poll_op, sync_op, tensor_to_base64_string, @@ -39,16 +40,18 @@ STARTING_POINT_ID_PATTERN = r"" class SupportedOpenAIModel(str, Enum): - o4_mini = "o4-mini" - o1 = "o1" - o3 = "o3" - o1_pro = "o1-pro" - gpt_4_1 = "gpt-4.1" - gpt_4_1_mini = "gpt-4.1-mini" - gpt_4_1_nano = "gpt-4.1-nano" + gpt_5_5_pro = "gpt-5.5-pro" + gpt_5_5 = "gpt-5.5" gpt_5 = "gpt-5" gpt_5_mini = "gpt-5-mini" gpt_5_nano = "gpt-5-nano" + gpt_4_1 = "gpt-4.1" + gpt_4_1_mini = "gpt-4.1-mini" + gpt_4_1_nano = "gpt-4.1-nano" + o4_mini = "o4-mini" + o3 = "o3" + o1_pro = "o1-pro" + o1 = "o1" async def validate_and_cast_response(response, timeout: int = None) -> torch.Tensor: @@ -357,15 +360,20 @@ def calculate_tokens_price_image_1_5(response: OpenAIImageGenerationResponse) -> return ((response.usage.input_tokens * 8.0) + (response.usage.output_tokens * 32.0)) / 1_000_000.0 +def calculate_tokens_price_image_2_0(response: OpenAIImageGenerationResponse) -> float | None: + return ((response.usage.input_tokens * 8.0) + (response.usage.output_tokens * 30.0)) / 1_000_000.0 + + class OpenAIGPTImage1(IO.ComfyNode): @classmethod def define_schema(cls): return IO.Schema( node_id="OpenAIGPTImage1", - display_name="OpenAI GPT Image 1.5", + display_name="OpenAI GPT Image 2", category="api node/image/OpenAI", description="Generates images synchronously via OpenAI's GPT Image endpoint.", + is_deprecated=True, inputs=[ IO.String.Input( "prompt", @@ -401,8 +409,19 @@ class OpenAIGPTImage1(IO.ComfyNode): IO.Combo.Input( "size", default="auto", - options=["auto", "1024x1024", "1024x1536", "1536x1024"], - tooltip="Image size", + options=[ + "auto", + "1024x1024", + "1024x1536", + "1536x1024", + "2048x2048", + "2048x1152", + "1152x2048", + "3840x2160", + "2160x3840", + "Custom", + ], + tooltip="Image size. Select 'Custom' to use the custom width and height (GPT Image 2 only).", optional=True, ), IO.Int.Input( @@ -427,8 +446,26 @@ class OpenAIGPTImage1(IO.ComfyNode): ), IO.Combo.Input( "model", - options=["gpt-image-1", "gpt-image-1.5"], - default="gpt-image-1.5", + options=["gpt-image-1", "gpt-image-1.5", "gpt-image-2"], + default="gpt-image-2", + optional=True, + ), + IO.Int.Input( + "custom_width", + default=1024, + min=1024, + max=3840, + step=16, + tooltip="Used only when `size` is 'Custom'. Must be a multiple of 16 (GPT Image 2 only).", + optional=True, + ), + IO.Int.Input( + "custom_height", + default=1024, + min=1024, + max=3840, + step=16, + tooltip="Used only when `size` is 'Custom'. Must be a multiple of 16 (GPT Image 2 only).", optional=True, ), ], @@ -442,23 +479,36 @@ class OpenAIGPTImage1(IO.ComfyNode): ], is_api_node=True, price_badge=IO.PriceBadge( - depends_on=IO.PriceBadgeDepends(widgets=["quality", "n"]), + depends_on=IO.PriceBadgeDepends(widgets=["quality", "n", "model"]), expr=""" ( $ranges := { - "low": [0.011, 0.02], - "medium": [0.046, 0.07], - "high": [0.167, 0.3] + "gpt-image-1": { + "low": [0.011, 0.02], + "medium": [0.042, 0.07], + "high": [0.167, 0.25] + }, + "gpt-image-1.5": { + "low": [0.009, 0.02], + "medium": [0.034, 0.062], + "high": [0.133, 0.22] + }, + "gpt-image-2": { + "low": [0.0048, 0.019], + "medium": [0.041, 0.168], + "high": [0.165, 0.67] + } }; - $range := $lookup($ranges, widgets.quality); - $n := widgets.n; + $range := $lookup($lookup($ranges, widgets.model), widgets.quality); + $nRaw := widgets.n; + $n := ($nRaw != null and $nRaw != 0) ? $nRaw : 1; ($n = 1) - ? {"type":"range_usd","min_usd": $range[0], "max_usd": $range[1]} + ? {"type":"range_usd","min_usd": $range[0], "max_usd": $range[1], "format": {"approximate": true}} : { "type":"range_usd", - "min_usd": $range[0], - "max_usd": $range[1], - "format": { "suffix": " x " & $string($n) & "/Run" } + "min_usd": $range[0] * $n, + "max_usd": $range[1] * $n, + "format": { "suffix": "/Run", "approximate": true } } ) """, @@ -476,6 +526,8 @@ class OpenAIGPTImage1(IO.ComfyNode): mask: Input.Image | None = None, n: int = 1, size: str = "1024x1024", + custom_width: int = 1024, + custom_height: int = 1024, model: str = "gpt-image-1", ) -> IO.NodeOutput: validate_string(prompt, strip_whitespace=False) @@ -483,10 +535,36 @@ class OpenAIGPTImage1(IO.ComfyNode): if mask is not None and image is None: raise ValueError("Cannot use a mask without an input image") + if size == "Custom": + if model != "gpt-image-2": + raise ValueError("Custom resolution is only supported by GPT Image 2 model") + if custom_width % 16 != 0 or custom_height % 16 != 0: + raise ValueError(f"Custom width and height must be multiples of 16, got {custom_width}x{custom_height}") + if max(custom_width, custom_height) > 3840: + raise ValueError(f"Custom resolution max edge must be <= 3840, got {custom_width}x{custom_height}") + ratio = max(custom_width, custom_height) / min(custom_width, custom_height) + if ratio > 3: + raise ValueError( + f"Custom resolution aspect ratio must not exceed 3:1, got {custom_width}x{custom_height}" + ) + total_pixels = custom_width * custom_height + if not 655_360 <= total_pixels <= 8_294_400: + raise ValueError( + f"Custom resolution total pixels must be between 655,360 and 8,294,400, got {total_pixels}" + ) + size = f"{custom_width}x{custom_height}" + elif model in ("gpt-image-1", "gpt-image-1.5"): + if size not in ("auto", "1024x1024", "1024x1536", "1536x1024"): + raise ValueError(f"Resolution {size} is only supported by GPT Image 2 model") + if model == "gpt-image-1": price_extractor = calculate_tokens_price_image_1 elif model == "gpt-image-1.5": price_extractor = calculate_tokens_price_image_1_5 + elif model == "gpt-image-2": + price_extractor = calculate_tokens_price_image_2_0 + if background == "transparent": + raise ValueError("Transparent background is not supported for GPT Image 2 model") else: raise ValueError(f"Unknown model: {model}") @@ -564,6 +642,316 @@ class OpenAIGPTImage1(IO.ComfyNode): return IO.NodeOutput(await validate_and_cast_response(response)) +def _gpt_image_shared_inputs(): + """Inputs shared by all GPT Image models (quality + reference images + mask).""" + return [ + IO.Combo.Input( + "quality", + default="low", + options=["low", "medium", "high"], + tooltip="Image quality, affects cost and generation time.", + ), + IO.Autogrow.Input( + "images", + template=IO.Autogrow.TemplateNames( + IO.Image.Input("image"), + names=[f"image_{i}" for i in range(1, 17)], + min=0, + ), + tooltip="Optional reference image(s) for image editing. Up to 16 images.", + ), + IO.Mask.Input( + "mask", + optional=True, + tooltip="Optional mask for inpainting (white areas will be replaced). " + "Requires exactly one reference image.", + ), + ] + + +def _gpt_image_legacy_model_inputs(): + """Per-model widget set for legacy gpt-image-1 / gpt-image-1.5 (4 base sizes, transparent bg allowed).""" + return [ + IO.Combo.Input( + "size", + default="auto", + options=["auto", "1024x1024", "1024x1536", "1536x1024"], + tooltip="Image size.", + ), + IO.Combo.Input( + "background", + default="auto", + options=["auto", "opaque", "transparent"], + tooltip="Return image with or without background.", + ), + *_gpt_image_shared_inputs(), + ] + + +class OpenAIGPTImageNodeV2(IO.ComfyNode): + + @classmethod + def define_schema(cls): + return IO.Schema( + node_id="OpenAIGPTImageNodeV2", + display_name="OpenAI GPT Image 2", + category="api node/image/OpenAI", + description="Generates images via OpenAI's GPT Image endpoint.", + inputs=[ + IO.String.Input( + "prompt", + default="", + multiline=True, + tooltip="Text prompt for GPT Image", + ), + IO.DynamicCombo.Input( + "model", + options=[ + IO.DynamicCombo.Option( + "gpt-image-2", + [ + IO.Combo.Input( + "size", + default="auto", + options=[ + "auto", + "1024x1024", + "1024x1536", + "1536x1024", + "2048x2048", + "2048x1152", + "1152x2048", + "3840x2160", + "2160x3840", + "Custom", + ], + tooltip="Image size. Select 'Custom' to use the custom width and height.", + ), + IO.Int.Input( + "custom_width", + default=1024, + min=1024, + max=3840, + step=16, + tooltip="Used only when `size` is 'Custom'. Must be a multiple of 16.", + ), + IO.Int.Input( + "custom_height", + default=1024, + min=1024, + max=3840, + step=16, + tooltip="Used only when `size` is 'Custom'. Must be a multiple of 16.", + ), + IO.Combo.Input( + "background", + default="auto", + options=["auto", "opaque"], + tooltip="Return image with or without background.", + ), + *_gpt_image_shared_inputs(), + ], + ), + IO.DynamicCombo.Option("gpt-image-1.5", _gpt_image_legacy_model_inputs()), + IO.DynamicCombo.Option("gpt-image-1", _gpt_image_legacy_model_inputs()), + ], + ), + IO.Int.Input( + "n", + default=1, + min=1, + max=8, + step=1, + tooltip="How many images to generate", + display_mode=IO.NumberDisplay.number, + ), + IO.Int.Input( + "seed", + default=0, + min=0, + max=2147483647, + step=1, + display_mode=IO.NumberDisplay.number, + control_after_generate=True, + tooltip="not implemented yet in backend", + ), + ], + outputs=[IO.Image.Output()], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + depends_on=IO.PriceBadgeDepends(widgets=["model", "model.quality", "n"]), + expr=""" + ( + $ranges := { + "gpt-image-1": { + "low": [0.011, 0.02], + "medium": [0.042, 0.07], + "high": [0.167, 0.25] + }, + "gpt-image-1.5": { + "low": [0.009, 0.02], + "medium": [0.034, 0.062], + "high": [0.133, 0.22] + }, + "gpt-image-2": { + "low": [0.0048, 0.019], + "medium": [0.041, 0.168], + "high": [0.165, 0.67] + } + }; + $range := $lookup($lookup($ranges, widgets.model), $lookup(widgets, "model.quality")); + $nRaw := widgets.n; + $n := ($nRaw != null and $nRaw != 0) ? $nRaw : 1; + ($n = 1) + ? {"type":"range_usd","min_usd": $range[0], "max_usd": $range[1], "format": {"approximate": true}} + : { + "type":"range_usd", + "min_usd": $range[0] * $n, + "max_usd": $range[1] * $n, + "format": { "suffix": "/Run", "approximate": true } + } + ) + """, + ), + ) + + @classmethod + async def execute( + cls, + prompt: str, + model: dict, + n: int, + seed: int, + ) -> IO.NodeOutput: + validate_string(prompt, strip_whitespace=False) + + model_id = model["model"] + size = model["size"] + background = model["background"] + quality = model["quality"] + custom_width = model.get("custom_width", 1024) + custom_height = model.get("custom_height", 1024) + + images_dict = model.get("images") or {} + image_tensors: list[Input.Image] = [t for t in images_dict.values() if t is not None] + n_images = sum(get_number_of_images(t) for t in image_tensors) + mask = model.get("mask") + + if mask is not None and n_images == 0: + raise ValueError("Cannot use a mask without an input image") + + if size == "Custom": + if custom_width % 16 != 0 or custom_height % 16 != 0: + raise ValueError( + f"Custom width and height must be multiples of 16, got {custom_width}x{custom_height}" + ) + if max(custom_width, custom_height) > 3840: + raise ValueError( + f"Custom resolution max edge must be <= 3840, got {custom_width}x{custom_height}" + ) + ratio = max(custom_width, custom_height) / min(custom_width, custom_height) + if ratio > 3: + raise ValueError( + f"Custom resolution aspect ratio must not exceed 3:1, got {custom_width}x{custom_height}" + ) + total_pixels = custom_width * custom_height + if not 655_360 <= total_pixels <= 8_294_400: + raise ValueError( + f"Custom resolution total pixels must be between 655,360 and 8,294,400, got {total_pixels}" + ) + size = f"{custom_width}x{custom_height}" + + if model_id == "gpt-image-1": + price_extractor = calculate_tokens_price_image_1 + elif model_id == "gpt-image-1.5": + price_extractor = calculate_tokens_price_image_1_5 + elif model_id == "gpt-image-2": + price_extractor = calculate_tokens_price_image_2_0 + else: + raise ValueError(f"Unknown model: {model_id}") + + if image_tensors: + flat: list[torch.Tensor] = [] + for tensor in image_tensors: + if len(tensor.shape) == 4: + flat.extend(tensor[i : i + 1] for i in range(tensor.shape[0])) + else: + flat.append(tensor.unsqueeze(0)) + + files = [] + for i, single_image in enumerate(flat): + scaled_image = downscale_image_tensor(single_image, total_pixels=2048 * 2048).squeeze() + image_np = (scaled_image.numpy() * 255).astype(np.uint8) + img = Image.fromarray(image_np) + img_byte_arr = BytesIO() + img.save(img_byte_arr, format="PNG") + img_byte_arr.seek(0) + + if len(flat) == 1: + files.append(("image", (f"image_{i}.png", img_byte_arr, "image/png"))) + else: + files.append(("image[]", (f"image_{i}.png", img_byte_arr, "image/png"))) + + if mask is not None: + if len(flat) != 1: + raise Exception("Cannot use a mask with multiple image") + ref_image = flat[0] + if mask.shape[1:] != ref_image.shape[1:-1]: + raise Exception("Mask and Image must be the same size") + _, height, width = mask.shape + rgba_mask = torch.zeros(height, width, 4, device="cpu") + rgba_mask[:, :, 3] = 1 - mask.squeeze().cpu() + scaled_mask = downscale_image_tensor( + rgba_mask.unsqueeze(0), total_pixels=2048 * 2048 + ).squeeze() + mask_np = (scaled_mask.numpy() * 255).astype(np.uint8) + mask_img = Image.fromarray(mask_np) + mask_img_byte_arr = BytesIO() + mask_img.save(mask_img_byte_arr, format="PNG") + mask_img_byte_arr.seek(0) + files.append(("mask", ("mask.png", mask_img_byte_arr, "image/png"))) + + response = await sync_op( + cls, + ApiEndpoint(path="/proxy/openai/images/edits", method="POST"), + response_model=OpenAIImageGenerationResponse, + data=OpenAIImageEditRequest( + model=model_id, + prompt=prompt, + quality=quality, + background=background, + n=n, + size=size, + moderation="low", + ), + content_type="multipart/form-data", + files=files, + price_extractor=price_extractor, + ) + else: + response = await sync_op( + cls, + ApiEndpoint(path="/proxy/openai/images/generations", method="POST"), + response_model=OpenAIImageGenerationResponse, + data=OpenAIImageGenerationRequest( + model=model_id, + prompt=prompt, + quality=quality, + background=background, + n=n, + size=size, + moderation="low", + ), + price_extractor=price_extractor, + ) + return IO.NodeOutput(await validate_and_cast_response(response)) + + class OpenAIChatNode(IO.ComfyNode): """ Node to generate text responses from an OpenAI model. @@ -665,6 +1053,16 @@ class OpenAIChatNode(IO.ComfyNode): "usd": [0.002, 0.008], "format": { "approximate": true, "separator": "-", "suffix": " per 1K tokens" } } + : $contains($m, "gpt-5.5-pro") ? { + "type": "list_usd", + "usd": [0.03, 0.18], + "format": { "approximate": true, "separator": "-", "suffix": " per 1K tokens" } + } + : $contains($m, "gpt-5.5") ? { + "type": "list_usd", + "usd": [0.005, 0.03], + "format": { "approximate": true, "separator": "-", "suffix": " per 1K tokens" } + } : $contains($m, "gpt-5-nano") ? { "type": "list_usd", "usd": [0.00005, 0.0004], @@ -913,6 +1311,7 @@ class OpenAIExtension(ComfyExtension): OpenAIDalle2, OpenAIDalle3, OpenAIGPTImage1, + OpenAIGPTImageNodeV2, OpenAIChatNode, OpenAIInputFiles, OpenAIChatConfig, diff --git a/comfy_api_nodes/nodes_quiver.py b/comfy_api_nodes/nodes_quiver.py index 61533263f..3269c0afe 100644 --- a/comfy_api_nodes/nodes_quiver.py +++ b/comfy_api_nodes/nodes_quiver.py @@ -17,6 +17,44 @@ from comfy_api_nodes.util import ( ) from comfy_extras.nodes_images import SVG +_ARROW_MODELS = ["arrow-1.1", "arrow-1.1-max", "arrow-preview"] + + +def _arrow_sampling_inputs(): + """Shared sampling inputs for all Arrow model variants.""" + return [ + IO.Float.Input( + "temperature", + default=1.0, + min=0.0, + max=2.0, + step=0.1, + display_mode=IO.NumberDisplay.slider, + tooltip="Randomness control. Higher values increase randomness.", + advanced=True, + ), + IO.Float.Input( + "top_p", + default=1.0, + min=0.05, + max=1.0, + step=0.05, + display_mode=IO.NumberDisplay.slider, + tooltip="Nucleus sampling parameter.", + advanced=True, + ), + IO.Float.Input( + "presence_penalty", + default=0.0, + min=-2.0, + max=2.0, + step=0.1, + display_mode=IO.NumberDisplay.slider, + tooltip="Token presence penalty.", + advanced=True, + ), + ] + class QuiverTextToSVGNode(IO.ComfyNode): @classmethod @@ -39,6 +77,7 @@ class QuiverTextToSVGNode(IO.ComfyNode): default="", tooltip="Additional style or formatting guidance.", optional=True, + advanced=True, ), IO.Autogrow.Input( "reference_images", @@ -53,43 +92,7 @@ class QuiverTextToSVGNode(IO.ComfyNode): ), IO.DynamicCombo.Input( "model", - options=[ - IO.DynamicCombo.Option( - "arrow-preview", - [ - IO.Float.Input( - "temperature", - default=1.0, - min=0.0, - max=2.0, - step=0.1, - display_mode=IO.NumberDisplay.slider, - tooltip="Randomness control. Higher values increase randomness.", - advanced=True, - ), - IO.Float.Input( - "top_p", - default=1.0, - min=0.05, - max=1.0, - step=0.05, - display_mode=IO.NumberDisplay.slider, - tooltip="Nucleus sampling parameter.", - advanced=True, - ), - IO.Float.Input( - "presence_penalty", - default=0.0, - min=-2.0, - max=2.0, - step=0.1, - display_mode=IO.NumberDisplay.slider, - tooltip="Token presence penalty.", - advanced=True, - ), - ], - ), - ], + options=[IO.DynamicCombo.Option(m, _arrow_sampling_inputs()) for m in _ARROW_MODELS], tooltip="Model to use for SVG generation.", ), IO.Int.Input( @@ -112,7 +115,16 @@ class QuiverTextToSVGNode(IO.ComfyNode): ], is_api_node=True, price_badge=IO.PriceBadge( - expr="""{"type":"usd","usd":0.429}""", + depends_on=IO.PriceBadgeDepends(widgets=["model"]), + expr=""" + ( + $contains(widgets.model, "max") + ? {"type":"usd","usd":0.3575} + : $contains(widgets.model, "preview") + ? {"type":"usd","usd":0.429} + : {"type":"usd","usd":0.286} + ) + """, ), ) @@ -131,7 +143,7 @@ class QuiverTextToSVGNode(IO.ComfyNode): if reference_images: references = [] for key in reference_images: - url = await upload_image_to_comfyapi(cls, reference_images[key]) + url = await upload_image_to_comfyapi(cls, reference_images[key], mime_type="image/png") references.append(QuiverImageObject(url=url)) if len(references) > 4: raise ValueError("Maximum 4 reference images are allowed.") @@ -176,12 +188,13 @@ class QuiverImageToSVGNode(IO.ComfyNode): "auto_crop", default=False, tooltip="Automatically crop to the dominant subject.", + advanced=True, ), IO.DynamicCombo.Input( "model", options=[ IO.DynamicCombo.Option( - "arrow-preview", + m, [ IO.Int.Input( "target_size", @@ -189,39 +202,12 @@ class QuiverImageToSVGNode(IO.ComfyNode): min=128, max=4096, tooltip="Square resize target in pixels.", - ), - IO.Float.Input( - "temperature", - default=1.0, - min=0.0, - max=2.0, - step=0.1, - display_mode=IO.NumberDisplay.slider, - tooltip="Randomness control. Higher values increase randomness.", - advanced=True, - ), - IO.Float.Input( - "top_p", - default=1.0, - min=0.05, - max=1.0, - step=0.05, - display_mode=IO.NumberDisplay.slider, - tooltip="Nucleus sampling parameter.", - advanced=True, - ), - IO.Float.Input( - "presence_penalty", - default=0.0, - min=-2.0, - max=2.0, - step=0.1, - display_mode=IO.NumberDisplay.slider, - tooltip="Token presence penalty.", advanced=True, ), + *_arrow_sampling_inputs(), ], - ), + ) + for m in _ARROW_MODELS ], tooltip="Model to use for SVG vectorization.", ), @@ -245,7 +231,16 @@ class QuiverImageToSVGNode(IO.ComfyNode): ], is_api_node=True, price_badge=IO.PriceBadge( - expr="""{"type":"usd","usd":0.429}""", + depends_on=IO.PriceBadgeDepends(widgets=["model"]), + expr=""" + ( + $contains(widgets.model, "max") + ? {"type":"usd","usd":0.3575} + : $contains(widgets.model, "preview") + ? {"type":"usd","usd":0.429} + : {"type":"usd","usd":0.286} + ) + """, ), ) @@ -257,7 +252,7 @@ class QuiverImageToSVGNode(IO.ComfyNode): model: dict, seed: int, ) -> IO.NodeOutput: - image_url = await upload_image_to_comfyapi(cls, image) + image_url = await upload_image_to_comfyapi(cls, image, mime_type="image/png") response = await sync_op( cls, diff --git a/comfy_api_nodes/nodes_sonilo.py b/comfy_api_nodes/nodes_sonilo.py new file mode 100644 index 000000000..5518f5902 --- /dev/null +++ b/comfy_api_nodes/nodes_sonilo.py @@ -0,0 +1,287 @@ +import base64 +import json +import logging +import time +from urllib.parse import urljoin + +import aiohttp +from typing_extensions import override + +from comfy_api.latest import IO, ComfyExtension, Input +from comfy_api_nodes.util import ( + ApiEndpoint, + audio_bytes_to_audio_input, + upload_video_to_comfyapi, + validate_string, +) +from comfy_api_nodes.util._helpers import ( + default_base_url, + get_auth_header, + get_node_id, + is_processing_interrupted, +) +from comfy_api_nodes.util.common_exceptions import ProcessingInterrupted +from server import PromptServer + +logger = logging.getLogger(__name__) + + +class SoniloVideoToMusic(IO.ComfyNode): + """Generate music from video using Sonilo's AI model.""" + + @classmethod + def define_schema(cls) -> IO.Schema: + return IO.Schema( + node_id="SoniloVideoToMusic", + display_name="Sonilo Video to Music", + category="api node/audio/Sonilo", + description="Generate music from video content using Sonilo's AI model. " + "Analyzes the video and creates matching music.", + inputs=[ + IO.Video.Input( + "video", + tooltip="Input video to generate music from. Maximum duration: 6 minutes.", + ), + IO.String.Input( + "prompt", + default="", + multiline=True, + tooltip="Optional text prompt to guide music generation. " + "Leave empty for best quality - the model will fully analyze the video content.", + ), + IO.Int.Input( + "seed", + default=0, + min=0, + max=0xFFFFFFFFFFFFFFFF, + control_after_generate=True, + tooltip="Seed for reproducibility. Currently ignored by the Sonilo " + "service but kept for graph consistency.", + ), + ], + outputs=[IO.Audio.Output()], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + expr='{"type":"usd","usd":0.009,"format":{"suffix":"/second"}}', + ), + ) + + @classmethod + async def execute( + cls, + video: Input.Video, + prompt: str = "", + seed: int = 0, + ) -> IO.NodeOutput: + video_url = await upload_video_to_comfyapi(cls, video, max_duration=360) + form = aiohttp.FormData() + form.add_field("video_url", video_url) + if prompt.strip(): + form.add_field("prompt", prompt.strip()) + audio_bytes = await _stream_sonilo_music( + cls, + ApiEndpoint(path="/proxy/sonilo/v2m/generate", method="POST"), + form, + ) + return IO.NodeOutput(audio_bytes_to_audio_input(audio_bytes)) + + +class SoniloTextToMusic(IO.ComfyNode): + """Generate music from a text prompt using Sonilo's AI model.""" + + @classmethod + def define_schema(cls) -> IO.Schema: + return IO.Schema( + node_id="SoniloTextToMusic", + display_name="Sonilo Text to Music", + category="api node/audio/Sonilo", + description="Generate music from a text prompt using Sonilo's AI model. " + "Leave duration at 0 to let the model infer it from the prompt.", + inputs=[ + IO.String.Input( + "prompt", + default="", + multiline=True, + tooltip="Text prompt describing the music to generate.", + ), + IO.Int.Input( + "duration", + default=0, + min=0, + max=360, + tooltip="Target duration in seconds. Set to 0 to let the model " + "infer the duration from the prompt. Maximum: 6 minutes.", + ), + IO.Int.Input( + "seed", + default=0, + min=0, + max=0xFFFFFFFFFFFFFFFF, + control_after_generate=True, + tooltip="Seed for reproducibility. Currently ignored by the Sonilo " + "service but kept for graph consistency.", + ), + ], + outputs=[IO.Audio.Output()], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + depends_on=IO.PriceBadgeDepends(widgets=["duration"]), + expr=""" + ( + widgets.duration > 0 + ? {"type":"usd","usd": 0.005 * widgets.duration} + : {"type":"usd","usd": 0.005, "format":{"suffix":"/second"}} + ) + """, + ), + ) + + @classmethod + async def execute( + cls, + prompt: str, + duration: int = 0, + seed: int = 0, + ) -> IO.NodeOutput: + validate_string(prompt, strip_whitespace=True, min_length=1) + form = aiohttp.FormData() + form.add_field("prompt", prompt) + if duration > 0: + form.add_field("duration", str(duration)) + audio_bytes = await _stream_sonilo_music( + cls, + ApiEndpoint(path="/proxy/sonilo/t2m/generate", method="POST"), + form, + ) + return IO.NodeOutput(audio_bytes_to_audio_input(audio_bytes)) + + +async def _stream_sonilo_music( + cls: type[IO.ComfyNode], + endpoint: ApiEndpoint, + form: aiohttp.FormData, +) -> bytes: + """POST ``form`` to Sonilo, read the NDJSON stream, and return the first stream's audio bytes.""" + url = urljoin(default_base_url().rstrip("/") + "/", endpoint.path.lstrip("/")) + + headers: dict[str, str] = {} + headers.update(get_auth_header(cls)) + headers.update(endpoint.headers) + + node_id = get_node_id(cls) + start_ts = time.monotonic() + last_chunk_status_ts = 0.0 + audio_streams: dict[int, list[bytes]] = {} + title: str | None = None + + timeout = aiohttp.ClientTimeout(total=1200.0, sock_read=300.0) + async with aiohttp.ClientSession(timeout=timeout) as session: + PromptServer.instance.send_progress_text("Status: Queued", node_id) + async with session.post(url, data=form, headers=headers) as resp: + if resp.status >= 400: + msg = await _extract_error_message(resp) + raise Exception(f"Sonilo API error ({resp.status}): {msg}") + + while True: + if is_processing_interrupted(): + raise ProcessingInterrupted("Task cancelled") + + raw_line = await resp.content.readline() + if not raw_line: + break + + line = raw_line.decode("utf-8").strip() + if not line: + continue + + try: + evt = json.loads(line) + except json.JSONDecodeError: + logger.warning("Sonilo: skipping malformed NDJSON line") + continue + + evt_type = evt.get("type") + if evt_type == "error": + code = evt.get("code", "UNKNOWN") + message = evt.get("message", "Unknown error") + raise Exception(f"Sonilo generation error ({code}): {message}") + if evt_type == "duration": + duration_sec = evt.get("duration_sec") + if duration_sec is not None: + PromptServer.instance.send_progress_text( + f"Status: Generating\nVideo duration: {duration_sec:.1f}s", + node_id, + ) + elif evt_type in ("titles", "title"): + # v2m sends a "titles" list, t2m sends a scalar "title" + if evt_type == "titles": + titles = evt.get("titles", []) + if titles: + title = titles[0] + else: + title = evt.get("title") or title + if title: + PromptServer.instance.send_progress_text( + f"Status: Generating\nTitle: {title}", + node_id, + ) + elif evt_type == "audio_chunk": + stream_idx = evt.get("stream_index", 0) + chunk_data = base64.b64decode(evt["data"]) + + if stream_idx not in audio_streams: + audio_streams[stream_idx] = [] + audio_streams[stream_idx].append(chunk_data) + + now = time.monotonic() + if now - last_chunk_status_ts >= 1.0: + total_chunks = sum(len(chunks) for chunks in audio_streams.values()) + elapsed = int(now - start_ts) + status_lines = ["Status: Receiving audio"] + if title: + status_lines.append(f"Title: {title}") + status_lines.append(f"Chunks received: {total_chunks}") + status_lines.append(f"Time elapsed: {elapsed}s") + PromptServer.instance.send_progress_text("\n".join(status_lines), node_id) + last_chunk_status_ts = now + elif evt_type == "complete": + break + + if not audio_streams: + raise Exception("Sonilo API returned no audio data.") + + PromptServer.instance.send_progress_text("Status: Completed", node_id) + selected_stream = 0 if 0 in audio_streams else min(audio_streams) + return b"".join(audio_streams[selected_stream]) + + +async def _extract_error_message(resp: aiohttp.ClientResponse) -> str: + """Extract a human-readable error message from an HTTP error response.""" + try: + error_body = await resp.json() + detail = error_body.get("detail", {}) + if isinstance(detail, dict): + return detail.get("message", str(detail)) + return str(detail) + except Exception: + return await resp.text() + + +class SoniloExtension(ComfyExtension): + @override + async def get_node_list(self) -> list[type[IO.ComfyNode]]: + return [SoniloVideoToMusic, SoniloTextToMusic] + + +async def comfy_entrypoint() -> SoniloExtension: + return SoniloExtension() diff --git a/comfy_api_nodes/nodes_sora.py b/comfy_api_nodes/nodes_sora.py index afc18bb25..c1d485188 100644 --- a/comfy_api_nodes/nodes_sora.py +++ b/comfy_api_nodes/nodes_sora.py @@ -33,9 +33,13 @@ class OpenAIVideoSora2(IO.ComfyNode): def define_schema(cls): return IO.Schema( node_id="OpenAIVideoSora2", - display_name="OpenAI Sora - Video", + display_name="OpenAI Sora - Video (DEPRECATED)", category="api node/video/Sora", - description="OpenAI video and audio generation.", + description=( + "OpenAI video and audio generation.\n\n" + "DEPRECATION NOTICE: OpenAI will stop serving the Sora v2 API in September 2026. " + "This node will be removed from ComfyUI at that time." + ), inputs=[ IO.Combo.Input( "model", diff --git a/comfy_api_nodes/nodes_stability.py b/comfy_api_nodes/nodes_stability.py index 9ef13c83b..906d8ff35 100644 --- a/comfy_api_nodes/nodes_stability.py +++ b/comfy_api_nodes/nodes_stability.py @@ -401,7 +401,7 @@ class StabilityUpscaleConservativeNode(IO.ComfyNode): ], is_api_node=True, price_badge=IO.PriceBadge( - expr="""{"type":"usd","usd":0.25}""", + expr="""{"type":"usd","usd":0.4}""", ), ) @@ -510,7 +510,7 @@ class StabilityUpscaleCreativeNode(IO.ComfyNode): ], is_api_node=True, price_badge=IO.PriceBadge( - expr="""{"type":"usd","usd":0.25}""", + expr="""{"type":"usd","usd":0.6}""", ), ) @@ -593,7 +593,7 @@ class StabilityUpscaleFastNode(IO.ComfyNode): ], is_api_node=True, price_badge=IO.PriceBadge( - expr="""{"type":"usd","usd":0.01}""", + expr="""{"type":"usd","usd":0.02}""", ), ) diff --git a/comfy_api_nodes/nodes_topaz.py b/comfy_api_nodes/nodes_topaz.py index b18b31af1..e79c16d3c 100644 --- a/comfy_api_nodes/nodes_topaz.py +++ b/comfy_api_nodes/nodes_topaz.py @@ -36,11 +36,15 @@ from comfy_api_nodes.util import ( ) UPSCALER_MODELS_MAP = { + "Astra 2": "ast-2", "Starlight (Astra) Fast": "slf-1", "Starlight (Astra) Creative": "slc-1", "Starlight Precise 2.5": "slp-2.5", } +AST2_MAX_FRAMES = 9000 +AST2_MAX_FRAMES_WITH_PROMPT = 450 + class TopazImageEnhance(IO.ComfyNode): @classmethod @@ -230,13 +234,20 @@ class TopazVideoEnhance(IO.ComfyNode): def define_schema(cls): return IO.Schema( node_id="TopazVideoEnhance", - display_name="Topaz Video Enhance", + display_name="Topaz Video Enhance (Legacy)", category="api node/video/Topaz", description="Breathe new life into video with powerful upscaling and recovery technology.", inputs=[ IO.Video.Input("video"), IO.Boolean.Input("upscaler_enabled", default=True), - IO.Combo.Input("upscaler_model", options=list(UPSCALER_MODELS_MAP.keys())), + IO.Combo.Input( + "upscaler_model", + options=[ + "Starlight (Astra) Fast", + "Starlight (Astra) Creative", + "Starlight Precise 2.5", + ], + ), IO.Combo.Input("upscaler_resolution", options=["FullHD (1080p)", "4K (2160p)"]), IO.Combo.Input( "upscaler_creativity", @@ -304,6 +315,7 @@ class TopazVideoEnhance(IO.ComfyNode): IO.Hidden.unique_id, ], is_api_node=True, + is_deprecated=True, ) @classmethod @@ -453,7 +465,350 @@ class TopazVideoEnhance(IO.ComfyNode): progress_extractor=lambda x: getattr(x, "progress", 0), price_extractor=lambda x: (x.estimates.cost[0] * 0.08 if x.estimates and x.estimates.cost[0] else None), poll_interval=10.0, - max_poll_attempts=320, + ) + return IO.NodeOutput(await download_url_to_video_output(final_response.download.url)) + + +class TopazVideoEnhanceV2(IO.ComfyNode): + @classmethod + def define_schema(cls): + return IO.Schema( + node_id="TopazVideoEnhanceV2", + display_name="Topaz Video Enhance", + category="api node/video/Topaz", + description="Breathe new life into video with powerful upscaling and recovery technology.", + inputs=[ + IO.Video.Input("video"), + IO.DynamicCombo.Input( + "upscaler_model", + options=[ + IO.DynamicCombo.Option( + "Astra 2", + [ + IO.Combo.Input("upscaler_resolution", options=["FullHD (1080p)", "4K (2160p)"]), + IO.Float.Input( + "creativity", + default=0.5, + min=0.0, + max=1.0, + step=0.1, + display_mode=IO.NumberDisplay.slider, + tooltip="Creative strength of the upscale.", + ), + IO.String.Input( + "prompt", + multiline=True, + default="", + tooltip="Optional descriptive (not instructive) scene prompt." + f"Capping input at {AST2_MAX_FRAMES_WITH_PROMPT} frames (~15s @ 30fps) when set.", + ), + IO.Float.Input( + "sharp", + default=0.5, + min=0.0, + max=1.0, + step=0.01, + display_mode=IO.NumberDisplay.slider, + tooltip="Pre-enhance sharpness: " + "0.0=Gaussian blur, 0.5=passthrough (default), 1.0=USM sharpening.", + advanced=True, + ), + IO.Float.Input( + "realism", + default=0.0, + min=0.0, + max=1.0, + step=0.01, + display_mode=IO.NumberDisplay.slider, + tooltip="Pulls output toward photographic realism." + "Leave at 0 for the model default.", + advanced=True, + ), + ], + ), + IO.DynamicCombo.Option( + "Starlight (Astra) Fast", + [IO.Combo.Input("upscaler_resolution", options=["FullHD (1080p)", "4K (2160p)"]),], + ), + IO.DynamicCombo.Option( + "Starlight (Astra) Creative", + [ + IO.Combo.Input("upscaler_resolution", options=["FullHD (1080p)", "4K (2160p)"]), + IO.Combo.Input( + "creativity", + options=["low", "middle", "high"], + default="low", + tooltip="Creative strength of the upscale.", + ), + ], + ), + IO.DynamicCombo.Option( + "Starlight Precise 2.5", + [IO.Combo.Input("upscaler_resolution", options=["FullHD (1080p)", "4K (2160p)"])], + ), + IO.DynamicCombo.Option("Disabled", []), + ], + ), + IO.DynamicCombo.Input( + "interpolation_model", + options=[ + IO.DynamicCombo.Option("Disabled", []), + IO.DynamicCombo.Option( + "apo-8", + [ + IO.Int.Input( + "interpolation_frame_rate", + default=60, + min=15, + max=240, + display_mode=IO.NumberDisplay.number, + tooltip="Output frame rate.", + ), + IO.Int.Input( + "interpolation_slowmo", + default=1, + min=1, + max=16, + display_mode=IO.NumberDisplay.number, + tooltip="Slow-motion factor applied to the input video. " + "For example, 2 makes the output twice as slow and doubles the duration.", + advanced=True, + ), + IO.Boolean.Input( + "interpolation_duplicate", + default=False, + tooltip="Analyze the input for duplicate frames and remove them.", + advanced=True, + ), + IO.Float.Input( + "interpolation_duplicate_threshold", + default=0.01, + min=0.001, + max=0.1, + step=0.001, + display_mode=IO.NumberDisplay.number, + tooltip="Detection sensitivity for duplicate frames.", + advanced=True, + ), + ], + ), + ], + ), + IO.Combo.Input( + "dynamic_compression_level", + options=["Low", "Mid", "High"], + default="Low", + tooltip="CQP level.", + optional=True, + ), + ], + outputs=[ + IO.Video.Output(), + ], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + depends_on=IO.PriceBadgeDepends(widgets=[ + "upscaler_model", + "upscaler_model.upscaler_resolution", + "interpolation_model", + ]), + expr=""" + ( + $model := $lookup(widgets, "upscaler_model"); + $res := $lookup(widgets, "upscaler_model.upscaler_resolution"); + $interp := $lookup(widgets, "interpolation_model"); + $is4k := $contains($res, "4k"); + $hasInterp := $interp != "disabled"; + $rates := { + "starlight (astra) fast": {"hd": 0.43, "uhd": 0.85}, + "starlight precise 2.5": {"hd": 0.70, "uhd": 1.54}, + "astra 2": {"hd": 1.72, "uhd": 2.85}, + "starlight (astra) creative": {"hd": 2.25, "uhd": 3.99} + }; + $surcharge := $is4k ? 0.28 : 0.14; + $entry := $lookup($rates, $model); + $base := $is4k ? $entry.uhd : $entry.hd; + $hi := $base + ($hasInterp ? $surcharge : 0); + $model = "disabled" + ? {"type":"text","text":"Interpolation only"} + : ($hasInterp + ? {"type":"text","text":"~" & $string($base) & "–" & $string($hi) & " credits/src frame"} + : {"type":"text","text":"~" & $string($base) & " credits/src frame"}) + ) + """, + ), + ) + + @classmethod + async def execute( + cls, + video: Input.Video, + upscaler_model: dict, + interpolation_model: dict, + dynamic_compression_level: str = "Low", + ) -> IO.NodeOutput: + upscaler_choice = upscaler_model["upscaler_model"] + interpolation_choice = interpolation_model["interpolation_model"] + if upscaler_choice == "Disabled" and interpolation_choice == "Disabled": + raise ValueError("There is nothing to do: both upscaling and interpolation are disabled.") + validate_container_format_is_mp4(video) + src_width, src_height = video.get_dimensions() + src_frame_rate = int(video.get_frame_rate()) + duration_sec = video.get_duration() + src_video_stream = video.get_stream_source() + target_width = src_width + target_height = src_height + target_frame_rate = src_frame_rate + filters = [] + if upscaler_choice != "Disabled": + if "1080p" in upscaler_model["upscaler_resolution"]: + target_pixel_p = 1080 + max_long_side = 1920 + else: + target_pixel_p = 2160 + max_long_side = 3840 + ar = src_width / src_height + if src_width >= src_height: + # Landscape or Square; Attempt to set height to target (e.g., 2160), calculate width + target_height = target_pixel_p + target_width = int(target_height * ar) + # Check if width exceeds standard bounds (for ultra-wide e.g., 21:9 ARs) + if target_width > max_long_side: + target_width = max_long_side + target_height = int(target_width / ar) + else: + # Portrait; Attempt to set width to target (e.g., 2160), calculate height + target_width = target_pixel_p + target_height = int(target_width / ar) + # Check if height exceeds standard bounds + if target_height > max_long_side: + target_height = max_long_side + target_width = int(target_height * ar) + if target_width % 2 != 0: + target_width += 1 + if target_height % 2 != 0: + target_height += 1 + model_id = UPSCALER_MODELS_MAP[upscaler_choice] + if model_id == "slc-1": + filters.append( + VideoEnhancementFilter( + model=model_id, + creativity=upscaler_model["creativity"], + isOptimizedMode=True, + ) + ) + elif model_id == "ast-2": + n_frames = video.get_frame_count() + ast2_prompt = (upscaler_model["prompt"] or "").strip() + if ast2_prompt and n_frames > AST2_MAX_FRAMES_WITH_PROMPT: + raise ValueError( + f"Astra 2 with a prompt is limited to {AST2_MAX_FRAMES_WITH_PROMPT} input frames " + f"(~15s @ 30fps); video has {n_frames}. Clear the prompt or shorten the clip." + ) + if n_frames > AST2_MAX_FRAMES: + raise ValueError(f"Astra 2 is limited to {AST2_MAX_FRAMES} input frames; video has {n_frames}.") + realism = upscaler_model["realism"] + filters.append( + VideoEnhancementFilter( + model=model_id, + creativity=upscaler_model["creativity"], + prompt=(ast2_prompt or None), + sharp=upscaler_model["sharp"], + realism=(realism if realism > 0 else None), + ) + ) + else: + filters.append(VideoEnhancementFilter(model=model_id)) + if interpolation_choice != "Disabled": + target_frame_rate = interpolation_model["interpolation_frame_rate"] + filters.append( + VideoFrameInterpolationFilter( + model=interpolation_choice, + slowmo=interpolation_model["interpolation_slowmo"], + fps=interpolation_model["interpolation_frame_rate"], + duplicate=interpolation_model["interpolation_duplicate"], + duplicate_threshold=interpolation_model["interpolation_duplicate_threshold"], + ), + ) + initial_res = await sync_op( + cls, + ApiEndpoint(path="/proxy/topaz/video/", method="POST"), + response_model=CreateVideoResponse, + data=CreateVideoRequest( + source=CreateVideoRequestSource( + container="mp4", + size=get_fs_object_size(src_video_stream), + duration=int(duration_sec), + frameCount=video.get_frame_count(), + frameRate=src_frame_rate, + resolution=Resolution(width=src_width, height=src_height), + ), + filters=filters, + output=OutputInformationVideo( + resolution=Resolution(width=target_width, height=target_height), + frameRate=target_frame_rate, + audioCodec="AAC", + audioTransfer="Copy", + dynamicCompressionLevel=dynamic_compression_level, + ), + ), + wait_label="Creating task", + final_label_on_success="Task created", + ) + upload_res = await sync_op( + cls, + ApiEndpoint( + path=f"/proxy/topaz/video/{initial_res.requestId}/accept", + method="PATCH", + ), + response_model=VideoAcceptResponse, + wait_label="Preparing upload", + final_label_on_success="Upload started", + ) + if len(upload_res.urls) > 1: + raise NotImplementedError( + "Large files are not currently supported. Please open an issue in the ComfyUI repository." + ) + async with aiohttp.ClientSession(headers={"Content-Type": "video/mp4"}) as session: + if isinstance(src_video_stream, BytesIO): + src_video_stream.seek(0) + async with session.put(upload_res.urls[0], data=src_video_stream, raise_for_status=True) as res: + upload_etag = res.headers["Etag"] + else: + with builtins.open(src_video_stream, "rb") as video_file: + async with session.put(upload_res.urls[0], data=video_file, raise_for_status=True) as res: + upload_etag = res.headers["Etag"] + await sync_op( + cls, + ApiEndpoint( + path=f"/proxy/topaz/video/{initial_res.requestId}/complete-upload", + method="PATCH", + ), + response_model=VideoCompleteUploadResponse, + data=VideoCompleteUploadRequest( + uploadResults=[ + VideoCompleteUploadRequestPart( + partNum=1, + eTag=upload_etag, + ), + ], + ), + wait_label="Finalizing upload", + final_label_on_success="Upload completed", + ) + final_response = await poll_op( + cls, + ApiEndpoint(path=f"/proxy/topaz/video/{initial_res.requestId}/status"), + response_model=VideoStatusResponse, + status_extractor=lambda x: x.status, + progress_extractor=lambda x: getattr(x, "progress", 0), + price_extractor=lambda x: (x.estimates.cost[0] * 0.08 if x.estimates and x.estimates.cost[0] else None), + poll_interval=10.0, ) return IO.NodeOutput(await download_url_to_video_output(final_response.download.url)) @@ -464,6 +819,7 @@ class TopazExtension(ComfyExtension): return [ TopazImageEnhance, TopazVideoEnhance, + TopazVideoEnhanceV2, ] diff --git a/comfy_api_nodes/nodes_tripo.py b/comfy_api_nodes/nodes_tripo.py index 9f4298dce..d6501dee4 100644 --- a/comfy_api_nodes/nodes_tripo.py +++ b/comfy_api_nodes/nodes_tripo.py @@ -60,6 +60,7 @@ async def poll_until_finished( ], status_extractor=lambda x: x.data.status, progress_extractor=lambda x: x.data.progress, + price_extractor=lambda x: x.data.consumed_credit * 0.01 if x.data.consumed_credit else None, estimated_duration=average_duration, ) if response_poll.data.status == TripoTaskStatus.SUCCESS: @@ -113,7 +114,6 @@ class TripoTextToModelNode(IO.ComfyNode): depends_on=IO.PriceBadgeDepends( widgets=[ "model_version", - "style", "texture", "pbr", "quad", @@ -124,20 +124,17 @@ class TripoTextToModelNode(IO.ComfyNode): expr=""" ( $isV14 := $contains(widgets.model_version,"v1.4"); - $style := widgets.style; - $hasStyle := ($style != "" and $style != "none"); + $isV3OrLater := $contains(widgets.model_version,"v3."); $withTexture := widgets.texture or widgets.pbr; $isHdTexture := (widgets.texture_quality = "detailed"); $isDetailedGeometry := (widgets.geometry_quality = "detailed"); - $baseCredits := - $isV14 ? 20 : ($withTexture ? 20 : 10); - $credits := - $baseCredits - + ($hasStyle ? 5 : 0) + $credits := $isV14 ? 20 : ( + ($withTexture ? 20 : 10) + (widgets.quad ? 5 : 0) + ($isHdTexture ? 10 : 0) - + ($isDetailedGeometry ? 20 : 0); - {"type":"usd","usd": $round($credits * 0.01, 2)} + + (($isDetailedGeometry and $isV3OrLater) ? 20 : 0) + ); + {"type":"usd","usd": $round($credits * 0.01, 2), "format": {"approximate": true}} ) """, ), @@ -239,7 +236,6 @@ class TripoImageToModelNode(IO.ComfyNode): depends_on=IO.PriceBadgeDepends( widgets=[ "model_version", - "style", "texture", "pbr", "quad", @@ -250,20 +246,17 @@ class TripoImageToModelNode(IO.ComfyNode): expr=""" ( $isV14 := $contains(widgets.model_version,"v1.4"); - $style := widgets.style; - $hasStyle := ($style != "" and $style != "none"); + $isV3OrLater := $contains(widgets.model_version,"v3."); $withTexture := widgets.texture or widgets.pbr; $isHdTexture := (widgets.texture_quality = "detailed"); $isDetailedGeometry := (widgets.geometry_quality = "detailed"); - $baseCredits := - $isV14 ? 30 : ($withTexture ? 30 : 20); - $credits := - $baseCredits - + ($hasStyle ? 5 : 0) + $credits := $isV14 ? 30 : ( + ($withTexture ? 30 : 20) + (widgets.quad ? 5 : 0) + ($isHdTexture ? 10 : 0) - + ($isDetailedGeometry ? 20 : 0); - {"type":"usd","usd": $round($credits * 0.01, 2)} + + (($isDetailedGeometry and $isV3OrLater) ? 20 : 0) + ); + {"type":"usd","usd": $round($credits * 0.01, 2), "format": {"approximate": true}} ) """, ), @@ -358,7 +351,7 @@ class TripoMultiviewToModelNode(IO.ComfyNode): "texture_alignment", default="original_image", options=["original_image", "geometry"], optional=True, advanced=True ), IO.Int.Input("face_limit", default=-1, min=-1, max=500000, optional=True, advanced=True), - IO.Boolean.Input("quad", default=False, optional=True, advanced=True), + IO.Boolean.Input("quad", default=False, optional=True, advanced=True, tooltip="This parameter is deprecated and does nothing."), IO.Combo.Input("geometry_quality", default="standard", options=["standard", "detailed"], optional=True, advanced=True), ], outputs=[ @@ -379,7 +372,6 @@ class TripoMultiviewToModelNode(IO.ComfyNode): "model_version", "texture", "pbr", - "quad", "texture_quality", "geometry_quality", ], @@ -387,17 +379,16 @@ class TripoMultiviewToModelNode(IO.ComfyNode): expr=""" ( $isV14 := $contains(widgets.model_version,"v1.4"); + $isV3OrLater := $contains(widgets.model_version,"v3."); $withTexture := widgets.texture or widgets.pbr; $isHdTexture := (widgets.texture_quality = "detailed"); $isDetailedGeometry := (widgets.geometry_quality = "detailed"); - $baseCredits := - $isV14 ? 30 : ($withTexture ? 30 : 20); - $credits := - $baseCredits - + (widgets.quad ? 5 : 0) + $credits := $isV14 ? 30 : ( + ($withTexture ? 30 : 20) + ($isHdTexture ? 10 : 0) - + ($isDetailedGeometry ? 20 : 0); - {"type":"usd","usd": $round($credits * 0.01, 2)} + + (($isDetailedGeometry and $isV3OrLater) ? 20 : 0) + ); + {"type":"usd","usd": $round($credits * 0.01, 2), "format": {"approximate": true}} ) """, ), @@ -457,7 +448,7 @@ class TripoMultiviewToModelNode(IO.ComfyNode): geometry_quality=geometry_quality, texture_alignment=texture_alignment, face_limit=face_limit if face_limit != -1 else None, - quad=quad, + quad=None, ), ) return await poll_until_finished(cls, response, average_duration=80) @@ -498,7 +489,7 @@ class TripoTextureNode(IO.ComfyNode): expr=""" ( $tq := widgets.texture_quality; - {"type":"usd","usd": ($contains($tq,"detailed") ? 0.2 : 0.1)} + {"type":"usd","usd": ($contains($tq,"detailed") ? 0.2 : 0.1), "format": {"approximate": true}} ) """, ), @@ -555,7 +546,7 @@ class TripoRefineNode(IO.ComfyNode): is_api_node=True, is_output_node=True, price_badge=IO.PriceBadge( - expr="""{"type":"usd","usd":0.3}""", + expr="""{"type":"usd","usd":0.3, "format": {"approximate": true}}""", ), ) @@ -592,7 +583,7 @@ class TripoRigNode(IO.ComfyNode): is_api_node=True, is_output_node=True, price_badge=IO.PriceBadge( - expr="""{"type":"usd","usd":0.25}""", + expr="""{"type":"usd","usd":0.25, "format": {"approximate": true}}""", ), ) @@ -652,7 +643,7 @@ class TripoRetargetNode(IO.ComfyNode): is_api_node=True, is_output_node=True, price_badge=IO.PriceBadge( - expr="""{"type":"usd","usd":0.1}""", + expr="""{"type":"usd","usd":0.1, "format": {"approximate": true}}""", ), ) @@ -761,19 +752,10 @@ class TripoConversionNode(IO.ComfyNode): "face_limit", "texture_size", "texture_format", - "force_symmetry", "flatten_bottom", "flatten_bottom_threshold", "pivot_to_center_bottom", "scale_factor", - "with_animation", - "pack_uv", - "bake", - "part_names", - "fbx_preset", - "export_vertex_colors", - "export_orientation", - "animate_in_place", ], ), expr=""" @@ -783,28 +765,16 @@ class TripoConversionNode(IO.ComfyNode): $flatThresh := (widgets.flatten_bottom_threshold != null) ? widgets.flatten_bottom_threshold : 0; $scale := (widgets.scale_factor != null) ? widgets.scale_factor : 1; $texFmt := (widgets.texture_format != "" ? widgets.texture_format : "jpeg"); - $part := widgets.part_names; - $fbx := (widgets.fbx_preset != "" ? widgets.fbx_preset : "blender"); - $orient := (widgets.export_orientation != "" ? widgets.export_orientation : "default"); $advanced := widgets.quad or - widgets.force_symmetry or widgets.flatten_bottom or widgets.pivot_to_center_bottom or - widgets.with_animation or - widgets.pack_uv or - widgets.bake or - widgets.export_vertex_colors or - widgets.animate_in_place or ($face != -1) or ($texSize != 4096) or ($flatThresh != 0) or ($scale != 1) or - ($texFmt != "jpeg") or - ($part != "") or - ($fbx != "blender") or - ($orient != "default"); - {"type":"usd","usd": ($advanced ? 0.1 : 0.05)} + ($texFmt != "jpeg"); + {"type":"usd","usd": ($advanced ? 0.1 : 0.05), "format": {"approximate": true}} ) """, ), diff --git a/comfy_api_nodes/nodes_veo2.py b/comfy_api_nodes/nodes_veo2.py index 13fc1cc36..2ff75d9b2 100644 --- a/comfy_api_nodes/nodes_veo2.py +++ b/comfy_api_nodes/nodes_veo2.py @@ -24,8 +24,9 @@ from comfy_api_nodes.util import ( AVERAGE_DURATION_VIDEO_GEN = 32 MODELS_MAP = { "veo-2.0-generate-001": "veo-2.0-generate-001", - "veo-3.1-generate": "veo-3.1-generate-preview", - "veo-3.1-fast-generate": "veo-3.1-fast-generate-preview", + "veo-3.1-generate": "veo-3.1-generate-001", + "veo-3.1-fast-generate": "veo-3.1-fast-generate-001", + "veo-3.1-lite": "veo-3.1-lite-generate-001", "veo-3.0-generate-001": "veo-3.0-generate-001", "veo-3.0-fast-generate-001": "veo-3.0-fast-generate-001", } @@ -247,17 +248,8 @@ class VeoVideoGenerationNode(IO.ComfyNode): raise Exception("Video generation completed but no video was returned") -class Veo3VideoGenerationNode(VeoVideoGenerationNode): - """ - Generates videos from text prompts using Google's Veo 3 API. - - Supported models: - - veo-3.0-generate-001 - - veo-3.0-fast-generate-001 - - This node extends the base Veo node with Veo 3 specific features including - audio generation and fixed 8-second duration. - """ +class Veo3VideoGenerationNode(IO.ComfyNode): + """Generates videos from text prompts using Google's Veo 3 API.""" @classmethod def define_schema(cls): @@ -279,6 +271,13 @@ class Veo3VideoGenerationNode(VeoVideoGenerationNode): default="16:9", tooltip="Aspect ratio of the output video", ), + IO.Combo.Input( + "resolution", + options=["720p", "1080p", "4k"], + default="720p", + tooltip="Output video resolution. 4K is not available for veo-3.1-lite and veo-3.0 models.", + optional=True, + ), IO.String.Input( "negative_prompt", multiline=True, @@ -289,11 +288,11 @@ class Veo3VideoGenerationNode(VeoVideoGenerationNode): IO.Int.Input( "duration_seconds", default=8, - min=8, + min=4, max=8, - step=1, + step=2, display_mode=IO.NumberDisplay.number, - tooltip="Duration of the output video in seconds (Veo 3 only supports 8 seconds)", + tooltip="Duration of the output video in seconds", optional=True, ), IO.Boolean.Input( @@ -332,10 +331,10 @@ class Veo3VideoGenerationNode(VeoVideoGenerationNode): options=[ "veo-3.1-generate", "veo-3.1-fast-generate", + "veo-3.1-lite", "veo-3.0-generate-001", "veo-3.0-fast-generate-001", ], - default="veo-3.0-generate-001", tooltip="Veo 3 model to use for video generation", optional=True, ), @@ -356,21 +355,111 @@ class Veo3VideoGenerationNode(VeoVideoGenerationNode): ], is_api_node=True, price_badge=IO.PriceBadge( - depends_on=IO.PriceBadgeDepends(widgets=["model", "generate_audio"]), + depends_on=IO.PriceBadgeDepends(widgets=["model", "generate_audio", "resolution", "duration_seconds"]), expr=""" ( $m := widgets.model; + $r := widgets.resolution; $a := widgets.generate_audio; - ($contains($m,"veo-3.0-fast-generate-001") or $contains($m,"veo-3.1-fast-generate")) - ? {"type":"usd","usd": ($a ? 1.2 : 0.8)} - : ($contains($m,"veo-3.0-generate-001") or $contains($m,"veo-3.1-generate")) - ? {"type":"usd","usd": ($a ? 3.2 : 1.6)} - : {"type":"range_usd","min_usd":0.8,"max_usd":3.2} + $seconds := widgets.duration_seconds; + $pps := + $contains($m, "lite") + ? ($r = "1080p" ? ($a ? 0.08 : 0.05) : ($a ? 0.05 : 0.03)) + : $contains($m, "3.1-fast") + ? ($r = "4k" ? ($a ? 0.30 : 0.25) : $r = "1080p" ? ($a ? 0.12 : 0.10) : ($a ? 0.10 : 0.08)) + : $contains($m, "3.1-generate") + ? ($r = "4k" ? ($a ? 0.60 : 0.40) : ($a ? 0.40 : 0.20)) + : $contains($m, "3.0-fast") + ? ($a ? 0.15 : 0.10) + : ($a ? 0.40 : 0.20); + {"type":"usd","usd": $pps * $seconds} ) """, ), ) + @classmethod + async def execute( + cls, + prompt, + aspect_ratio="16:9", + resolution="720p", + negative_prompt="", + duration_seconds=8, + enhance_prompt=True, + person_generation="ALLOW", + seed=0, + image=None, + model="veo-3.0-generate-001", + generate_audio=False, + ): + if resolution == "4k" and ("lite" in model or "3.0" in model): + raise Exception("4K resolution is not supported by the veo-3.1-lite or veo-3.0 models.") + + model = MODELS_MAP[model] + + instances = [{"prompt": prompt}] + if image is not None: + image_base64 = tensor_to_base64_string(image) + if image_base64: + instances[0]["image"] = {"bytesBase64Encoded": image_base64, "mimeType": "image/png"} + + parameters = { + "aspectRatio": aspect_ratio, + "personGeneration": person_generation, + "durationSeconds": duration_seconds, + "enhancePrompt": True, + "generateAudio": generate_audio, + } + if negative_prompt: + parameters["negativePrompt"] = negative_prompt + if seed > 0: + parameters["seed"] = seed + if "veo-3.1" in model: + parameters["resolution"] = resolution + + initial_response = await sync_op( + cls, + ApiEndpoint(path=f"/proxy/veo/{model}/generate", method="POST"), + response_model=VeoGenVidResponse, + data=VeoGenVidRequest( + instances=instances, + parameters=parameters, + ), + ) + + poll_response = await poll_op( + cls, + ApiEndpoint(path=f"/proxy/veo/{model}/poll", method="POST"), + response_model=VeoGenVidPollResponse, + status_extractor=lambda r: "completed" if r.done else "pending", + data=VeoGenVidPollRequest(operationName=initial_response.name), + poll_interval=9.0, + estimated_duration=AVERAGE_DURATION_VIDEO_GEN, + ) + + if poll_response.error: + raise Exception(f"Veo API error: {poll_response.error.message} (code: {poll_response.error.code})") + + response = poll_response.response + filtered_count = response.raiMediaFilteredCount + if filtered_count: + reasons = response.raiMediaFilteredReasons or [] + reason_part = f": {reasons[0]}" if reasons else "" + raise Exception( + f"Content blocked by Google's Responsible AI filters{reason_part} " + f"({filtered_count} video{'s' if filtered_count != 1 else ''} filtered)." + ) + + if response.videos: + video = response.videos[0] + if video.bytesBase64Encoded: + return IO.NodeOutput(InputImpl.VideoFromFile(BytesIO(base64.b64decode(video.bytesBase64Encoded)))) + if video.gcsUri: + return IO.NodeOutput(await download_url_to_video_output(video.gcsUri)) + raise Exception("Video returned but no data or URL was provided") + raise Exception("Video generation completed but no video was returned") + class Veo3FirstLastFrameNode(IO.ComfyNode): @@ -394,7 +483,7 @@ class Veo3FirstLastFrameNode(IO.ComfyNode): default="", tooltip="Negative text prompt to guide what to avoid in the video", ), - IO.Combo.Input("resolution", options=["720p", "1080p"]), + IO.Combo.Input("resolution", options=["720p", "1080p", "4k"]), IO.Combo.Input( "aspect_ratio", options=["16:9", "9:16"], @@ -424,8 +513,7 @@ class Veo3FirstLastFrameNode(IO.ComfyNode): IO.Image.Input("last_frame", tooltip="End frame"), IO.Combo.Input( "model", - options=["veo-3.1-generate", "veo-3.1-fast-generate"], - default="veo-3.1-fast-generate", + options=["veo-3.1-generate", "veo-3.1-fast-generate", "veo-3.1-lite"], ), IO.Boolean.Input( "generate_audio", @@ -443,26 +531,20 @@ class Veo3FirstLastFrameNode(IO.ComfyNode): ], is_api_node=True, price_badge=IO.PriceBadge( - depends_on=IO.PriceBadgeDepends(widgets=["model", "generate_audio", "duration"]), + depends_on=IO.PriceBadgeDepends(widgets=["model", "generate_audio", "duration", "resolution"]), expr=""" ( - $prices := { - "veo-3.1-fast-generate": { "audio": 0.15, "no_audio": 0.10 }, - "veo-3.1-generate": { "audio": 0.40, "no_audio": 0.20 } - }; $m := widgets.model; - $ga := (widgets.generate_audio = "true"); + $r := widgets.resolution; + $ga := widgets.generate_audio; $seconds := widgets.duration; - $modelKey := - $contains($m, "veo-3.1-fast-generate") ? "veo-3.1-fast-generate" : - $contains($m, "veo-3.1-generate") ? "veo-3.1-generate" : - ""; - $audioKey := $ga ? "audio" : "no_audio"; - $modelPrices := $lookup($prices, $modelKey); - $pps := $lookup($modelPrices, $audioKey); - ($pps != null) - ? {"type":"usd","usd": $pps * $seconds} - : {"type":"range_usd","min_usd": 0.4, "max_usd": 3.2} + $pps := + $contains($m, "lite") + ? ($r = "1080p" ? ($ga ? 0.08 : 0.05) : ($ga ? 0.05 : 0.03)) + : $contains($m, "fast") + ? ($r = "4k" ? ($ga ? 0.30 : 0.25) : $r = "1080p" ? ($ga ? 0.12 : 0.10) : ($ga ? 0.10 : 0.08)) + : ($r = "4k" ? ($ga ? 0.60 : 0.40) : ($ga ? 0.40 : 0.20)); + {"type":"usd","usd": $pps * $seconds} ) """, ), @@ -482,6 +564,9 @@ class Veo3FirstLastFrameNode(IO.ComfyNode): model: str, generate_audio: bool, ): + if "lite" in model and resolution == "4k": + raise Exception("4K resolution is not supported by the veo-3.1-lite model.") + model = MODELS_MAP[model] initial_response = await sync_op( cls, @@ -519,7 +604,7 @@ class Veo3FirstLastFrameNode(IO.ComfyNode): data=VeoGenVidPollRequest( operationName=initial_response.name, ), - poll_interval=5.0, + poll_interval=9.0, estimated_duration=AVERAGE_DURATION_VIDEO_GEN, ) diff --git a/comfy_api_nodes/nodes_vidu.py b/comfy_api_nodes/nodes_vidu.py index f04407eb5..8d90cefeb 100644 --- a/comfy_api_nodes/nodes_vidu.py +++ b/comfy_api_nodes/nodes_vidu.py @@ -38,7 +38,7 @@ async def execute_task( cls: type[IO.ComfyNode], vidu_endpoint: str, payload: TaskCreationRequest | TaskExtendCreationRequest | TaskMultiFrameCreationRequest, - max_poll_attempts: int = 320, + max_poll_attempts: int = 480, ) -> list[TaskResult]: task_creation_response = await sync_op( cls, @@ -1097,7 +1097,6 @@ class ViduExtendVideoNode(IO.ComfyNode): video_url=await upload_video_to_comfyapi(cls, video, wait_label="Uploading video"), images=[image_url] if image_url else None, ), - max_poll_attempts=480, ) return IO.NodeOutput(await download_url_to_video_output(results[0].url)) diff --git a/comfy_api_nodes/nodes_wan.py b/comfy_api_nodes/nodes_wan.py index d1470894a..68061bb5c 100644 --- a/comfy_api_nodes/nodes_wan.py +++ b/comfy_api_nodes/nodes_wan.py @@ -818,7 +818,6 @@ class WanReferenceVideoApi(IO.ComfyNode): response_model=VideoTaskStatusResponse, status_extractor=lambda x: x.output.task_status, poll_interval=6, - max_poll_attempts=280, ) return IO.NodeOutput(await download_url_to_video_output(response.output.video_url)) @@ -1646,6 +1645,557 @@ class Wan2ReferenceVideoApi(IO.ComfyNode): return IO.NodeOutput(await download_url_to_video_output(response.output.video_url)) +class HappyHorseTextToVideoApi(IO.ComfyNode): + @classmethod + def define_schema(cls): + return IO.Schema( + node_id="HappyHorseTextToVideoApi", + display_name="HappyHorse Text to Video", + category="api node/video/Wan", + description="Generates a video based on a text prompt using the HappyHorse model.", + inputs=[ + IO.DynamicCombo.Input( + "model", + options=[ + IO.DynamicCombo.Option( + "happyhorse-1.0-t2v", + [ + IO.String.Input( + "prompt", + multiline=True, + default="", + tooltip="Prompt describing the elements and visual features. " + "Supports English and Chinese.", + ), + IO.Combo.Input( + "resolution", + options=["720P", "1080P"], + ), + IO.Combo.Input( + "ratio", + options=["16:9", "9:16", "1:1", "4:3", "3:4"], + ), + IO.Int.Input( + "duration", + default=5, + min=3, + max=15, + step=1, + display_mode=IO.NumberDisplay.number, + ), + ], + ), + ], + ), + IO.Int.Input( + "seed", + default=0, + min=0, + max=2147483647, + step=1, + display_mode=IO.NumberDisplay.number, + control_after_generate=True, + tooltip="Seed to use for generation.", + ), + IO.Boolean.Input( + "watermark", + default=False, + tooltip="Whether to add an AI-generated watermark to the result.", + advanced=True, + ), + ], + outputs=[ + IO.Video.Output(), + ], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + depends_on=IO.PriceBadgeDepends(widgets=["model", "model.resolution", "model.duration"]), + expr=""" + ( + $res := $lookup(widgets, "model.resolution"); + $dur := $lookup(widgets, "model.duration"); + $ppsTable := { "720p": 0.14, "1080p": 0.24 }; + $pps := $lookup($ppsTable, $res); + { "type": "usd", "usd": $pps * $dur } + ) + """, + ), + ) + + @classmethod + async def execute( + cls, + model: dict, + seed: int, + watermark: bool, + ): + validate_string(model["prompt"], strip_whitespace=False, min_length=1) + initial_response = await sync_op( + cls, + ApiEndpoint( + path="/proxy/wan/api/v1/services/aigc/video-generation/video-synthesis", + method="POST", + ), + response_model=TaskCreationResponse, + data=Wan27Text2VideoTaskCreationRequest( + model=model["model"], + input=Text2VideoInputField( + prompt=model["prompt"], + negative_prompt=None, + ), + parameters=Wan27Text2VideoParametersField( + resolution=model["resolution"], + ratio=model["ratio"], + duration=model["duration"], + seed=seed, + watermark=watermark, + ), + ), + ) + if not initial_response.output: + raise Exception(f"An unknown error occurred: {initial_response.code} - {initial_response.message}") + response = await poll_op( + cls, + ApiEndpoint(path=f"/proxy/wan/api/v1/tasks/{initial_response.output.task_id}"), + response_model=VideoTaskStatusResponse, + status_extractor=lambda x: x.output.task_status, + poll_interval=7, + ) + return IO.NodeOutput(await download_url_to_video_output(response.output.video_url)) + + +class HappyHorseImageToVideoApi(IO.ComfyNode): + @classmethod + def define_schema(cls): + return IO.Schema( + node_id="HappyHorseImageToVideoApi", + display_name="HappyHorse Image to Video", + category="api node/video/Wan", + description="Generate a video from a first-frame image using the HappyHorse model.", + inputs=[ + IO.DynamicCombo.Input( + "model", + options=[ + IO.DynamicCombo.Option( + "happyhorse-1.0-i2v", + [ + IO.String.Input( + "prompt", + multiline=True, + default="", + tooltip="Prompt describing the elements and visual features. " + "Supports English and Chinese.", + ), + IO.Combo.Input( + "resolution", + options=["720P", "1080P"], + ), + IO.Int.Input( + "duration", + default=5, + min=3, + max=15, + step=1, + display_mode=IO.NumberDisplay.number, + ), + ], + ), + ], + ), + IO.Image.Input( + "first_frame", + tooltip="First frame image. The output aspect ratio is derived from this image.", + ), + IO.Int.Input( + "seed", + default=0, + min=0, + max=2147483647, + step=1, + display_mode=IO.NumberDisplay.number, + control_after_generate=True, + tooltip="Seed to use for generation.", + ), + IO.Boolean.Input( + "watermark", + default=False, + tooltip="Whether to add an AI-generated watermark to the result.", + advanced=True, + ), + ], + outputs=[ + IO.Video.Output(), + ], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + depends_on=IO.PriceBadgeDepends(widgets=["model", "model.resolution", "model.duration"]), + expr=""" + ( + $res := $lookup(widgets, "model.resolution"); + $dur := $lookup(widgets, "model.duration"); + $ppsTable := { "720p": 0.14, "1080p": 0.24 }; + $pps := $lookup($ppsTable, $res); + { "type": "usd", "usd": $pps * $dur } + ) + """, + ), + ) + + @classmethod + async def execute( + cls, + model: dict, + first_frame: Input.Image, + seed: int, + watermark: bool, + ): + media = [ + Wan27MediaItem( + type="first_frame", + url=await upload_image_to_comfyapi(cls, image=first_frame), + ) + ] + initial_response = await sync_op( + cls, + ApiEndpoint( + path="/proxy/wan/api/v1/services/aigc/video-generation/video-synthesis", + method="POST", + ), + response_model=TaskCreationResponse, + data=Wan27ImageToVideoTaskCreationRequest( + model=model["model"], + input=Wan27ImageToVideoInputField( + prompt=model["prompt"] or None, + negative_prompt=None, + media=media, + ), + parameters=Wan27ImageToVideoParametersField( + resolution=model["resolution"], + duration=model["duration"], + seed=seed, + watermark=watermark, + ), + ), + ) + if not initial_response.output: + raise Exception(f"An unknown error occurred: {initial_response.code} - {initial_response.message}") + response = await poll_op( + cls, + ApiEndpoint(path=f"/proxy/wan/api/v1/tasks/{initial_response.output.task_id}"), + response_model=VideoTaskStatusResponse, + status_extractor=lambda x: x.output.task_status, + poll_interval=7, + ) + return IO.NodeOutput(await download_url_to_video_output(response.output.video_url)) + + +class HappyHorseVideoEditApi(IO.ComfyNode): + @classmethod + def define_schema(cls): + return IO.Schema( + node_id="HappyHorseVideoEditApi", + display_name="HappyHorse Video Edit", + category="api node/video/Wan", + description="Edit a video using text instructions or reference images with the HappyHorse model. " + "Output duration is 3-15s and matches the input video; inputs longer than 15s are truncated.", + inputs=[ + IO.DynamicCombo.Input( + "model", + options=[ + IO.DynamicCombo.Option( + "happyhorse-1.0-video-edit", + [ + IO.String.Input( + "prompt", + multiline=True, + default="", + tooltip="Editing instructions or style transfer requirements.", + ), + IO.Combo.Input( + "resolution", + options=["720P", "1080P"], + ), + IO.Combo.Input( + "ratio", + options=["16:9", "9:16", "1:1", "4:3", "3:4"], + tooltip="Aspect ratio. If not changed, approximates the input video ratio.", + ), + IO.Autogrow.Input( + "reference_images", + template=IO.Autogrow.TemplateNames( + IO.Image.Input("reference_image"), + names=[ + "image1", + "image2", + "image3", + "image4", + "image5", + ], + min=0, + ), + ), + ], + ), + ], + ), + IO.Video.Input( + "video", + tooltip="The video to edit.", + ), + IO.Int.Input( + "seed", + default=0, + min=0, + max=2147483647, + step=1, + display_mode=IO.NumberDisplay.number, + control_after_generate=True, + tooltip="Seed to use for generation.", + ), + IO.Boolean.Input( + "watermark", + default=False, + tooltip="Whether to add an AI-generated watermark to the result.", + advanced=True, + ), + ], + outputs=[ + IO.Video.Output(), + ], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + depends_on=IO.PriceBadgeDepends(widgets=["model", "model.resolution"]), + expr=""" + ( + $res := $lookup(widgets, "model.resolution"); + $ppsTable := { "720p": 0.14, "1080p": 0.24 }; + $pps := $lookup($ppsTable, $res); + { "type": "usd", "usd": $pps, "format": { "suffix": "/second" } } + ) + """, + ), + ) + + @classmethod + async def execute( + cls, + model: dict, + video: Input.Video, + seed: int, + watermark: bool, + ): + validate_string(model["prompt"], strip_whitespace=False, min_length=1) + validate_video_duration(video, min_duration=3, max_duration=60) + media = [Wan27MediaItem(type="video", url=await upload_video_to_comfyapi(cls, video))] + reference_images = model.get("reference_images", {}) + for key in reference_images: + media.append( + Wan27MediaItem( + type="reference_image", url=await upload_image_to_comfyapi(cls, image=reference_images[key]) + ) + ) + initial_response = await sync_op( + cls, + ApiEndpoint( + path="/proxy/wan/api/v1/services/aigc/video-generation/video-synthesis", + method="POST", + ), + response_model=TaskCreationResponse, + data=Wan27VideoEditTaskCreationRequest( + model=model["model"], + input=Wan27VideoEditInputField(prompt=model["prompt"], media=media), + parameters=Wan27VideoEditParametersField( + resolution=model["resolution"], + ratio=model["ratio"], + duration=None, + watermark=watermark, + seed=seed, + ), + ), + ) + if not initial_response.output: + raise Exception(f"An unknown error occurred: {initial_response.code} - {initial_response.message}") + response = await poll_op( + cls, + ApiEndpoint(path=f"/proxy/wan/api/v1/tasks/{initial_response.output.task_id}"), + response_model=VideoTaskStatusResponse, + status_extractor=lambda x: x.output.task_status, + poll_interval=7, + ) + return IO.NodeOutput(await download_url_to_video_output(response.output.video_url)) + + +class HappyHorseReferenceVideoApi(IO.ComfyNode): + @classmethod + def define_schema(cls): + return IO.Schema( + node_id="HappyHorseReferenceVideoApi", + display_name="HappyHorse Reference to Video", + category="api node/video/Wan", + description="Generate a video featuring a person or object from reference materials with the HappyHorse " + "model. Supports single-character performances and multi-character interactions.", + inputs=[ + IO.DynamicCombo.Input( + "model", + options=[ + IO.DynamicCombo.Option( + "happyhorse-1.0-r2v", + [ + IO.String.Input( + "prompt", + multiline=True, + default="", + tooltip="Prompt describing the video. Use identifiers such as 'character1' and " + "'character2' to refer to the reference characters.", + ), + IO.Combo.Input( + "resolution", + options=["720P", "1080P"], + ), + IO.Combo.Input( + "ratio", + options=["16:9", "9:16", "1:1", "4:3", "3:4"], + ), + IO.Int.Input( + "duration", + default=5, + min=3, + max=15, + step=1, + display_mode=IO.NumberDisplay.number, + ), + IO.Autogrow.Input( + "reference_images", + template=IO.Autogrow.TemplateNames( + IO.Image.Input("reference_image"), + names=[ + "image1", + "image2", + "image3", + "image4", + "image5", + "image6", + "image7", + "image8", + "image9", + ], + min=1, + ), + ), + ], + ), + ], + ), + IO.Int.Input( + "seed", + default=0, + min=0, + max=2147483647, + step=1, + display_mode=IO.NumberDisplay.number, + control_after_generate=True, + tooltip="Seed to use for generation.", + ), + IO.Boolean.Input( + "watermark", + default=False, + tooltip="Whether to add an AI-generated watermark to the result.", + advanced=True, + ), + ], + outputs=[ + IO.Video.Output(), + ], + hidden=[ + IO.Hidden.auth_token_comfy_org, + IO.Hidden.api_key_comfy_org, + IO.Hidden.unique_id, + ], + is_api_node=True, + price_badge=IO.PriceBadge( + depends_on=IO.PriceBadgeDepends(widgets=["model", "model.resolution", "model.duration"]), + expr=""" + ( + $res := $lookup(widgets, "model.resolution"); + $dur := $lookup(widgets, "model.duration"); + $ppsTable := { "720p": 0.14, "1080p": 0.24 }; + $pps := $lookup($ppsTable, $res); + { "type": "usd", "usd": $pps * $dur } + ) + """, + ), + ) + + @classmethod + async def execute( + cls, + model: dict, + seed: int, + watermark: bool, + ): + validate_string(model["prompt"], strip_whitespace=False, min_length=1) + media = [] + reference_images = model.get("reference_images", {}) + for key in reference_images: + media.append( + Wan27MediaItem( + type="reference_image", + url=await upload_image_to_comfyapi(cls, image=reference_images[key]), + ) + ) + if not media: + raise ValueError("At least one reference reference image must be provided.") + + initial_response = await sync_op( + cls, + ApiEndpoint( + path="/proxy/wan/api/v1/services/aigc/video-generation/video-synthesis", + method="POST", + ), + response_model=TaskCreationResponse, + data=Wan27ReferenceVideoTaskCreationRequest( + model=model["model"], + input=Wan27ReferenceVideoInputField( + prompt=model["prompt"], + negative_prompt=None, + media=media, + ), + parameters=Wan27ReferenceVideoParametersField( + resolution=model["resolution"], + ratio=model["ratio"], + duration=model["duration"], + watermark=watermark, + seed=seed, + ), + ), + ) + if not initial_response.output: + raise Exception(f"An unknown error occurred: {initial_response.code} - {initial_response.message}") + response = await poll_op( + cls, + ApiEndpoint(path=f"/proxy/wan/api/v1/tasks/{initial_response.output.task_id}"), + response_model=VideoTaskStatusResponse, + status_extractor=lambda x: x.output.task_status, + poll_interval=7, + ) + return IO.NodeOutput(await download_url_to_video_output(response.output.video_url)) + + class WanApiExtension(ComfyExtension): @override async def get_node_list(self) -> list[type[IO.ComfyNode]]: @@ -1660,6 +2210,10 @@ class WanApiExtension(ComfyExtension): Wan2VideoContinuationApi, Wan2VideoEditApi, Wan2ReferenceVideoApi, + HappyHorseTextToVideoApi, + HappyHorseImageToVideoApi, + HappyHorseVideoEditApi, + HappyHorseReferenceVideoApi, ] diff --git a/comfy_api_nodes/nodes_wavespeed.py b/comfy_api_nodes/nodes_wavespeed.py index c59fafd3b..65e45f60a 100644 --- a/comfy_api_nodes/nodes_wavespeed.py +++ b/comfy_api_nodes/nodes_wavespeed.py @@ -84,7 +84,6 @@ class WavespeedFlashVSRNode(IO.ComfyNode): response_model=TaskResultResponse, status_extractor=lambda x: "failed" if x.data is None else x.data.status, poll_interval=10.0, - max_poll_attempts=480, ) if final_response.code != 200: raise ValueError( @@ -156,7 +155,6 @@ class WavespeedImageUpscaleNode(IO.ComfyNode): response_model=TaskResultResponse, status_extractor=lambda x: "failed" if x.data is None else x.data.status, poll_interval=10.0, - max_poll_attempts=480, ) if final_response.code != 200: raise ValueError( diff --git a/comfy_api_nodes/util/__init__.py b/comfy_api_nodes/util/__init__.py index 0cb9a47c7..f3584aba9 100644 --- a/comfy_api_nodes/util/__init__.py +++ b/comfy_api_nodes/util/__init__.py @@ -19,6 +19,7 @@ from .conversions import ( image_tensor_pair_to_batch, pil_to_bytesio, resize_mask_to_image, + resize_video_to_pixel_budget, tensor_to_base64_string, tensor_to_bytesio, tensor_to_pil, @@ -90,6 +91,7 @@ __all__ = [ "image_tensor_pair_to_batch", "pil_to_bytesio", "resize_mask_to_image", + "resize_video_to_pixel_budget", "tensor_to_base64_string", "tensor_to_bytesio", "tensor_to_pil", diff --git a/comfy_api_nodes/util/client.py b/comfy_api_nodes/util/client.py index 9d730b81a..052301c33 100644 --- a/comfy_api_nodes/util/client.py +++ b/comfy_api_nodes/util/client.py @@ -19,6 +19,8 @@ from comfy import utils from comfy_api.latest import IO from server import PromptServer +from comfy.deploy_environment import get_deploy_environment + from . import request_logger from ._helpers import ( default_base_url, @@ -148,7 +150,7 @@ async def poll_op( queued_statuses: list[str | int] | None = None, data: BaseModel | None = None, poll_interval: float = 5.0, - max_poll_attempts: int = 160, + max_poll_attempts: int = 480, timeout_per_poll: float = 120.0, max_retries_per_poll: int = 10, retry_delay_per_poll: float = 1.0, @@ -156,6 +158,7 @@ async def poll_op( estimated_duration: int | None = None, cancel_endpoint: ApiEndpoint | None = None, cancel_timeout: float = 10.0, + extra_text: str | None = None, ) -> M: raw = await poll_op_raw( cls, @@ -176,6 +179,7 @@ async def poll_op( estimated_duration=estimated_duration, cancel_endpoint=cancel_endpoint, cancel_timeout=cancel_timeout, + extra_text=extra_text, ) if not isinstance(raw, dict): raise Exception("Expected JSON response to validate into a Pydantic model, got non-JSON (binary or text).") @@ -252,7 +256,7 @@ async def poll_op_raw( queued_statuses: list[str | int] | None = None, data: dict[str, Any] | BaseModel | None = None, poll_interval: float = 5.0, - max_poll_attempts: int = 160, + max_poll_attempts: int = 480, timeout_per_poll: float = 120.0, max_retries_per_poll: int = 10, retry_delay_per_poll: float = 1.0, @@ -260,6 +264,7 @@ async def poll_op_raw( estimated_duration: int | None = None, cancel_endpoint: ApiEndpoint | None = None, cancel_timeout: float = 10.0, + extra_text: str | None = None, ) -> dict[str, Any]: """ Polls an endpoint until the task reaches a terminal state. Displays time while queued/processing, @@ -299,6 +304,7 @@ async def poll_op_raw( price=state.price, is_queued=state.is_queued, processing_elapsed_seconds=int(proc_elapsed), + extra_text=extra_text, ) await asyncio.sleep(1.0) except Exception as exc: @@ -389,6 +395,7 @@ async def poll_op_raw( price=state.price, is_queued=False, processing_elapsed_seconds=int(state.base_processing_elapsed), + extra_text=extra_text, ) return resp_json @@ -462,6 +469,7 @@ def _display_time_progress( price: float | None = None, is_queued: bool | None = None, processing_elapsed_seconds: int | None = None, + extra_text: str | None = None, ) -> None: if estimated_total is not None and estimated_total > 0 and is_queued is False: pe = processing_elapsed_seconds if processing_elapsed_seconds is not None else elapsed_seconds @@ -469,7 +477,8 @@ def _display_time_progress( time_line = f"Time elapsed: {int(elapsed_seconds)}s (~{remaining}s remaining)" else: time_line = f"Time elapsed: {int(elapsed_seconds)}s" - _display_text(node_cls, time_line, status=status, price=price) + text = f"{time_line}\n\n{extra_text}" if extra_text else time_line + _display_text(node_cls, text, status=status, price=price) async def _diagnose_connectivity() -> dict[str, bool]: @@ -479,10 +488,30 @@ async def _diagnose_connectivity() -> dict[str, bool]: "api_accessible": False, } timeout = aiohttp.ClientTimeout(total=5.0) + + # Probe Google and Baidu in parallel: Google is blocked by the GFW in mainland China, so a Baidu probe is required + # to correctly detect that Chinese users with working internet do have working internet. + internet_probe_urls = ("https://www.google.com", "https://www.baidu.com") + async with aiohttp.ClientSession(timeout=timeout) as session: - with contextlib.suppress(ClientError, OSError): - async with session.get("https://www.google.com") as resp: - results["internet_accessible"] = resp.status < 500 + async def _probe(url: str) -> bool: + try: + async with session.get(url) as resp: + return resp.status < 500 + except (ClientError, OSError, asyncio.TimeoutError): + return False + + probe_tasks = [asyncio.create_task(_probe(u)) for u in internet_probe_urls] + try: + for fut in asyncio.as_completed(probe_tasks): + if await fut: + results["internet_accessible"] = True + break + finally: + for t in probe_tasks: + if not t.done(): + t.cancel() + await asyncio.gather(*probe_tasks, return_exceptions=True) if not results["internet_accessible"]: return results @@ -617,6 +646,7 @@ async def _request_base(cfg: _RequestConfig, expect_binary: bool): payload_headers = {"Accept": "*/*"} if expect_binary else {"Accept": "application/json"} if not parsed_url.scheme and not parsed_url.netloc: # is URL relative? payload_headers.update(get_auth_header(cfg.node_cls)) + payload_headers["Comfy-Env"] = get_deploy_environment() if cfg.endpoint.headers: payload_headers.update(cfg.endpoint.headers) diff --git a/comfy_api_nodes/util/conversions.py b/comfy_api_nodes/util/conversions.py index 82b6d22a5..be5d5719b 100644 --- a/comfy_api_nodes/util/conversions.py +++ b/comfy_api_nodes/util/conversions.py @@ -129,22 +129,38 @@ def pil_to_bytesio(img: Image.Image, mime_type: str = "image/png") -> BytesIO: return img_byte_arr +def _compute_downscale_dims(src_w: int, src_h: int, total_pixels: int) -> tuple[int, int] | None: + """Return downscaled (w, h) with even dims fitting ``total_pixels``, or None if already fits. + + Source aspect ratio is preserved; output may drift by a fraction of a percent because both dimensions + are rounded down to even values (many codecs require divisible-by-2). + """ + pixels = src_w * src_h + if pixels <= total_pixels: + return None + scale = math.sqrt(total_pixels / pixels) + new_w = max(2, int(src_w * scale)) + new_h = max(2, int(src_h * scale)) + new_w -= new_w % 2 + new_h -= new_h % 2 + return new_w, new_h + + def downscale_image_tensor(image: torch.Tensor, total_pixels: int = 1536 * 1024) -> torch.Tensor: - """Downscale input image tensor to roughly the specified total pixels.""" + """Downscale input image tensor to roughly the specified total pixels. + + Output dimensions are rounded down to even values so that the result is guaranteed to fit within ``total_pixels`` + and is compatible with codecs that require even dimensions (e.g. yuv420p). + """ samples = image.movedim(-1, 1) - total = int(total_pixels) - scale_by = math.sqrt(total / (samples.shape[3] * samples.shape[2])) - if scale_by >= 1: + dims = _compute_downscale_dims(samples.shape[3], samples.shape[2], int(total_pixels)) + if dims is None: return image - width = round(samples.shape[3] * scale_by) - height = round(samples.shape[2] * scale_by) - - s = common_upscale(samples, width, height, "lanczos", "disabled") - s = s.movedim(1, -1) - return s + new_w, new_h = dims + return common_upscale(samples, new_w, new_h, "lanczos", "disabled").movedim(1, -1) -def downscale_image_tensor_by_max_side(image: torch.Tensor, *, max_side: int) -> torch.Tensor: +def downscale_image_tensor_by_max_side(image: torch.Tensor, *, max_side: int) -> torch.Tensor: """Downscale input image tensor so the largest dimension is at most max_side pixels.""" samples = image.movedim(-1, 1) height, width = samples.shape[2], samples.shape[3] @@ -399,6 +415,72 @@ def trim_video(video: Input.Video, duration_sec: float) -> Input.Video: raise RuntimeError(f"Failed to trim video: {str(e)}") from e +def resize_video_to_pixel_budget(video: Input.Video, total_pixels: int) -> Input.Video: + """Downscale a video to fit within ``total_pixels`` (w * h), preserving aspect ratio. + + Returns the original video object untouched when it already fits. Preserves frame rate, duration, and audio. + Aspect ratio is preserved up to a fraction of a percent (even-dim rounding). + """ + src_w, src_h = video.get_dimensions() + scale_dims = _compute_downscale_dims(src_w, src_h, total_pixels) + if scale_dims is None: + return video + return _apply_video_scale(video, scale_dims) + + +def _apply_video_scale(video: Input.Video, scale_dims: tuple[int, int]) -> Input.Video: + """Re-encode ``video`` scaled to ``scale_dims`` with a single decode/encode pass.""" + out_w, out_h = scale_dims + output_buffer = BytesIO() + input_container = None + output_container = None + + try: + input_source = video.get_stream_source() + input_container = av.open(input_source, mode="r") + output_container = av.open(output_buffer, mode="w", format="mp4") + + video_stream = output_container.add_stream("h264", rate=video.get_frame_rate()) + video_stream.width = out_w + video_stream.height = out_h + video_stream.pix_fmt = "yuv420p" + + audio_stream = None + for stream in input_container.streams: + if isinstance(stream, av.AudioStream): + audio_stream = output_container.add_stream("aac", rate=stream.sample_rate) + audio_stream.sample_rate = stream.sample_rate + audio_stream.layout = stream.layout + break + + for frame in input_container.decode(video=0): + frame = frame.reformat(width=out_w, height=out_h, format="yuv420p") + for packet in video_stream.encode(frame): + output_container.mux(packet) + for packet in video_stream.encode(): + output_container.mux(packet) + + if audio_stream is not None: + input_container.seek(0) + for audio_frame in input_container.decode(audio=0): + for packet in audio_stream.encode(audio_frame): + output_container.mux(packet) + for packet in audio_stream.encode(): + output_container.mux(packet) + + output_container.close() + input_container.close() + output_buffer.seek(0) + return InputImpl.VideoFromFile(output_buffer) + + except Exception as e: + if input_container is not None: + input_container.close() + if output_container is not None: + output_container.close() + raise RuntimeError(f"Failed to resize video: {str(e)}") from e + + def _f32_pcm(wav: torch.Tensor) -> torch.Tensor: """Convert audio to float 32 bits PCM format. Copy-paste from nodes_audio.py file.""" if wav.dtype.is_floating_point: diff --git a/comfy_execution/caching.py b/comfy_execution/caching.py index f9c913bdb..ba1e8bc84 100644 --- a/comfy_execution/caching.py +++ b/comfy_execution/caching.py @@ -5,6 +5,7 @@ import psutil import time import torch from typing import Sequence, Mapping, Dict +from comfy.model_patcher import ModelPatcher from comfy_execution.graph import DynamicPrompt from abc import ABC, abstractmethod @@ -523,13 +524,15 @@ class RAMPressureCache(LRUCache): self.timestamps[self.cache_key_set.get_data_key(node_id)] = time.time() super().set_local(node_id, value) - def ram_release(self, target): + def ram_release(self, target, free_active=False): if psutil.virtual_memory().available >= target: return clean_list = [] for key, cache_entry in self.cache.items(): + if not free_active and self.used_generation[key] == self.generation: + continue oom_score = RAM_CACHE_OLD_WORKFLOW_OOM_MULTIPLIER ** (self.generation - self.used_generation[key]) ram_usage = RAM_CACHE_DEFAULT_RAM_USAGE @@ -542,6 +545,9 @@ class RAMPressureCache(LRUCache): scan_list_for_ram_usage(output) elif isinstance(output, torch.Tensor) and output.device.type == 'cpu': ram_usage += output.numel() * output.element_size() + elif isinstance(output, ModelPatcher) and self.used_generation[key] != self.generation: + #old ModelPatchers are the first to go + ram_usage = 1e30 scan_list_for_ram_usage(cache_entry.outputs) oom_score *= ram_usage diff --git a/comfy_extras/frame_interpolation_models/film_net.py b/comfy_extras/frame_interpolation_models/film_net.py new file mode 100644 index 000000000..36bc79dc3 --- /dev/null +++ b/comfy_extras/frame_interpolation_models/film_net.py @@ -0,0 +1,261 @@ +"""FILM: Frame Interpolation for Large Motion (ECCV 2022).""" + +import torch +import torch.nn as nn +import torch.nn.functional as F + +import comfy.ops + +ops = comfy.ops.disable_weight_init + + +class FilmConv2d(nn.Module): + """Conv2d with optional LeakyReLU and FILM-style padding.""" + + def __init__(self, in_channels, out_channels, size, activation=True, device=None, dtype=None, operations=ops): + super().__init__() + self.even_pad = not size % 2 + self.conv = operations.Conv2d(in_channels, out_channels, kernel_size=size, padding=size // 2 if size % 2 else 0, device=device, dtype=dtype) + self.activation = nn.LeakyReLU(0.2) if activation else None + + def forward(self, x): + if self.even_pad: + x = F.pad(x, (0, 1, 0, 1)) + x = self.conv(x) + if self.activation is not None: + x = self.activation(x) + return x + + +def _warp_core(image, flow, grid_x, grid_y): + dtype = image.dtype + H, W = flow.shape[2], flow.shape[3] + dx = flow[:, 0].float() / (W * 0.5) + dy = flow[:, 1].float() / (H * 0.5) + grid = torch.stack([grid_x[None, None, :] + dx, grid_y[None, :, None] + dy], dim=3) + return F.grid_sample(image.float(), grid, mode="bilinear", padding_mode="border", align_corners=False).to(dtype) + + +def build_image_pyramid(image, pyramid_levels): + pyramid = [image] + for _ in range(1, pyramid_levels): + image = F.avg_pool2d(image, 2, 2) + pyramid.append(image) + return pyramid + + +def flow_pyramid_synthesis(residual_pyramid): + flow = residual_pyramid[-1] + flow_pyramid = [flow] + for residual_flow in residual_pyramid[:-1][::-1]: + flow = F.interpolate(flow, size=residual_flow.shape[2:4], mode="bilinear", scale_factor=None).mul_(2).add_(residual_flow) + flow_pyramid.append(flow) + flow_pyramid.reverse() + return flow_pyramid + + +def multiply_pyramid(pyramid, scalar): + return [image * scalar[:, None, None, None] for image in pyramid] + + +def pyramid_warp(feature_pyramid, flow_pyramid, warp_fn): + return [warp_fn(features, flow) for features, flow in zip(feature_pyramid, flow_pyramid)] + + +def concatenate_pyramids(pyramid1, pyramid2): + return [torch.cat([f1, f2], dim=1) for f1, f2 in zip(pyramid1, pyramid2)] + + +class SubTreeExtractor(nn.Module): + def __init__(self, in_channels=3, channels=64, n_layers=4, device=None, dtype=None, operations=ops): + super().__init__() + convs = [] + for i in range(n_layers): + out_ch = channels << i + convs.append(nn.Sequential( + FilmConv2d(in_channels, out_ch, 3, device=device, dtype=dtype, operations=operations), + FilmConv2d(out_ch, out_ch, 3, device=device, dtype=dtype, operations=operations))) + in_channels = out_ch + self.convs = nn.ModuleList(convs) + + def forward(self, image, n): + head = image + pyramid = [] + for i, layer in enumerate(self.convs): + head = layer(head) + pyramid.append(head) + if i < n - 1: + head = F.avg_pool2d(head, 2, 2) + return pyramid + + +class FeatureExtractor(nn.Module): + def __init__(self, in_channels=3, channels=64, sub_levels=4, device=None, dtype=None, operations=ops): + super().__init__() + self.extract_sublevels = SubTreeExtractor(in_channels, channels, sub_levels, device=device, dtype=dtype, operations=operations) + self.sub_levels = sub_levels + + def forward(self, image_pyramid): + sub_pyramids = [self.extract_sublevels(image_pyramid[i], min(len(image_pyramid) - i, self.sub_levels)) + for i in range(len(image_pyramid))] + feature_pyramid = [] + for i in range(len(image_pyramid)): + features = sub_pyramids[i][0] + for j in range(1, self.sub_levels): + if j <= i: + features = torch.cat([features, sub_pyramids[i - j][j]], dim=1) + feature_pyramid.append(features) + # Free sub-pyramids no longer needed by future levels + if i >= self.sub_levels - 1: + sub_pyramids[i - self.sub_levels + 1] = None + return feature_pyramid + + +class FlowEstimator(nn.Module): + def __init__(self, in_channels, num_convs, num_filters, device=None, dtype=None, operations=ops): + super().__init__() + self._convs = nn.ModuleList() + for _ in range(num_convs): + self._convs.append(FilmConv2d(in_channels, num_filters, 3, device=device, dtype=dtype, operations=operations)) + in_channels = num_filters + self._convs.append(FilmConv2d(in_channels, num_filters // 2, 1, device=device, dtype=dtype, operations=operations)) + self._convs.append(FilmConv2d(num_filters // 2, 2, 1, activation=False, device=device, dtype=dtype, operations=operations)) + + def forward(self, features_a, features_b): + net = torch.cat([features_a, features_b], dim=1) + for conv in self._convs: + net = conv(net) + return net + + +class PyramidFlowEstimator(nn.Module): + def __init__(self, filters=64, flow_convs=(3, 3, 3, 3), flow_filters=(32, 64, 128, 256), device=None, dtype=None, operations=ops): + super().__init__() + in_channels = filters << 1 + predictors = [] + for i in range(len(flow_convs)): + predictors.append(FlowEstimator(in_channels, flow_convs[i], flow_filters[i], device=device, dtype=dtype, operations=operations)) + in_channels += filters << (i + 2) + self._predictor = predictors[-1] + self._predictors = nn.ModuleList(predictors[:-1][::-1]) + + def forward(self, feature_pyramid_a, feature_pyramid_b, warp_fn): + levels = len(feature_pyramid_a) + v = self._predictor(feature_pyramid_a[-1], feature_pyramid_b[-1]) + residuals = [v] + # Coarse-to-fine: shared predictor for deep levels, then specialized predictors for fine levels + steps = [(i, self._predictor) for i in range(levels - 2, len(self._predictors) - 1, -1)] + steps += [(len(self._predictors) - 1 - k, p) for k, p in enumerate(self._predictors)] + for i, predictor in steps: + v = F.interpolate(v, size=feature_pyramid_a[i].shape[2:4], mode="bilinear").mul_(2) + v_residual = predictor(feature_pyramid_a[i], warp_fn(feature_pyramid_b[i], v)) + residuals.append(v_residual) + v = v.add_(v_residual) + residuals.reverse() + return residuals + + +def _get_fusion_channels(level, filters): + # Per direction: multi-scale features + RGB image (3ch) + flow (2ch), doubled for both directions + return (sum(filters << i for i in range(level)) + 3 + 2) * 2 + + +class Fusion(nn.Module): + def __init__(self, n_layers=4, specialized_layers=3, filters=64, device=None, dtype=None, operations=ops): + super().__init__() + self.output_conv = operations.Conv2d(filters, 3, kernel_size=1, device=device, dtype=dtype) + self.convs = nn.ModuleList() + in_channels = _get_fusion_channels(n_layers, filters) + increase = 0 + for i in range(n_layers)[::-1]: + num_filters = (filters << i) if i < specialized_layers else (filters << specialized_layers) + self.convs.append(nn.ModuleList([ + FilmConv2d(in_channels, num_filters, 2, activation=False, device=device, dtype=dtype, operations=operations), + FilmConv2d(in_channels + (increase or num_filters), num_filters, 3, device=device, dtype=dtype, operations=operations), + FilmConv2d(num_filters, num_filters, 3, device=device, dtype=dtype, operations=operations)])) + in_channels = num_filters + increase = _get_fusion_channels(i, filters) - num_filters // 2 + + def forward(self, pyramid): + net = pyramid[-1] + for k, layers in enumerate(self.convs): + i = len(self.convs) - 1 - k + net = layers[0](F.interpolate(net, size=pyramid[i].shape[2:4], mode="nearest")) + net = layers[2](layers[1](torch.cat([pyramid[i], net], dim=1))) + return self.output_conv(net) + + +class FILMNet(nn.Module): + def __init__(self, pyramid_levels=7, fusion_pyramid_levels=5, specialized_levels=3, sub_levels=4, + filters=64, flow_convs=(3, 3, 3, 3), flow_filters=(32, 64, 128, 256), device=None, dtype=None, operations=ops): + super().__init__() + self.pyramid_levels = pyramid_levels + self.fusion_pyramid_levels = fusion_pyramid_levels + self.extract = FeatureExtractor(3, filters, sub_levels, device=device, dtype=dtype, operations=operations) + self.predict_flow = PyramidFlowEstimator(filters, flow_convs, flow_filters, device=device, dtype=dtype, operations=operations) + self.fuse = Fusion(sub_levels, specialized_levels, filters, device=device, dtype=dtype, operations=operations) + self._warp_grids = {} + + def get_dtype(self): + return self.extract.extract_sublevels.convs[0][0].conv.weight.dtype + + def memory_used_forward(self, shape, dtype): + return 1700 * shape[1] * shape[2] * dtype.itemsize + + def _build_warp_grids(self, H, W, device): + """Pre-compute warp grids for all pyramid levels.""" + if (H, W) in self._warp_grids: + return + self._warp_grids = {} # clear old resolution grids to prevent memory leaks + for _ in range(self.pyramid_levels): + self._warp_grids[(H, W)] = ( + torch.linspace(-(1 - 1 / W), 1 - 1 / W, W, dtype=torch.float32, device=device), + torch.linspace(-(1 - 1 / H), 1 - 1 / H, H, dtype=torch.float32, device=device), + ) + H, W = H // 2, W // 2 + + def warp(self, image, flow): + grid_x, grid_y = self._warp_grids[(flow.shape[2], flow.shape[3])] + return _warp_core(image, flow, grid_x, grid_y) + + def extract_features(self, img): + """Extract image and feature pyramids for a single frame. Can be cached across pairs.""" + image_pyramid = build_image_pyramid(img, self.pyramid_levels) + feature_pyramid = self.extract(image_pyramid) + return image_pyramid, feature_pyramid + + def forward(self, img0, img1, timestep=0.5, cache=None): + # FILM uses a scalar timestep per batch element (spatially-varying timesteps not supported) + t = timestep.mean(dim=(1, 2, 3)).item() if isinstance(timestep, torch.Tensor) else timestep + return self.forward_multi_timestep(img0, img1, [t], cache=cache) + + def forward_multi_timestep(self, img0, img1, timesteps, cache=None): + """Compute flow once, synthesize at multiple timesteps. Expects batch=1 inputs.""" + self._build_warp_grids(img0.shape[2], img0.shape[3], img0.device) + + image_pyr0, feat_pyr0 = cache["img0"] if cache and "img0" in cache else self.extract_features(img0) + image_pyr1, feat_pyr1 = cache["img1"] if cache and "img1" in cache else self.extract_features(img1) + + fwd_flow = flow_pyramid_synthesis(self.predict_flow(feat_pyr0, feat_pyr1, self.warp))[:self.fusion_pyramid_levels] + bwd_flow = flow_pyramid_synthesis(self.predict_flow(feat_pyr1, feat_pyr0, self.warp))[:self.fusion_pyramid_levels] + + # Build warp targets and free full pyramids (only first fpl levels needed from here) + fpl = self.fusion_pyramid_levels + p2w = [concatenate_pyramids(image_pyr0[:fpl], feat_pyr0[:fpl]), + concatenate_pyramids(image_pyr1[:fpl], feat_pyr1[:fpl])] + del image_pyr0, image_pyr1, feat_pyr0, feat_pyr1 + + results = [] + dt_tensors = torch.tensor(timesteps, device=img0.device, dtype=img0.dtype) + for idx in range(len(timesteps)): + batch_dt = dt_tensors[idx:idx + 1] + bwd_scaled = multiply_pyramid(bwd_flow, batch_dt) + fwd_scaled = multiply_pyramid(fwd_flow, 1 - batch_dt) + fwd_warped = pyramid_warp(p2w[0], bwd_scaled, self.warp) + bwd_warped = pyramid_warp(p2w[1], fwd_scaled, self.warp) + aligned = [torch.cat([fw, bw, bf, ff], dim=1) + for fw, bw, bf, ff in zip(fwd_warped, bwd_warped, bwd_scaled, fwd_scaled)] + del fwd_warped, bwd_warped, bwd_scaled, fwd_scaled + results.append(self.fuse(aligned)) + del aligned + return torch.cat(results, dim=0) diff --git a/comfy_extras/frame_interpolation_models/ifnet.py b/comfy_extras/frame_interpolation_models/ifnet.py new file mode 100644 index 000000000..ad6edbec9 --- /dev/null +++ b/comfy_extras/frame_interpolation_models/ifnet.py @@ -0,0 +1,131 @@ +import torch +import torch.nn as nn +import torch.nn.functional as F + +import comfy.ops + +ops = comfy.ops.disable_weight_init + + +def _warp(img, flow, warp_grids): + B, _, H, W = img.shape + base_grid, flow_div = warp_grids[(H, W)] + flow_norm = torch.cat([flow[:, 0:1] / flow_div[0], flow[:, 1:2] / flow_div[1]], 1).float() + grid = (base_grid.expand(B, -1, -1, -1) + flow_norm).permute(0, 2, 3, 1) + return F.grid_sample(img.float(), grid, mode="bilinear", padding_mode="border", align_corners=True).to(img.dtype) + + +class Head(nn.Module): + def __init__(self, out_ch=4, device=None, dtype=None, operations=ops): + super().__init__() + self.cnn0 = operations.Conv2d(3, 16, 3, 2, 1, device=device, dtype=dtype) + self.cnn1 = operations.Conv2d(16, 16, 3, 1, 1, device=device, dtype=dtype) + self.cnn2 = operations.Conv2d(16, 16, 3, 1, 1, device=device, dtype=dtype) + self.cnn3 = operations.ConvTranspose2d(16, out_ch, 4, 2, 1, device=device, dtype=dtype) + self.relu = nn.LeakyReLU(0.2, True) + + def forward(self, x): + x = self.relu(self.cnn0(x)) + x = self.relu(self.cnn1(x)) + x = self.relu(self.cnn2(x)) + return self.cnn3(x) + + +class ResConv(nn.Module): + def __init__(self, c, device=None, dtype=None, operations=ops): + super().__init__() + self.conv = operations.Conv2d(c, c, 3, 1, 1, device=device, dtype=dtype) + self.beta = nn.Parameter(torch.ones((1, c, 1, 1), device=device, dtype=dtype)) + self.relu = nn.LeakyReLU(0.2, True) + + def forward(self, x): + return self.relu(torch.addcmul(x, self.conv(x), self.beta)) + + +class IFBlock(nn.Module): + def __init__(self, in_planes, c=64, device=None, dtype=None, operations=ops): + super().__init__() + self.conv0 = nn.Sequential( + nn.Sequential(operations.Conv2d(in_planes, c // 2, 3, 2, 1, device=device, dtype=dtype), nn.LeakyReLU(0.2, True)), + nn.Sequential(operations.Conv2d(c // 2, c, 3, 2, 1, device=device, dtype=dtype), nn.LeakyReLU(0.2, True))) + self.convblock = nn.Sequential(*(ResConv(c, device=device, dtype=dtype, operations=operations) for _ in range(8))) + self.lastconv = nn.Sequential(operations.ConvTranspose2d(c, 4 * 13, 4, 2, 1, device=device, dtype=dtype), nn.PixelShuffle(2)) + + def forward(self, x, flow=None, scale=1): + x = F.interpolate(x, scale_factor=1.0 / scale, mode="bilinear") + if flow is not None: + flow = F.interpolate(flow, scale_factor=1.0 / scale, mode="bilinear").div_(scale) + x = torch.cat((x, flow), 1) + feat = self.convblock(self.conv0(x)) + tmp = F.interpolate(self.lastconv(feat), scale_factor=scale, mode="bilinear") + return tmp[:, :4] * scale, tmp[:, 4:5], tmp[:, 5:] + + +class IFNet(nn.Module): + def __init__(self, head_ch=4, channels=(192, 128, 96, 64, 32), device=None, dtype=None, operations=ops): + super().__init__() + self.encode = Head(out_ch=head_ch, device=device, dtype=dtype, operations=operations) + block_in = [7 + 2 * head_ch] + [8 + 4 + 8 + 2 * head_ch] * 4 + self.blocks = nn.ModuleList([IFBlock(block_in[i], channels[i], device=device, dtype=dtype, operations=operations) for i in range(5)]) + self.scale_list = [16, 8, 4, 2, 1] + self.pad_align = 64 + self._warp_grids = {} + + def get_dtype(self): + return self.encode.cnn0.weight.dtype + + def memory_used_forward(self, shape, dtype): + return 300 * shape[1] * shape[2] * dtype.itemsize + + def _build_warp_grids(self, H, W, device): + if (H, W) in self._warp_grids: + return + self._warp_grids = {} # clear old resolution grids to prevent memory leaks + grid_y, grid_x = torch.meshgrid( + torch.linspace(-1.0, 1.0, H, device=device, dtype=torch.float32), + torch.linspace(-1.0, 1.0, W, device=device, dtype=torch.float32), indexing="ij") + self._warp_grids[(H, W)] = ( + torch.stack((grid_x, grid_y), dim=0).unsqueeze(0), + torch.tensor([(W - 1.0) / 2.0, (H - 1.0) / 2.0], dtype=torch.float32, device=device)) + + def warp(self, img, flow): + return _warp(img, flow, self._warp_grids) + + def extract_features(self, img): + """Extract head features for a single frame. Can be cached across pairs.""" + return self.encode(img) + + def forward(self, img0, img1, timestep=0.5, cache=None): + if not isinstance(timestep, torch.Tensor): + timestep = torch.full((img0.shape[0], 1, img0.shape[2], img0.shape[3]), timestep, device=img0.device, dtype=img0.dtype) + + self._build_warp_grids(img0.shape[2], img0.shape[3], img0.device) + + B = img0.shape[0] + f0 = cache["img0"].expand(B, -1, -1, -1) if cache and "img0" in cache else self.encode(img0) + f1 = cache["img1"].expand(B, -1, -1, -1) if cache and "img1" in cache else self.encode(img1) + flow = mask = feat = None + warped_img0, warped_img1 = img0, img1 + for i, block in enumerate(self.blocks): + if flow is None: + flow, mask, feat = block(torch.cat((img0, img1, f0, f1, timestep), 1), None, scale=self.scale_list[i]) + else: + fd, mask, feat = block( + torch.cat((warped_img0, warped_img1, self.warp(f0, flow[:, :2]), self.warp(f1, flow[:, 2:4]), timestep, mask, feat), 1), + flow, scale=self.scale_list[i]) + flow = flow.add_(fd) + warped_img0 = self.warp(img0, flow[:, :2]) + warped_img1 = self.warp(img1, flow[:, 2:4]) + return torch.lerp(warped_img1, warped_img0, torch.sigmoid(mask)) + + +def detect_rife_config(state_dict): + head_ch = state_dict["encode.cnn3.weight"].shape[1] # ConvTranspose2d: (in_ch, out_ch, kH, kW) + channels = [] + for i in range(5): + key = f"blocks.{i}.conv0.1.0.weight" + if key in state_dict: + channels.append(state_dict[key].shape[0]) + if len(channels) != 5: + raise ValueError(f"Unsupported RIFE model: expected 5 blocks, found {len(channels)}") + return head_ch, channels diff --git a/comfy_extras/nodes_ace.py b/comfy_extras/nodes_ace.py index cbfaf913d..affcf3b71 100644 --- a/comfy_extras/nodes_ace.py +++ b/comfy_extras/nodes_ace.py @@ -3,136 +3,136 @@ from typing_extensions import override import comfy.model_management import node_helpers -from comfy_api.latest import ComfyExtension, io +from comfy_api.latest import ComfyExtension, IO -class TextEncodeAceStepAudio(io.ComfyNode): +class TextEncodeAceStepAudio(IO.ComfyNode): @classmethod def define_schema(cls): - return io.Schema( + return IO.Schema( node_id="TextEncodeAceStepAudio", category="conditioning", inputs=[ - io.Clip.Input("clip"), - io.String.Input("tags", multiline=True, dynamic_prompts=True), - io.String.Input("lyrics", multiline=True, dynamic_prompts=True), - io.Float.Input("lyrics_strength", default=1.0, min=0.0, max=10.0, step=0.01), + IO.Clip.Input("clip"), + IO.String.Input("tags", multiline=True, dynamic_prompts=True), + IO.String.Input("lyrics", multiline=True, dynamic_prompts=True), + IO.Float.Input("lyrics_strength", default=1.0, min=0.0, max=10.0, step=0.01), ], - outputs=[io.Conditioning.Output()], + outputs=[IO.Conditioning.Output()], ) @classmethod - def execute(cls, clip, tags, lyrics, lyrics_strength) -> io.NodeOutput: + def execute(cls, clip, tags, lyrics, lyrics_strength) -> IO.NodeOutput: tokens = clip.tokenize(tags, lyrics=lyrics) conditioning = clip.encode_from_tokens_scheduled(tokens) conditioning = node_helpers.conditioning_set_values(conditioning, {"lyrics_strength": lyrics_strength}) - return io.NodeOutput(conditioning) + return IO.NodeOutput(conditioning) -class TextEncodeAceStepAudio15(io.ComfyNode): +class TextEncodeAceStepAudio15(IO.ComfyNode): @classmethod def define_schema(cls): - return io.Schema( + return IO.Schema( node_id="TextEncodeAceStepAudio1.5", category="conditioning", inputs=[ - io.Clip.Input("clip"), - io.String.Input("tags", multiline=True, dynamic_prompts=True), - io.String.Input("lyrics", multiline=True, dynamic_prompts=True), - io.Int.Input("seed", default=0, min=0, max=0xffffffffffffffff, control_after_generate=True), - io.Int.Input("bpm", default=120, min=10, max=300), - io.Float.Input("duration", default=120.0, min=0.0, max=2000.0, step=0.1), - io.Combo.Input("timesignature", options=['2', '3', '4', '6']), - io.Combo.Input("language", options=["en", "ja", "zh", "es", "de", "fr", "pt", "ru", "it", "nl", "pl", "tr", "vi", "cs", "fa", "id", "ko", "uk", "hu", "ar", "sv", "ro", "el"]), - io.Combo.Input("keyscale", options=[f"{root} {quality}" for quality in ["major", "minor"] for root in ["C", "C#", "Db", "D", "D#", "Eb", "E", "F", "F#", "Gb", "G", "G#", "Ab", "A", "A#", "Bb", "B"]]), - io.Boolean.Input("generate_audio_codes", default=True, tooltip="Enable the LLM that generates audio codes. This can be slow but will increase the quality of the generated audio. Turn this off if you are giving the model an audio reference.", advanced=True), - io.Float.Input("cfg_scale", default=2.0, min=0.0, max=100.0, step=0.1, advanced=True), - io.Float.Input("temperature", default=0.85, min=0.0, max=2.0, step=0.01, advanced=True), - io.Float.Input("top_p", default=0.9, min=0.0, max=2000.0, step=0.01, advanced=True), - io.Int.Input("top_k", default=0, min=0, max=100, advanced=True), - io.Float.Input("min_p", default=0.000, min=0.0, max=1.0, step=0.001, advanced=True), + IO.Clip.Input("clip"), + IO.String.Input("tags", multiline=True, dynamic_prompts=True), + IO.String.Input("lyrics", multiline=True, dynamic_prompts=True), + IO.Int.Input("seed", default=0, min=0, max=0xffffffffffffffff, control_after_generate=True), + IO.Int.Input("bpm", default=120, min=10, max=300), + IO.Float.Input("duration", default=120.0, min=0.0, max=2000.0, step=0.1), + IO.Combo.Input("timesignature", options=['2', '3', '4', '6']), + IO.Combo.Input("language", options=['ar', 'az', 'bg', 'bn', 'ca', 'cs', 'da', 'de', 'el', 'en', 'es', 'fa', 'fi', 'fr', 'he', 'hi', 'hr', 'ht', 'hu', 'id', 'is', 'it', 'ja', 'ko', 'la', 'lt', 'ms', 'ne', 'nl', 'no', 'pa', 'pl', 'pt', 'ro', 'ru', 'sa', 'sk', 'sr', 'sv', 'sw', 'ta', 'te', 'th', 'tl', 'tr', 'uk', 'ur', 'vi', 'yue', 'zh', 'unknown'], default='en'), + IO.Combo.Input("keyscale", options=[f"{root} {quality}" for quality in ["major", "minor"] for root in ["C", "C#", "Db", "D", "D#", "Eb", "E", "F", "F#", "Gb", "G", "G#", "Ab", "A", "A#", "Bb", "B"]]), + IO.Boolean.Input("generate_audio_codes", default=True, tooltip="Enable the LLM that generates audio codes. This can be slow but will increase the quality of the generated audio. Turn this off if you are giving the model an audio reference.", advanced=True), + IO.Float.Input("cfg_scale", default=2.0, min=0.0, max=100.0, step=0.1, advanced=True), + IO.Float.Input("temperature", default=0.85, min=0.0, max=2.0, step=0.01, advanced=True), + IO.Float.Input("top_p", default=0.9, min=0.0, max=2000.0, step=0.01, advanced=True), + IO.Int.Input("top_k", default=0, min=0, max=100, advanced=True), + IO.Float.Input("min_p", default=0.000, min=0.0, max=1.0, step=0.001, advanced=True), ], - outputs=[io.Conditioning.Output()], + outputs=[IO.Conditioning.Output()], ) @classmethod - def execute(cls, clip, tags, lyrics, seed, bpm, duration, timesignature, language, keyscale, generate_audio_codes, cfg_scale, temperature, top_p, top_k, min_p) -> io.NodeOutput: + def execute(cls, clip, tags, lyrics, seed, bpm, duration, timesignature, language, keyscale, generate_audio_codes, cfg_scale, temperature, top_p, top_k, min_p) -> IO.NodeOutput: tokens = clip.tokenize(tags, lyrics=lyrics, bpm=bpm, duration=duration, timesignature=int(timesignature), language=language, keyscale=keyscale, seed=seed, generate_audio_codes=generate_audio_codes, cfg_scale=cfg_scale, temperature=temperature, top_p=top_p, top_k=top_k, min_p=min_p) conditioning = clip.encode_from_tokens_scheduled(tokens) - return io.NodeOutput(conditioning) + return IO.NodeOutput(conditioning) -class EmptyAceStepLatentAudio(io.ComfyNode): +class EmptyAceStepLatentAudio(IO.ComfyNode): @classmethod def define_schema(cls): - return io.Schema( + return IO.Schema( node_id="EmptyAceStepLatentAudio", display_name="Empty Ace Step 1.0 Latent Audio", category="latent/audio", inputs=[ - io.Float.Input("seconds", default=120.0, min=1.0, max=1000.0, step=0.1), - io.Int.Input( + IO.Float.Input("seconds", default=120.0, min=1.0, max=1000.0, step=0.1), + IO.Int.Input( "batch_size", default=1, min=1, max=4096, tooltip="The number of latent images in the batch." ), ], - outputs=[io.Latent.Output()], + outputs=[IO.Latent.Output()], ) @classmethod - def execute(cls, seconds, batch_size) -> io.NodeOutput: + def execute(cls, seconds, batch_size) -> IO.NodeOutput: length = int(seconds * 44100 / 512 / 8) latent = torch.zeros([batch_size, 8, 16, length], device=comfy.model_management.intermediate_device(), dtype=comfy.model_management.intermediate_dtype()) - return io.NodeOutput({"samples": latent, "type": "audio"}) + return IO.NodeOutput({"samples": latent, "type": "audio"}) -class EmptyAceStep15LatentAudio(io.ComfyNode): +class EmptyAceStep15LatentAudio(IO.ComfyNode): @classmethod def define_schema(cls): - return io.Schema( + return IO.Schema( node_id="EmptyAceStep1.5LatentAudio", display_name="Empty Ace Step 1.5 Latent Audio", category="latent/audio", inputs=[ - io.Float.Input("seconds", default=120.0, min=1.0, max=1000.0, step=0.01), - io.Int.Input( + IO.Float.Input("seconds", default=120.0, min=1.0, max=1000.0, step=0.01), + IO.Int.Input( "batch_size", default=1, min=1, max=4096, tooltip="The number of latent images in the batch." ), ], - outputs=[io.Latent.Output()], + outputs=[IO.Latent.Output()], ) @classmethod - def execute(cls, seconds, batch_size) -> io.NodeOutput: + def execute(cls, seconds, batch_size) -> IO.NodeOutput: length = round((seconds * 48000 / 1920)) latent = torch.zeros([batch_size, 64, length], device=comfy.model_management.intermediate_device(), dtype=comfy.model_management.intermediate_dtype()) - return io.NodeOutput({"samples": latent, "type": "audio"}) + return IO.NodeOutput({"samples": latent, "type": "audio"}) -class ReferenceAudio(io.ComfyNode): +class ReferenceAudio(IO.ComfyNode): @classmethod def define_schema(cls): - return io.Schema( + return IO.Schema( node_id="ReferenceTimbreAudio", display_name="Reference Audio", category="advanced/conditioning/audio", is_experimental=True, description="This node sets the reference audio for ace step 1.5", inputs=[ - io.Conditioning.Input("conditioning"), - io.Latent.Input("latent", optional=True), + IO.Conditioning.Input("conditioning"), + IO.Latent.Input("latent", optional=True), ], outputs=[ - io.Conditioning.Output(), + IO.Conditioning.Output(), ] ) @classmethod - def execute(cls, conditioning, latent=None) -> io.NodeOutput: + def execute(cls, conditioning, latent=None) -> IO.NodeOutput: if latent is not None: conditioning = node_helpers.conditioning_set_values(conditioning, {"reference_audio_timbre_latents": [latent["samples"]]}, append=True) - return io.NodeOutput(conditioning) + return IO.NodeOutput(conditioning) class AceExtension(ComfyExtension): @override - async def get_node_list(self) -> list[type[io.ComfyNode]]: + async def get_node_list(self) -> list[type[IO.ComfyNode]]: return [ TextEncodeAceStepAudio, EmptyAceStepLatentAudio, diff --git a/comfy_extras/nodes_advanced_samplers.py b/comfy_extras/nodes_advanced_samplers.py index 7f716cd76..567c37be0 100644 --- a/comfy_extras/nodes_advanced_samplers.py +++ b/comfy_extras/nodes_advanced_samplers.py @@ -86,13 +86,44 @@ def sample_euler_pp(model, x, sigmas, extra_args=None, callback=None, disable=No return x +class SamplerLCM(io.ComfyNode): + @classmethod + def define_schema(cls) -> io.Schema: + return io.Schema( + node_id="SamplerLCM", + category="sampling/samplers", + description=("LCM sampler with tunable per-step noise. s_noise is a multiplier on the model's training noise scale"), + inputs=[ + io.Float.Input("s_noise", default=1.0, min=0.0, max=64.0, step=0.01, + tooltip="Per-step noise multiplier at the first step (1.0 = match training)."), + io.Float.Input("s_noise_end", default=1.0, min=0.0, max=64.0, step=0.01, + tooltip="Per-step noise multiplier at the last step. Set equal to s_noise for a constant schedule."), + io.Float.Input("noise_clip_std", default=0.0, min=0.0, max=10.0, step=0.01, + tooltip="Clamp per-step noise to +/- N*std. 0 disables."), + ], + outputs=[io.Sampler.Output()], + ) + + @classmethod + def execute(cls, s_noise, s_noise_end, noise_clip_std) -> io.NodeOutput: + sampler = comfy.samplers.ksampler( + "lcm", + { + "s_noise": float(s_noise), + "s_noise_end": float(s_noise_end), + "noise_clip_std": float(noise_clip_std), + }, + ) + return io.NodeOutput(sampler) + + class SamplerEulerCFGpp(io.ComfyNode): @classmethod def define_schema(cls) -> io.Schema: return io.Schema( node_id="SamplerEulerCFGpp", display_name="SamplerEulerCFG++", - category="_for_testing", # "sampling/custom_sampling/samplers" + category="experimental", # "sampling/custom_sampling/samplers" inputs=[ io.Combo.Input("version", options=["regular", "alternative"], advanced=True), ], @@ -114,6 +145,7 @@ class AdvancedSamplersExtension(ComfyExtension): async def get_node_list(self) -> list[type[io.ComfyNode]]: return [ SamplerLCMUpscale, + SamplerLCM, SamplerEulerCFGpp, ] diff --git a/comfy_extras/nodes_ar_video.py b/comfy_extras/nodes_ar_video.py new file mode 100644 index 000000000..b36588b14 --- /dev/null +++ b/comfy_extras/nodes_ar_video.py @@ -0,0 +1,136 @@ +""" +ComfyUI nodes for autoregressive video generation (Causal Forcing, Self-Forcing, etc.). + - EmptyARVideoLatent: create 5D [B, C, T, H, W] video latent tensors + - SamplerARVideo: SAMPLER for the block-by-block autoregressive denoising loop + - ARVideoI2V: image-to-video conditioning for AR models (seeds KV cache with start image) +""" + +import torch +from typing_extensions import override + +import comfy.model_management +import comfy.samplers +import comfy.utils +from comfy_api.latest import ComfyExtension, io + + +class EmptyARVideoLatent(io.ComfyNode): + @classmethod + def define_schema(cls): + return io.Schema( + node_id="EmptyARVideoLatent", + category="latent/video", + inputs=[ + io.Int.Input("width", default=832, min=16, max=8192, step=16), + io.Int.Input("height", default=480, min=16, max=8192, step=16), + io.Int.Input("length", default=81, min=1, max=1024, step=4), + io.Int.Input("batch_size", default=1, min=1, max=64), + ], + outputs=[ + io.Latent.Output(display_name="LATENT"), + ], + ) + + @classmethod + def execute(cls, width, height, length, batch_size) -> io.NodeOutput: + lat_t = ((length - 1) // 4) + 1 + latent = torch.zeros( + [batch_size, 16, lat_t, height // 8, width // 8], + device=comfy.model_management.intermediate_device(), + ) + return io.NodeOutput({"samples": latent}) + + +class SamplerARVideo(io.ComfyNode): + """Sampler for autoregressive video models (Causal Forcing, Self-Forcing). + + All AR-loop parameters are owned by this node so they live in the workflow. + Add new widgets here as the AR sampler grows new options. + """ + + @classmethod + def define_schema(cls): + return io.Schema( + node_id="SamplerARVideo", + display_name="Sampler AR Video", + category="sampling/custom_sampling/samplers", + inputs=[ + io.Int.Input( + "num_frame_per_block", + default=1, min=1, max=64, + tooltip="Frames per autoregressive block. 1 = framewise, " + "3 = chunkwise. Must match the checkpoint's training mode.", + ), + ], + outputs=[io.Sampler.Output()], + ) + + @classmethod + def execute(cls, num_frame_per_block) -> io.NodeOutput: + extra_options = { + "num_frame_per_block": num_frame_per_block, + } + return io.NodeOutput(comfy.samplers.ksampler("ar_video", extra_options)) + + +class ARVideoI2V(io.ComfyNode): + """Image-to-video setup for AR video models (Causal Forcing, Self-Forcing). + + VAE-encodes the start image and stores it in the model's transformer_options + so that sample_ar_video can seed the KV cache before denoising. + Uses the same T2V model checkpoint -- no separate I2V architecture needed. + """ + + @classmethod + def define_schema(cls): + return io.Schema( + node_id="ARVideoI2V", + category="conditioning/video_models", + inputs=[ + io.Model.Input("model"), + io.Vae.Input("vae"), + io.Image.Input("start_image"), + io.Int.Input("width", default=832, min=16, max=8192, step=16), + io.Int.Input("height", default=480, min=16, max=8192, step=16), + io.Int.Input("length", default=81, min=1, max=1024, step=4), + io.Int.Input("batch_size", default=1, min=1, max=64), + ], + outputs=[ + io.Model.Output(display_name="MODEL"), + io.Latent.Output(display_name="LATENT"), + ], + ) + + @classmethod + def execute(cls, model, vae, start_image, width, height, length, batch_size) -> io.NodeOutput: + start_image = comfy.utils.common_upscale( + start_image[:1].movedim(-1, 1), width, height, "bilinear", "center" + ).movedim(1, -1) + + initial_latent = vae.encode(start_image[:, :, :, :3]) + + m = model.clone() + to = m.model_options.setdefault("transformer_options", {}) + ar_cfg = to.setdefault("ar_config", {}) + ar_cfg["initial_latent"] = initial_latent + + lat_t = ((length - 1) // 4) + 1 + latent = torch.zeros( + [batch_size, 16, lat_t, height // 8, width // 8], + device=comfy.model_management.intermediate_device(), + ) + return io.NodeOutput(m, {"samples": latent}) + + +class ARVideoExtension(ComfyExtension): + @override + async def get_node_list(self) -> list[type[io.ComfyNode]]: + return [ + EmptyARVideoLatent, + SamplerARVideo, + ARVideoI2V, + ] + + +async def comfy_entrypoint() -> ARVideoExtension: + return ARVideoExtension() diff --git a/comfy_extras/nodes_attention_multiply.py b/comfy_extras/nodes_attention_multiply.py index 060a5c9be..f4ee6a689 100644 --- a/comfy_extras/nodes_attention_multiply.py +++ b/comfy_extras/nodes_attention_multiply.py @@ -25,7 +25,7 @@ class UNetSelfAttentionMultiply(io.ComfyNode): def define_schema(cls) -> io.Schema: return io.Schema( node_id="UNetSelfAttentionMultiply", - category="_for_testing/attention_experiments", + category="experimental/attention_experiments", inputs=[ io.Model.Input("model"), io.Float.Input("q", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True), @@ -48,7 +48,7 @@ class UNetCrossAttentionMultiply(io.ComfyNode): def define_schema(cls) -> io.Schema: return io.Schema( node_id="UNetCrossAttentionMultiply", - category="_for_testing/attention_experiments", + category="experimental/attention_experiments", inputs=[ io.Model.Input("model"), io.Float.Input("q", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True), @@ -72,7 +72,7 @@ class CLIPAttentionMultiply(io.ComfyNode): return io.Schema( node_id="CLIPAttentionMultiply", search_aliases=["clip attention scale", "text encoder attention"], - category="_for_testing/attention_experiments", + category="experimental/attention_experiments", inputs=[ io.Clip.Input("clip"), io.Float.Input("q", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True), @@ -106,7 +106,7 @@ class UNetTemporalAttentionMultiply(io.ComfyNode): def define_schema(cls) -> io.Schema: return io.Schema( node_id="UNetTemporalAttentionMultiply", - category="_for_testing/attention_experiments", + category="experimental/attention_experiments", inputs=[ io.Model.Input("model"), io.Float.Input("self_structural", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True), diff --git a/comfy_extras/nodes_audio.py b/comfy_extras/nodes_audio.py index a395392d8..6382dd618 100644 --- a/comfy_extras/nodes_audio.py +++ b/comfy_extras/nodes_audio.py @@ -104,7 +104,7 @@ def vae_decode_audio(vae, samples, tile=None, overlap=None): std = torch.std(audio, dim=[1, 2], keepdim=True) * 5.0 std[std < 1.0] = 1.0 audio /= std - vae_sample_rate = getattr(vae, "audio_sample_rate", 44100) + vae_sample_rate = getattr(vae, "audio_sample_rate_output", getattr(vae, "audio_sample_rate", 44100)) return {"waveform": audio, "sample_rate": vae_sample_rate if "sample_rate" not in samples else samples["sample_rate"]} @@ -297,6 +297,7 @@ class LoadAudio(IO.ComfyNode): @classmethod def define_schema(cls): input_dir = folder_paths.get_input_directory() + os.makedirs(input_dir, exist_ok=True) files = folder_paths.filter_files_content_types(os.listdir(input_dir), ["audio", "video"]) return IO.Schema( node_id="LoadAudio", diff --git a/comfy_extras/nodes_audio_encoder.py b/comfy_extras/nodes_audio_encoder.py index 13aacd41a..6a85da89b 100644 --- a/comfy_extras/nodes_audio_encoder.py +++ b/comfy_extras/nodes_audio_encoder.py @@ -10,6 +10,7 @@ class AudioEncoderLoader(io.ComfyNode): def define_schema(cls) -> io.Schema: return io.Schema( node_id="AudioEncoderLoader", + display_name="Load Audio Encoder", category="loaders", inputs=[ io.Combo.Input( diff --git a/comfy_extras/nodes_bg_removal.py b/comfy_extras/nodes_bg_removal.py new file mode 100644 index 000000000..8d046b8d4 --- /dev/null +++ b/comfy_extras/nodes_bg_removal.py @@ -0,0 +1,60 @@ +import folder_paths +from typing_extensions import override +from comfy_api.latest import ComfyExtension, IO +from comfy.bg_removal_model import load + + +class LoadBackgroundRemovalModel(IO.ComfyNode): + @classmethod + def define_schema(cls): + files = folder_paths.get_filename_list("background_removal") + return IO.Schema( + node_id="LoadBackgroundRemovalModel", + display_name="Load Background Removal Model", + category="loaders", + inputs=[ + IO.Combo.Input("bg_removal_name", options=sorted(files), tooltip="The model used to remove backgrounds from images"), + ], + outputs=[ + IO.BackgroundRemoval.Output("bg_model") + ] + ) + @classmethod + def execute(cls, bg_removal_name): + path = folder_paths.get_full_path_or_raise("background_removal", bg_removal_name) + bg = load(path) + if bg is None: + raise RuntimeError("ERROR: background model file is invalid and does not contain a valid background removal model.") + return IO.NodeOutput(bg) + +class RemoveBackground(IO.ComfyNode): + @classmethod + def define_schema(cls): + return IO.Schema( + node_id="RemoveBackground", + display_name="Remove Background", + category="image/background removal", + inputs=[ + IO.Image.Input("image", tooltip="Input image to remove the background from"), + IO.BackgroundRemoval.Input("bg_removal_model", tooltip="Background removal model used to generate the mask") + ], + outputs=[ + IO.Mask.Output("mask", tooltip="Generated foreground mask") + ] + ) + @classmethod + def execute(cls, image, bg_removal_model): + mask = bg_removal_model.encode_image(image) + return IO.NodeOutput(mask) + +class BackgroundRemovalExtension(ComfyExtension): + @override + async def get_node_list(self) -> list[type[IO.ComfyNode]]: + return [ + LoadBackgroundRemovalModel, + RemoveBackground + ] + + +async def comfy_entrypoint() -> BackgroundRemovalExtension: + return BackgroundRemovalExtension() diff --git a/comfy_extras/nodes_camera_trajectory.py b/comfy_extras/nodes_camera_trajectory.py index e7efa29ba..34b78e81b 100644 --- a/comfy_extras/nodes_camera_trajectory.py +++ b/comfy_extras/nodes_camera_trajectory.py @@ -153,7 +153,7 @@ class WanCameraEmbedding(io.ComfyNode): def define_schema(cls): return io.Schema( node_id="WanCameraEmbedding", - category="camera", + category="conditioning/video_models", inputs=[ io.Combo.Input( "camera_pose", diff --git a/comfy_extras/nodes_compositing.py b/comfy_extras/nodes_compositing.py index 3bc9fccb3..720efc629 100644 --- a/comfy_extras/nodes_compositing.py +++ b/comfy_extras/nodes_compositing.py @@ -202,14 +202,11 @@ class JoinImageWithAlpha(io.ComfyNode): @classmethod def execute(cls, image: torch.Tensor, alpha: torch.Tensor) -> io.NodeOutput: - batch_size = min(len(image), len(alpha)) - out_images = [] - - alpha = 1.0 - resize_mask(alpha, image.shape[1:]) - for i in range(batch_size): - out_images.append(torch.cat((image[i][:,:,:3], alpha[i].unsqueeze(2)), dim=2)) - - return io.NodeOutput(torch.stack(out_images)) + batch_size = max(len(image), len(alpha)) + alpha = 1.0 - resize_mask(alpha.to(image), image.shape[1:]) + alpha = comfy.utils.repeat_to_batch_size(alpha, batch_size) + image = comfy.utils.repeat_to_batch_size(image, batch_size) + return io.NodeOutput(torch.cat((image[..., :3], alpha.unsqueeze(-1)), dim=-1)) class CompositingExtension(ComfyExtension): diff --git a/comfy_extras/nodes_cond.py b/comfy_extras/nodes_cond.py index 86426a780..b745a43af 100644 --- a/comfy_extras/nodes_cond.py +++ b/comfy_extras/nodes_cond.py @@ -8,7 +8,7 @@ class CLIPTextEncodeControlnet(io.ComfyNode): def define_schema(cls) -> io.Schema: return io.Schema( node_id="CLIPTextEncodeControlnet", - category="_for_testing/conditioning", + category="experimental/conditioning", inputs=[ io.Clip.Input("clip"), io.Conditioning.Input("conditioning"), @@ -35,7 +35,7 @@ class T5TokenizerOptions(io.ComfyNode): def define_schema(cls) -> io.Schema: return io.Schema( node_id="T5TokenizerOptions", - category="_for_testing/conditioning", + category="experimental/conditioning", inputs=[ io.Clip.Input("clip"), io.Int.Input("min_padding", default=0, min=0, max=10000, step=1, advanced=True), diff --git a/comfy_extras/nodes_context_windows.py b/comfy_extras/nodes_context_windows.py index 0e43f2e44..f7ca833dc 100644 --- a/comfy_extras/nodes_context_windows.py +++ b/comfy_extras/nodes_context_windows.py @@ -10,7 +10,7 @@ class ContextWindowsManualNode(io.ComfyNode): return io.Schema( node_id="ContextWindowsManual", display_name="Context Windows (Manual)", - category="context", + category="model_patches", description="Manually set context windows.", inputs=[ io.Model.Input("model", tooltip="The model to apply context windows to during sampling."), @@ -29,6 +29,7 @@ class ContextWindowsManualNode(io.ComfyNode): io.Boolean.Input("freenoise", default=False, tooltip="Whether to apply FreeNoise noise shuffling, improves window blending."), io.String.Input("cond_retain_index_list", default="", tooltip="List of latent indices to retain in the conditioning tensors for each window, for example setting this to '0' will use the initial start image for each window."), io.Boolean.Input("split_conds_to_windows", default=False, tooltip="Whether to split multiple conditionings (created by ConditionCombine) to each window based on region index."), + io.Boolean.Input("causal_window_fix", default=True, tooltip="Whether to add a causal fix frame to non-0-indexed context windows."), ], outputs=[ io.Model.Output(tooltip="The model with context windows applied during sampling."), @@ -38,7 +39,7 @@ class ContextWindowsManualNode(io.ComfyNode): @classmethod def execute(cls, model: io.Model.Type, context_length: int, context_overlap: int, context_schedule: str, context_stride: int, closed_loop: bool, fuse_method: str, dim: int, freenoise: bool, - cond_retain_index_list: list[int]=[], split_conds_to_windows: bool=False) -> io.Model: + cond_retain_index_list: list[int]=[], split_conds_to_windows: bool=False, causal_window_fix: bool=True) -> io.Model: model = model.clone() model.model_options["context_handler"] = comfy.context_windows.IndexListContextHandler( context_schedule=comfy.context_windows.get_matching_context_schedule(context_schedule), @@ -50,7 +51,8 @@ class ContextWindowsManualNode(io.ComfyNode): dim=dim, freenoise=freenoise, cond_retain_index_list=cond_retain_index_list, - split_conds_to_windows=split_conds_to_windows + split_conds_to_windows=split_conds_to_windows, + causal_window_fix=causal_window_fix, ) # make memory usage calculation only take into account the context window latents comfy.context_windows.create_prepare_sampling_wrapper(model) diff --git a/comfy_extras/nodes_custom_sampler.py b/comfy_extras/nodes_custom_sampler.py index 1e957c09b..c67145d2d 100644 --- a/comfy_extras/nodes_custom_sampler.py +++ b/comfy_extras/nodes_custom_sampler.py @@ -984,7 +984,7 @@ class AddNoise(io.ComfyNode): def define_schema(cls): return io.Schema( node_id="AddNoise", - category="_for_testing/custom_sampling/noise", + category="experimental/custom_sampling/noise", is_experimental=True, inputs=[ io.Model.Input("model"), @@ -1034,7 +1034,7 @@ class ManualSigmas(io.ComfyNode): return io.Schema( node_id="ManualSigmas", search_aliases=["custom noise schedule", "define sigmas"], - category="_for_testing/custom_sampling", + category="experimental/custom_sampling", is_experimental=True, inputs=[ io.String.Input("sigmas", default="1, 0.5", multiline=False) diff --git a/comfy_extras/nodes_differential_diffusion.py b/comfy_extras/nodes_differential_diffusion.py index 34ffb9a89..4fa61ad0e 100644 --- a/comfy_extras/nodes_differential_diffusion.py +++ b/comfy_extras/nodes_differential_diffusion.py @@ -13,7 +13,7 @@ class DifferentialDiffusion(io.ComfyNode): node_id="DifferentialDiffusion", search_aliases=["inpaint gradient", "variable denoise strength"], display_name="Differential Diffusion", - category="_for_testing", + category="experimental", inputs=[ io.Model.Input("model"), io.Float.Input( diff --git a/comfy_extras/nodes_flux.py b/comfy_extras/nodes_flux.py index 3a23c7d04..5e04a5f77 100644 --- a/comfy_extras/nodes_flux.py +++ b/comfy_extras/nodes_flux.py @@ -102,7 +102,7 @@ class FluxDisableGuidance(io.ComfyNode): append = execute # TODO: remove -PREFERED_KONTEXT_RESOLUTIONS = [ +PREFERRED_KONTEXT_RESOLUTIONS = [ (672, 1568), (688, 1504), (720, 1456), @@ -143,7 +143,7 @@ class FluxKontextImageScale(io.ComfyNode): width = image.shape[2] height = image.shape[1] aspect_ratio = width / height - _, width, height = min((abs(aspect_ratio - w / h), w, h) for w, h in PREFERED_KONTEXT_RESOLUTIONS) + _, width, height = min((abs(aspect_ratio - w / h), w, h) for w, h in PREFERRED_KONTEXT_RESOLUTIONS) image = comfy.utils.common_upscale(image.movedim(-1, 1), width, height, "lanczos", "center").movedim(1, -1) return io.NodeOutput(image) diff --git a/comfy_extras/nodes_frame_interpolation.py b/comfy_extras/nodes_frame_interpolation.py new file mode 100644 index 000000000..9dd34cfb8 --- /dev/null +++ b/comfy_extras/nodes_frame_interpolation.py @@ -0,0 +1,208 @@ +import torch +from tqdm import tqdm +from typing_extensions import override + +import comfy.model_patcher +import comfy.utils +import folder_paths +from comfy import model_management +from comfy_extras.frame_interpolation_models.ifnet import IFNet, detect_rife_config +from comfy_extras.frame_interpolation_models.film_net import FILMNet +from comfy_api.latest import ComfyExtension, io + +FrameInterpolationModel = io.Custom("INTERP_MODEL") + + +class FrameInterpolationModelLoader(io.ComfyNode): + @classmethod + def define_schema(cls): + return io.Schema( + node_id="FrameInterpolationModelLoader", + display_name="Load Frame Interpolation Model", + category="loaders", + inputs=[ + io.Combo.Input("model_name", options=folder_paths.get_filename_list("frame_interpolation"), + tooltip="Select a frame interpolation model to load. Models must be placed in the 'frame_interpolation' folder."), + ], + outputs=[ + FrameInterpolationModel.Output(), + ], + ) + + @classmethod + def execute(cls, model_name) -> io.NodeOutput: + model_path = folder_paths.get_full_path_or_raise("frame_interpolation", model_name) + sd = comfy.utils.load_torch_file(model_path, safe_load=True) + + model = cls._detect_and_load(sd) + dtype = torch.float16 if model_management.should_use_fp16(model_management.get_torch_device()) else torch.float32 + model.eval().to(dtype) + patcher = comfy.model_patcher.CoreModelPatcher( + model, + load_device=model_management.get_torch_device(), + offload_device=model_management.unet_offload_device(), + ) + return io.NodeOutput(patcher) + + @classmethod + def _detect_and_load(cls, sd): + # Try FILM + if "extract.extract_sublevels.convs.0.0.conv.weight" in sd: + model = FILMNet() + model.load_state_dict(sd) + return model + + # Try RIFE (needs key remapping for raw checkpoints) + sd = comfy.utils.state_dict_prefix_replace(sd, {"module.": "", "flownet.": ""}) + key_map = {} + for k in sd: + for i in range(5): + if k.startswith(f"block{i}."): + key_map[k] = f"blocks.{i}.{k[len(f'block{i}.'):]}" + if key_map: + sd = {key_map.get(k, k): v for k, v in sd.items()} + sd = {k: v for k, v in sd.items() if not k.startswith(("teacher.", "caltime."))} + + try: + head_ch, channels = detect_rife_config(sd) + except (KeyError, ValueError): + raise ValueError("Unrecognized frame interpolation model format") + model = IFNet(head_ch=head_ch, channels=channels) + model.load_state_dict(sd) + return model + + +class FrameInterpolate(io.ComfyNode): + @classmethod + def define_schema(cls): + return io.Schema( + node_id="FrameInterpolate", + display_name="Frame Interpolate", + category="video", + search_aliases=["rife", "film", "frame interpolation", "slow motion", "interpolate frames", "vfi"], + inputs=[ + FrameInterpolationModel.Input("interp_model"), + io.Image.Input("images"), + io.Int.Input("multiplier", default=2, min=2, max=16), + ], + outputs=[ + io.Image.Output(), + ], + ) + + @classmethod + def execute(cls, interp_model, images, multiplier) -> io.NodeOutput: + offload_device = model_management.intermediate_device() + + num_frames = images.shape[0] + if num_frames < 2 or multiplier < 2: + return io.NodeOutput(images) + + device = interp_model.load_device + dtype = interp_model.model_dtype() + inference_model = interp_model.model + activation_mem = inference_model.memory_used_forward(images.shape, dtype) + model_management.load_models_gpu([interp_model], memory_required=activation_mem) + align = getattr(inference_model, "pad_align", 1) + H, W = images.shape[1], images.shape[2] + + # Prepare a single padded frame on device for determining output dimensions + def prepare_frame(idx): + frame = images[idx:idx + 1].movedim(-1, 1).to(dtype=dtype, device=device) + if align > 1: + from comfy.ldm.common_dit import pad_to_patch_size + frame = pad_to_patch_size(frame, (align, align), padding_mode="reflect") + return frame + + # Count total interpolation passes for progress bar + total_pairs = num_frames - 1 + num_interp = multiplier - 1 + total_steps = total_pairs * num_interp + pbar = comfy.utils.ProgressBar(total_steps) + tqdm_bar = tqdm(total=total_steps, desc="Frame interpolation") + + batch = num_interp # reduced on OOM and persists across pairs (same resolution = same limit) + t_values = [t / multiplier for t in range(1, multiplier)] + + out_dtype = model_management.intermediate_dtype() + total_out_frames = total_pairs * multiplier + 1 + result = torch.empty((total_out_frames, 3, H, W), dtype=out_dtype, device=offload_device) + result[0] = images[0].movedim(-1, 0).to(out_dtype) + out_idx = 1 + + # Pre-compute timestep tensor on device (padded dimensions needed) + sample = prepare_frame(0) + pH, pW = sample.shape[2], sample.shape[3] + ts_full = torch.tensor(t_values, device=device, dtype=dtype).reshape(num_interp, 1, 1, 1) + ts_full = ts_full.expand(-1, 1, pH, pW) + del sample + + multi_fn = getattr(inference_model, "forward_multi_timestep", None) + feat_cache = {} + prev_frame = None + + try: + for i in range(total_pairs): + img0_single = prev_frame if prev_frame is not None else prepare_frame(i) + img1_single = prepare_frame(i + 1) + prev_frame = img1_single + + # Cache features: img1 of pair N becomes img0 of pair N+1 + feat_cache["img0"] = feat_cache.pop("next") if "next" in feat_cache else inference_model.extract_features(img0_single) + feat_cache["img1"] = inference_model.extract_features(img1_single) + feat_cache["next"] = feat_cache["img1"] + + used_multi = False + if multi_fn is not None: + # Models with timestep-independent flow can compute it once for all timesteps + try: + mids = multi_fn(img0_single, img1_single, t_values, cache=feat_cache) + result[out_idx:out_idx + num_interp] = mids[:, :, :H, :W].to(out_dtype) + out_idx += num_interp + pbar.update(num_interp) + tqdm_bar.update(num_interp) + used_multi = True + except model_management.OOM_EXCEPTION: + model_management.soft_empty_cache() + multi_fn = None # fall through to single-timestep path + + if not used_multi: + j = 0 + while j < num_interp: + b = min(batch, num_interp - j) + try: + img0 = img0_single.expand(b, -1, -1, -1) + img1 = img1_single.expand(b, -1, -1, -1) + mids = inference_model(img0, img1, timestep=ts_full[j:j + b], cache=feat_cache) + result[out_idx:out_idx + b] = mids[:, :, :H, :W].to(out_dtype) + out_idx += b + pbar.update(b) + tqdm_bar.update(b) + j += b + except model_management.OOM_EXCEPTION: + if batch <= 1: + raise + batch = max(1, batch // 2) + model_management.soft_empty_cache() + + result[out_idx] = images[i + 1].movedim(-1, 0).to(out_dtype) + out_idx += 1 + finally: + tqdm_bar.close() + + # BCHW -> BHWC + result = result.movedim(1, -1).clamp_(0.0, 1.0) + return io.NodeOutput(result) + + +class FrameInterpolationExtension(ComfyExtension): + @override + async def get_node_list(self) -> list[type[io.ComfyNode]]: + return [ + FrameInterpolationModelLoader, + FrameInterpolate, + ] + + +async def comfy_entrypoint() -> FrameInterpolationExtension: + return FrameInterpolationExtension() diff --git a/comfy_extras/nodes_fresca.py b/comfy_extras/nodes_fresca.py index eab4f303f..173f42154 100644 --- a/comfy_extras/nodes_fresca.py +++ b/comfy_extras/nodes_fresca.py @@ -60,7 +60,7 @@ class FreSca(io.ComfyNode): node_id="FreSca", search_aliases=["frequency guidance"], display_name="FreSca", - category="_for_testing", + category="experimental", description="Applies frequency-dependent scaling to the guidance", inputs=[ io.Model.Input("model"), diff --git a/comfy_extras/nodes_hidream_o1.py b/comfy_extras/nodes_hidream_o1.py new file mode 100644 index 000000000..f393745f6 --- /dev/null +++ b/comfy_extras/nodes_hidream_o1.py @@ -0,0 +1,256 @@ +from typing_extensions import override + +import torch + +import comfy.model_management +import comfy.patcher_extension +import node_helpers +from comfy_api.latest import ComfyExtension, io + + +class EmptyHiDreamO1LatentImage(io.ComfyNode): + @classmethod + def define_schema(cls) -> io.Schema: + return io.Schema( + node_id="EmptyHiDreamO1LatentImage", + display_name="Empty HiDream-O1 Latent Image", + category="latent/image", + description=( + "Empty pixel-space latent for HiDream-O1-Image. The model was " + "trained at ~4 megapixels; lower resolutions go off-distribution " + "and quality regresses noticeably. Trained resolutions: " + "2048x2048, 2304x1728, 1728x2304, 2560x1440, 1440x2560, " + "2496x1664, 1664x2496, 3104x1312, 1312x3104, 2304x1792, 1792x2304." + ), + inputs=[ + io.Int.Input(id="width", default=2048, min=64, max=4096, step=32), + io.Int.Input(id="height", default=2048, min=64, max=4096, step=32), + io.Int.Input(id="batch_size", default=1, min=1, max=64), + ], + outputs=[io.Latent().Output()], + ) + + @classmethod + def execute(cls, *, width: int, height: int, batch_size: int = 1) -> io.NodeOutput: + latent = torch.zeros( + (batch_size, 3, height, width), + device=comfy.model_management.intermediate_device(), + ) + return io.NodeOutput({"samples": latent}) + + +class HiDreamO1ReferenceImages(io.ComfyNode): + """Attach reference images to both positive and negative conditioning.""" + + @classmethod + def define_schema(cls) -> io.Schema: + return io.Schema( + node_id="HiDreamO1ReferenceImages", + display_name="HiDream-O1 Reference Images", + category="conditioning/image", + description=( + "Attach 1-10 reference images to conditioning, one for edit instruction" + "or multiple for subject-driven personalization." + ), + inputs=[ + io.Conditioning.Input(id="positive"), + io.Conditioning.Input(id="negative"), + io.Autogrow.Input( + "images", + template=io.Autogrow.TemplateNames( + io.Image.Input("image"), + names=[f"image_{i}" for i in range(1, 11)], + min=1, + ), + tooltip=("Reference images. 1 image = instruction edit; 2-10 images = multi reference." + ), + ), + ], + outputs=[ + io.Conditioning.Output(display_name="positive"), + io.Conditioning.Output(display_name="negative"), + ], + ) + + @classmethod + def execute(cls, *, positive, negative, images: io.Autogrow.Type) -> io.NodeOutput: + refs = [images[f"image_{i}"] for i in range(1, 11) if f"image_{i}" in images] + positive = node_helpers.conditioning_set_values(positive, {"reference_latents": refs}, append=True) + negative = node_helpers.conditioning_set_values(negative, {"reference_latents": refs}, append=True) + return io.NodeOutput(positive, negative) + + +class HiDreamO1PatchSeamSmoothing(io.ComfyNode): + PATCH_SIZE = 32 + EDGE_FEATHER = 4 + + # Shift presets per (pattern, N). 8-pass = 4-quadrant + 4 quarter-patch offsets. + SHIFTS_BY_PATTERN = { + ("single_shift", 2): [(0, 0), (16, 16)], + ("single_shift", 4): [(0, 0), (16, 0), (0, 16), (16, 16)], + ("single_shift", 8): [(0, 0), (16, 0), (0, 16), (16, 16), + (8, 8), (24, 8), (8, 24), (24, 24)], + ("symmetric", 2): [(-8, -8), (8, 8)], + ("symmetric", 4): [(-8, -8), (8, -8), (-8, 8), (8, 8)], + ("symmetric", 8): [(-12, -12), (4, -12), (-12, 4), (4, 4), + (-4, -4), (12, -4), (-4, 12), (12, 12)], + } + RAMP_LEVELS = { + "2": [2], + "4": [4], + "ramp_2_4": [2, 4], + "ramp_2_4_8": [2, 4, 8], + } + + @staticmethod + def _hann_tile(cy: int, cx: int, size: int = 32) -> torch.Tensor: + """size x size Hann tile peaking at (cy, cx) within a patch.""" + half = size // 2 + yy = torch.arange(size).view(size, 1) + xx = torch.arange(size).view(1, size) + dy = ((yy - cy + half) % size) - half + dx = ((xx - cx + half) % size) - half + return 0.25 * (1 + torch.cos(torch.pi * dy / half)) * (1 + torch.cos(torch.pi * dx / half)) + + @classmethod + def define_schema(cls) -> io.Schema: + return io.Schema( + node_id="HiDreamO1PatchSeamSmoothing", + display_name="HiDream-O1 Patch Seam Smoothing", + category="advanced/model", + is_experimental=True, + description=( + "Average the model output across multiple shifted patch-grid " + "positions during the late portion of sampling. Cancels seams." + ), + inputs=[ + io.Model.Input(id="model"), + io.Float.Input(id="start_percent", default=0.8, min=0.0, max=1.0, step=0.01, + tooltip="Sampling progress (0=start, 1=end) at which the blend turns ON.", + ), + io.Float.Input(id="end_percent", default=1.0, min=0.0, max=1.0, step=0.01, + tooltip="Sampling progress at which the blend turns OFF.", + ), + io.Combo.Input( + id="pattern", + options=["single_shift", "symmetric"], + default="single_shift", + tooltip="Shift layout. single_shift: one pass at the natural patch grid + others offset. symmetric: all passes off-grid, shifts split around origin.", + ), + io.Combo.Input( + id="passes", + options=["2", "4", "ramp_2_4", "ramp_2_4_8"], + default="2", + tooltip="Number of passes per gated step. 2/4 = fixed. ramp_*: pass count increases as sampling approaches end (more smoothing where seams are most visible).", + ), + io.Combo.Input( + id="blend", + options=["average", "window", "median"], + default="average", + tooltip="average: equal-weight mean. window: Hann-windowed weighting favoring each pass away from its patch boundaries. median: per-pixel median, rejects wraparound-outlier passes.", + ), + io.Float.Input(id="strength", default=1.0, min=0.0, max=1.0, step=0.01, + tooltip="Interpolation between the natural-grid pred (0) and the averaged result (1).", + ), + ], + outputs=[io.Model.Output()], + ) + + @classmethod + def execute(cls, *, model, start_percent: float, end_percent: float, pattern: str, passes: str, blend: str, strength: float) -> io.NodeOutput: + if strength <= 0.0 or end_percent <= start_percent: + return io.NodeOutput(model) + + P = cls.PATCH_SIZE + half = P // 2 + shift_levels = [cls.SHIFTS_BY_PATTERN[(pattern, n)] for n in cls.RAMP_LEVELS[passes]] + + if blend == "window": + window_tile_levels = [ + torch.stack([cls._hann_tile((half - sy) % P, (half - sx) % P, P) for sy, sx in lst], dim=0) + for lst in shift_levels + ] + else: + window_tile_levels = [None] * len(shift_levels) + + m = model.clone() + model_sampling = m.get_model_object("model_sampling") + multiplier = float(model_sampling.multiplier) + start_t = float(model_sampling.percent_to_sigma(start_percent)) * multiplier + end_t = float(model_sampling.percent_to_sigma(end_percent)) * multiplier + + edge_ramp_cache: dict = {} + + def get_edge_ramp(H: int, W: int, device, dtype) -> torch.Tensor: + key = (H, W, device, dtype) + cached = edge_ramp_cache.get(key) + if cached is not None: + return cached + feather = cls.EDGE_FEATHER + ys = torch.minimum(torch.arange(H, device=device, dtype=torch.float32), + (H - 1) - torch.arange(H, device=device, dtype=torch.float32)) + xs = torch.minimum(torch.arange(W, device=device, dtype=torch.float32), + (W - 1) - torch.arange(W, device=device, dtype=torch.float32)) + y_mask = ((ys - P) / feather).clamp(0, 1) + x_mask = ((xs - P) / feather).clamp(0, 1) + ramp = (y_mask[:, None] * x_mask[None, :]).to(dtype) + edge_ramp_cache[key] = ramp + return ramp + + def smoothing_wrapper(executor, *args, **kwargs): + x = args[0] + t = float(args[1][0]) + pred = executor(*args, **kwargs) + if not (end_t <= t <= start_t): + return pred + # Pick shift-level by sigma phase across the gated range. + if len(shift_levels) == 1: + level_idx = 0 + else: + phase = (start_t - t) / max(start_t - end_t, 1e-8) + level_idx = min(int(phase * len(shift_levels)), len(shift_levels) - 1) + shifts = shift_levels[level_idx] + window_tiles = window_tile_levels[level_idx] + + preds = [] + for sy, sx in shifts: + if sy == 0 and sx == 0: + preds.append(pred) + continue + x_rolled = torch.roll(x, shifts=(sy, sx), dims=(-2, -1)) + pred_rolled = executor(x_rolled, *args[1:], **kwargs) + preds.append(torch.roll(pred_rolled, shifts=(-sy, -sx), dims=(-2, -1))) + stacked = torch.stack(preds, dim=0) # (N, B, C, H, W) + _, _, _, H, W = stacked.shape + if blend == "window": + N = stacked.shape[0] + tiles = window_tiles.to(device=stacked.device, dtype=stacked.dtype) + w = tiles.repeat(1, H // P, W // P)[:, :H, :W] + sum_w = w.sum(dim=0, keepdim=True) + w = torch.where(sum_w < 1e-3, torch.full_like(w, 1.0 / N), w / sum_w.clamp(min=1e-8)) + avg = (stacked * w[:, None, None, :, :]).sum(dim=0) + elif blend == "median": + avg = torch.median(stacked, dim=0).values + else: + avg = stacked.mean(dim=0) + + # Mask out the P-px wraparound contamination strip at each edge. + mask = get_edge_ramp(H, W, pred.device, pred.dtype) + return pred * (1.0 - mask * strength) + avg * (mask * strength) + + m.add_wrapper_with_key(comfy.patcher_extension.WrappersMP.DIFFUSION_MODEL, "hidream_o1_patch_seam_smoothing", smoothing_wrapper) + return io.NodeOutput(m) + + +class HiDreamO1Extension(ComfyExtension): + @override + async def get_node_list(self) -> list[type[io.ComfyNode]]: + return [ + EmptyHiDreamO1LatentImage, + HiDreamO1ReferenceImages, + HiDreamO1PatchSeamSmoothing, + ] + + +async def comfy_entrypoint() -> HiDreamO1Extension: + return HiDreamO1Extension() diff --git a/comfy_extras/nodes_hunyuan.py b/comfy_extras/nodes_hunyuan.py index 4ea93a499..9e4873be5 100644 --- a/comfy_extras/nodes_hunyuan.py +++ b/comfy_extras/nodes_hunyuan.py @@ -131,6 +131,8 @@ class HunyuanVideo15SuperResolution(io.ComfyNode): def define_schema(cls): return io.Schema( node_id="HunyuanVideo15SuperResolution", + display_name="Hunyuan Video 1.5 Super Resolution", + category="conditioning/video_models", inputs=[ io.Conditioning.Input("positive"), io.Conditioning.Input("negative"), @@ -381,6 +383,8 @@ class HunyuanRefinerLatent(io.ComfyNode): def define_schema(cls): return io.Schema( node_id="HunyuanRefinerLatent", + display_name="Hunyuan Latent Refiner", + category="conditioning/video_models", inputs=[ io.Conditioning.Input("positive"), io.Conditioning.Input("negative"), diff --git a/comfy_extras/nodes_hunyuan3d.py b/comfy_extras/nodes_hunyuan3d.py index df0c3e4b1..403eb855b 100644 --- a/comfy_extras/nodes_hunyuan3d.py +++ b/comfy_extras/nodes_hunyuan3d.py @@ -1,12 +1,7 @@ import torch -import os -import json -import struct -import numpy as np from comfy.ldm.modules.diffusionmodules.mmdit import get_1d_sincos_pos_embed_from_grid_torch -import folder_paths import comfy.model_management -from comfy.cli_args import args +from comfy_extras.nodes_save_3d import pack_variable_mesh_batch from typing_extensions import override from comfy_api.latest import ComfyExtension, IO, Types from comfy_api.latest._util import MESH, VOXEL # only for backward compatibility if someone import it from this file (will be removed later) # noqa @@ -40,7 +35,7 @@ class Hunyuan3Dv2Conditioning(IO.ComfyNode): def define_schema(cls): return IO.Schema( node_id="Hunyuan3Dv2Conditioning", - category="conditioning/video_models", + category="conditioning/3d_models", inputs=[ IO.ClipVisionOutput.Input("clip_vision_output"), ], @@ -65,7 +60,7 @@ class Hunyuan3Dv2ConditioningMultiView(IO.ComfyNode): def define_schema(cls): return IO.Schema( node_id="Hunyuan3Dv2ConditioningMultiView", - category="conditioning/video_models", + category="conditioning/3d_models", inputs=[ IO.ClipVisionOutput.Input("front", optional=True), IO.ClipVisionOutput.Input("left", optional=True), @@ -424,6 +419,7 @@ class VoxelToMeshBasic(IO.ComfyNode): def define_schema(cls): return IO.Schema( node_id="VoxelToMeshBasic", + display_name="Voxel to Mesh (Basic)", category="3d", inputs=[ IO.Voxel.Input("voxel"), @@ -443,7 +439,9 @@ class VoxelToMeshBasic(IO.ComfyNode): vertices.append(v) faces.append(f) - return IO.NodeOutput(Types.MESH(torch.stack(vertices), torch.stack(faces))) + if vertices and all(v.shape == vertices[0].shape for v in vertices) and all(f.shape == faces[0].shape for f in faces): + return IO.NodeOutput(Types.MESH(torch.stack(vertices), torch.stack(faces))) + return IO.NodeOutput(pack_variable_mesh_batch(vertices, faces)) decode = execute # TODO: remove @@ -453,6 +451,7 @@ class VoxelToMesh(IO.ComfyNode): def define_schema(cls): return IO.Schema( node_id="VoxelToMesh", + display_name="Voxel to Mesh", category="3d", inputs=[ IO.Voxel.Input("voxel"), @@ -479,206 +478,13 @@ class VoxelToMesh(IO.ComfyNode): vertices.append(v) faces.append(f) - return IO.NodeOutput(Types.MESH(torch.stack(vertices), torch.stack(faces))) + if vertices and all(v.shape == vertices[0].shape for v in vertices) and all(f.shape == faces[0].shape for f in faces): + return IO.NodeOutput(Types.MESH(torch.stack(vertices), torch.stack(faces))) + return IO.NodeOutput(pack_variable_mesh_batch(vertices, faces)) decode = execute # TODO: remove -def save_glb(vertices, faces, filepath, metadata=None): - """ - Save PyTorch tensor vertices and faces as a GLB file without external dependencies. - - Parameters: - vertices: torch.Tensor of shape (N, 3) - The vertex coordinates - faces: torch.Tensor of shape (M, 3) - The face indices (triangle faces) - filepath: str - Output filepath (should end with .glb) - """ - - # Convert tensors to numpy arrays - vertices_np = vertices.cpu().numpy().astype(np.float32) - faces_np = faces.cpu().numpy().astype(np.uint32) - - vertices_buffer = vertices_np.tobytes() - indices_buffer = faces_np.tobytes() - - def pad_to_4_bytes(buffer): - padding_length = (4 - (len(buffer) % 4)) % 4 - return buffer + b'\x00' * padding_length - - vertices_buffer_padded = pad_to_4_bytes(vertices_buffer) - indices_buffer_padded = pad_to_4_bytes(indices_buffer) - - buffer_data = vertices_buffer_padded + indices_buffer_padded - - vertices_byte_length = len(vertices_buffer) - vertices_byte_offset = 0 - indices_byte_length = len(indices_buffer) - indices_byte_offset = len(vertices_buffer_padded) - - gltf = { - "asset": {"version": "2.0", "generator": "ComfyUI"}, - "buffers": [ - { - "byteLength": len(buffer_data) - } - ], - "bufferViews": [ - { - "buffer": 0, - "byteOffset": vertices_byte_offset, - "byteLength": vertices_byte_length, - "target": 34962 # ARRAY_BUFFER - }, - { - "buffer": 0, - "byteOffset": indices_byte_offset, - "byteLength": indices_byte_length, - "target": 34963 # ELEMENT_ARRAY_BUFFER - } - ], - "accessors": [ - { - "bufferView": 0, - "byteOffset": 0, - "componentType": 5126, # FLOAT - "count": len(vertices_np), - "type": "VEC3", - "max": vertices_np.max(axis=0).tolist(), - "min": vertices_np.min(axis=0).tolist() - }, - { - "bufferView": 1, - "byteOffset": 0, - "componentType": 5125, # UNSIGNED_INT - "count": faces_np.size, - "type": "SCALAR" - } - ], - "meshes": [ - { - "primitives": [ - { - "attributes": { - "POSITION": 0 - }, - "indices": 1, - "mode": 4 # TRIANGLES - } - ] - } - ], - "nodes": [ - { - "mesh": 0 - } - ], - "scenes": [ - { - "nodes": [0] - } - ], - "scene": 0 - } - - if metadata is not None: - gltf["asset"]["extras"] = metadata - - # Convert the JSON to bytes - gltf_json = json.dumps(gltf).encode('utf8') - - def pad_json_to_4_bytes(buffer): - padding_length = (4 - (len(buffer) % 4)) % 4 - return buffer + b' ' * padding_length - - gltf_json_padded = pad_json_to_4_bytes(gltf_json) - - # Create the GLB header - # Magic glTF - glb_header = struct.pack('<4sII', b'glTF', 2, 12 + 8 + len(gltf_json_padded) + 8 + len(buffer_data)) - - # Create JSON chunk header (chunk type 0) - json_chunk_header = struct.pack('