Merge branch 'master' into node-essentials-category

Add 24 non-cloud essential blueprints from comfyui-wiki/Subgraph-Blueprints.
These cover common workflows: text/image/video generation, editing,
inpainting, outpainting, upscaling, depth maps, pose, captioning, and more.

Cloud-only blueprints (5) are excluded and will be added once
client-side distribution filtering lands.

Amp-Thread-ID: https://ampcode.com/threads/T-019c6f43-6212-7308-bea6-bfc35a486cbf

.coderabbit.yaml

@@ -0,0 +1,114 @@
+# yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json
+language: "en-US"
+early_access: false
+
+reviews:
+  profile: "chill"
+  request_changes_workflow: false
+  high_level_summary: false
+  poem: false
+  review_status: false
+  review_details: false
+  commit_status: true
+  collapse_walkthrough: true
+  changed_files_summary: false
+  sequence_diagrams: false
+  estimate_code_review_effort: false
+  assess_linked_issues: false
+  related_issues: false
+  related_prs: false
+  suggested_labels: false
+  auto_apply_labels: false
+  suggested_reviewers: false
+  auto_assign_reviewers: false
+  in_progress_fortune: false
+  enable_prompt_for_ai_agents: true
+
+  path_filters:
+    - "!comfy_api_nodes/apis/**"
+    - "!**/generated/*.pyi"
+    - "!.ci/**"
+    - "!script_examples/**"
+    - "!**/__pycache__/**"
+    - "!**/*.ipynb"
+    - "!**/*.png"
+    - "!**/*.bat"
+
+  path_instructions:
+    - path: "comfy/**"
+      instructions: |
+        Core ML/diffusion engine. Focus on:
+        - Backward compatibility (breaking changes affect all custom nodes)
+        - Memory management and GPU resource handling
+        - Performance implications in hot paths
+        - Thread safety for concurrent execution
+    - path: "comfy_api_nodes/**"
+      instructions: |
+        Third-party API integration nodes. Focus on:
+        - No hardcoded API keys or secrets
+        - Proper error handling for API failures (timeouts, rate limits, auth errors)
+        - Correct Pydantic model usage
+        - Security of user data passed to external APIs
+    - path: "comfy_extras/**"
+      instructions: |
+        Community-contributed extra nodes. Focus on:
+        - Consistency with node patterns (INPUT_TYPES, RETURN_TYPES, FUNCTION, CATEGORY)
+        - No breaking changes to existing node interfaces
+    - path: "comfy_execution/**"
+      instructions: |
+        Execution engine (graph execution, caching, jobs). Focus on:
+        - Caching correctness
+        - Concurrent execution safety
+        - Graph validation edge cases
+    - path: "nodes.py"
+      instructions: |
+        Core node definitions (2500+ lines). Focus on:
+        - Backward compatibility of NODE_CLASS_MAPPINGS
+        - Consistency of INPUT_TYPES return format
+    - path: "alembic_db/**"
+      instructions: |
+        Database migrations. Focus on:
+        - Migration safety and rollback support
+        - Data preservation during schema changes
+
+  auto_review:
+    enabled: true
+    auto_incremental_review: true
+    drafts: true
+
+  finishing_touches:
+    docstrings:
+      enabled: false
+    unit_tests:
+      enabled: false
+
+  tools:
+    ruff:
+      enabled: true
+    pylint:
+      enabled: false
+    flake8:
+      enabled: false
+    gitleaks:
+      enabled: true
+    shellcheck:
+      enabled: false
+    markdownlint:
+      enabled: false
+    yamllint:
+      enabled: false
+    languagetool:
+      enabled: false
+    github-checks:
+      enabled: true
+      timeout_ms: 90000
+    ast-grep:
+      essential_rules: true
+
+chat:
+  auto_reply: true
+
+knowledge_base:
+  opt_out: false
+  learnings:
+    scope: "auto"

.gitignore

@@ -11,7 +11,7 @@ extra_model_paths.yaml
 /.vs
 .vscode/
 .idea/
-venv/
+venv*/
 .venv/
 /web/extensions/*
 !/web/extensions/logging.js.example

README.md

@@ -229,9 +229,9 @@ AMD users can install rocm and pytorch with pip if you don't have it already ins
 
 ```pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm7.1```
 
-This is the command to install the nightly with ROCm 7.1 which might have some performance improvements:
+This is the command to install the nightly with ROCm 7.2 which might have some performance improvements:
 
-```pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/rocm7.1```
+```pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/rocm7.2```
 
 
 ### AMD GPUs (Experimental: Windows and Linux), RDNA 3, 3.5 and 4 only.

app/node_replace_manager.py

@@ -0,0 +1,105 @@
+from __future__ import annotations
+
+from aiohttp import web
+
+from typing import TYPE_CHECKING, TypedDict
+if TYPE_CHECKING:
+    from comfy_api.latest._io_public import NodeReplace
+
+from comfy_execution.graph_utils import is_link
+import nodes
+
+class NodeStruct(TypedDict):
+    inputs: dict[str, str | int | float | bool | tuple[str, int]]
+    class_type: str
+    _meta: dict[str, str]
+
+def copy_node_struct(node_struct: NodeStruct, empty_inputs: bool = False) -> NodeStruct:
+    new_node_struct = node_struct.copy()
+    if empty_inputs:
+        new_node_struct["inputs"] = {}
+    else:
+        new_node_struct["inputs"] = node_struct["inputs"].copy()
+    new_node_struct["_meta"] = node_struct["_meta"].copy()
+    return new_node_struct
+
+
+class NodeReplaceManager:
+    """Manages node replacement registrations."""
+
+    def __init__(self):
+        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)
+
+    def get_replacement(self, old_node_id: str) -> list[NodeReplace] | None:
+        """Get replacements for an old node ID."""
+        return self._replacements.get(old_node_id)
+
+    def has_replacement(self, old_node_id: str) -> bool:
+        """Check if a replacement exists for an old node ID."""
+        return old_node_id in self._replacements
+
+    def apply_replacements(self, prompt: dict[str, NodeStruct]):
+        connections: dict[str, list[tuple[str, str, int]]] = {}
+        need_replacement: set[str] = set()
+        for node_number, node_struct in prompt.items():
+            class_type = node_struct["class_type"]
+            # need replacement if not in NODE_CLASS_MAPPINGS and has replacement
+            if class_type not in nodes.NODE_CLASS_MAPPINGS.keys() and self.has_replacement(class_type):
+                need_replacement.add(node_number)
+            # keep track of connections
+            for input_id, input_value in node_struct["inputs"].items():
+                if is_link(input_value):
+                    conn_number = input_value[0]
+                    connections.setdefault(conn_number, []).append((node_number, input_id, input_value[1]))
+        for node_number in need_replacement:
+            node_struct = prompt[node_number]
+            class_type = node_struct["class_type"]
+            replacements = self.get_replacement(class_type)
+            if replacements is None:
+                continue
+            # just use the first replacement
+            replacement = replacements[0]
+            new_node_id = replacement.new_node_id
+            # if replacement is not a valid node, skip trying to replace it as will only cause confusion
+            if new_node_id not in nodes.NODE_CLASS_MAPPINGS.keys():
+                continue
+            # first, replace node id (class_type)
+            new_node_struct = copy_node_struct(node_struct, empty_inputs=True)
+            new_node_struct["class_type"] = new_node_id
+            # TODO: consider replacing display_name in _meta as well for error reporting purposes; would need to query node schema
+            # second, replace inputs
+            if replacement.input_mapping is not None:
+                for input_map in replacement.input_mapping:
+                    if "set_value" in input_map:
+                        new_node_struct["inputs"][input_map["new_id"]] = input_map["set_value"]
+                    elif "old_id" in input_map:
+                        new_node_struct["inputs"][input_map["new_id"]] = node_struct["inputs"][input_map["old_id"]]
+            # finalize input replacement
+            prompt[node_number] = new_node_struct
+            # third, replace outputs
+            if replacement.output_mapping is not None:
+                # re-mapping outputs requires changing the input values of nodes that receive connections from this one
+                if node_number in connections:
+                    for conns in connections[node_number]:
+                        conn_node_number, conn_input_id, old_output_idx = conns
+                        for output_map in replacement.output_mapping:
+                            if output_map["old_idx"] == old_output_idx:
+                                new_output_idx = output_map["new_idx"]
+                                previous_input = prompt[conn_node_number]["inputs"][conn_input_id]
+                                previous_input[1] = new_output_idx
+
+    def as_dict(self):
+        """Serialize all replacements to dict."""
+        return {
+            k: [v.as_dict() for v in v_list]
+            for k, v_list in self._replacements.items()
+        }
+
+    def add_routes(self, routes):
+        @routes.get("/node_replacements")
+        async def get_node_replacements(request):
+            return web.json_response(self.as_dict())

blueprints/.glsl/Hue_and_Saturation_1.frag

@@ -169,7 +169,7 @@ vec3 colorize(vec3 rgb, float hue, float sat, float light) {
     float lum = dot(rgb, vec3(0.299, 0.587, 0.114));
     float l = adjustLightness(lum, light);
 
-    vec3 hsl = vec3(fract(hue), clamp(abs(sat), 0.0, 1.0), clamp(l, 0.0, 1.0));
+    vec3 hsl = vec3(fract(hue), clamp(sat, 0.0, 1.0), clamp(l, 0.0, 1.0));
     return hsl2rgb(hsl);
 }
 

blueprints/Image Channels.json

@@ -1 +1 @@
-{"revision": 0, "last_node_id": 29, "last_link_id": 0, "nodes": [{"id": 29, "type": "4c9d6ea4-b912-40e5-8766-6793a9758c53", "pos": [1970, -230], "size": [180, 86], "flags": {}, "order": 5, "mode": 0, "inputs": [{"label": "image", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": null}], "outputs": [{"label": "R", "localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": []}, {"label": "G", "localized_name": "IMAGE1", "name": "IMAGE1", "type": "IMAGE", "links": []}, {"label": "B", "localized_name": "IMAGE2", "name": "IMAGE2", "type": "IMAGE", "links": []}, {"label": "A", "localized_name": "IMAGE3", "name": "IMAGE3", "type": "IMAGE", "links": []}], "title": "Image Channels", "properties": {"proxyWidgets": []}, "widgets_values": []}], "links": [], "version": 0.4, "definitions": {"subgraphs": [{"id": "4c9d6ea4-b912-40e5-8766-6793a9758c53", "version": 1, "state": {"lastGroupId": 0, "lastNodeId": 28, "lastLinkId": 39, "lastRerouteId": 0}, "revision": 0, "config": {}, "name": "Image Channels", "inputNode": {"id": -10, "bounding": [1820, -185, 120, 60]}, "outputNode": {"id": -20, "bounding": [2460, -215, 120, 120]}, "inputs": [{"id": "3522932b-2d86-4a1f-a02a-cb29f3a9d7fe", "name": "images.image0", "type": "IMAGE", "linkIds": [39], "localized_name": "images.image0", "label": "image", "pos": [1920, -165]}], "outputs": [{"id": "605cb9c3-b065-4d9b-81d2-3ec331889b2b", "name": "IMAGE0", "type": "IMAGE", "linkIds": [26], "localized_name": "IMAGE0", "label": "R", "pos": [2480, -195]}, {"id": "fb44a77e-0522-43e9-9527-82e7465b3596", "name": "IMAGE1", "type": "IMAGE", "linkIds": [27], "localized_name": "IMAGE1", "label": "G", "pos": [2480, -175]}, {"id": "81460ee6-0131-402a-874f-6bf3001fc4ff", "name": "IMAGE2", "type": "IMAGE", "linkIds": [28], "localized_name": "IMAGE2", "label": "B", "pos": [2480, -155]}, {"id": "ae690246-80d4-4951-b1d9-9306d8a77417", "name": "IMAGE3", "type": "IMAGE", "linkIds": [29], "localized_name": "IMAGE3", "label": "A", "pos": [2480, -135]}], "widgets": [], "nodes": [{"id": 23, "type": "GLSLShader", "pos": [2000, -330], "size": [400, 172], "flags": {}, "order": 0, "mode": 0, "inputs": [{"label": "image", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": 39}, {"localized_name": "fragment_shader", "name": "fragment_shader", "type": "STRING", "widget": {"name": "fragment_shader"}, "link": null}, {"localized_name": "size_mode", "name": "size_mode", "type": "COMFY_DYNAMICCOMBO_V3", "widget": {"name": "size_mode"}, "link": null}, {"label": "image1", "localized_name": "images.image1", "name": "images.image1", "shape": 7, "type": "IMAGE", "link": null}], "outputs": [{"label": "R", "localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": [26]}, {"label": "G", "localized_name": "IMAGE1", "name": "IMAGE1", "type": "IMAGE", "links": [27]}, {"label": "B", "localized_name": "IMAGE2", "name": "IMAGE2", "type": "IMAGE", "links": [28]}, {"label": "A", "localized_name": "IMAGE3", "name": "IMAGE3", "type": "IMAGE", "links": [29]}], "properties": {"Node name for S&R": "GLSLShader"}, "widgets_values": ["#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\nlayout(location = 1) out vec4 fragColor1;\nlayout(location = 2) out vec4 fragColor2;\nlayout(location = 3) out vec4 fragColor3;\n\nvoid main() {\n  vec4 color = texture(u_image0, v_texCoord);\n  // Output each channel as grayscale to separate render targets\n  fragColor0 = vec4(vec3(color.r), 1.0);  // Red channel\n  fragColor1 = vec4(vec3(color.g), 1.0);  // Green channel\n  fragColor2 = vec4(vec3(color.b), 1.0);  // Blue channel\n  fragColor3 = vec4(vec3(color.a), 1.0);  // Alpha channel\n}\n", "from_input"]}], "groups": [], "links": [{"id": 39, "origin_id": -10, "origin_slot": 0, "target_id": 23, "target_slot": 0, "type": "IMAGE"}, {"id": 26, "origin_id": 23, "origin_slot": 0, "target_id": -20, "target_slot": 0, "type": "IMAGE"}, {"id": 27, "origin_id": 23, "origin_slot": 1, "target_id": -20, "target_slot": 1, "type": "IMAGE"}, {"id": 28, "origin_id": 23, "origin_slot": 2, "target_id": -20, "target_slot": 2, "type": "IMAGE"}, {"id": 29, "origin_id": 23, "origin_slot": 3, "target_id": -20, "target_slot": 3, "type": "IMAGE"}], "extra": {"workflowRendererVersion": "LG"}}]}}
+{"revision": 0, "last_node_id": 29, "last_link_id": 0, "nodes": [{"id": 29, "type": "4c9d6ea4-b912-40e5-8766-6793a9758c53", "pos": [1970, -230], "size": [180, 86], "flags": {}, "order": 5, "mode": 0, "inputs": [{"label": "image", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": null}], "outputs": [{"label": "R", "localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": []}, {"label": "G", "localized_name": "IMAGE1", "name": "IMAGE1", "type": "IMAGE", "links": []}, {"label": "B", "localized_name": "IMAGE2", "name": "IMAGE2", "type": "IMAGE", "links": []}, {"label": "A", "localized_name": "IMAGE3", "name": "IMAGE3", "type": "IMAGE", "links": []}], "title": "Image Channels", "properties": {"proxyWidgets": []}, "widgets_values": []}], "links": [], "version": 0.4, "definitions": {"subgraphs": [{"id": "4c9d6ea4-b912-40e5-8766-6793a9758c53", "version": 1, "state": {"lastGroupId": 0, "lastNodeId": 28, "lastLinkId": 39, "lastRerouteId": 0}, "revision": 0, "config": {}, "name": "Image Channels", "inputNode": {"id": -10, "bounding": [1820, -185, 120, 60]}, "outputNode": {"id": -20, "bounding": [2460, -215, 120, 120]}, "inputs": [{"id": "3522932b-2d86-4a1f-a02a-cb29f3a9d7fe", "name": "images.image0", "type": "IMAGE", "linkIds": [39], "localized_name": "images.image0", "label": "image", "pos": [1920, -165]}], "outputs": [{"id": "605cb9c3-b065-4d9b-81d2-3ec331889b2b", "name": "IMAGE0", "type": "IMAGE", "linkIds": [26], "localized_name": "IMAGE0", "label": "R", "pos": [2480, -195]}, {"id": "fb44a77e-0522-43e9-9527-82e7465b3596", "name": "IMAGE1", "type": "IMAGE", "linkIds": [27], "localized_name": "IMAGE1", "label": "G", "pos": [2480, -175]}, {"id": "81460ee6-0131-402a-874f-6bf3001fc4ff", "name": "IMAGE2", "type": "IMAGE", "linkIds": [28], "localized_name": "IMAGE2", "label": "B", "pos": [2480, -155]}, {"id": "ae690246-80d4-4951-b1d9-9306d8a77417", "name": "IMAGE3", "type": "IMAGE", "linkIds": [29], "localized_name": "IMAGE3", "label": "A", "pos": [2480, -135]}], "widgets": [], "nodes": [{"id": 23, "type": "GLSLShader", "pos": [2000, -330], "size": [400, 172], "flags": {}, "order": 0, "mode": 0, "inputs": [{"label": "image", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": 39}, {"localized_name": "fragment_shader", "name": "fragment_shader", "type": "STRING", "widget": {"name": "fragment_shader"}, "link": null}, {"localized_name": "size_mode", "name": "size_mode", "type": "COMFY_DYNAMICCOMBO_V3", "widget": {"name": "size_mode"}, "link": null}, {"label": "image1", "localized_name": "images.image1", "name": "images.image1", "shape": 7, "type": "IMAGE", "link": null}], "outputs": [{"label": "R", "localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": [26]}, {"label": "G", "localized_name": "IMAGE1", "name": "IMAGE1", "type": "IMAGE", "links": [27]}, {"label": "B", "localized_name": "IMAGE2", "name": "IMAGE2", "type": "IMAGE", "links": [28]}, {"label": "A", "localized_name": "IMAGE3", "name": "IMAGE3", "type": "IMAGE", "links": [29]}], "properties": {"Node name for S&R": "GLSLShader"}, "widgets_values": ["#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\nlayout(location = 1) out vec4 fragColor1;\nlayout(location = 2) out vec4 fragColor2;\nlayout(location = 3) out vec4 fragColor3;\n\nvoid main() {\n  vec4 color = texture(u_image0, v_texCoord);\n  // Output each channel as grayscale to separate render targets\n  fragColor0 = vec4(vec3(color.r), 1.0);  // Red channel\n  fragColor1 = vec4(vec3(color.g), 1.0);  // Green channel\n  fragColor2 = vec4(vec3(color.b), 1.0);  // Blue channel\n  fragColor3 = vec4(vec3(color.a), 1.0);  // Alpha channel\n}\n", "from_input"]}], "groups": [], "links": [{"id": 39, "origin_id": -10, "origin_slot": 0, "target_id": 23, "target_slot": 0, "type": "IMAGE"}, {"id": 26, "origin_id": 23, "origin_slot": 0, "target_id": -20, "target_slot": 0, "type": "IMAGE"}, {"id": 27, "origin_id": 23, "origin_slot": 1, "target_id": -20, "target_slot": 1, "type": "IMAGE"}, {"id": 28, "origin_id": 23, "origin_slot": 2, "target_id": -20, "target_slot": 2, "type": "IMAGE"}, {"id": 29, "origin_id": 23, "origin_slot": 3, "target_id": -20, "target_slot": 3, "type": "IMAGE"}], "extra": {"workflowRendererVersion": "LG"}, "category": "Image Tools/Color adjust"}]}}

blueprints/Prompt Enhance.json

@@ -0,0 +1 @@
+{"revision": 0, "last_node_id": 15, "last_link_id": 0, "nodes": [{"id": 15, "type": "24d8bbfd-39d4-4774-bff0-3de40cc7a471", "pos": [-1490, 2040], "size": [400, 260], "flags": {}, "order": 0, "mode": 0, "inputs": [{"name": "prompt", "type": "STRING", "widget": {"name": "prompt"}, "link": null}, {"label": "reference images", "name": "images", "type": "IMAGE", "link": null}], "outputs": [{"name": "STRING", "type": "STRING", "links": null}], "title": "Prompt Enhance", "properties": {"proxyWidgets": [["-1", "prompt"]], "cnr_id": "comfy-core", "ver": "0.14.1"}, "widgets_values": [""]}], "links": [], "version": 0.4, "definitions": {"subgraphs": [{"id": "24d8bbfd-39d4-4774-bff0-3de40cc7a471", "version": 1, "state": {"lastGroupId": 0, "lastNodeId": 15, "lastLinkId": 14, "lastRerouteId": 0}, "revision": 0, "config": {}, "name": "Prompt Enhance", "inputNode": {"id": -10, "bounding": [-2170, 2110, 138.876953125, 80]}, "outputNode": {"id": -20, "bounding": [-640, 2110, 120, 60]}, "inputs": [{"id": "aeab7216-00e0-4528-a09b-bba50845c5a6", "name": "prompt", "type": "STRING", "linkIds": [11], "pos": [-2051.123046875, 2130]}, {"id": "7b73fd36-aa31-4771-9066-f6c83879994b", "name": "images", "type": "IMAGE", "linkIds": [14], "label": "reference images", "pos": [-2051.123046875, 2150]}], "outputs": [{"id": "c7b0d930-68a1-48d1-b496-0519e5837064", "name": "STRING", "type": "STRING", "linkIds": [13], "pos": [-620, 2130]}], "widgets": [], "nodes": [{"id": 11, "type": "GeminiNode", "pos": [-1560, 1990], "size": [470, 470], "flags": {}, "order": 0, "mode": 0, "inputs": [{"localized_name": "images", "name": "images", "shape": 7, "type": "IMAGE", "link": 14}, {"localized_name": "audio", "name": "audio", "shape": 7, "type": "AUDIO", "link": null}, {"localized_name": "video", "name": "video", "shape": 7, "type": "VIDEO", "link": null}, {"localized_name": "files", "name": "files", "shape": 7, "type": "GEMINI_INPUT_FILES", "link": null}, {"localized_name": "prompt", "name": "prompt", "type": "STRING", "widget": {"name": "prompt"}, "link": 11}, {"localized_name": "model", "name": "model", "type": "COMBO", "widget": {"name": "model"}, "link": null}, {"localized_name": "seed", "name": "seed", "type": "INT", "widget": {"name": "seed"}, "link": null}, {"localized_name": "system_prompt", "name": "system_prompt", "shape": 7, "type": "STRING", "widget": {"name": "system_prompt"}, "link": null}], "outputs": [{"localized_name": "STRING", "name": "STRING", "type": "STRING", "links": [13]}], "properties": {"cnr_id": "comfy-core", "ver": "0.14.1", "Node name for S&R": "GeminiNode"}, "widgets_values": ["", "gemini-3-pro-preview", 42, "randomize", "You are an expert in prompt writing.\nBased on the input, rewrite the user's input into a detailed prompt.\nincluding camera settings, lighting, composition, and style.\nReturn the prompt only"], "color": "#432", "bgcolor": "#653"}], "groups": [], "links": [{"id": 11, "origin_id": -10, "origin_slot": 0, "target_id": 11, "target_slot": 4, "type": "STRING"}, {"id": 13, "origin_id": 11, "origin_slot": 0, "target_id": -20, "target_slot": 0, "type": "STRING"}, {"id": 14, "origin_id": -10, "origin_slot": 1, "target_id": 11, "target_slot": 0, "type": "IMAGE"}], "extra": {"workflowRendererVersion": "LG"}, "category": "Text generation/Prompt enhance"}]}, "extra": {}}

blueprints/Sharpen.json

@@ -1 +1 @@
-{"revision":0,"last_node_id":25,"last_link_id":0,"nodes":[{"id":25,"type":"621ba4e2-22a8-482d-a369-023753198b7b","pos":[4610,-790],"size":[230,58],"flags":{},"order":4,"mode":0,"inputs":[{"label":"image","localized_name":"images.image0","name":"images.image0","type":"IMAGE","link":null}],"outputs":[{"label":"IMAGE","localized_name":"IMAGE0","name":"IMAGE0","type":"IMAGE","links":[]}],"title":"Sharpen","properties":{"proxyWidgets":[["24","value"]]},"widgets_values":[]}],"links":[],"version":0.4,"definitions":{"subgraphs":[{"id":"621ba4e2-22a8-482d-a369-023753198b7b","version":1,"state":{"lastGroupId":0,"lastNodeId":24,"lastLinkId":36,"lastRerouteId":0},"revision":0,"config":{},"name":"Sharpen","inputNode":{"id":-10,"bounding":[4090,-825,120,60]},"outputNode":{"id":-20,"bounding":[5150,-825,120,60]},"inputs":[{"id":"37011fb7-14b7-4e0e-b1a0-6a02e8da1fd7","name":"images.image0","type":"IMAGE","linkIds":[34],"localized_name":"images.image0","label":"image","pos":[4190,-805]}],"outputs":[{"id":"e9182b3f-635c-4cd4-a152-4b4be17ae4b9","name":"IMAGE0","type":"IMAGE","linkIds":[35],"localized_name":"IMAGE0","label":"IMAGE","pos":[5170,-805]}],"widgets":[],"nodes":[{"id":24,"type":"PrimitiveFloat","pos":[4280,-1240],"size":[270,58],"flags":{},"order":0,"mode":0,"inputs":[{"label":"strength","localized_name":"value","name":"value","type":"FLOAT","widget":{"name":"value"},"link":null}],"outputs":[{"localized_name":"FLOAT","name":"FLOAT","type":"FLOAT","links":[36]}],"properties":{"Node name for S&R":"PrimitiveFloat","min":0,"max":3,"precision":2,"step":0.05},"widgets_values":[0.5]},{"id":23,"type":"GLSLShader","pos":[4570,-1240],"size":[370,192],"flags":{},"order":1,"mode":0,"inputs":[{"label":"image0","localized_name":"images.image0","name":"images.image0","type":"IMAGE","link":34},{"label":"image1","localized_name":"images.image1","name":"images.image1","shape":7,"type":"IMAGE","link":null},{"label":"u_float0","localized_name":"floats.u_float0","name":"floats.u_float0","shape":7,"type":"FLOAT","link":36},{"label":"u_float1","localized_name":"floats.u_float1","name":"floats.u_float1","shape":7,"type":"FLOAT","link":null},{"label":"u_int0","localized_name":"ints.u_int0","name":"ints.u_int0","shape":7,"type":"INT","link":null},{"localized_name":"fragment_shader","name":"fragment_shader","type":"STRING","widget":{"name":"fragment_shader"},"link":null},{"localized_name":"size_mode","name":"size_mode","type":"COMFY_DYNAMICCOMBO_V3","widget":{"name":"size_mode"},"link":null}],"outputs":[{"localized_name":"IMAGE0","name":"IMAGE0","type":"IMAGE","links":[35]},{"localized_name":"IMAGE1","name":"IMAGE1","type":"IMAGE","links":null},{"localized_name":"IMAGE2","name":"IMAGE2","type":"IMAGE","links":null},{"localized_name":"IMAGE3","name":"IMAGE3","type":"IMAGE","links":null}],"properties":{"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}","from_input"]}],"groups":[],"links":[{"id":36,"origin_id":24,"origin_slot":0,"target_id":23,"target_slot":2,"type":"FLOAT"},{"id":34,"origin_id":-10,"origin_slot":0,"target_id":23,"target_slot":0,"type":"IMAGE"},{"id":35,"origin_id":23,"origin_slot":0,"target_id":-20,"target_slot":0,"type":"IMAGE"}],"extra":{"workflowRendererVersion":"LG"}}]}}
+{"revision": 0, "last_node_id": 25, "last_link_id": 0, "nodes": [{"id": 25, "type": "621ba4e2-22a8-482d-a369-023753198b7b", "pos": [4610, -790], "size": [230, 58], "flags": {}, "order": 4, "mode": 0, "inputs": [{"label": "image", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": null}], "outputs": [{"label": "IMAGE", "localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": []}], "title": "Sharpen", "properties": {"proxyWidgets": [["24", "value"]]}, "widgets_values": []}], "links": [], "version": 0.4, "definitions": {"subgraphs": [{"id": "621ba4e2-22a8-482d-a369-023753198b7b", "version": 1, "state": {"lastGroupId": 0, "lastNodeId": 24, "lastLinkId": 36, "lastRerouteId": 0}, "revision": 0, "config": {}, "name": "Sharpen", "inputNode": {"id": -10, "bounding": [4090, -825, 120, 60]}, "outputNode": {"id": -20, "bounding": [5150, -825, 120, 60]}, "inputs": [{"id": "37011fb7-14b7-4e0e-b1a0-6a02e8da1fd7", "name": "images.image0", "type": "IMAGE", "linkIds": [34], "localized_name": "images.image0", "label": "image", "pos": [4190, -805]}], "outputs": [{"id": "e9182b3f-635c-4cd4-a152-4b4be17ae4b9", "name": "IMAGE0", "type": "IMAGE", "linkIds": [35], "localized_name": "IMAGE0", "label": "IMAGE", "pos": [5170, -805]}], "widgets": [], "nodes": [{"id": 24, "type": "PrimitiveFloat", "pos": [4280, -1240], "size": [270, 58], "flags": {}, "order": 0, "mode": 0, "inputs": [{"label": "strength", "localized_name": "value", "name": "value", "type": "FLOAT", "widget": {"name": "value"}, "link": null}], "outputs": [{"localized_name": "FLOAT", "name": "FLOAT", "type": "FLOAT", "links": [36]}], "properties": {"Node name for S&R": "PrimitiveFloat", "min": 0, "max": 3, "precision": 2, "step": 0.05}, "widgets_values": [0.5]}, {"id": 23, "type": "GLSLShader", "pos": [4570, -1240], "size": [370, 192], "flags": {}, "order": 1, "mode": 0, "inputs": [{"label": "image0", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": 34}, {"label": "image1", "localized_name": "images.image1", "name": "images.image1", "shape": 7, "type": "IMAGE", "link": null}, {"label": "u_float0", "localized_name": "floats.u_float0", "name": "floats.u_float0", "shape": 7, "type": "FLOAT", "link": 36}, {"label": "u_float1", "localized_name": "floats.u_float1", "name": "floats.u_float1", "shape": 7, "type": "FLOAT", "link": null}, {"label": "u_int0", "localized_name": "ints.u_int0", "name": "ints.u_int0", "shape": 7, "type": "INT", "link": null}, {"localized_name": "fragment_shader", "name": "fragment_shader", "type": "STRING", "widget": {"name": "fragment_shader"}, "link": null}, {"localized_name": "size_mode", "name": "size_mode", "type": "COMFY_DYNAMICCOMBO_V3", "widget": {"name": "size_mode"}, "link": null}], "outputs": [{"localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": [35]}, {"localized_name": "IMAGE1", "name": "IMAGE1", "type": "IMAGE", "links": null}, {"localized_name": "IMAGE2", "name": "IMAGE2", "type": "IMAGE", "links": null}, {"localized_name": "IMAGE3", "name": "IMAGE3", "type": "IMAGE", "links": null}], "properties": {"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}", "from_input"]}], "groups": [], "links": [{"id": 36, "origin_id": 24, "origin_slot": 0, "target_id": 23, "target_slot": 2, "type": "FLOAT"}, {"id": 34, "origin_id": -10, "origin_slot": 0, "target_id": 23, "target_slot": 0, "type": "IMAGE"}, {"id": 35, "origin_id": 23, "origin_slot": 0, "target_id": -20, "target_slot": 0, "type": "IMAGE"}], "extra": {"workflowRendererVersion": "LG"}, "category": "Image Tools/Sharpen"}]}}

blueprints/Video Upscale(GAN x4).json

@@ -0,0 +1 @@
+{"revision": 0, "last_node_id": 13, "last_link_id": 0, "nodes": [{"id": 13, "type": "cf95b747-3e17-46cb-8097-cac60ff9b2e1", "pos": [1120, 330], "size": [240, 58], "flags": {}, "order": 3, "mode": 0, "inputs": [{"localized_name": "video", "name": "video", "type": "VIDEO", "link": null}, {"name": "model_name", "type": "COMBO", "widget": {"name": "model_name"}, "link": null}], "outputs": [{"localized_name": "VIDEO", "name": "VIDEO", "type": "VIDEO", "links": []}], "title": "Video Upscale(GAN x4)", "properties": {"proxyWidgets": [["-1", "model_name"]], "cnr_id": "comfy-core", "ver": "0.14.1"}, "widgets_values": ["RealESRGAN_x4plus.safetensors"]}], "links": [], "version": 0.4, "definitions": {"subgraphs": [{"id": "cf95b747-3e17-46cb-8097-cac60ff9b2e1", "version": 1, "state": {"lastGroupId": 0, "lastNodeId": 13, "lastLinkId": 19, "lastRerouteId": 0}, "revision": 0, "config": {}, "name": "Video Upscale(GAN x4)", "inputNode": {"id": -10, "bounding": [550, 460, 120, 80]}, "outputNode": {"id": -20, "bounding": [1490, 460, 120, 60]}, "inputs": [{"id": "666d633e-93e7-42dc-8d11-2b7b99b0f2a6", "name": "video", "type": "VIDEO", "linkIds": [10], "localized_name": "video", "pos": [650, 480]}, {"id": "2e23a087-caa8-4d65-99e6-662761aa905a", "name": "model_name", "type": "COMBO", "linkIds": [19], "pos": [650, 500]}], "outputs": [{"id": "0c1768ea-3ec2-412f-9af6-8e0fa36dae70", "name": "VIDEO", "type": "VIDEO", "linkIds": [15], "localized_name": "VIDEO", "pos": [1510, 480]}], "widgets": [], "nodes": [{"id": 2, "type": "ImageUpscaleWithModel", "pos": [1110, 450], "size": [320, 46], "flags": {}, "order": 1, "mode": 0, "inputs": [{"localized_name": "upscale_model", "name": "upscale_model", "type": "UPSCALE_MODEL", "link": 1}, {"localized_name": "image", "name": "image", "type": "IMAGE", "link": 14}], "outputs": [{"localized_name": "IMAGE", "name": "IMAGE", "type": "IMAGE", "links": [13]}], "properties": {"cnr_id": "comfy-core", "ver": "0.10.0", "Node name for S&R": "ImageUpscaleWithModel"}}, {"id": 11, "type": "CreateVideo", "pos": [1110, 550], "size": [320, 78], "flags": {}, "order": 3, "mode": 0, "inputs": [{"localized_name": "images", "name": "images", "type": "IMAGE", "link": 13}, {"localized_name": "audio", "name": "audio", "shape": 7, "type": "AUDIO", "link": 16}, {"localized_name": "fps", "name": "fps", "type": "FLOAT", "widget": {"name": "fps"}, "link": 12}], "outputs": [{"localized_name": "VIDEO", "name": "VIDEO", "type": "VIDEO", "links": [15]}], "properties": {"cnr_id": "comfy-core", "ver": "0.10.0", "Node name for S&R": "CreateVideo"}, "widgets_values": [30]}, {"id": 10, "type": "GetVideoComponents", "pos": [1110, 330], "size": [320, 70], "flags": {}, "order": 2, "mode": 0, "inputs": [{"localized_name": "video", "name": "video", "type": "VIDEO", "link": 10}], "outputs": [{"localized_name": "images", "name": "images", "type": "IMAGE", "links": [14]}, {"localized_name": "audio", "name": "audio", "type": "AUDIO", "links": [16]}, {"localized_name": "fps", "name": "fps", "type": "FLOAT", "links": [12]}], "properties": {"cnr_id": "comfy-core", "ver": "0.10.0", "Node name for S&R": "GetVideoComponents"}}, {"id": 1, "type": "UpscaleModelLoader", "pos": [750, 450], "size": [280, 60], "flags": {}, "order": 0, "mode": 0, "inputs": [{"localized_name": "model_name", "name": "model_name", "type": "COMBO", "widget": {"name": "model_name"}, "link": 19}], "outputs": [{"localized_name": "UPSCALE_MODEL", "name": "UPSCALE_MODEL", "type": "UPSCALE_MODEL", "links": [1]}], "properties": {"cnr_id": "comfy-core", "ver": "0.10.0", "Node name for S&R": "UpscaleModelLoader", "models": [{"name": "RealESRGAN_x4plus.safetensors", "url": "https://huggingface.co/Comfy-Org/Real-ESRGAN_repackaged/resolve/main/RealESRGAN_x4plus.safetensors", "directory": "upscale_models"}]}, "widgets_values": ["RealESRGAN_x4plus.safetensors"]}], "groups": [], "links": [{"id": 1, "origin_id": 1, "origin_slot": 0, "target_id": 2, "target_slot": 0, "type": "UPSCALE_MODEL"}, {"id": 14, "origin_id": 10, "origin_slot": 0, "target_id": 2, "target_slot": 1, "type": "IMAGE"}, {"id": 13, "origin_id": 2, "origin_slot": 0, "target_id": 11, "target_slot": 0, "type": "IMAGE"}, {"id": 16, "origin_id": 10, "origin_slot": 1, "target_id": 11, "target_slot": 1, "type": "AUDIO"}, {"id": 12, "origin_id": 10, "origin_slot": 2, "target_id": 11, "target_slot": 2, "type": "FLOAT"}, {"id": 10, "origin_id": -10, "origin_slot": 0, "target_id": 10, "target_slot": 0, "type": "VIDEO"}, {"id": 15, "origin_id": 11, "origin_slot": 0, "target_id": -20, "target_slot": 0, "type": "VIDEO"}, {"id": 19, "origin_id": -10, "origin_slot": 1, "target_id": 1, "target_slot": 0, "type": "COMBO"}], "extra": {"workflowRendererVersion": "LG"}, "category": "Video generation and editing/Enhance video"}]}, "extra": {}}

comfy/comfy_types/node_typing.py

@@ -176,6 +176,8 @@ class InputTypeOptions(TypedDict):
     """COMBO type only. Specifies the configuration for a multi-select widget.
     Available after ComfyUI frontend v1.13.4
     https://github.com/Comfy-Org/ComfyUI_frontend/pull/2987"""
+    gradient_stops: NotRequired[list[list[float]]]
+    """Gradient color stops for gradientslider display mode. Each stop is [offset, r, g, b] (``FLOAT``)."""
 
 
 class HiddenInputTypeDict(TypedDict):

comfy/ldm/anima/model.py

@@ -179,8 +179,8 @@ class LLMAdapter(nn.Module):
             if source_attention_mask.ndim == 2:
                 source_attention_mask = source_attention_mask.unsqueeze(1).unsqueeze(1)
 
-        x = self.in_proj(self.embed(target_input_ids))
         context = source_hidden_states
+        x = self.in_proj(self.embed(target_input_ids, out_dtype=context.dtype))
         position_ids = torch.arange(x.shape[1], device=x.device).unsqueeze(0)
         position_ids_context = torch.arange(context.shape[1], device=x.device).unsqueeze(0)
         position_embeddings = self.rotary_emb(x, position_ids)

comfy/ldm/chroma/model.py

@@ -152,6 +152,7 @@ class Chroma(nn.Module):
         transformer_options={},
         attn_mask: Tensor = None,
     ) -> Tensor:
+        transformer_options = transformer_options.copy()
         patches_replace = transformer_options.get("patches_replace", {})
 
         # running on sequences img
@@ -228,6 +229,7 @@ class Chroma(nn.Module):
 
         transformer_options["total_blocks"] = len(self.single_blocks)
         transformer_options["block_type"] = "single"
+        transformer_options["img_slice"] = [txt.shape[1], img.shape[1]]
         for i, block in enumerate(self.single_blocks):
             transformer_options["block_index"] = i
             if i not in self.skip_dit:

comfy/ldm/flux/layers.py

@@ -196,6 +196,9 @@ class DoubleStreamBlock(nn.Module):
         else:
             (img_mod1, img_mod2), (txt_mod1, txt_mod2) = vec
 
+        transformer_patches = transformer_options.get("patches", {})
+        extra_options = transformer_options.copy()
+
         # prepare image for attention
         img_modulated = self.img_norm1(img)
         img_modulated = apply_mod(img_modulated, (1 + img_mod1.scale), img_mod1.shift, modulation_dims_img)
@@ -224,6 +227,12 @@ class DoubleStreamBlock(nn.Module):
         attn = attention(q, k, v, pe=pe, mask=attn_mask, transformer_options=transformer_options)
         del q, k, v
 
+        if "attn1_output_patch" in transformer_patches:
+            extra_options["img_slice"] = [txt.shape[1], attn.shape[1]]
+            patch = transformer_patches["attn1_output_patch"]
+            for p in patch:
+                attn = p(attn, extra_options)
+
         txt_attn, img_attn = attn[:, : txt.shape[1]], attn[:, txt.shape[1]:]
 
         # calculate the img bloks
@@ -303,6 +312,9 @@ class SingleStreamBlock(nn.Module):
         else:
             mod = vec
 
+        transformer_patches = transformer_options.get("patches", {})
+        extra_options = transformer_options.copy()
+
         qkv, mlp = torch.split(self.linear1(apply_mod(self.pre_norm(x), (1 + mod.scale), mod.shift, modulation_dims)), [3 * self.hidden_size, self.mlp_hidden_dim_first], dim=-1)
 
         q, k, v = qkv.view(qkv.shape[0], qkv.shape[1], 3, self.num_heads, -1).permute(2, 0, 3, 1, 4)
@@ -312,6 +324,12 @@ class SingleStreamBlock(nn.Module):
         # compute attention
         attn = attention(q, k, v, pe=pe, mask=attn_mask, transformer_options=transformer_options)
         del q, k, v
+
+        if "attn1_output_patch" in transformer_patches:
+            patch = transformer_patches["attn1_output_patch"]
+            for p in patch:
+                attn = p(attn, extra_options)
+
         # compute activation in mlp stream, cat again and run second linear layer
         if self.yak_mlp:
             mlp = self.mlp_act(mlp[..., self.mlp_hidden_dim_first // 2:]) * mlp[..., :self.mlp_hidden_dim_first // 2]

comfy/ldm/flux/model.py

@@ -142,6 +142,7 @@ class Flux(nn.Module):
         attn_mask: Tensor = None,
     ) -> Tensor:
 
+        transformer_options = transformer_options.copy()
         patches = transformer_options.get("patches", {})
         patches_replace = transformer_options.get("patches_replace", {})
         if img.ndim != 3 or txt.ndim != 3:
@@ -231,6 +232,7 @@ class Flux(nn.Module):
 
         transformer_options["total_blocks"] = len(self.single_blocks)
         transformer_options["block_type"] = "single"
+        transformer_options["img_slice"] = [txt.shape[1], img.shape[1]]
         for i, block in enumerate(self.single_blocks):
             transformer_options["block_index"] = i
             if ("single_block", i) in blocks_replace:

comfy/ldm/hunyuan_video/model.py

@@ -304,6 +304,7 @@ class HunyuanVideo(nn.Module):
         control=None,
         transformer_options={},
     ) -> Tensor:
+        transformer_options = transformer_options.copy()
         patches_replace = transformer_options.get("patches_replace", {})
 
         initial_shape = list(img.shape)
@@ -416,6 +417,7 @@ class HunyuanVideo(nn.Module):
 
         transformer_options["total_blocks"] = len(self.single_blocks)
         transformer_options["block_type"] = "single"
+        transformer_options["img_slice"] = [txt.shape[1], img.shape[1]]
         for i, block in enumerate(self.single_blocks):
             transformer_options["block_index"] = i
             if ("single_block", i) in blocks_replace:

comfy/ldm/modules/diffusionmodules/model.py

@@ -102,19 +102,7 @@ class VideoConv3d(nn.Module):
         return self.conv(x)
 
 def interpolate_up(x, scale_factor):
-    try:
-        return torch.nn.functional.interpolate(x, scale_factor=scale_factor, mode="nearest")
-    except: #operation not implemented for bf16
-        orig_shape = list(x.shape)
-        out_shape = orig_shape[:2]
-        for i in range(len(orig_shape) - 2):
-            out_shape.append(round(orig_shape[i + 2] * scale_factor[i]))
-        out = torch.empty(out_shape, dtype=x.dtype, layout=x.layout, device=x.device)
-        split = 8
-        l = out.shape[1] // split
-        for i in range(0, out.shape[1], l):
-            out[:,i:i+l] = torch.nn.functional.interpolate(x[:,i:i+l].to(torch.float32), scale_factor=scale_factor, mode="nearest").to(x.dtype)
-        return out
+    return torch.nn.functional.interpolate(x, scale_factor=scale_factor, mode="nearest")
 
 class Upsample(nn.Module):
     def __init__(self, in_channels, with_conv, conv_op=ops.Conv2d, scale_factor=2.0):

comfy/lora.py

@@ -374,6 +374,31 @@ def pad_tensor_to_shape(tensor: torch.Tensor, new_shape: list[int]) -> torch.Ten
 
     return padded_tensor
 
+def calculate_shape(patches, weight, key, original_weights=None):
+    current_shape = weight.shape
+
+    for p in patches:
+        v = p[1]
+        offset = p[3]
+
+        # Offsets restore the old shape; lists force a diff without metadata
+        if offset is not None or isinstance(v, list):
+            continue
+
+        if isinstance(v, weight_adapter.WeightAdapterBase):
+            adapter_shape = v.calculate_shape(key)
+            if adapter_shape is not None:
+                current_shape = adapter_shape
+            continue
+
+        # Standard diff logic with padding
+        if len(v) == 2:
+            patch_type, patch_data = v[0], v[1]
+            if patch_type == "diff" and len(patch_data) > 1 and patch_data[1]['pad_weight']:
+                current_shape = patch_data[0].shape
+
+    return current_shape
+
 def calculate_weight(patches, weight, key, intermediate_dtype=torch.float32, original_weights=None):
     for p in patches:
         strength = p[0]

comfy/model_base.py

@@ -178,10 +178,7 @@ class BaseModel(torch.nn.Module):
             xc = torch.cat([xc] + [comfy.model_management.cast_to_device(c_concat, xc.device, xc.dtype)], dim=1)
 
         context = c_crossattn
-        dtype = self.get_dtype()
-
-        if self.manual_cast_dtype is not None:
-            dtype = self.manual_cast_dtype
+        dtype = self.get_dtype_inference()
 
         xc = xc.to(dtype)
         device = xc.device
@@ -218,6 +215,13 @@ class BaseModel(torch.nn.Module):
     def get_dtype(self):
         return self.diffusion_model.dtype
 
+    def get_dtype_inference(self):
+        dtype = self.get_dtype()
+
+        if self.manual_cast_dtype is not None:
+            dtype = self.manual_cast_dtype
+        return dtype
+
     def encode_adm(self, **kwargs):
         return None
 
@@ -372,9 +376,7 @@ class BaseModel(torch.nn.Module):
                     input_shapes += shape
 
         if comfy.model_management.xformers_enabled() or comfy.model_management.pytorch_attention_flash_attention():
-            dtype = self.get_dtype()
-            if self.manual_cast_dtype is not None:
-                dtype = self.manual_cast_dtype
+            dtype = self.get_dtype_inference()
             #TODO: this needs to be tweaked
             area = sum(map(lambda input_shape: input_shape[0] * math.prod(input_shape[2:]), input_shapes))
             return (area * comfy.model_management.dtype_size(dtype) * 0.01 * self.memory_usage_factor) * (1024 * 1024)
@@ -1165,7 +1167,7 @@ class Anima(BaseModel):
                 t5xxl_ids = t5xxl_ids.unsqueeze(0)
 
                 if torch.is_inference_mode_enabled():  # if not we are training
-                    cross_attn = self.diffusion_model.preprocess_text_embeds(cross_attn.to(device=device, dtype=self.get_dtype()), t5xxl_ids.to(device=device), t5xxl_weights=t5xxl_weights.to(device=device, dtype=self.get_dtype()))
+                    cross_attn = self.diffusion_model.preprocess_text_embeds(cross_attn.to(device=device, dtype=self.get_dtype_inference()), t5xxl_ids.to(device=device), t5xxl_weights=t5xxl_weights.to(device=device, dtype=self.get_dtype_inference()))
                 else:
                     out['t5xxl_ids'] = comfy.conds.CONDRegular(t5xxl_ids)
                     out['t5xxl_weights'] = comfy.conds.CONDRegular(t5xxl_weights)

comfy/model_patcher.py

@@ -406,13 +406,16 @@ class ModelPatcher:
     def memory_required(self, input_shape):
         return self.model.memory_required(input_shape=input_shape)
 
+    def disable_model_cfg1_optimization(self):
+        self.model_options["disable_cfg1_optimization"] = True
+
     def set_model_sampler_cfg_function(self, sampler_cfg_function, disable_cfg1_optimization=False):
         if len(inspect.signature(sampler_cfg_function).parameters) == 3:
             self.model_options["sampler_cfg_function"] = lambda args: sampler_cfg_function(args["cond"], args["uncond"], args["cond_scale"]) #Old way
         else:
             self.model_options["sampler_cfg_function"] = sampler_cfg_function
         if disable_cfg1_optimization:
-            self.model_options["disable_cfg1_optimization"] = True
+            self.disable_model_cfg1_optimization()
 
     def set_model_sampler_post_cfg_function(self, post_cfg_function, disable_cfg1_optimization=False):
         self.model_options = set_model_options_post_cfg_function(self.model_options, post_cfg_function, disable_cfg1_optimization)
@@ -1514,8 +1517,10 @@ class ModelPatcherDynamic(ModelPatcher):
 
                     weight, _, _ = get_key_weight(self.model, key)
                     if weight is None:
-                        return 0
+                        return (False, 0)
                     if key in self.patches:
+                        if comfy.lora.calculate_shape(self.patches[key], weight, key) != weight.shape:
+                            return (True, 0)
                         setattr(m, param_key + "_lowvram_function", LowVramPatch(key, self.patches))
                         num_patches += 1
                     else:
@@ -1529,7 +1534,13 @@ class ModelPatcherDynamic(ModelPatcher):
                         model_dtype = getattr(m, param_key + "_comfy_model_dtype", None) or weight.dtype
                         weight._model_dtype = model_dtype
                         geometry = comfy.memory_management.TensorGeometry(shape=weight.shape, dtype=model_dtype)
-                    return comfy.memory_management.vram_aligned_size(geometry)
+                    return (False, comfy.memory_management.vram_aligned_size(geometry))
+
+                def force_load_param(self, param_key, device_to):
+                    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)
 
                 if hasattr(m, "comfy_cast_weights"):
                     m.comfy_cast_weights = True
@@ -1537,13 +1548,19 @@ class ModelPatcherDynamic(ModelPatcher):
                     m.seed_key = n
                     set_dirty(m, dirty)
 
-                    v_weight_size = 0
-                    v_weight_size += setup_param(self, m, n, "weight")
-                    v_weight_size += setup_param(self, m, n, "bias")
+                    force_load, v_weight_size = setup_param(self, m, n, "weight")
+                    force_load_bias, v_weight_bias = setup_param(self, m, n, "bias")
+                    force_load = force_load or force_load_bias
+                    v_weight_size += v_weight_bias
 
-                    if vbar is not None and not hasattr(m, "_v"):
-                        m._v = vbar.alloc(v_weight_size)
-                    allocated_size += v_weight_size
+                    if force_load:
+                        logging.info(f"Module {n} has resizing Lora - force loading")
+                        force_load_param(self, "weight", device_to)
+                        force_load_param(self, "bias", device_to)
+                    else:
+                        if vbar is not None and not hasattr(m, "_v"):
+                            m._v = vbar.alloc(v_weight_size)
+                        allocated_size += v_weight_size
 
                 else:
                     for param in params:
@@ -1606,6 +1623,11 @@ class ModelPatcherDynamic(ModelPatcher):
             for m in self.model.modules():
                 move_weight_functions(m, device_to)
 
+            keys = list(self.backup.keys())
+            for k in keys:
+                bk = self.backup[k]
+                comfy.utils.set_attr_param(self.model, k, bk.weight)
+
     def partially_load(self, device_to, extra_memory=0, force_patch_weights=False):
         assert not force_patch_weights #See above
         with self.use_ejected(skip_and_inject_on_exit_only=True):

comfy/ops.py

@@ -21,7 +21,6 @@ import logging
 import comfy.model_management
 from comfy.cli_args import args, PerformanceFeature, enables_dynamic_vram
 import comfy.float
-import comfy.rmsnorm
 import json
 import comfy.memory_management
 import comfy.pinned_memory
@@ -80,7 +79,7 @@ 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):
+def cast_bias_weight_with_vbar(s, dtype, device, bias_dtype, non_blocking, compute_dtype, want_requant):
     offload_stream = None
     xfer_dest = None
 
@@ -171,10 +170,10 @@ def cast_bias_weight_with_vbar(s, dtype, device, bias_dtype, non_blocking, compu
             #FIXME: this is not accurate, we need to be sensitive to the compute dtype
             x = lowvram_fn(x)
             if (isinstance(orig, QuantizedTensor) and
-                (orig.dtype == dtype and len(fns) == 0 or update_weight)):
+                (want_requant and len(fns) == 0 or update_weight)):
                 seed = comfy.utils.string_to_seed(s.seed_key)
                 y = QuantizedTensor.from_float(x, s.layout_type, scale="recalculate", stochastic_rounding=seed)
-                if orig.dtype == dtype and len(fns) == 0:
+                if want_requant and len(fns) == 0:
                     #The layer actually wants our freshly saved QT
                     x = y
             elif update_weight:
@@ -195,7 +194,7 @@ def cast_bias_weight_with_vbar(s, dtype, device, bias_dtype, non_blocking, compu
     return weight, bias, (offload_stream, device if signature is not None else None, None)
 
 
-def cast_bias_weight(s, input=None, dtype=None, device=None, bias_dtype=None, offloadable=False, compute_dtype=None):
+def cast_bias_weight(s, input=None, dtype=None, device=None, bias_dtype=None, offloadable=False, compute_dtype=None, want_requant=False):
     # NOTE: offloadable=False is a a legacy and if you are a custom node author reading this please pass
     # offloadable=True and call uncast_bias_weight() after your last usage of the weight/bias. This
     # will add async-offload support to your cast and improve performance.
@@ -213,7 +212,7 @@ def cast_bias_weight(s, input=None, dtype=None, device=None, bias_dtype=None, of
     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)
+        return cast_bias_weight_with_vbar(s, dtype, device, bias_dtype, non_blocking, compute_dtype, want_requant)
 
     if offloadable and (device != s.weight.device or
                         (s.bias is not None and device != s.bias.device)):
@@ -463,7 +462,7 @@ class disable_weight_init:
             else:
                 return super().forward(*args, **kwargs)
 
-    class RMSNorm(comfy.rmsnorm.RMSNorm, CastWeightBiasOp):
+    class RMSNorm(torch.nn.RMSNorm, CastWeightBiasOp):
         def reset_parameters(self):
             self.bias = None
             return None
@@ -475,8 +474,7 @@ class disable_weight_init:
                 weight = None
                 bias = None
                 offload_stream = None
-            x = comfy.rmsnorm.rms_norm(input, weight, self.eps)  # TODO: switch to commented out line when old torch is deprecated
-            # x = torch.nn.functional.rms_norm(input, self.normalized_shape, weight, self.eps)
+            x = torch.nn.functional.rms_norm(input, self.normalized_shape, weight, self.eps)
             uncast_bias_weight(self, weight, bias, offload_stream)
             return x
 
@@ -852,8 +850,8 @@ def mixed_precision_ops(quant_config={}, compute_dtype=torch.bfloat16, full_prec
             def _forward(self, input, weight, bias):
                 return torch.nn.functional.linear(input, weight, bias)
 
-            def forward_comfy_cast_weights(self, input, compute_dtype=None):
-                weight, bias, offload_stream = cast_bias_weight(self, input, offloadable=True, compute_dtype=compute_dtype)
+            def forward_comfy_cast_weights(self, input, compute_dtype=None, want_requant=False):
+                weight, bias, offload_stream = cast_bias_weight(self, input, offloadable=True, compute_dtype=compute_dtype, want_requant=want_requant)
                 x = self._forward(input, weight, bias)
                 uncast_bias_weight(self, weight, bias, offload_stream)
                 return x
@@ -883,8 +881,7 @@ def mixed_precision_ops(quant_config={}, compute_dtype=torch.bfloat16, full_prec
                             scale = comfy.model_management.cast_to_device(scale, input.device, None)
                         input = QuantizedTensor.from_float(input_reshaped, self.layout_type, scale=scale)
 
-
-                output = self.forward_comfy_cast_weights(input, compute_dtype)
+                output = self.forward_comfy_cast_weights(input, compute_dtype, want_requant=isinstance(input, QuantizedTensor))
 
                 # Reshape output back to 3D if input was 3D
                 if reshaped_3d:

comfy/rmsnorm.py

@@ -1,57 +1,10 @@
 import torch
 import comfy.model_management
-import numbers
-import logging
-
-RMSNorm = None
-
-try:
-    rms_norm_torch = torch.nn.functional.rms_norm
-    RMSNorm = torch.nn.RMSNorm
-except:
-    rms_norm_torch = None
-    logging.warning("Please update pytorch to use native RMSNorm")
 
+RMSNorm = torch.nn.RMSNorm
 
 def rms_norm(x, weight=None, eps=1e-6):
-    if rms_norm_torch is not None and not (torch.jit.is_tracing() or torch.jit.is_scripting()):
-        if weight is None:
-            return rms_norm_torch(x, (x.shape[-1],), eps=eps)
-        else:
-            return rms_norm_torch(x, weight.shape, weight=comfy.model_management.cast_to(weight, dtype=x.dtype, device=x.device), eps=eps)
+    if weight is None:
+        return torch.nn.functional.rms_norm(x, (x.shape[-1],), eps=eps)
     else:
-        r = x * torch.rsqrt(torch.mean(x**2, dim=-1, keepdim=True) + eps)
-        if weight is None:
-            return r
-        else:
-            return r * comfy.model_management.cast_to(weight, dtype=x.dtype, device=x.device)
-
-
-if RMSNorm is None:
-    class RMSNorm(torch.nn.Module):
-        def __init__(
-            self,
-            normalized_shape,
-            eps=1e-6,
-            elementwise_affine=True,
-            device=None,
-            dtype=None,
-        ):
-            factory_kwargs = {"device": device, "dtype": dtype}
-            super().__init__()
-            if isinstance(normalized_shape, numbers.Integral):
-                # mypy error: incompatible types in assignment
-                normalized_shape = (normalized_shape,)  # type: ignore[assignment]
-            self.normalized_shape = tuple(normalized_shape)  # type: ignore[arg-type]
-            self.eps = eps
-            self.elementwise_affine = elementwise_affine
-            if self.elementwise_affine:
-                self.weight = torch.nn.Parameter(
-                    torch.empty(self.normalized_shape, **factory_kwargs)
-                )
-            else:
-                self.register_parameter("weight", None)
-            self.bias = None
-
-        def forward(self, x):
-            return rms_norm(x, self.weight, self.eps)
+        return torch.nn.functional.rms_norm(x, weight.shape, weight=comfy.model_management.cast_to(weight, dtype=x.dtype, device=x.device), eps=eps)

comfy/sd.py

@@ -423,6 +423,17 @@ class CLIP:
     def get_key_patches(self):
         return self.patcher.get_key_patches()
 
+    def generate(self, tokens, do_sample=True, max_length=256, temperature=1.0, top_k=50, top_p=0.95, min_p=0.0, repetition_penalty=1.0, seed=None):
+        self.cond_stage_model.reset_clip_options()
+
+        self.load_model()
+        self.cond_stage_model.set_clip_options({"layer": None})
+        self.cond_stage_model.set_clip_options({"execution_device": self.patcher.load_device})
+        return self.cond_stage_model.generate(tokens, do_sample=do_sample, max_length=max_length, temperature=temperature, top_k=top_k, top_p=top_p, min_p=min_p, repetition_penalty=repetition_penalty, seed=seed)
+
+    def decode(self, token_ids, skip_special_tokens=True):
+        return self.tokenizer.decode(token_ids, skip_special_tokens=skip_special_tokens)
+
 class VAE:
     def __init__(self, sd=None, device=None, config=None, dtype=None, metadata=None):
         if 'decoder.up_blocks.0.resnets.0.norm1.weight' in sd.keys(): #diffusers format
@@ -1182,6 +1193,7 @@ class TEModel(Enum):
     JINA_CLIP_2 = 19
     QWEN3_8B = 20
     QWEN3_06B = 21
+    GEMMA_3_4B_VISION = 22
 
 
 def detect_te_model(sd):
@@ -1210,7 +1222,10 @@ def detect_te_model(sd):
         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:
-            return TEModel.GEMMA_3_4B
+            if 'vision_model.embeddings.patch_embedding.weight' in sd:
+                return TEModel.GEMMA_3_4B_VISION
+            else:
+                return TEModel.GEMMA_3_4B
         return TEModel.GEMMA_2_2B
     if 'model.layers.0.self_attn.k_proj.bias' in sd:
         weight = sd['model.layers.0.self_attn.k_proj.bias']
@@ -1270,6 +1285,8 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip
         else:
             if "text_projection" in clip_data[i]:
                 clip_data[i]["text_projection.weight"] = clip_data[i]["text_projection"].transpose(0, 1) #old models saved with the CLIPSave node
+        if "lm_head.weight" in clip_data[i]:
+            clip_data[i]["model.lm_head.weight"] = clip_data[i].pop("lm_head.weight") # prefix missing in some models
 
     tokenizer_data = {}
     clip_target = EmptyClass()
@@ -1335,6 +1352,14 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip
             clip_target.clip = comfy.text_encoders.lumina2.te(**llama_detect(clip_data), model_type="gemma3_4b")
             clip_target.tokenizer = comfy.text_encoders.lumina2.NTokenizer
             tokenizer_data["spiece_model"] = clip_data[0].get("spiece_model", None)
+        elif te_model == TEModel.GEMMA_3_4B_VISION:
+            clip_target.clip = comfy.text_encoders.lumina2.te(**llama_detect(clip_data), model_type="gemma3_4b_vision")
+            clip_target.tokenizer = comfy.text_encoders.lumina2.NTokenizer
+            tokenizer_data["spiece_model"] = clip_data[0].get("spiece_model", None)
+        elif te_model == TEModel.GEMMA_3_12B:
+            clip_target.clip = comfy.text_encoders.lt.gemma3_te(**llama_detect(clip_data))
+            clip_target.tokenizer = comfy.text_encoders.lt.Gemma3_12BTokenizer
+            tokenizer_data["spiece_model"] = clip_data[0].get("spiece_model", None)
         elif te_model == TEModel.LLAMA3_8:
             clip_target.clip = comfy.text_encoders.hidream.hidream_clip(**llama_detect(clip_data),
                                                                         clip_l=False, clip_g=False, t5=False, llama=True, dtype_t5=None)

comfy/sd1_clip.py

@@ -308,6 +308,15 @@ class SDClipModel(torch.nn.Module, ClipTokenWeightEncoder):
     def load_sd(self, sd):
         return self.transformer.load_state_dict(sd, strict=False, assign=getattr(self, "can_assign_sd", False))
 
+    def generate(self, tokens, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed):
+        if isinstance(tokens, dict):
+            tokens_only = next(iter(tokens.values())) # todo: get this better?
+        else:
+            tokens_only = tokens
+        tokens_only = [[t[0] for t in b] for b in tokens_only]
+        embeds = self.process_tokens(tokens_only, device=self.execution_device)[0]
+        return self.transformer.generate(embeds, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed)
+
 def parse_parentheses(string):
     result = []
     current_item = ""
@@ -564,6 +573,8 @@ class SDTokenizer:
         min_length = tokenizer_options.get("{}_min_length".format(self.embedding_key), self.min_length)
         min_padding = tokenizer_options.get("{}_min_padding".format(self.embedding_key), self.min_padding)
 
+        min_length = kwargs.get("min_length", min_length)
+
         text = escape_important(text)
         if kwargs.get("disable_weights", self.disable_weights):
             parsed_weights = [(text, 1.0)]
@@ -663,6 +674,9 @@ class SDTokenizer:
     def state_dict(self):
         return {}
 
+    def decode(self, token_ids, skip_special_tokens=True):
+        return self.tokenizer.decode(token_ids, skip_special_tokens=skip_special_tokens)
+
 class SD1Tokenizer:
     def __init__(self, embedding_directory=None, tokenizer_data={}, clip_name="l", tokenizer=SDTokenizer, name=None):
         if name is not None:
@@ -686,6 +700,9 @@ class SD1Tokenizer:
     def state_dict(self):
         return getattr(self, self.clip).state_dict()
 
+    def decode(self, token_ids, skip_special_tokens=True):
+        return getattr(self, self.clip).decode(token_ids, skip_special_tokens=skip_special_tokens)
+
 class SD1CheckpointClipModel(SDClipModel):
     def __init__(self, device="cpu", dtype=None, model_options={}):
         super().__init__(device=device, return_projected_pooled=False, dtype=dtype, model_options=model_options)
@@ -722,3 +739,6 @@ class SD1ClipModel(torch.nn.Module):
 
     def load_sd(self, sd):
         return getattr(self, self.clip).load_sd(sd)
+
+    def generate(self, tokens, do_sample=True, max_length=256, temperature=1.0, top_k=50, top_p=0.95, min_p=0.0, repetition_penalty=1.0, seed=None):
+        return getattr(self, self.clip).generate(tokens, do_sample=do_sample, max_length=max_length, temperature=temperature, top_k=top_k, top_p=top_p, min_p=min_p, repetition_penalty=repetition_penalty, seed=seed)

comfy/text_encoders/anima.py

@@ -33,6 +33,8 @@ class AnimaTokenizer:
     def state_dict(self):
         return {}
 
+    def decode(self, token_ids, **kwargs):
+        return self.qwen3_06b.decode(token_ids, **kwargs)
 
 class Qwen3_06BModel(sd1_clip.SDClipModel):
     def __init__(self, device="cpu", layer="last", layer_idx=None, dtype=None, attention_mask=True, model_options={}):

comfy/text_encoders/llama.py

@@ -3,6 +3,8 @@ import torch.nn as nn
 from dataclasses import dataclass
 from typing import Optional, Any, Tuple
 import math
+from tqdm import tqdm
+import comfy.utils
 
 from comfy.ldm.modules.attention import optimized_attention_for_device
 import comfy.model_management
@@ -103,6 +105,7 @@ class Qwen3_06BConfig:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
+    stop_tokens = [151643, 151645]
 
 @dataclass
 class Qwen3_06B_ACE15_Config:
@@ -126,6 +129,7 @@ class Qwen3_06B_ACE15_Config:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
+    stop_tokens = [151643, 151645]
 
 @dataclass
 class Qwen3_2B_ACE15_lm_Config:
@@ -149,6 +153,7 @@ class Qwen3_2B_ACE15_lm_Config:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
+    stop_tokens = [151643, 151645]
 
 @dataclass
 class Qwen3_4B_ACE15_lm_Config:
@@ -172,6 +177,7 @@ class Qwen3_4B_ACE15_lm_Config:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
+    stop_tokens = [151643, 151645]
 
 @dataclass
 class Qwen3_4BConfig:
@@ -195,6 +201,7 @@ class Qwen3_4BConfig:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
+    stop_tokens = [151643, 151645]
 
 @dataclass
 class Qwen3_8BConfig:
@@ -218,6 +225,7 @@ class Qwen3_8BConfig:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
+    stop_tokens = [151643, 151645]
 
 @dataclass
 class Ovis25_2BConfig:
@@ -288,6 +296,7 @@ class Gemma2_2B_Config:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
+    stop_tokens = [1]
 
 @dataclass
 class Gemma3_4B_Config:
@@ -312,6 +321,14 @@ class Gemma3_4B_Config:
     rope_scale = [8.0, 1.0]
     final_norm: bool = True
     lm_head: bool = False
+    stop_tokens = [1, 106]
+
+GEMMA3_VISION_CONFIG = {"num_channels": 3, "hidden_act": "gelu_pytorch_tanh", "hidden_size": 1152, "image_size": 896, "intermediate_size": 4304, "model_type": "siglip_vision_model", "num_attention_heads": 16, "num_hidden_layers": 27, "patch_size": 14}
+
+@dataclass
+class Gemma3_4B_Vision_Config(Gemma3_4B_Config):
+    vision_config = GEMMA3_VISION_CONFIG
+    mm_tokens_per_image = 256
 
 @dataclass
 class Gemma3_12B_Config:
@@ -336,8 +353,9 @@ class Gemma3_12B_Config:
     rope_scale = [8.0, 1.0]
     final_norm: bool = True
     lm_head: bool = False
-    vision_config = {"num_channels": 3, "hidden_act": "gelu_pytorch_tanh", "hidden_size": 1152, "image_size": 896, "intermediate_size": 4304, "model_type": "siglip_vision_model", "num_attention_heads": 16, "num_hidden_layers": 27, "patch_size": 14}
+    vision_config = GEMMA3_VISION_CONFIG
     mm_tokens_per_image = 256
+    stop_tokens = [1, 106]
 
 class RMSNorm(nn.Module):
     def __init__(self, dim: int, eps: float = 1e-5, add=False, device=None, dtype=None):
@@ -441,8 +459,10 @@ class Attention(nn.Module):
         freqs_cis: Optional[torch.Tensor] = None,
         optimized_attention=None,
         past_key_value: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
+        sliding_window: Optional[int] = None,
     ):
         batch_size, seq_length, _ = hidden_states.shape
+
         xq = self.q_proj(hidden_states)
         xk = self.k_proj(hidden_states)
         xv = self.v_proj(hidden_states)
@@ -477,6 +497,11 @@ 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:
+                xk = xk[:, :, -sliding_window:]
+                xv = xv[:, :, -sliding_window:]
+                attention_mask = attention_mask[..., -sliding_window:] if attention_mask is not None else None
+
         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)
 
@@ -559,10 +584,12 @@ class TransformerBlockGemma2(nn.Module):
         optimized_attention=None,
         past_key_value: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
     ):
+        sliding_window = None
         if self.transformer_type == 'gemma3':
             if self.sliding_attention:
+                sliding_window = self.sliding_attention
                 if x.shape[1] > self.sliding_attention:
-                    sliding_mask = torch.full((x.shape[1], x.shape[1]), float("-inf"), device=x.device, dtype=x.dtype)
+                    sliding_mask = torch.full((x.shape[1], x.shape[1]), torch.finfo(x.dtype).min, device=x.device, dtype=x.dtype)
                     sliding_mask.tril_(diagonal=-self.sliding_attention)
                     if attention_mask is not None:
                         attention_mask = attention_mask + sliding_mask
@@ -581,6 +608,7 @@ class TransformerBlockGemma2(nn.Module):
             freqs_cis=freqs_cis,
             optimized_attention=optimized_attention,
             past_key_value=past_key_value,
+            sliding_window=sliding_window,
         )
 
         x = self.post_attention_layernorm(x)
@@ -765,6 +793,107 @@ class BaseLlama:
     def forward(self, input_ids, *args, **kwargs):
         return self.model(input_ids, *args, **kwargs)
 
+class BaseGenerate:
+    def logits(self, x):
+        input = x[:, -1:]
+        if hasattr(self.model, "lm_head"):
+            module = self.model.lm_head
+        else:
+            module = self.model.embed_tokens
+
+        offload_stream = None
+        if module.comfy_cast_weights:
+            weight, _, offload_stream = comfy.ops.cast_bias_weight(module, input, offloadable=True)
+        else:
+            weight = self.model.embed_tokens.weight.to(x)
+
+        x = torch.nn.functional.linear(input, weight, None)
+
+        comfy.ops.uncast_bias_weight(module, weight, None, offload_stream)
+        return x
+
+    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):
+        device = embeds.device
+        model_config = self.model.config
+
+        if stop_tokens is None:
+            stop_tokens = self.model.config.stop_tokens
+
+        if execution_dtype is None:
+            if comfy.model_management.should_use_bf16(device):
+                execution_dtype = torch.bfloat16
+            else:
+                execution_dtype = torch.float32
+        embeds = embeds.to(execution_dtype)
+
+        if embeds.ndim == 2:
+            embeds = embeds.unsqueeze(0)
+
+        past_key_values = [] #kv_cache init
+        max_cache_len = embeds.shape[1] + max_length
+        for x in range(model_config.num_hidden_layers):
+            past_key_values.append((torch.empty([embeds.shape[0], model_config.num_key_value_heads, max_cache_len, model_config.head_dim], device=device, dtype=execution_dtype),
+                                    torch.empty([embeds.shape[0], model_config.num_key_value_heads, max_cache_len, model_config.head_dim], device=device, dtype=execution_dtype), 0))
+
+        generator = torch.Generator(device=device).manual_seed(seed) if do_sample else None
+
+        generated_token_ids = []
+        pbar = comfy.utils.ProgressBar(max_length)
+
+        # Generation loop
+        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)
+            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)
+            token_id = next_token[0].item()
+            generated_token_ids.append(token_id)
+
+            embeds = self.model.embed_tokens(next_token).to(execution_dtype)
+            pbar.update(1)
+
+            if token_id in stop_tokens:
+                break
+
+        return generated_token_ids
+
+    def sample_token(self, logits, temperature, top_k, top_p, min_p, repetition_penalty, token_history, generator, do_sample=True):
+
+        if not do_sample or temperature == 0.0:
+            return torch.argmax(logits, dim=-1, keepdim=True)
+
+        # Sampling mode
+        if repetition_penalty != 1.0:
+            for i in range(logits.shape[0]):
+                for token_id in set(token_history):
+                    logits[i, token_id] *= repetition_penalty if logits[i, token_id] < 0 else 1/repetition_penalty
+
+        if temperature != 1.0:
+            logits = logits / temperature
+
+        if top_k > 0:
+            indices_to_remove = logits < torch.topk(logits, top_k)[0][..., -1, None]
+            logits[indices_to_remove] = torch.finfo(logits.dtype).min
+
+        if min_p > 0.0:
+            probs_before_filter = torch.nn.functional.softmax(logits, dim=-1)
+            top_probs, _ = probs_before_filter.max(dim=-1, keepdim=True)
+            min_threshold = min_p * top_probs
+            indices_to_remove = probs_before_filter < min_threshold
+            logits[indices_to_remove] = torch.finfo(logits.dtype).min
+
+        if top_p < 1.0:
+            sorted_logits, sorted_indices = torch.sort(logits, descending=True)
+            cumulative_probs = torch.cumsum(torch.nn.functional.softmax(sorted_logits, dim=-1), dim=-1)
+            sorted_indices_to_remove = cumulative_probs > top_p
+            sorted_indices_to_remove[..., 0] = False
+            indices_to_remove = torch.zeros_like(logits, dtype=torch.bool)
+            indices_to_remove.scatter_(1, sorted_indices, sorted_indices_to_remove)
+            logits[indices_to_remove] = torch.finfo(logits.dtype).min
+
+        probs = torch.nn.functional.softmax(logits, dim=-1)
+
+        return torch.multinomial(probs, num_samples=1, generator=generator)
+
 class BaseQwen3:
     def logits(self, x):
         input = x[:, -1:]
@@ -808,7 +937,7 @@ class Qwen25_3B(BaseLlama, torch.nn.Module):
         self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
         self.dtype = dtype
 
-class Qwen3_06B(BaseLlama, BaseQwen3, torch.nn.Module):
+class Qwen3_06B(BaseLlama, BaseQwen3, BaseGenerate, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()
         config = Qwen3_06BConfig(**config_dict)
@@ -835,7 +964,7 @@ class Qwen3_2B_ACE15_lm(BaseLlama, BaseQwen3, torch.nn.Module):
         self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
         self.dtype = dtype
 
-class Qwen3_4B(BaseLlama, BaseQwen3, torch.nn.Module):
+class Qwen3_4B(BaseLlama, BaseQwen3, BaseGenerate, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()
         config = Qwen3_4BConfig(**config_dict)
@@ -853,7 +982,7 @@ class Qwen3_4B_ACE15_lm(BaseLlama, BaseQwen3, torch.nn.Module):
         self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
         self.dtype = dtype
 
-class Qwen3_8B(BaseLlama, BaseQwen3, torch.nn.Module):
+class Qwen3_8B(BaseLlama, BaseQwen3, BaseGenerate, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()
         config = Qwen3_8BConfig(**config_dict)
@@ -871,7 +1000,7 @@ class Ovis25_2B(BaseLlama, torch.nn.Module):
         self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
         self.dtype = dtype
 
-class Qwen25_7BVLI(BaseLlama, torch.nn.Module):
+class Qwen25_7BVLI(BaseLlama, BaseGenerate, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()
         config = Qwen25_7BVLI_Config(**config_dict)
@@ -881,6 +1010,9 @@ class Qwen25_7BVLI(BaseLlama, torch.nn.Module):
         self.visual = qwen_vl.Qwen2VLVisionTransformer(hidden_size=1280, output_hidden_size=config.hidden_size, device=device, dtype=dtype, ops=operations)
         self.dtype = dtype
 
+        # todo: should this be tied or not?
+        #self.lm_head = operations.Linear(config.hidden_size, config.vocab_size, bias=False, device=device, dtype=dtype)
+
     def preprocess_embed(self, embed, device):
         if embed["type"] == "image":
             image, grid = qwen_vl.process_qwen2vl_images(embed["data"])
@@ -914,7 +1046,7 @@ class Qwen25_7BVLI(BaseLlama, torch.nn.Module):
 
         return super().forward(x, attention_mask=attention_mask, embeds=embeds, num_tokens=num_tokens, intermediate_output=intermediate_output, final_layer_norm_intermediate=final_layer_norm_intermediate, dtype=dtype, position_ids=position_ids)
 
-class Gemma2_2B(BaseLlama, torch.nn.Module):
+class Gemma2_2B(BaseLlama, BaseGenerate, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()
         config = Gemma2_2B_Config(**config_dict)
@@ -923,7 +1055,7 @@ class Gemma2_2B(BaseLlama, torch.nn.Module):
         self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
         self.dtype = dtype
 
-class Gemma3_4B(BaseLlama, torch.nn.Module):
+class Gemma3_4B(BaseLlama, BaseGenerate, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()
         config = Gemma3_4B_Config(**config_dict)
@@ -932,7 +1064,25 @@ class Gemma3_4B(BaseLlama, torch.nn.Module):
         self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
         self.dtype = dtype
 
-class Gemma3_12B(BaseLlama, torch.nn.Module):
+class Gemma3_4B_Vision(BaseLlama, BaseGenerate, torch.nn.Module):
+    def __init__(self, config_dict, dtype, device, operations):
+        super().__init__()
+        config = Gemma3_4B_Vision_Config(**config_dict)
+        self.num_layers = config.num_hidden_layers
+
+        self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
+        self.dtype = dtype
+        self.multi_modal_projector = Gemma3MultiModalProjector(config, dtype, device, operations)
+        self.vision_model = comfy.clip_model.CLIPVision(config.vision_config, dtype, device, operations)
+        self.image_size = config.vision_config["image_size"]
+
+    def preprocess_embed(self, embed, device):
+        if embed["type"] == "image":
+            image = comfy.clip_model.clip_preprocess(embed["data"], size=self.image_size, mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5], crop=True)
+            return self.multi_modal_projector(self.vision_model(image.to(device, dtype=torch.float32))[0]), None
+        return None, None
+
+class Gemma3_12B(BaseLlama, BaseGenerate, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()
         config = Gemma3_12B_Config(**config_dict)

comfy/text_encoders/lt.py

@@ -6,6 +6,7 @@ import comfy.text_encoders.genmo
 from comfy.ldm.lightricks.embeddings_connector import Embeddings1DConnector
 import torch
 import comfy.utils
+import math
 
 class T5XXLTokenizer(sd1_clip.SDTokenizer):
     def __init__(self, embedding_directory=None, tokenizer_data={}):
@@ -22,40 +23,79 @@ def ltxv_te(*args, **kwargs):
     return comfy.text_encoders.genmo.mochi_te(*args, **kwargs)
 
 
-class Gemma3_12BTokenizer(sd1_clip.SDTokenizer):
-    def __init__(self, embedding_directory=None, tokenizer_data={}):
-        tokenizer = tokenizer_data.get("spiece_model", None)
-        super().__init__(tokenizer, pad_with_end=False, embedding_size=3840, embedding_key='gemma3_12b', tokenizer_class=SPieceTokenizer, has_end_token=False, pad_to_max_length=False, max_length=99999999, min_length=512, pad_left=True, disable_weights=True, tokenizer_args={"add_bos": True, "add_eos": False}, tokenizer_data=tokenizer_data)
-
+class Gemma3_Tokenizer():
     def state_dict(self):
         return {"spiece_model": self.tokenizer.serialize_model()}
 
+    def tokenize_with_weights(self, text, return_word_ids=False, image=None, llama_template=None, skip_template=True, **kwargs):
+        self.llama_template = "<start_of_turn>system\nYou are a helpful assistant.<end_of_turn>\n<start_of_turn>user\n{}<end_of_turn>\n<start_of_turn>model\n"
+        self.llama_template_images = "<start_of_turn>system\nYou are a helpful assistant.<end_of_turn>\n<start_of_turn>user\n\n<image_soft_token>{}<end_of_turn>\n\n<start_of_turn>model\n"
+
+        if image is None:
+            images = []
+        else:
+            samples = image.movedim(-1, 1)
+            total = int(896 * 896)
+
+            scale_by = math.sqrt(total / (samples.shape[3] * samples.shape[2]))
+            width = round(samples.shape[3] * scale_by)
+            height = round(samples.shape[2] * scale_by)
+
+            s = comfy.utils.common_upscale(samples, width, height, "area", "disabled").movedim(1, -1)
+            images = [s[:, :, :, :3]]
+
+        if text.startswith('<start_of_turn>'):
+            skip_template = True
+
+        if skip_template:
+            llama_text = text
+        else:
+            if llama_template is None:
+                if len(images) > 0:
+                    llama_text = self.llama_template_images.format(text)
+                else:
+                    llama_text = self.llama_template.format(text)
+            else:
+                llama_text = llama_template.format(text)
+
+        text_tokens = super().tokenize_with_weights(llama_text, return_word_ids)
+
+        if len(images) > 0:
+            embed_count = 0
+            for r in text_tokens:
+                for i, token in enumerate(r):
+                    if token[0] == 262144 and embed_count < len(images):
+                        r[i] = ({"type": "image", "data": images[embed_count]},) + token[1:]
+                        embed_count += 1
+        return text_tokens
+
+class Gemma3_12BTokenizer(Gemma3_Tokenizer, sd1_clip.SDTokenizer):
+    def __init__(self, embedding_directory=None, tokenizer_data={}):
+        tokenizer = tokenizer_data.get("spiece_model", None)
+        special_tokens = {"<image_soft_token>": 262144, "<end_of_turn>": 106}
+        super().__init__(tokenizer, pad_with_end=False, embedding_size=3840, embedding_key='gemma3_12b', tokenizer_class=SPieceTokenizer, has_end_token=False, pad_to_max_length=False, max_length=99999999, min_length=512, pad_left=True, disable_weights=True, tokenizer_args={"add_bos": True, "add_eos": False, "special_tokens": special_tokens}, tokenizer_data=tokenizer_data)
+
+
 class LTXAVGemmaTokenizer(sd1_clip.SD1Tokenizer):
     def __init__(self, embedding_directory=None, tokenizer_data={}):
         super().__init__(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data, name="gemma3_12b", tokenizer=Gemma3_12BTokenizer)
 
+
 class Gemma3_12BModel(sd1_clip.SDClipModel):
     def __init__(self, device="cpu", layer="all", layer_idx=None, dtype=None, attention_mask=True, model_options={}):
         llama_quantization_metadata = model_options.get("llama_quantization_metadata", None)
         if llama_quantization_metadata is not None:
             model_options = model_options.copy()
             model_options["quantization_metadata"] = llama_quantization_metadata
-
+        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=comfy.text_encoders.llama.Gemma3_12B, enable_attention_masks=attention_mask, return_attention_masks=attention_mask, model_options=model_options)
 
-    def tokenize_with_weights(self, text, return_word_ids=False, llama_template="{}", image_embeds=None, **kwargs):
-        text = llama_template.format(text)
-        text_tokens = super().tokenize_with_weights(text, return_word_ids)
-        embed_count = 0
-        for k in text_tokens:
-            tt = text_tokens[k]
-            for r in tt:
-                for i in range(len(r)):
-                    if r[i][0] == 262144:
-                        if image_embeds is not None and embed_count < image_embeds.shape[0]:
-                            r[i] = ({"type": "embedding", "data": image_embeds[embed_count], "original_type": "image"},) + r[i][1:]
-                            embed_count += 1
-        return text_tokens
+    def generate(self, tokens, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed):
+        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)
+        return self.transformer.generate(embeds, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed, stop_tokens=[106])  # 106 is <end_of_turn>
 
 class LTXAVTEModel(torch.nn.Module):
     def __init__(self, dtype_llama=None, device="cpu", dtype=None, model_options={}):
@@ -112,6 +152,9 @@ class LTXAVTEModel(torch.nn.Module):
 
         return out.to(out_device), pooled
 
+    def generate(self, tokens, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed):
+        return self.gemma3_12b.generate(tokens["gemma3_12b"], do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed)
+
     def load_sd(self, sd):
         if "model.layers.47.self_attn.q_norm.weight" in sd:
             return self.gemma3_12b.load_sd(sd)
@@ -152,3 +195,14 @@ def ltxav_te(dtype_llama=None, llama_quantization_metadata=None):
                 dtype = dtype_llama
             super().__init__(dtype_llama=dtype_llama, device=device, dtype=dtype, model_options=model_options)
     return LTXAVTEModel_
+
+def gemma3_te(dtype_llama=None, llama_quantization_metadata=None):
+    class Gemma3_12BModel_(Gemma3_12BModel):
+        def __init__(self, device="cpu", dtype=None, model_options={}):
+            if llama_quantization_metadata is not None:
+                model_options = model_options.copy()
+                model_options["llama_quantization_metadata"] = llama_quantization_metadata
+            if dtype_llama is not None:
+                dtype = dtype_llama
+            super().__init__(device=device, dtype=dtype, model_options=model_options)
+    return Gemma3_12BModel_

comfy/text_encoders/lumina2.py

@@ -1,23 +1,23 @@
 from comfy import sd1_clip
 from .spiece_tokenizer import SPieceTokenizer
 import comfy.text_encoders.llama
-
+from comfy.text_encoders.lt import Gemma3_Tokenizer
+import comfy.utils
 
 class Gemma2BTokenizer(sd1_clip.SDTokenizer):
     def __init__(self, embedding_directory=None, tokenizer_data={}):
         tokenizer = tokenizer_data.get("spiece_model", None)
-        super().__init__(tokenizer, pad_with_end=False, embedding_size=2304, embedding_key='gemma2_2b', tokenizer_class=SPieceTokenizer, has_end_token=False, pad_to_max_length=False, max_length=99999999, min_length=1, tokenizer_args={"add_bos": True, "add_eos": False}, tokenizer_data=tokenizer_data)
+        special_tokens = {"<end_of_turn>": 107}
+        super().__init__(tokenizer, pad_with_end=False, embedding_size=2304, embedding_key='gemma2_2b', tokenizer_class=SPieceTokenizer, has_end_token=False, pad_to_max_length=False, max_length=99999999, min_length=1, tokenizer_args={"add_bos": True, "add_eos": False, "special_tokens": special_tokens}, tokenizer_data=tokenizer_data)
 
     def state_dict(self):
         return {"spiece_model": self.tokenizer.serialize_model()}
 
-class Gemma3_4BTokenizer(sd1_clip.SDTokenizer):
+class Gemma3_4BTokenizer(Gemma3_Tokenizer, sd1_clip.SDTokenizer):
     def __init__(self, embedding_directory=None, tokenizer_data={}):
         tokenizer = tokenizer_data.get("spiece_model", None)
-        super().__init__(tokenizer, pad_with_end=False, embedding_size=2560, embedding_key='gemma3_4b', tokenizer_class=SPieceTokenizer, has_end_token=False, pad_to_max_length=False, max_length=99999999, min_length=1, tokenizer_args={"add_bos": True, "add_eos": False}, disable_weights=True, tokenizer_data=tokenizer_data)
-
-    def state_dict(self):
-        return {"spiece_model": self.tokenizer.serialize_model()}
+        special_tokens = {"<image_soft_token>": 262144, "<end_of_turn>": 106}
+        super().__init__(tokenizer, pad_with_end=False, embedding_size=2560, embedding_key='gemma3_4b', tokenizer_class=SPieceTokenizer, has_end_token=False, pad_to_max_length=False, max_length=99999999, min_length=1, tokenizer_args={"add_bos": True, "add_eos": False, "special_tokens": special_tokens}, disable_weights=True, tokenizer_data=tokenizer_data)
 
 class LuminaTokenizer(sd1_clip.SD1Tokenizer):
     def __init__(self, embedding_directory=None, tokenizer_data={}):
@@ -40,6 +40,20 @@ class Gemma3_4BModel(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, enable_attention_masks=attention_mask, return_attention_masks=attention_mask, model_options=model_options)
 
+class Gemma3_4B_Vision_Model(sd1_clip.SDClipModel):
+    def __init__(self, device="cpu", layer="hidden", layer_idx=-2, dtype=None, attention_mask=True, model_options={}):
+        llama_quantization_metadata = model_options.get("llama_quantization_metadata", None)
+        if llama_quantization_metadata is not None:
+            model_options = model_options.copy()
+            model_options["quantization_metadata"] = llama_quantization_metadata
+
+        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)
+        return embeds
+
 class LuminaModel(sd1_clip.SD1ClipModel):
     def __init__(self, device="cpu", dtype=None, model_options={}, name="gemma2_2b", clip_model=Gemma2_2BModel):
         super().__init__(device=device, dtype=dtype, name=name, clip_model=clip_model, model_options=model_options)
@@ -50,6 +64,8 @@ def te(dtype_llama=None, llama_quantization_metadata=None, model_type="gemma2_2b
         model = Gemma2_2BModel
     elif model_type == "gemma3_4b":
         model = Gemma3_4BModel
+    elif model_type == "gemma3_4b_vision":
+        model = Gemma3_4B_Vision_Model
 
     class LuminaTEModel_(LuminaModel):
         def __init__(self, device="cpu", dtype=None, model_options={}):

comfy/text_encoders/spiece_tokenizer.py

@@ -6,9 +6,10 @@ class SPieceTokenizer:
     def from_pretrained(path, **kwargs):
         return SPieceTokenizer(path, **kwargs)
 
-    def __init__(self, tokenizer_path, add_bos=False, add_eos=True):
+    def __init__(self, tokenizer_path, add_bos=False, add_eos=True, special_tokens=None):
         self.add_bos = add_bos
         self.add_eos = add_eos
+        self.special_tokens = special_tokens
         import sentencepiece
         if torch.is_tensor(tokenizer_path):
             tokenizer_path = tokenizer_path.numpy().tobytes()
@@ -27,8 +28,32 @@ class SPieceTokenizer:
         return out
 
     def __call__(self, string):
+        if self.special_tokens is not None:
+            import re
+            special_tokens_pattern = '|'.join(re.escape(token) for token in self.special_tokens.keys())
+            if special_tokens_pattern and re.search(special_tokens_pattern, string):
+                parts = re.split(f'({special_tokens_pattern})', string)
+                result = []
+                for part in parts:
+                    if not part:
+                        continue
+                    if part in self.special_tokens:
+                        result.append(self.special_tokens[part])
+                    else:
+                        encoded = self.tokenizer.encode(part, add_bos=False, add_eos=False)
+                        result.extend(encoded)
+                return {"input_ids": result}
+
         out = self.tokenizer.encode(string)
         return {"input_ids": out}
 
+    def decode(self, token_ids, skip_special_tokens=False):
+
+        if skip_special_tokens and self.special_tokens:
+            special_token_ids = set(self.special_tokens.values())
+            token_ids = [tid for tid in token_ids if tid not in special_token_ids]
+
+        return self.tokenizer.decode(token_ids)
+
     def serialize_model(self):
         return torch.ByteTensor(list(self.tokenizer.serialized_model_proto()))

comfy/utils.py

@@ -1418,3 +1418,11 @@ 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

comfy/weight_adapter/base.py

@@ -49,6 +49,12 @@ class WeightAdapterBase:
         """
         raise NotImplementedError
 
+    def calculate_shape(
+        self,
+        key
+    ):
+        return None
+
     def calculate_weight(
         self,
         weight,

comfy/weight_adapter/lora.py

@@ -214,6 +214,13 @@ class LoRAAdapter(WeightAdapterBase):
         else:
             return None
 
+    def calculate_shape(
+        self,
+        key
+    ):
+        reshape = self.weights[5]
+        return tuple(reshape) if reshape is not None else None
+
     def calculate_weight(
         self,
         weight,

comfy_api/feature_flags.py

@@ -14,6 +14,7 @@ SERVER_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}},
+    "node_replacements": True,
 }
 
 

comfy_api/latest/__init__.py

@@ -21,6 +21,17 @@ class ComfyAPI_latest(ComfyAPIBase):
     VERSION = "latest"
     STABLE = False
 
+    def __init__(self):
+        super().__init__()
+        self.node_replacement = self.NodeReplacement()
+        self.execution = self.Execution()
+
+    class NodeReplacement(ProxiedSingleton):
+        async def register(self, node_replace: io.NodeReplace) -> None:
+            """Register a node replacement mapping."""
+            from server import PromptServer
+            PromptServer.instance.node_replace_manager.register(node_replace)
+
     class Execution(ProxiedSingleton):
         async def set_progress(
             self,
@@ -73,8 +84,6 @@ class ComfyAPI_latest(ComfyAPIBase):
                 image=to_display,
             )
 
-    execution: Execution
-
 class ComfyExtension(ABC):
     async def on_load(self) -> None:
         """

comfy_api/latest/_input_impl/video_types.py

@@ -444,7 +444,7 @@ class VideoFromComponents(VideoInput):
             output.mux(packet)
 
             if audio_stream and self.__components.audio:
-                frame = av.AudioFrame.from_ndarray(waveform.float().cpu().numpy(), format='fltp', layout=layout)
+                frame = av.AudioFrame.from_ndarray(waveform.float().cpu().contiguous().numpy(), format='fltp', layout=layout)
                 frame.sample_rate = audio_sample_rate
                 frame.pts = 0
                 output.mux(audio_stream.encode(frame))

comfy_api/latest/_io.py

@@ -73,8 +73,15 @@ class RemoteOptions:
 class NumberDisplay(str, Enum):
     number = "number"
     slider = "slider"
+    gradient_slider = "gradientslider"
 
 
+class ControlAfterGenerate(str, Enum):
+    fixed = "fixed"
+    increment = "increment"
+    decrement = "decrement"
+    randomize = "randomize"
+
 class _ComfyType(ABC):
     Type = Any
     io_type: str = None
@@ -263,7 +270,7 @@ class Int(ComfyTypeIO):
     class Input(WidgetInput):
         '''Integer input.'''
         def __init__(self, id: str, display_name: str=None, optional=False, tooltip: str=None, lazy: bool=None,
-                    default: int=None, min: int=None, max: int=None, step: int=None, control_after_generate: bool=None,
+                    default: int=None, min: int=None, max: int=None, step: int=None, control_after_generate: bool | ControlAfterGenerate=None,
                     display_mode: NumberDisplay=None, socketless: bool=None, force_input: bool=None, extra_dict=None, raw_link: bool=None, advanced: bool=None):
             super().__init__(id, display_name, optional, tooltip, lazy, default, socketless, None, force_input, extra_dict, raw_link, advanced)
             self.min = min
@@ -290,13 +297,15 @@ class Float(ComfyTypeIO):
         '''Float input.'''
         def __init__(self, id: str, display_name: str=None, optional=False, tooltip: str=None, lazy: bool=None,
                     default: float=None, min: float=None, max: float=None, step: float=None, round: float=None,
-                    display_mode: NumberDisplay=None, socketless: bool=None, force_input: bool=None, extra_dict=None, raw_link: bool=None, advanced: bool=None):
+                    display_mode: NumberDisplay=None, gradient_stops: list[list[float]]=None,
+                    socketless: bool=None, force_input: bool=None, extra_dict=None, raw_link: bool=None, advanced: bool=None):
             super().__init__(id, display_name, optional, tooltip, lazy, default, socketless, None, force_input, extra_dict, raw_link, advanced)
             self.min = min
             self.max = max
             self.step = step
             self.round = round
             self.display_mode = display_mode
+            self.gradient_stops = gradient_stops
             self.default: float
 
         def as_dict(self):
@@ -306,6 +315,7 @@ class Float(ComfyTypeIO):
                 "step": self.step,
                 "round": self.round,
                 "display": self.display_mode,
+                "gradient_stops": self.gradient_stops,
             })
 
 @comfytype(io_type="STRING")
@@ -345,7 +355,7 @@ class Combo(ComfyTypeIO):
             tooltip: str=None,
             lazy: bool=None,
             default: str | int | Enum = None,
-            control_after_generate: bool=None,
+            control_after_generate: bool | ControlAfterGenerate=None,
             upload: UploadType=None,
             image_folder: FolderType=None,
             remote: RemoteOptions=None,
@@ -389,7 +399,7 @@ class MultiCombo(ComfyTypeI):
     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,
-                    default: list[str]=None, placeholder: str=None, chip: bool=None, control_after_generate: bool=None,
+                    default: list[str]=None, placeholder: str=None, chip: bool=None, control_after_generate: bool | ControlAfterGenerate=None,
                     socketless: bool=None, extra_dict=None, raw_link: bool=None, advanced: bool=None):
             super().__init__(id, options, display_name, optional, tooltip, lazy, default, control_after_generate, socketless=socketless, extra_dict=extra_dict, raw_link=raw_link, advanced=advanced)
             self.multiselect = True
@@ -1203,6 +1213,30 @@ class Color(ComfyTypeIO):
       def as_dict(self):
           return super().as_dict()
 
+@comfytype(io_type="BOUNDING_BOX")
+class BoundingBox(ComfyTypeIO):
+    class BoundingBoxDict(TypedDict):
+        x: int
+        y: int
+        width: int
+        height: int
+    Type = BoundingBoxDict
+
+    class Input(WidgetInput):
+        def __init__(self, id: str, display_name: str=None, optional=False, tooltip: str=None,
+                     socketless: bool=True, default: dict=None, component: str=None):
+            super().__init__(id, display_name, optional, tooltip, None, default, socketless)
+            self.component = component
+            if default is None:
+                self.default = {"x": 0, "y": 0, "width": 512, "height": 512}
+
+        def as_dict(self):
+            d = super().as_dict()
+            if self.component:
+                d["component"] = self.component
+            return d
+
+
 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
@@ -1309,6 +1343,7 @@ class NodeInfoV1:
     api_node: bool=None
     price_badge: dict | None = None
     search_aliases: list[str]=None
+    essentials_category: str=None
 
 
 @dataclass
@@ -1430,6 +1465,8 @@ class Schema:
     """Flags a node as expandable, allowing NodeOutput to include 'expand' property."""
     accept_all_inputs: bool=False
     """When True, all inputs from the prompt will be passed to the node as kwargs, even if not defined in the schema."""
+    essentials_category: str | None = None
+    """Optional category for the Essentials tab. Path-based like category field (e.g., 'Basic', 'Image Tools/Editing')."""
 
     def validate(self):
         '''Validate the schema:
@@ -1536,6 +1573,7 @@ class Schema:
             python_module=getattr(cls, "RELATIVE_PYTHON_MODULE", "nodes"),
             price_badge=self.price_badge.as_dict(self.inputs) if self.price_badge is not None else None,
             search_aliases=self.search_aliases if self.search_aliases else None,
+            essentials_category=self.essentials_category,
         )
         return info
 
@@ -2030,11 +2068,74 @@ class _UIOutput(ABC):
         ...
 
 
+class InputMapOldId(TypedDict):
+    """Map an old node input to a new node input by ID."""
+    new_id: str
+    old_id: str
+
+class InputMapSetValue(TypedDict):
+    """Set a specific value for a new node input."""
+    new_id: str
+    set_value: Any
+
+InputMap = InputMapOldId | InputMapSetValue
+"""
+Input mapping for node replacement. Type is inferred by dictionary keys:
+- {"new_id": str, "old_id": str} - maps old input to new input
+- {"new_id": str, "set_value": Any} - sets a specific value for new input
+"""
+
+class OutputMap(TypedDict):
+    """Map outputs of node replacement via indexes."""
+    new_idx: int
+    old_idx: int
+
+class NodeReplace:
+    """
+    Defines a possible node replacement, mapping inputs and outputs of the old node to the new node.
+
+    Also supports assigning specific values to the input widgets of the new node.
+
+    Args:
+        new_node_id: The class name of the new replacement node.
+        old_node_id: The class name of the deprecated node.
+        old_widget_ids: Ordered list of input IDs for widgets that may not have an input slot
+            connected. The workflow JSON stores widget values by their relative position index,
+            not by ID. This list maps those positional indexes to input IDs, enabling the
+            replacement system to correctly identify widget values during node migration.
+        input_mapping: List of input mappings from old node to new node.
+        output_mapping: List of output mappings from old node to new node.
+    """
+    def __init__(self,
+        new_node_id: str,
+        old_node_id: str,
+        old_widget_ids: list[str] | None=None,
+        input_mapping: list[InputMap] | None=None,
+        output_mapping: list[OutputMap] | None=None,
+    ):
+        self.new_node_id = new_node_id
+        self.old_node_id = old_node_id
+        self.old_widget_ids = old_widget_ids
+        self.input_mapping = input_mapping
+        self.output_mapping = output_mapping
+
+    def as_dict(self):
+        """Create serializable representation of the node replacement."""
+        return {
+            "new_node_id": self.new_node_id,
+            "old_node_id": self.old_node_id,
+            "old_widget_ids": self.old_widget_ids,
+            "input_mapping": list(self.input_mapping) if self.input_mapping else None,
+            "output_mapping": list(self.output_mapping) if self.output_mapping else None,
+        }
+
+
 __all__ = [
     "FolderType",
     "UploadType",
     "RemoteOptions",
     "NumberDisplay",
+    "ControlAfterGenerate",
 
     "comfytype",
     "Custom",
@@ -2121,4 +2222,6 @@ __all__ = [
     "ImageCompare",
     "PriceBadgeDepends",
     "PriceBadge",
+    "BoundingBox",
+    "NodeReplace",
 ]

comfy_api_nodes/apis/bria.py

@@ -45,17 +45,55 @@ class BriaEditImageRequest(BaseModel):
     )
 
 
+class BriaRemoveBackgroundRequest(BaseModel):
+    image: str = Field(...)
+    sync: bool = Field(False)
+    visual_input_content_moderation: bool = Field(
+        False, description="If true, returns 422 on input image moderation failure."
+    )
+    visual_output_content_moderation: bool = Field(
+        False, description="If true, returns 422 on visual output moderation failure."
+    )
+    seed: int = Field(...)
+
+
 class BriaStatusResponse(BaseModel):
     request_id: str = Field(...)
     status_url: str = Field(...)
     warning: str | None = Field(None)
 
 
-class BriaResult(BaseModel):
+class BriaRemoveBackgroundResult(BaseModel):
+    image_url: str = Field(...)
+
+
+class BriaRemoveBackgroundResponse(BaseModel):
+    status: str = Field(...)
+    result: BriaRemoveBackgroundResult | None = Field(None)
+
+
+class BriaImageEditResult(BaseModel):
     structured_prompt: str = Field(...)
     image_url: str = Field(...)
 
 
-class BriaResponse(BaseModel):
+class BriaImageEditResponse(BaseModel):
     status: str = Field(...)
-    result: BriaResult | None = Field(None)
+    result: BriaImageEditResult | None = Field(None)
+
+
+class BriaRemoveVideoBackgroundRequest(BaseModel):
+    video: str = Field(...)
+    background_color: str = Field(default="transparent", description="Background color for the output video.")
+    output_container_and_codec: str = Field(...)
+    preserve_audio: bool = Field(True)
+    seed: int = Field(...)
+
+
+class BriaRemoveVideoBackgroundResult(BaseModel):
+    video_url: str = Field(...)
+
+
+class BriaRemoveVideoBackgroundResponse(BaseModel):
+    status: str = Field(...)
+    result: BriaRemoveVideoBackgroundResult | None = Field(None)

comfy_api_nodes/apis/elevenlabs.py

@@ -0,0 +1,88 @@
+from pydantic import BaseModel, Field
+
+
+class SpeechToTextRequest(BaseModel):
+    model_id: str = Field(...)
+    cloud_storage_url: str = Field(...)
+    language_code: str | None = Field(None, description="ISO-639-1 or ISO-639-3 language code")
+    tag_audio_events: bool | None = Field(None, description="Annotate sounds like (laughter) in transcript")
+    num_speakers: int | None = Field(None, description="Max speakers predicted")
+    timestamps_granularity: str = Field(default="word", description="Timing precision: none, word, or character")
+    diarize: bool | None = Field(None, description="Annotate which speaker is talking")
+    diarization_threshold: float | None = Field(None, description="Speaker separation sensitivity")
+    temperature: float | None = Field(None, description="Randomness control")
+    seed: int = Field(..., description="Seed for deterministic sampling")
+
+
+class SpeechToTextWord(BaseModel):
+    text: str = Field(..., description="The word text")
+    type: str = Field(default="word", description="Type of text element (word, spacing, etc.)")
+    start: float | None = Field(None, description="Start time in seconds (when timestamps enabled)")
+    end: float | None = Field(None, description="End time in seconds (when timestamps enabled)")
+    speaker_id: str | None = Field(None, description="Speaker identifier when diarization is enabled")
+    logprob: float | None = Field(None, description="Log probability of the word")
+
+
+class SpeechToTextResponse(BaseModel):
+    language_code: str = Field(..., description="Detected or specified language code")
+    language_probability: float | None = Field(None, description="Confidence of language detection")
+    text: str = Field(..., description="Full transcript text")
+    words: list[SpeechToTextWord] | None = Field(None, description="Word-level timing information")
+
+
+class TextToSpeechVoiceSettings(BaseModel):
+    stability: float | None = Field(None, description="Voice stability")
+    similarity_boost: float | None = Field(None, description="Similarity boost")
+    style: float | None = Field(None, description="Style exaggeration")
+    use_speaker_boost: bool | None = Field(None, description="Boost similarity to original speaker")
+    speed: float | None = Field(None, description="Speech speed")
+
+
+class TextToSpeechRequest(BaseModel):
+    text: str = Field(..., description="Text to convert to speech")
+    model_id: str = Field(..., description="Model ID for TTS")
+    language_code: str | None = Field(None, description="ISO-639-1 or ISO-639-3 language code")
+    voice_settings: TextToSpeechVoiceSettings | None = Field(None, description="Voice settings")
+    seed: int = Field(..., description="Seed for deterministic sampling")
+    apply_text_normalization: str | None = Field(None, description="Text normalization mode: auto, on, off")
+
+
+class TextToSoundEffectsRequest(BaseModel):
+    text: str = Field(..., description="Text prompt to convert into a sound effect")
+    duration_seconds: float = Field(..., description="Duration of generated sound in seconds")
+    prompt_influence: float = Field(..., description="How closely generation follows the prompt")
+    loop: bool | None = Field(None, description="Whether to create a smoothly looping sound effect")
+
+
+class AddVoiceRequest(BaseModel):
+    name: str = Field(..., description="Name that identifies the voice")
+    remove_background_noise: bool = Field(..., description="Remove background noise from voice samples")
+
+
+class AddVoiceResponse(BaseModel):
+    voice_id: str = Field(..., description="The newly created voice's unique identifier")
+
+
+class SpeechToSpeechRequest(BaseModel):
+    model_id: str = Field(..., description="Model ID for speech-to-speech")
+    voice_settings: str = Field(..., description="JSON string of voice settings")
+    seed: int = Field(..., description="Seed for deterministic sampling")
+    remove_background_noise: bool = Field(..., description="Remove background noise from input audio")
+
+
+class DialogueInput(BaseModel):
+    text: str = Field(..., description="Text content to convert to speech")
+    voice_id: str = Field(..., description="Voice identifier for this dialogue segment")
+
+
+class DialogueSettings(BaseModel):
+    stability: float | None = Field(None, description="Voice stability (0-1)")
+
+
+class TextToDialogueRequest(BaseModel):
+    inputs: list[DialogueInput] = Field(..., description="List of dialogue segments")
+    model_id: str = Field(..., description="Model ID for dialogue generation")
+    language_code: str | None = Field(None, description="ISO-639-1 language code")
+    settings: DialogueSettings | None = Field(None, description="Voice settings")
+    seed: int | None = Field(None, description="Seed for deterministic sampling")
+    apply_text_normalization: str | None = Field(None, description="Text normalization mode: auto, on, off")

comfy_api_nodes/apis/gemini.py

@@ -116,9 +116,15 @@ class GeminiGenerationConfig(BaseModel):
     topP: float | None = Field(None, ge=0.0, le=1.0)
 
 
+class GeminiImageOutputOptions(BaseModel):
+    mimeType: str = Field("image/png")
+    compressionQuality: int | None = Field(None)
+
+
 class GeminiImageConfig(BaseModel):
     aspectRatio: str | None = Field(None)
     imageSize: str | None = Field(None)
+    imageOutputOptions: GeminiImageOutputOptions = Field(default_factory=GeminiImageOutputOptions)
 
 
 class GeminiImageGenerationConfig(GeminiGenerationConfig):

comfy_api_nodes/apis/hunyuan3d.py

@@ -64,3 +64,23 @@ class To3DProTaskResultResponse(BaseModel):
 
 class To3DProTaskQueryRequest(BaseModel):
     JobId: str = Field(...)
+
+
+class To3DUVFileInput(BaseModel):
+    Type: str = Field(..., description="File type: GLB, OBJ, or FBX")
+    Url: str = Field(...)
+
+
+class To3DUVTaskRequest(BaseModel):
+    File: To3DUVFileInput = Field(...)
+
+
+class TextureEditImageInfo(BaseModel):
+    Url: str = Field(...)
+
+
+class TextureEditTaskRequest(BaseModel):
+    File3D: To3DUVFileInput = Field(...)
+    Image: TextureEditImageInfo | None = Field(None)
+    Prompt: str | None = Field(None)
+    EnablePBR: bool | None = Field(None)

comfy_api_nodes/apis/recraft.py

@@ -198,11 +198,6 @@ dict_recraft_substyles_v3 = {
 }
 
 
-class RecraftModel(str, Enum):
-    recraftv3 = 'recraftv3'
-    recraftv2 = 'recraftv2'
-
-
 class RecraftImageSize(str, Enum):
     res_1024x1024 = '1024x1024'
     res_1365x1024 = '1365x1024'
@@ -221,6 +216,41 @@ class RecraftImageSize(str, Enum):
     res_1707x1024 = '1707x1024'
 
 
+RECRAFT_V4_SIZES = [
+    "1024x1024",
+    "1536x768",
+    "768x1536",
+    "1280x832",
+    "832x1280",
+    "1216x896",
+    "896x1216",
+    "1152x896",
+    "896x1152",
+    "832x1344",
+    "1280x896",
+    "896x1280",
+    "1344x768",
+    "768x1344",
+]
+
+RECRAFT_V4_PRO_SIZES = [
+    "2048x2048",
+    "3072x1536",
+    "1536x3072",
+    "2560x1664",
+    "1664x2560",
+    "2432x1792",
+    "1792x2432",
+    "2304x1792",
+    "1792x2304",
+    "1664x2688",
+    "1434x1024",
+    "1024x1434",
+    "2560x1792",
+    "1792x2560",
+]
+
+
 class RecraftColorObject(BaseModel):
     rgb: list[int] = Field(..., description='An array of 3 integer values in range of 0...255 defining RGB Color Model')
 
@@ -234,17 +264,16 @@ class RecraftControlsObject(BaseModel):
 
 class RecraftImageGenerationRequest(BaseModel):
     prompt: str = Field(..., description='The text prompt describing the image to generate')
-    size: RecraftImageSize | None = Field(None, description='The size of the generated image (e.g., "1024x1024")')
+    size: str | None = Field(None, description='The size of the generated image (e.g., "1024x1024")')
     n: int = Field(..., description='The number of images to generate')
     negative_prompt: str | None = Field(None, description='A text description of undesired elements on an image')
-    model: RecraftModel | None = Field(RecraftModel.recraftv3, description='The model to use for generation (e.g., "recraftv3")')
+    model: str = Field(...)
     style: str | None = Field(None, description='The style to apply to the generated image (e.g., "digital_illustration")')
     substyle: str | None = Field(None, description='The substyle to apply to the generated image, depending on the style input')
     controls: RecraftControlsObject | None = Field(None, description='A set of custom parameters to tweak generation process')
     style_id: str | None = Field(None, description='Use a previously uploaded style as a reference; UUID')
     strength: float | None = Field(None, description='Defines the difference with the original image, should lie in [0, 1], where 0 means almost identical, and 1 means miserable similarity')
     random_seed: int | None = Field(None, description="Seed for video generation")
-    # text_layout
 
 
 class RecraftReturnedObject(BaseModel):

comfy_api_nodes/nodes_bfl.py

@@ -57,6 +57,7 @@ class FluxProUltraImageNode(IO.ComfyNode):
                     tooltip="Whether to perform upsampling on the prompt. "
                     "If active, automatically modifies the prompt for more creative generation, "
                     "but results are nondeterministic (same seed will not produce exactly the same result).",
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -200,6 +201,7 @@ class FluxKontextProImageNode(IO.ComfyNode):
                     "prompt_upsampling",
                     default=False,
                     tooltip="Whether to perform upsampling on the prompt. If active, automatically modifies the prompt for more creative generation, but results are nondeterministic (same seed will not produce exactly the same result).",
+                    advanced=True,
                 ),
                 IO.Image.Input(
                     "input_image",
@@ -296,6 +298,7 @@ class FluxProExpandNode(IO.ComfyNode):
                     tooltip="Whether to perform upsampling on the prompt. "
                     "If active, automatically modifies the prompt for more creative generation, "
                     "but results are nondeterministic (same seed will not produce exactly the same result).",
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "top",
@@ -433,6 +436,7 @@ class FluxProFillNode(IO.ComfyNode):
                     tooltip="Whether to perform upsampling on the prompt. "
                     "If active, automatically modifies the prompt for more creative generation, "
                     "but results are nondeterministic (same seed will not produce exactly the same result).",
+                    advanced=True,
                 ),
                 IO.Float.Input(
                     "guidance",
@@ -577,6 +581,7 @@ class Flux2ProImageNode(IO.ComfyNode):
                     default=True,
                     tooltip="Whether to perform upsampling on the prompt. "
                     "If active, automatically modifies the prompt for more creative generation.",
+                    advanced=True,
                 ),
                 IO.Image.Input("images", optional=True, tooltip="Up to 9 images to be used as references."),
             ],

comfy_api_nodes/nodes_bria.py

@@ -3,7 +3,11 @@ from typing_extensions import override
 from comfy_api.latest import IO, ComfyExtension, Input
 from comfy_api_nodes.apis.bria import (
     BriaEditImageRequest,
-    BriaResponse,
+    BriaRemoveBackgroundRequest,
+    BriaRemoveBackgroundResponse,
+    BriaRemoveVideoBackgroundRequest,
+    BriaRemoveVideoBackgroundResponse,
+    BriaImageEditResponse,
     BriaStatusResponse,
     InputModerationSettings,
 )
@@ -11,10 +15,12 @@ from comfy_api_nodes.util import (
     ApiEndpoint,
     convert_mask_to_image,
     download_url_to_image_tensor,
-    get_number_of_images,
+    download_url_to_video_output,
     poll_op,
     sync_op,
-    upload_images_to_comfyapi,
+    upload_image_to_comfyapi,
+    upload_video_to_comfyapi,
+    validate_video_duration,
 )
 
 
@@ -73,21 +79,15 @@ class BriaImageEditNode(IO.ComfyNode):
                 IO.DynamicCombo.Input(
                     "moderation",
                     options=[
+                        IO.DynamicCombo.Option("false", []),
                         IO.DynamicCombo.Option(
                             "true",
                             [
-                                IO.Boolean.Input(
-                                    "prompt_content_moderation", default=False
-                                ),
-                                IO.Boolean.Input(
-                                    "visual_input_moderation", default=False
-                                ),
-                                IO.Boolean.Input(
-                                    "visual_output_moderation", default=True
-                                ),
+                                IO.Boolean.Input("prompt_content_moderation", default=False),
+                                IO.Boolean.Input("visual_input_moderation", default=False),
+                                IO.Boolean.Input("visual_output_moderation", default=True),
                             ],
                         ),
-                        IO.DynamicCombo.Option("false", []),
                     ],
                     tooltip="Moderation settings",
                 ),
@@ -127,50 +127,26 @@ class BriaImageEditNode(IO.ComfyNode):
         mask: Input.Image | None = None,
     ) -> IO.NodeOutput:
         if not prompt and not structured_prompt:
-            raise ValueError(
-                "One of prompt or structured_prompt is required to be non-empty."
-            )
-        if get_number_of_images(image) != 1:
-            raise ValueError("Exactly one input image is required.")
+            raise ValueError("One of prompt or structured_prompt is required to be non-empty.")
         mask_url = None
         if mask is not None:
-            mask_url = (
-                await upload_images_to_comfyapi(
-                    cls,
-                    convert_mask_to_image(mask),
-                    max_images=1,
-                    mime_type="image/png",
-                    wait_label="Uploading mask",
-                )
-            )[0]
+            mask_url = await upload_image_to_comfyapi(cls, convert_mask_to_image(mask), wait_label="Uploading mask")
         response = await sync_op(
             cls,
             ApiEndpoint(path="proxy/bria/v2/image/edit", method="POST"),
             data=BriaEditImageRequest(
                 instruction=prompt if prompt else None,
                 structured_instruction=structured_prompt if structured_prompt else None,
-                images=await upload_images_to_comfyapi(
-                    cls,
-                    image,
-                    max_images=1,
-                    mime_type="image/png",
-                    wait_label="Uploading image",
-                ),
+                images=[await upload_image_to_comfyapi(cls, image, wait_label="Uploading image")],
                 mask=mask_url,
                 negative_prompt=negative_prompt if negative_prompt else None,
                 guidance_scale=guidance_scale,
                 seed=seed,
                 model_version=model,
                 steps_num=steps,
-                prompt_content_moderation=moderation.get(
-                    "prompt_content_moderation", False
-                ),
-                visual_input_content_moderation=moderation.get(
-                    "visual_input_moderation", False
-                ),
-                visual_output_content_moderation=moderation.get(
-                    "visual_output_moderation", False
-                ),
+                prompt_content_moderation=moderation.get("prompt_content_moderation", False),
+                visual_input_content_moderation=moderation.get("visual_input_moderation", False),
+                visual_output_content_moderation=moderation.get("visual_output_moderation", False),
             ),
             response_model=BriaStatusResponse,
         )
@@ -178,7 +154,7 @@ class BriaImageEditNode(IO.ComfyNode):
             cls,
             ApiEndpoint(path=f"/proxy/bria/v2/status/{response.request_id}"),
             status_extractor=lambda r: r.status,
-            response_model=BriaResponse,
+            response_model=BriaImageEditResponse,
         )
         return IO.NodeOutput(
             await download_url_to_image_tensor(response.result.image_url),
@@ -186,11 +162,167 @@ class BriaImageEditNode(IO.ComfyNode):
         )
 
 
+class BriaRemoveImageBackground(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls):
+        return IO.Schema(
+            node_id="BriaRemoveImageBackground",
+            display_name="Bria Remove Image Background",
+            category="api node/image/Bria",
+            description="Remove the background from an image using Bria RMBG 2.0.",
+            inputs=[
+                IO.Image.Input("image"),
+                IO.DynamicCombo.Input(
+                    "moderation",
+                    options=[
+                        IO.DynamicCombo.Option("false", []),
+                        IO.DynamicCombo.Option(
+                            "true",
+                            [
+                                IO.Boolean.Input("visual_input_moderation", default=False),
+                                IO.Boolean.Input("visual_output_moderation", default=True),
+                            ],
+                        ),
+                    ],
+                    tooltip="Moderation settings",
+                ),
+                IO.Int.Input(
+                    "seed",
+                    default=0,
+                    min=0,
+                    max=2147483647,
+                    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.",
+                ),
+            ],
+            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(
+                expr="""{"type":"usd","usd":0.018}""",
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        image: Input.Image,
+        moderation: dict,
+        seed: int,
+    ) -> IO.NodeOutput:
+        response = await sync_op(
+            cls,
+            ApiEndpoint(path="/proxy/bria/v2/image/edit/remove_background", method="POST"),
+            data=BriaRemoveBackgroundRequest(
+                image=await upload_image_to_comfyapi(cls, image, wait_label="Uploading image"),
+                sync=False,
+                visual_input_content_moderation=moderation.get("visual_input_moderation", False),
+                visual_output_content_moderation=moderation.get("visual_output_moderation", False),
+                seed=seed,
+            ),
+            response_model=BriaStatusResponse,
+        )
+        response = await poll_op(
+            cls,
+            ApiEndpoint(path=f"/proxy/bria/v2/status/{response.request_id}"),
+            status_extractor=lambda r: r.status,
+            response_model=BriaRemoveBackgroundResponse,
+        )
+        return IO.NodeOutput(await download_url_to_image_tensor(response.result.image_url))
+
+
+class BriaRemoveVideoBackground(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls):
+        return IO.Schema(
+            node_id="BriaRemoveVideoBackground",
+            display_name="Bria Remove Video Background",
+            category="api node/video/Bria",
+            description="Remove the background from a video using Bria. ",
+            inputs=[
+                IO.Video.Input("video"),
+                IO.Combo.Input(
+                    "background_color",
+                    options=[
+                        "Black",
+                        "White",
+                        "Gray",
+                        "Red",
+                        "Green",
+                        "Blue",
+                        "Yellow",
+                        "Cyan",
+                        "Magenta",
+                        "Orange",
+                    ],
+                    tooltip="Background color for the output video.",
+                ),
+                IO.Int.Input(
+                    "seed",
+                    default=0,
+                    min=0,
+                    max=2147483647,
+                    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.",
+                ),
+            ],
+            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(
+                expr="""{"type":"usd","usd":0.14,"format":{"suffix":"/second"}}""",
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        video: Input.Video,
+        background_color: str,
+        seed: int,
+    ) -> IO.NodeOutput:
+        validate_video_duration(video, max_duration=60.0)
+        response = await sync_op(
+            cls,
+            ApiEndpoint(path="/proxy/bria/v2/video/edit/remove_background", method="POST"),
+            data=BriaRemoveVideoBackgroundRequest(
+                video=await upload_video_to_comfyapi(cls, video),
+                background_color=background_color,
+                output_container_and_codec="mp4_h264",
+                seed=seed,
+            ),
+            response_model=BriaStatusResponse,
+        )
+        response = await poll_op(
+            cls,
+            ApiEndpoint(path=f"/proxy/bria/v2/status/{response.request_id}"),
+            status_extractor=lambda r: r.status,
+            response_model=BriaRemoveVideoBackgroundResponse,
+        )
+        return IO.NodeOutput(await download_url_to_video_output(response.result.video_url))
+
+
 class BriaExtension(ComfyExtension):
     @override
     async def get_node_list(self) -> list[type[IO.ComfyNode]]:
         return [
             BriaImageEditNode,
+            BriaRemoveImageBackground,
+            BriaRemoveVideoBackground,
         ]
 
 

comfy_api_nodes/nodes_bytedance.py

@@ -114,6 +114,7 @@ class ByteDanceImageNode(IO.ComfyNode):
                     default=False,
                     tooltip='Whether to add an "AI generated" watermark to the image',
                     optional=True,
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -259,12 +260,14 @@ class ByteDanceSeedreamNode(IO.ComfyNode):
                     default=False,
                     tooltip='Whether to add an "AI generated" watermark to the image.',
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "fail_on_partial",
                     default=True,
                     tooltip="If enabled, abort execution if any requested images are missing or return an error.",
                     optional=True,
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -432,18 +435,21 @@ class ByteDanceTextToVideoNode(IO.ComfyNode):
                     tooltip="Specifies whether to fix the camera. The platform appends an instruction "
                     "to fix the camera to your prompt, but does not guarantee the actual effect.",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "watermark",
                     default=False,
                     tooltip='Whether to add an "AI generated" watermark to the video.',
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "generate_audio",
                     default=False,
                     tooltip="This parameter is ignored for any model except seedance-1-5-pro.",
                     optional=True,
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -561,18 +567,21 @@ class ByteDanceImageToVideoNode(IO.ComfyNode):
                     tooltip="Specifies whether to fix the camera. The platform appends an instruction "
                     "to fix the camera to your prompt, but does not guarantee the actual effect.",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "watermark",
                     default=False,
                     tooltip='Whether to add an "AI generated" watermark to the video.',
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "generate_audio",
                     default=False,
                     tooltip="This parameter is ignored for any model except seedance-1-5-pro.",
                     optional=True,
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -694,18 +703,21 @@ class ByteDanceFirstLastFrameNode(IO.ComfyNode):
                     tooltip="Specifies whether to fix the camera. The platform appends an instruction "
                     "to fix the camera to your prompt, but does not guarantee the actual effect.",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "watermark",
                     default=False,
                     tooltip='Whether to add an "AI generated" watermark to the video.',
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "generate_audio",
                     default=False,
                     tooltip="This parameter is ignored for any model except seedance-1-5-pro.",
                     optional=True,
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -834,6 +846,7 @@ class ByteDanceImageReferenceNode(IO.ComfyNode):
                     default=False,
                     tooltip='Whether to add an "AI generated" watermark to the video.',
                     optional=True,
+                    advanced=True,
                 ),
             ],
             outputs=[

comfy_api_nodes/nodes_elevenlabs.py

@@ -0,0 +1,924 @@
+import json
+import uuid
+
+from typing_extensions import override
+
+from comfy_api.latest import IO, ComfyExtension, Input
+from comfy_api_nodes.apis.elevenlabs import (
+    AddVoiceRequest,
+    AddVoiceResponse,
+    DialogueInput,
+    DialogueSettings,
+    SpeechToSpeechRequest,
+    SpeechToTextRequest,
+    SpeechToTextResponse,
+    TextToDialogueRequest,
+    TextToSoundEffectsRequest,
+    TextToSpeechRequest,
+    TextToSpeechVoiceSettings,
+)
+from comfy_api_nodes.util import (
+    ApiEndpoint,
+    audio_bytes_to_audio_input,
+    audio_ndarray_to_bytesio,
+    audio_tensor_to_contiguous_ndarray,
+    sync_op,
+    sync_op_raw,
+    upload_audio_to_comfyapi,
+    validate_string,
+)
+
+ELEVENLABS_MUSIC_SECTIONS = "ELEVENLABS_MUSIC_SECTIONS"  # Custom type for music sections
+ELEVENLABS_COMPOSITION_PLAN = "ELEVENLABS_COMPOSITION_PLAN"  # Custom type for composition plan
+ELEVENLABS_VOICE = "ELEVENLABS_VOICE"  # Custom type for voice selection
+
+# Predefined ElevenLabs voices: (voice_id, display_name, gender, accent)
+ELEVENLABS_VOICES = [
+    ("CwhRBWXzGAHq8TQ4Fs17", "Roger", "male", "american"),
+    ("EXAVITQu4vr4xnSDxMaL", "Sarah", "female", "american"),
+    ("FGY2WhTYpPnrIDTdsKH5", "Laura", "female", "american"),
+    ("IKne3meq5aSn9XLyUdCD", "Charlie", "male", "australian"),
+    ("JBFqnCBsd6RMkjVDRZzb", "George", "male", "british"),
+    ("N2lVS1w4EtoT3dr4eOWO", "Callum", "male", "american"),
+    ("SAz9YHcvj6GT2YYXdXww", "River", "neutral", "american"),
+    ("SOYHLrjzK2X1ezoPC6cr", "Harry", "male", "american"),
+    ("TX3LPaxmHKxFdv7VOQHJ", "Liam", "male", "american"),
+    ("Xb7hH8MSUJpSbSDYk0k2", "Alice", "female", "british"),
+    ("XrExE9yKIg1WjnnlVkGX", "Matilda", "female", "american"),
+    ("bIHbv24MWmeRgasZH58o", "Will", "male", "american"),
+    ("cgSgspJ2msm6clMCkdW9", "Jessica", "female", "american"),
+    ("cjVigY5qzO86Huf0OWal", "Eric", "male", "american"),
+    ("hpp4J3VqNfWAUOO0d1Us", "Bella", "female", "american"),
+    ("iP95p4xoKVk53GoZ742B", "Chris", "male", "american"),
+    ("nPczCjzI2devNBz1zQrb", "Brian", "male", "american"),
+    ("onwK4e9ZLuTAKqWW03F9", "Daniel", "male", "british"),
+    ("pFZP5JQG7iQjIQuC4Bku", "Lily", "female", "british"),
+    ("pNInz6obpgDQGcFmaJgB", "Adam", "male", "american"),
+    ("pqHfZKP75CvOlQylNhV4", "Bill", "male", "american"),
+]
+
+ELEVENLABS_VOICE_OPTIONS = [f"{name} ({gender}, {accent})" for _, name, gender, accent in ELEVENLABS_VOICES]
+ELEVENLABS_VOICE_MAP = {
+    f"{name} ({gender}, {accent})": voice_id for voice_id, name, gender, accent in ELEVENLABS_VOICES
+}
+
+
+class ElevenLabsSpeechToText(IO.ComfyNode):
+    @classmethod
+    def define_schema(cls) -> IO.Schema:
+        return IO.Schema(
+            node_id="ElevenLabsSpeechToText",
+            display_name="ElevenLabs Speech to Text",
+            category="api node/audio/ElevenLabs",
+            description="Transcribe audio to text. "
+            "Supports automatic language detection, speaker diarization, and audio event tagging.",
+            inputs=[
+                IO.Audio.Input(
+                    "audio",
+                    tooltip="Audio to transcribe.",
+                ),
+                IO.DynamicCombo.Input(
+                    "model",
+                    options=[
+                        IO.DynamicCombo.Option(
+                            "scribe_v2",
+                            [
+                                IO.Boolean.Input(
+                                    "tag_audio_events",
+                                    default=False,
+                                    tooltip="Annotate sounds like (laughter), (music), etc. in transcript.",
+                                ),
+                                IO.Boolean.Input(
+                                    "diarize",
+                                    default=False,
+                                    tooltip="Annotate which speaker is talking.",
+                                ),
+                                IO.Float.Input(
+                                    "diarization_threshold",
+                                    default=0.22,
+                                    min=0.1,
+                                    max=0.4,
+                                    step=0.01,
+                                    display_mode=IO.NumberDisplay.slider,
+                                    tooltip="Speaker separation sensitivity. "
+                                    "Lower values are more sensitive to speaker changes.",
+                                ),
+                                IO.Float.Input(
+                                    "temperature",
+                                    default=0.0,
+                                    min=0.0,
+                                    max=2.0,
+                                    step=0.01,
+                                    display_mode=IO.NumberDisplay.slider,
+                                    tooltip="Randomness control. "
+                                    "0.0 uses model default. Higher values increase randomness.",
+                                ),
+                                IO.Combo.Input(
+                                    "timestamps_granularity",
+                                    options=["word", "character", "none"],
+                                    default="word",
+                                    tooltip="Timing precision for transcript words.",
+                                ),
+                            ],
+                        ),
+                    ],
+                    tooltip="Model to use for transcription.",
+                ),
+                IO.String.Input(
+                    "language_code",
+                    default="",
+                    tooltip="ISO-639-1 or ISO-639-3 language code (e.g., 'en', 'es', 'fra'). "
+                    "Leave empty for automatic detection.",
+                ),
+                IO.Int.Input(
+                    "num_speakers",
+                    default=0,
+                    min=0,
+                    max=32,
+                    display_mode=IO.NumberDisplay.slider,
+                    tooltip="Maximum number of speakers to predict. Set to 0 for automatic detection.",
+                ),
+                IO.Int.Input(
+                    "seed",
+                    default=1,
+                    min=0,
+                    max=2147483647,
+                    tooltip="Seed for reproducibility (determinism not guaranteed).",
+                ),
+            ],
+            outputs=[
+                IO.String.Output(display_name="text"),
+                IO.String.Output(display_name="language_code"),
+                IO.String.Output(display_name="words_json"),
+            ],
+            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":0.0073,"format":{"approximate":true,"suffix":"/minute"}}""",
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        audio: Input.Audio,
+        model: dict,
+        language_code: str,
+        num_speakers: int,
+        seed: int,
+    ) -> IO.NodeOutput:
+        if model["diarize"] and num_speakers:
+            raise ValueError(
+                "Number of speakers cannot be specified when diarization is enabled. "
+                "Either disable diarization or set num_speakers to 0."
+            )
+        request = SpeechToTextRequest(
+            model_id=model["model"],
+            cloud_storage_url=await upload_audio_to_comfyapi(
+                cls, audio, container_format="mp4", codec_name="aac", mime_type="audio/mp4"
+            ),
+            language_code=language_code if language_code.strip() else None,
+            tag_audio_events=model["tag_audio_events"],
+            num_speakers=num_speakers if num_speakers > 0 else None,
+            timestamps_granularity=model["timestamps_granularity"],
+            diarize=model["diarize"],
+            diarization_threshold=model["diarization_threshold"] if model["diarize"] else None,
+            seed=seed,
+            temperature=model["temperature"],
+        )
+        response = await sync_op(
+            cls,
+            ApiEndpoint(path="/proxy/elevenlabs/v1/speech-to-text", method="POST"),
+            response_model=SpeechToTextResponse,
+            data=request,
+            content_type="multipart/form-data",
+        )
+        words_json = json.dumps(
+            [w.model_dump(exclude_none=True) for w in response.words] if response.words else [],
+            indent=2,
+        )
+        return IO.NodeOutput(response.text, response.language_code, words_json)
+
+
+class ElevenLabsVoiceSelector(IO.ComfyNode):
+    @classmethod
+    def define_schema(cls) -> IO.Schema:
+        return IO.Schema(
+            node_id="ElevenLabsVoiceSelector",
+            display_name="ElevenLabs Voice Selector",
+            category="api node/audio/ElevenLabs",
+            description="Select a predefined ElevenLabs voice for text-to-speech generation.",
+            inputs=[
+                IO.Combo.Input(
+                    "voice",
+                    options=ELEVENLABS_VOICE_OPTIONS,
+                    tooltip="Choose a voice from the predefined ElevenLabs voices.",
+                ),
+            ],
+            outputs=[
+                IO.Custom(ELEVENLABS_VOICE).Output(display_name="voice"),
+            ],
+            is_api_node=False,
+        )
+
+    @classmethod
+    def execute(cls, voice: str) -> IO.NodeOutput:
+        voice_id = ELEVENLABS_VOICE_MAP.get(voice)
+        if not voice_id:
+            raise ValueError(f"Unknown voice: {voice}")
+        return IO.NodeOutput(voice_id)
+
+
+class ElevenLabsTextToSpeech(IO.ComfyNode):
+    @classmethod
+    def define_schema(cls) -> IO.Schema:
+        return IO.Schema(
+            node_id="ElevenLabsTextToSpeech",
+            display_name="ElevenLabs Text to Speech",
+            category="api node/audio/ElevenLabs",
+            description="Convert text to speech.",
+            inputs=[
+                IO.Custom(ELEVENLABS_VOICE).Input(
+                    "voice",
+                    tooltip="Voice to use for speech synthesis. Connect from Voice Selector or Instant Voice Clone.",
+                ),
+                IO.String.Input(
+                    "text",
+                    multiline=True,
+                    default="",
+                    tooltip="The text to convert to speech.",
+                ),
+                IO.Float.Input(
+                    "stability",
+                    default=0.5,
+                    min=0.0,
+                    max=1.0,
+                    step=0.01,
+                    display_mode=IO.NumberDisplay.slider,
+                    tooltip="Voice stability. Lower values give broader emotional range, "
+                    "higher values produce more consistent but potentially monotonous speech.",
+                ),
+                IO.Combo.Input(
+                    "apply_text_normalization",
+                    options=["auto", "on", "off"],
+                    tooltip="Text normalization mode. 'auto' lets the system decide, "
+                    "'on' always applies normalization, 'off' skips it.",
+                ),
+                IO.DynamicCombo.Input(
+                    "model",
+                    options=[
+                        IO.DynamicCombo.Option(
+                            "eleven_multilingual_v2",
+                            [
+                                IO.Float.Input(
+                                    "speed",
+                                    default=1.0,
+                                    min=0.7,
+                                    max=1.3,
+                                    step=0.01,
+                                    display_mode=IO.NumberDisplay.slider,
+                                    tooltip="Speech speed. 1.0 is normal, <1.0 slower, >1.0 faster.",
+                                ),
+                                IO.Float.Input(
+                                    "similarity_boost",
+                                    default=0.75,
+                                    min=0.0,
+                                    max=1.0,
+                                    step=0.01,
+                                    display_mode=IO.NumberDisplay.slider,
+                                    tooltip="Similarity boost. Higher values make the voice more similar to the original.",
+                                ),
+                                IO.Boolean.Input(
+                                    "use_speaker_boost",
+                                    default=False,
+                                    tooltip="Boost similarity to the original speaker voice.",
+                                ),
+                                IO.Float.Input(
+                                    "style",
+                                    default=0.0,
+                                    min=0.0,
+                                    max=0.2,
+                                    step=0.01,
+                                    display_mode=IO.NumberDisplay.slider,
+                                    tooltip="Style exaggeration. Higher values increase stylistic expression "
+                                    "but may reduce stability.",
+                                ),
+                            ],
+                        ),
+                        IO.DynamicCombo.Option(
+                            "eleven_v3",
+                            [
+                                IO.Float.Input(
+                                    "speed",
+                                    default=1.0,
+                                    min=0.7,
+                                    max=1.3,
+                                    step=0.01,
+                                    display_mode=IO.NumberDisplay.slider,
+                                    tooltip="Speech speed. 1.0 is normal, <1.0 slower, >1.0 faster.",
+                                ),
+                                IO.Float.Input(
+                                    "similarity_boost",
+                                    default=0.75,
+                                    min=0.0,
+                                    max=1.0,
+                                    step=0.01,
+                                    display_mode=IO.NumberDisplay.slider,
+                                    tooltip="Similarity boost. Higher values make the voice more similar to the original.",
+                                ),
+                            ],
+                        ),
+                    ],
+                    tooltip="Model to use for text-to-speech.",
+                ),
+                IO.String.Input(
+                    "language_code",
+                    default="",
+                    tooltip="ISO-639-1 or ISO-639-3 language code (e.g., 'en', 'es', 'fra'). "
+                    "Leave empty for automatic detection.",
+                ),
+                IO.Int.Input(
+                    "seed",
+                    default=1,
+                    min=0,
+                    max=2147483647,
+                    tooltip="Seed for reproducibility (determinism not guaranteed).",
+                ),
+                IO.Combo.Input(
+                    "output_format",
+                    options=["mp3_44100_192", "opus_48000_192"],
+                    tooltip="Audio output format.",
+                ),
+            ],
+            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":0.24,"format":{"approximate":true,"suffix":"/1K chars"}}""",
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        voice: str,
+        text: str,
+        stability: float,
+        apply_text_normalization: str,
+        model: dict,
+        language_code: str,
+        seed: int,
+        output_format: str,
+    ) -> IO.NodeOutput:
+        validate_string(text, min_length=1)
+        request = TextToSpeechRequest(
+            text=text,
+            model_id=model["model"],
+            language_code=language_code if language_code.strip() else None,
+            voice_settings=TextToSpeechVoiceSettings(
+                stability=stability,
+                similarity_boost=model["similarity_boost"],
+                speed=model["speed"],
+                use_speaker_boost=model.get("use_speaker_boost", None),
+                style=model.get("style", None),
+            ),
+            seed=seed,
+            apply_text_normalization=apply_text_normalization,
+        )
+        response = await sync_op_raw(
+            cls,
+            ApiEndpoint(
+                path=f"/proxy/elevenlabs/v1/text-to-speech/{voice}",
+                method="POST",
+                query_params={"output_format": output_format},
+            ),
+            data=request,
+            as_binary=True,
+        )
+        return IO.NodeOutput(audio_bytes_to_audio_input(response))
+
+
+class ElevenLabsAudioIsolation(IO.ComfyNode):
+    @classmethod
+    def define_schema(cls) -> IO.Schema:
+        return IO.Schema(
+            node_id="ElevenLabsAudioIsolation",
+            display_name="ElevenLabs Voice Isolation",
+            category="api node/audio/ElevenLabs",
+            description="Remove background noise from audio, isolating vocals or speech.",
+            inputs=[
+                IO.Audio.Input(
+                    "audio",
+                    tooltip="Audio to process for background noise removal.",
+                ),
+            ],
+            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":0.24,"format":{"approximate":true,"suffix":"/minute"}}""",
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        audio: Input.Audio,
+    ) -> IO.NodeOutput:
+        audio_data_np = audio_tensor_to_contiguous_ndarray(audio["waveform"])
+        audio_bytes_io = audio_ndarray_to_bytesio(audio_data_np, audio["sample_rate"], "mp4", "aac")
+        response = await sync_op_raw(
+            cls,
+            ApiEndpoint(path="/proxy/elevenlabs/v1/audio-isolation", method="POST"),
+            files={"audio": ("audio.mp4", audio_bytes_io, "audio/mp4")},
+            content_type="multipart/form-data",
+            as_binary=True,
+        )
+        return IO.NodeOutput(audio_bytes_to_audio_input(response))
+
+
+class ElevenLabsTextToSoundEffects(IO.ComfyNode):
+    @classmethod
+    def define_schema(cls) -> IO.Schema:
+        return IO.Schema(
+            node_id="ElevenLabsTextToSoundEffects",
+            display_name="ElevenLabs Text to Sound Effects",
+            category="api node/audio/ElevenLabs",
+            description="Generate sound effects from text descriptions.",
+            inputs=[
+                IO.String.Input(
+                    "text",
+                    multiline=True,
+                    default="",
+                    tooltip="Text description of the sound effect to generate.",
+                ),
+                IO.DynamicCombo.Input(
+                    "model",
+                    options=[
+                        IO.DynamicCombo.Option(
+                            "eleven_sfx_v2",
+                            [
+                                IO.Float.Input(
+                                    "duration",
+                                    default=5.0,
+                                    min=0.5,
+                                    max=30.0,
+                                    step=0.1,
+                                    display_mode=IO.NumberDisplay.slider,
+                                    tooltip="Duration of generated sound in seconds.",
+                                ),
+                                IO.Boolean.Input(
+                                    "loop",
+                                    default=False,
+                                    tooltip="Create a smoothly looping sound effect.",
+                                ),
+                                IO.Float.Input(
+                                    "prompt_influence",
+                                    default=0.3,
+                                    min=0.0,
+                                    max=1.0,
+                                    step=0.01,
+                                    display_mode=IO.NumberDisplay.slider,
+                                    tooltip="How closely generation follows the prompt. "
+                                    "Higher values make the sound follow the text more closely.",
+                                ),
+                            ],
+                        ),
+                    ],
+                    tooltip="Model to use for sound effect generation.",
+                ),
+                IO.Combo.Input(
+                    "output_format",
+                    options=["mp3_44100_192", "opus_48000_192"],
+                    tooltip="Audio output format.",
+                ),
+            ],
+            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":0.14,"format":{"approximate":true,"suffix":"/minute"}}""",
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        text: str,
+        model: dict,
+        output_format: str,
+    ) -> IO.NodeOutput:
+        validate_string(text, min_length=1)
+        response = await sync_op_raw(
+            cls,
+            ApiEndpoint(
+                path="/proxy/elevenlabs/v1/sound-generation",
+                method="POST",
+                query_params={"output_format": output_format},
+            ),
+            data=TextToSoundEffectsRequest(
+                text=text,
+                duration_seconds=model["duration"],
+                prompt_influence=model["prompt_influence"],
+                loop=model.get("loop", None),
+            ),
+            as_binary=True,
+        )
+        return IO.NodeOutput(audio_bytes_to_audio_input(response))
+
+
+class ElevenLabsInstantVoiceClone(IO.ComfyNode):
+    @classmethod
+    def define_schema(cls) -> IO.Schema:
+        return IO.Schema(
+            node_id="ElevenLabsInstantVoiceClone",
+            display_name="ElevenLabs Instant Voice Clone",
+            category="api node/audio/ElevenLabs",
+            description="Create a cloned voice from audio samples. "
+            "Provide 1-8 audio recordings of the voice to clone.",
+            inputs=[
+                IO.Autogrow.Input(
+                    "files",
+                    template=IO.Autogrow.TemplatePrefix(
+                        IO.Audio.Input("audio"),
+                        prefix="audio",
+                        min=1,
+                        max=8,
+                    ),
+                    tooltip="Audio recordings for voice cloning.",
+                ),
+                IO.Boolean.Input(
+                    "remove_background_noise",
+                    default=False,
+                    tooltip="Remove background noise from voice samples using audio isolation.",
+                ),
+            ],
+            outputs=[
+                IO.Custom(ELEVENLABS_VOICE).Output(display_name="voice"),
+            ],
+            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":0.15}"""),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        files: IO.Autogrow.Type,
+        remove_background_noise: bool,
+    ) -> IO.NodeOutput:
+        file_tuples: list[tuple[str, tuple[str, bytes, str]]] = []
+        for key in files:
+            audio = files[key]
+            sample_rate: int = audio["sample_rate"]
+            waveform = audio["waveform"]
+            audio_data_np = audio_tensor_to_contiguous_ndarray(waveform)
+            audio_bytes_io = audio_ndarray_to_bytesio(audio_data_np, sample_rate, "mp4", "aac")
+            file_tuples.append(("files", (f"{key}.mp4", audio_bytes_io.getvalue(), "audio/mp4")))
+
+        response = await sync_op(
+            cls,
+            ApiEndpoint(path="/proxy/elevenlabs/v1/voices/add", method="POST"),
+            response_model=AddVoiceResponse,
+            data=AddVoiceRequest(
+                name=str(uuid.uuid4()),
+                remove_background_noise=remove_background_noise,
+            ),
+            files=file_tuples,
+            content_type="multipart/form-data",
+        )
+        return IO.NodeOutput(response.voice_id)
+
+
+ELEVENLABS_STS_VOICE_SETTINGS = [
+    IO.Float.Input(
+        "speed",
+        default=1.0,
+        min=0.7,
+        max=1.3,
+        step=0.01,
+        display_mode=IO.NumberDisplay.slider,
+        tooltip="Speech speed. 1.0 is normal, <1.0 slower, >1.0 faster.",
+    ),
+    IO.Float.Input(
+        "similarity_boost",
+        default=0.75,
+        min=0.0,
+        max=1.0,
+        step=0.01,
+        display_mode=IO.NumberDisplay.slider,
+        tooltip="Similarity boost. Higher values make the voice more similar to the original.",
+    ),
+    IO.Boolean.Input(
+        "use_speaker_boost",
+        default=False,
+        tooltip="Boost similarity to the original speaker voice.",
+    ),
+    IO.Float.Input(
+        "style",
+        default=0.0,
+        min=0.0,
+        max=0.2,
+        step=0.01,
+        display_mode=IO.NumberDisplay.slider,
+        tooltip="Style exaggeration. Higher values increase stylistic expression but may reduce stability.",
+    ),
+]
+
+
+class ElevenLabsSpeechToSpeech(IO.ComfyNode):
+    @classmethod
+    def define_schema(cls) -> IO.Schema:
+        return IO.Schema(
+            node_id="ElevenLabsSpeechToSpeech",
+            display_name="ElevenLabs Speech to Speech",
+            category="api node/audio/ElevenLabs",
+            description="Transform speech from one voice to another while preserving the original content and emotion.",
+            inputs=[
+                IO.Custom(ELEVENLABS_VOICE).Input(
+                    "voice",
+                    tooltip="Target voice for the transformation. "
+                    "Connect from Voice Selector or Instant Voice Clone.",
+                ),
+                IO.Audio.Input(
+                    "audio",
+                    tooltip="Source audio to transform.",
+                ),
+                IO.Float.Input(
+                    "stability",
+                    default=0.5,
+                    min=0.0,
+                    max=1.0,
+                    step=0.01,
+                    display_mode=IO.NumberDisplay.slider,
+                    tooltip="Voice stability. Lower values give broader emotional range, "
+                    "higher values produce more consistent but potentially monotonous speech.",
+                ),
+                IO.DynamicCombo.Input(
+                    "model",
+                    options=[
+                        IO.DynamicCombo.Option(
+                            "eleven_multilingual_sts_v2",
+                            ELEVENLABS_STS_VOICE_SETTINGS,
+                        ),
+                        IO.DynamicCombo.Option(
+                            "eleven_english_sts_v2",
+                            ELEVENLABS_STS_VOICE_SETTINGS,
+                        ),
+                    ],
+                    tooltip="Model to use for speech-to-speech transformation.",
+                ),
+                IO.Combo.Input(
+                    "output_format",
+                    options=["mp3_44100_192", "opus_48000_192"],
+                    tooltip="Audio output format.",
+                ),
+                IO.Int.Input(
+                    "seed",
+                    default=0,
+                    min=0,
+                    max=4294967295,
+                    tooltip="Seed for reproducibility.",
+                ),
+                IO.Boolean.Input(
+                    "remove_background_noise",
+                    default=False,
+                    tooltip="Remove background noise from input audio using audio isolation.",
+                ),
+            ],
+            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":0.24,"format":{"approximate":true,"suffix":"/minute"}}""",
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        voice: str,
+        audio: Input.Audio,
+        stability: float,
+        model: dict,
+        output_format: str,
+        seed: int,
+        remove_background_noise: bool,
+    ) -> IO.NodeOutput:
+        audio_data_np = audio_tensor_to_contiguous_ndarray(audio["waveform"])
+        audio_bytes_io = audio_ndarray_to_bytesio(audio_data_np, audio["sample_rate"], "mp4", "aac")
+        voice_settings = TextToSpeechVoiceSettings(
+            stability=stability,
+            similarity_boost=model["similarity_boost"],
+            style=model["style"],
+            use_speaker_boost=model["use_speaker_boost"],
+            speed=model["speed"],
+        )
+        response = await sync_op_raw(
+            cls,
+            ApiEndpoint(
+                path=f"/proxy/elevenlabs/v1/speech-to-speech/{voice}",
+                method="POST",
+                query_params={"output_format": output_format},
+            ),
+            data=SpeechToSpeechRequest(
+                model_id=model["model"],
+                voice_settings=voice_settings.model_dump_json(exclude_none=True),
+                seed=seed,
+                remove_background_noise=remove_background_noise,
+            ),
+            files={"audio": ("audio.mp4", audio_bytes_io.getvalue(), "audio/mp4")},
+            content_type="multipart/form-data",
+            as_binary=True,
+        )
+        return IO.NodeOutput(audio_bytes_to_audio_input(response))
+
+
+def _generate_dialogue_inputs(count: int) -> list:
+    """Generate input widgets for a given number of dialogue entries."""
+    inputs = []
+    for i in range(1, count + 1):
+        inputs.extend(
+            [
+                IO.String.Input(
+                    f"text{i}",
+                    multiline=True,
+                    default="",
+                    tooltip=f"Text content for dialogue entry {i}.",
+                ),
+                IO.Custom(ELEVENLABS_VOICE).Input(
+                    f"voice{i}",
+                    tooltip=f"Voice for dialogue entry {i}. Connect from Voice Selector or Instant Voice Clone.",
+                ),
+            ]
+        )
+    return inputs
+
+
+class ElevenLabsTextToDialogue(IO.ComfyNode):
+    @classmethod
+    def define_schema(cls) -> IO.Schema:
+        return IO.Schema(
+            node_id="ElevenLabsTextToDialogue",
+            display_name="ElevenLabs Text to Dialogue",
+            category="api node/audio/ElevenLabs",
+            description="Generate multi-speaker dialogue from text. Each dialogue entry has its own text and voice.",
+            inputs=[
+                IO.Float.Input(
+                    "stability",
+                    default=0.5,
+                    min=0.0,
+                    max=1.0,
+                    step=0.5,
+                    display_mode=IO.NumberDisplay.slider,
+                    tooltip="Voice stability. Lower values give broader emotional range, "
+                    "higher values produce more consistent but potentially monotonous speech.",
+                ),
+                IO.Combo.Input(
+                    "apply_text_normalization",
+                    options=["auto", "on", "off"],
+                    tooltip="Text normalization mode. 'auto' lets the system decide, "
+                    "'on' always applies normalization, 'off' skips it.",
+                ),
+                IO.Combo.Input(
+                    "model",
+                    options=["eleven_v3"],
+                    tooltip="Model to use for dialogue generation.",
+                ),
+                IO.DynamicCombo.Input(
+                    "inputs",
+                    options=[
+                        IO.DynamicCombo.Option("1", _generate_dialogue_inputs(1)),
+                        IO.DynamicCombo.Option("2", _generate_dialogue_inputs(2)),
+                        IO.DynamicCombo.Option("3", _generate_dialogue_inputs(3)),
+                        IO.DynamicCombo.Option("4", _generate_dialogue_inputs(4)),
+                        IO.DynamicCombo.Option("5", _generate_dialogue_inputs(5)),
+                        IO.DynamicCombo.Option("6", _generate_dialogue_inputs(6)),
+                        IO.DynamicCombo.Option("7", _generate_dialogue_inputs(7)),
+                        IO.DynamicCombo.Option("8", _generate_dialogue_inputs(8)),
+                        IO.DynamicCombo.Option("9", _generate_dialogue_inputs(9)),
+                        IO.DynamicCombo.Option("10", _generate_dialogue_inputs(10)),
+                    ],
+                    tooltip="Number of dialogue entries.",
+                ),
+                IO.String.Input(
+                    "language_code",
+                    default="",
+                    tooltip="ISO-639-1 or ISO-639-3 language code (e.g., 'en', 'es', 'fra'). "
+                    "Leave empty for automatic detection.",
+                ),
+                IO.Int.Input(
+                    "seed",
+                    default=1,
+                    min=0,
+                    max=4294967295,
+                    tooltip="Seed for reproducibility.",
+                ),
+                IO.Combo.Input(
+                    "output_format",
+                    options=["mp3_44100_192", "opus_48000_192"],
+                    tooltip="Audio output format.",
+                ),
+            ],
+            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":0.24,"format":{"approximate":true,"suffix":"/1K chars"}}""",
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        stability: float,
+        apply_text_normalization: str,
+        model: str,
+        inputs: dict,
+        language_code: str,
+        seed: int,
+        output_format: str,
+    ) -> IO.NodeOutput:
+        num_entries = int(inputs["inputs"])
+        dialogue_inputs: list[DialogueInput] = []
+        for i in range(1, num_entries + 1):
+            text = inputs[f"text{i}"]
+            voice_id = inputs[f"voice{i}"]
+            validate_string(text, min_length=1)
+            dialogue_inputs.append(DialogueInput(text=text, voice_id=voice_id))
+        request = TextToDialogueRequest(
+            inputs=dialogue_inputs,
+            model_id=model,
+            language_code=language_code if language_code.strip() else None,
+            settings=DialogueSettings(stability=stability),
+            seed=seed,
+            apply_text_normalization=apply_text_normalization,
+        )
+        response = await sync_op_raw(
+            cls,
+            ApiEndpoint(
+                path="/proxy/elevenlabs/v1/text-to-dialogue",
+                method="POST",
+                query_params={"output_format": output_format},
+            ),
+            data=request,
+            as_binary=True,
+        )
+        return IO.NodeOutput(audio_bytes_to_audio_input(response))
+
+
+class ElevenLabsExtension(ComfyExtension):
+    @override
+    async def get_node_list(self) -> list[type[IO.ComfyNode]]:
+        return [
+            ElevenLabsSpeechToText,
+            ElevenLabsVoiceSelector,
+            ElevenLabsTextToSpeech,
+            ElevenLabsAudioIsolation,
+            ElevenLabsTextToSoundEffects,
+            ElevenLabsInstantVoiceClone,
+            ElevenLabsSpeechToSpeech,
+            ElevenLabsTextToDialogue,
+        ]
+
+
+async def comfy_entrypoint() -> ElevenLabsExtension:
+    return ElevenLabsExtension()

comfy_api_nodes/nodes_gemini.py

@@ -6,6 +6,7 @@ See: https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/infer
 import base64
 import os
 from enum import Enum
+from fnmatch import fnmatch
 from io import BytesIO
 from typing import Literal
 
@@ -119,6 +120,13 @@ async def create_image_parts(
     return image_parts
 
 
+def _mime_matches(mime: GeminiMimeType | None, pattern: str) -> bool:
+    """Check if a MIME type matches a pattern. Supports fnmatch globs (e.g. 'image/*')."""
+    if mime is None:
+        return False
+    return fnmatch(mime.value, pattern)
+
+
 def get_parts_by_type(response: GeminiGenerateContentResponse, part_type: Literal["text"] | str) -> list[GeminiPart]:
     """
     Filter response parts by their type.
@@ -151,9 +159,9 @@ def get_parts_by_type(response: GeminiGenerateContentResponse, part_type: Litera
         for part in candidate.content.parts:
             if part_type == "text" and part.text:
                 parts.append(part)
-            elif part.inlineData and part.inlineData.mimeType == part_type:
+            elif part.inlineData and _mime_matches(part.inlineData.mimeType, part_type):
                 parts.append(part)
-            elif part.fileData and part.fileData.mimeType == part_type:
+            elif part.fileData and _mime_matches(part.fileData.mimeType, part_type):
                 parts.append(part)
 
     if not parts and blocked_reasons:
@@ -178,7 +186,7 @@ def get_text_from_response(response: GeminiGenerateContentResponse) -> str:
 
 async def get_image_from_response(response: GeminiGenerateContentResponse) -> Input.Image:
     image_tensors: list[Input.Image] = []
-    parts = get_parts_by_type(response, "image/png")
+    parts = get_parts_by_type(response, "image/*")
     for part in parts:
         if part.inlineData:
             image_data = base64.b64decode(part.inlineData.data)
@@ -308,6 +316,7 @@ class GeminiNode(IO.ComfyNode):
                     default="",
                     optional=True,
                     tooltip="Foundational instructions that dictate an AI's behavior.",
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -585,6 +594,7 @@ class GeminiImage(IO.ComfyNode):
                     tooltip="Choose 'IMAGE' for image-only output, or "
                     "'IMAGE+TEXT' to return both the generated image and a text response.",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.String.Input(
                     "system_prompt",
@@ -592,6 +602,7 @@ class GeminiImage(IO.ComfyNode):
                     default=GEMINI_IMAGE_SYS_PROMPT,
                     optional=True,
                     tooltip="Foundational instructions that dictate an AI's behavior.",
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -626,7 +637,7 @@ class GeminiImage(IO.ComfyNode):
 
         if not aspect_ratio:
             aspect_ratio = "auto"  # for backward compatability with old workflows; to-do remove this in December
-        image_config = GeminiImageConfig(aspectRatio=aspect_ratio)
+        image_config = GeminiImageConfig() if aspect_ratio == "auto" else GeminiImageConfig(aspectRatio=aspect_ratio)
 
         if images is not None:
             parts.extend(await create_image_parts(cls, images))
@@ -646,7 +657,7 @@ class GeminiImage(IO.ComfyNode):
                 ],
                 generationConfig=GeminiImageGenerationConfig(
                     responseModalities=(["IMAGE"] if response_modalities == "IMAGE" else ["TEXT", "IMAGE"]),
-                    imageConfig=None if aspect_ratio == "auto" else image_config,
+                    imageConfig=image_config,
                 ),
                 systemInstruction=gemini_system_prompt,
             ),
@@ -706,6 +717,7 @@ class GeminiImage2(IO.ComfyNode):
                     options=["IMAGE+TEXT", "IMAGE"],
                     tooltip="Choose 'IMAGE' for image-only output, or "
                     "'IMAGE+TEXT' to return both the generated image and a text response.",
+                    advanced=True,
                 ),
                 IO.Image.Input(
                     "images",
@@ -725,6 +737,7 @@ class GeminiImage2(IO.ComfyNode):
                     default=GEMINI_IMAGE_SYS_PROMPT,
                     optional=True,
                     tooltip="Foundational instructions that dictate an AI's behavior.",
+                    advanced=True,
                 ),
             ],
             outputs=[

comfy_api_nodes/nodes_hunyuan3d.py

@@ -1,31 +1,48 @@
 from typing_extensions import override
 
-from comfy_api.latest import IO, ComfyExtension, Input
+from comfy_api.latest import IO, ComfyExtension, Input, Types
 from comfy_api_nodes.apis.hunyuan3d import (
     Hunyuan3DViewImage,
     InputGenerateType,
     ResultFile3D,
+    TextureEditTaskRequest,
     To3DProTaskCreateResponse,
     To3DProTaskQueryRequest,
     To3DProTaskRequest,
     To3DProTaskResultResponse,
+    To3DUVFileInput,
+    To3DUVTaskRequest,
 )
 from comfy_api_nodes.util import (
     ApiEndpoint,
     download_url_to_file_3d,
+    download_url_to_image_tensor,
     downscale_image_tensor_by_max_side,
     poll_op,
     sync_op,
+    upload_3d_model_to_comfyapi,
     upload_image_to_comfyapi,
     validate_image_dimensions,
     validate_string,
 )
 
 
-def get_file_from_response(response_objs: list[ResultFile3D], file_type: str) -> ResultFile3D | None:
+def _is_tencent_rate_limited(status: int, body: object) -> bool:
+    return (
+        status == 400
+        and isinstance(body, dict)
+        and "RequestLimitExceeded" in str(body.get("Response", {}).get("Error", {}).get("Code", ""))
+    )
+
+
+def get_file_from_response(
+    response_objs: list[ResultFile3D], file_type: str, raise_if_not_found: bool = True
+) -> ResultFile3D | None:
     for i in response_objs:
         if i.Type.lower() == file_type.lower():
             return i
+    if raise_if_not_found:
+        raise ValueError(f"'{file_type}' file type is not found in the response.")
     return None
 
 
@@ -35,8 +52,9 @@ class TencentTextToModelNode(IO.ComfyNode):
     def define_schema(cls):
         return IO.Schema(
             node_id="TencentTextToModelNode",
-            display_name="Hunyuan3D: Text to Model (Pro)",
+            display_name="Hunyuan3D: Text to Model",
             category="api node/3d/Tencent",
+            essentials_category="3D",
             inputs=[
                 IO.Combo.Input(
                     "model",
@@ -120,6 +138,7 @@ class TencentTextToModelNode(IO.ComfyNode):
                 EnablePBR=generate_type.get("pbr", None),
                 PolygonType=generate_type.get("polygon_type", None),
             ),
+            is_rate_limited=_is_tencent_rate_limited,
         )
         if response.Error:
             raise ValueError(f"Task creation failed with code {response.Error.Code}: {response.Error.Message}")
@@ -131,11 +150,14 @@ class TencentTextToModelNode(IO.ComfyNode):
             response_model=To3DProTaskResultResponse,
             status_extractor=lambda r: r.Status,
         )
-        glb_result = get_file_from_response(result.ResultFile3Ds, "glb")
-        obj_result = get_file_from_response(result.ResultFile3Ds, "obj")
-        file_glb = await download_url_to_file_3d(glb_result.Url, "glb", task_id=task_id) if glb_result else None
         return IO.NodeOutput(
-            file_glb, file_glb, await download_url_to_file_3d(obj_result.Url, "obj", task_id=task_id) if obj_result else None
+            f"{task_id}.glb",
+            await download_url_to_file_3d(
+                get_file_from_response(result.ResultFile3Ds, "glb").Url, "glb", task_id=task_id
+            ),
+            await download_url_to_file_3d(
+                get_file_from_response(result.ResultFile3Ds, "obj").Url, "obj", task_id=task_id
+            ),
         )
 
 
@@ -145,8 +167,9 @@ class TencentImageToModelNode(IO.ComfyNode):
     def define_schema(cls):
         return IO.Schema(
             node_id="TencentImageToModelNode",
-            display_name="Hunyuan3D: Image(s) to Model (Pro)",
+            display_name="Hunyuan3D: Image(s) to Model",
             category="api node/3d/Tencent",
+            essentials_category="3D",
             inputs=[
                 IO.Combo.Input(
                     "model",
@@ -268,6 +291,7 @@ class TencentImageToModelNode(IO.ComfyNode):
                 EnablePBR=generate_type.get("pbr", None),
                 PolygonType=generate_type.get("polygon_type", None),
             ),
+            is_rate_limited=_is_tencent_rate_limited,
         )
         if response.Error:
             raise ValueError(f"Task creation failed with code {response.Error.Code}: {response.Error.Message}")
@@ -279,11 +303,257 @@ class TencentImageToModelNode(IO.ComfyNode):
             response_model=To3DProTaskResultResponse,
             status_extractor=lambda r: r.Status,
         )
-        glb_result = get_file_from_response(result.ResultFile3Ds, "glb")
-        obj_result = get_file_from_response(result.ResultFile3Ds, "obj")
-        file_glb = await download_url_to_file_3d(glb_result.Url, "glb", task_id=task_id) if glb_result else None
         return IO.NodeOutput(
-            file_glb, file_glb, await download_url_to_file_3d(obj_result.Url, "obj", task_id=task_id) if obj_result else None
+            f"{task_id}.glb",
+            await download_url_to_file_3d(
+                get_file_from_response(result.ResultFile3Ds, "glb").Url, "glb", task_id=task_id
+            ),
+            await download_url_to_file_3d(
+                get_file_from_response(result.ResultFile3Ds, "obj").Url, "obj", task_id=task_id
+            ),
+        )
+
+
+class TencentModelTo3DUVNode(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls):
+        return IO.Schema(
+            node_id="TencentModelTo3DUVNode",
+            display_name="Hunyuan3D: Model to UV",
+            category="api node/3d/Tencent",
+            description="Perform UV unfolding on a 3D model to generate UV texture. "
+            "Input model must have less than 30000 faces.",
+            inputs=[
+                IO.MultiType.Input(
+                    "model_3d",
+                    types=[IO.File3DGLB, IO.File3DOBJ, IO.File3DFBX, IO.File3DAny],
+                    tooltip="Input 3D model (GLB, OBJ, or FBX)",
+                ),
+                IO.Int.Input(
+                    "seed",
+                    default=1,
+                    min=0,
+                    max=2147483647,
+                    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.",
+                ),
+            ],
+            outputs=[
+                IO.File3DOBJ.Output(display_name="OBJ"),
+                IO.File3DFBX.Output(display_name="FBX"),
+                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(expr='{"type":"usd","usd":0.2}'),
+        )
+
+    SUPPORTED_FORMATS = {"glb", "obj", "fbx"}
+
+    @classmethod
+    async def execute(
+        cls,
+        model_3d: Types.File3D,
+        seed: int,
+    ) -> IO.NodeOutput:
+        _ = seed
+        file_format = model_3d.format.lower()
+        if file_format not in cls.SUPPORTED_FORMATS:
+            raise ValueError(
+                f"Unsupported file format: '{file_format}'. "
+                f"Supported formats: {', '.join(sorted(cls.SUPPORTED_FORMATS))}."
+            )
+        response = await sync_op(
+            cls,
+            ApiEndpoint(path="/proxy/tencent/hunyuan/3d-uv", method="POST"),
+            response_model=To3DProTaskCreateResponse,
+            data=To3DUVTaskRequest(
+                File=To3DUVFileInput(
+                    Type=file_format.upper(),
+                    Url=await upload_3d_model_to_comfyapi(cls, model_3d, file_format),
+                )
+            ),
+            is_rate_limited=_is_tencent_rate_limited,
+        )
+        if response.Error:
+            raise ValueError(f"Task creation failed with code {response.Error.Code}: {response.Error.Message}")
+        result = await poll_op(
+            cls,
+            ApiEndpoint(path="/proxy/tencent/hunyuan/3d-uv/query", method="POST"),
+            data=To3DProTaskQueryRequest(JobId=response.JobId),
+            response_model=To3DProTaskResultResponse,
+            status_extractor=lambda r: r.Status,
+        )
+        return IO.NodeOutput(
+            await download_url_to_file_3d(get_file_from_response(result.ResultFile3Ds, "obj").Url, "obj"),
+            await download_url_to_file_3d(get_file_from_response(result.ResultFile3Ds, "fbx").Url, "fbx"),
+            await download_url_to_image_tensor(get_file_from_response(result.ResultFile3Ds, "image").Url),
+        )
+
+
+class Tencent3DTextureEditNode(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls):
+        return IO.Schema(
+            node_id="Tencent3DTextureEditNode",
+            display_name="Hunyuan3D: 3D Texture Edit",
+            category="api node/3d/Tencent",
+            description="After inputting the 3D model, perform 3D model texture redrawing.",
+            inputs=[
+                IO.MultiType.Input(
+                    "model_3d",
+                    types=[IO.File3DFBX, IO.File3DAny],
+                    tooltip="3D model in FBX format. Model should have less than 100000 faces.",
+                ),
+                IO.String.Input(
+                    "prompt",
+                    multiline=True,
+                    default="",
+                    tooltip="Describes texture editing. Supports up to 1024 UTF-8 characters.",
+                ),
+                IO.Int.Input(
+                    "seed",
+                    default=0,
+                    min=0,
+                    max=2147483647,
+                    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.",
+                ),
+            ],
+            outputs=[
+                IO.File3DGLB.Output(display_name="GLB"),
+                IO.File3DFBX.Output(display_name="FBX"),
+            ],
+            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": 0.6}""",
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        model_3d: Types.File3D,
+        prompt: str,
+        seed: int,
+    ) -> IO.NodeOutput:
+        _ = seed
+        file_format = model_3d.format.lower()
+        if file_format != "fbx":
+            raise ValueError(f"Unsupported file format: '{file_format}'. Only FBX format is supported.")
+        validate_string(prompt, field_name="prompt", min_length=1, max_length=1024)
+        model_url = await upload_3d_model_to_comfyapi(cls, model_3d, file_format)
+        response = await sync_op(
+            cls,
+            ApiEndpoint(path="/proxy/tencent/hunyuan/3d-texture-edit", method="POST"),
+            response_model=To3DProTaskCreateResponse,
+            data=TextureEditTaskRequest(
+                File3D=To3DUVFileInput(Type=file_format.upper(), Url=model_url),
+                Prompt=prompt,
+                EnablePBR=True,
+            ),
+            is_rate_limited=_is_tencent_rate_limited,
+        )
+        if response.Error:
+            raise ValueError(f"Task creation failed with code {response.Error.Code}: {response.Error.Message}")
+
+        result = await poll_op(
+            cls,
+            ApiEndpoint(path="/proxy/tencent/hunyuan/3d-texture-edit/query", method="POST"),
+            data=To3DProTaskQueryRequest(JobId=response.JobId),
+            response_model=To3DProTaskResultResponse,
+            status_extractor=lambda r: r.Status,
+        )
+        return IO.NodeOutput(
+            await download_url_to_file_3d(get_file_from_response(result.ResultFile3Ds, "glb").Url, "glb"),
+            await download_url_to_file_3d(get_file_from_response(result.ResultFile3Ds, "fbx").Url, "fbx"),
+        )
+
+
+class Tencent3DPartNode(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls):
+        return IO.Schema(
+            node_id="Tencent3DPartNode",
+            display_name="Hunyuan3D: 3D Part",
+            category="api node/3d/Tencent",
+            description="Automatically perform component identification and generation based on the model structure.",
+            inputs=[
+                IO.MultiType.Input(
+                    "model_3d",
+                    types=[IO.File3DFBX, IO.File3DAny],
+                    tooltip="3D model in FBX format. Model should have less than 30000 faces.",
+                ),
+                IO.Int.Input(
+                    "seed",
+                    default=0,
+                    min=0,
+                    max=2147483647,
+                    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.",
+                ),
+            ],
+            outputs=[
+                IO.File3DFBX.Output(display_name="FBX"),
+            ],
+            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":0.6}'),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        model_3d: Types.File3D,
+        seed: int,
+    ) -> IO.NodeOutput:
+        _ = seed
+        file_format = model_3d.format.lower()
+        if file_format != "fbx":
+            raise ValueError(f"Unsupported file format: '{file_format}'. Only FBX format is supported.")
+        model_url = await upload_3d_model_to_comfyapi(cls, model_3d, file_format)
+        response = await sync_op(
+            cls,
+            ApiEndpoint(path="/proxy/tencent/hunyuan/3d-part", method="POST"),
+            response_model=To3DProTaskCreateResponse,
+            data=To3DUVTaskRequest(
+                File=To3DUVFileInput(Type=file_format.upper(), Url=model_url),
+            ),
+            is_rate_limited=_is_tencent_rate_limited,
+        )
+        if response.Error:
+            raise ValueError(f"Task creation failed with code {response.Error.Code}: {response.Error.Message}")
+        result = await poll_op(
+            cls,
+            ApiEndpoint(path="/proxy/tencent/hunyuan/3d-part/query", method="POST"),
+            data=To3DProTaskQueryRequest(JobId=response.JobId),
+            response_model=To3DProTaskResultResponse,
+            status_extractor=lambda r: r.Status,
+        )
+        return IO.NodeOutput(
+            await download_url_to_file_3d(get_file_from_response(result.ResultFile3Ds, "fbx").Url, "fbx"),
         )
 
 
@@ -293,6 +563,9 @@ class TencentHunyuan3DExtension(ComfyExtension):
         return [
             TencentTextToModelNode,
             TencentImageToModelNode,
+            # TencentModelTo3DUVNode,
+            # Tencent3DTextureEditNode,
+            Tencent3DPartNode,
         ]
 
 

comfy_api_nodes/nodes_ideogram.py

@@ -261,6 +261,7 @@ class IdeogramV1(IO.ComfyNode):
                     default="AUTO",
                     tooltip="Determine if MagicPrompt should be used in generation",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -394,6 +395,7 @@ class IdeogramV2(IO.ComfyNode):
                     default="AUTO",
                     tooltip="Determine if MagicPrompt should be used in generation",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -411,6 +413,7 @@ class IdeogramV2(IO.ComfyNode):
                     default="NONE",
                     tooltip="Style type for generation (V2 only)",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.String.Input(
                     "negative_prompt",
@@ -564,6 +567,7 @@ class IdeogramV3(IO.ComfyNode):
                     default="AUTO",
                     tooltip="Determine if MagicPrompt should be used in generation",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -590,6 +594,7 @@ class IdeogramV3(IO.ComfyNode):
                     default="DEFAULT",
                     tooltip="Controls the trade-off between generation speed and quality",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Image.Input(
                     "character_image",

comfy_api_nodes/nodes_kling.py

@@ -2262,6 +2262,7 @@ class KlingLipSyncAudioToVideoNode(IO.ComfyNode):
             node_id="KlingLipSyncAudioToVideoNode",
             display_name="Kling Lip Sync Video with Audio",
             category="api node/video/Kling",
+            essentials_category="Video Generation",
             description="Kling Lip Sync Audio to Video Node. Syncs mouth movements in a video file to the audio content of an audio file. When using, ensure that the audio contains clearly distinguishable vocals and that the video contains a distinct face. The audio file should not be larger than 5MB. The video file should not be larger than 100MB, should have height/width between 720px and 1920px, and should be between 2s and 10s in length.",
             inputs=[
                 IO.Video.Input("video"),
@@ -2333,6 +2334,7 @@ class KlingLipSyncTextToVideoNode(IO.ComfyNode):
                     max=2.0,
                     display_mode=IO.NumberDisplay.slider,
                     tooltip="Speech Rate. Valid range: 0.8~2.0, accurate to one decimal place.",
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -2454,6 +2456,7 @@ class KlingImageGenerationNode(IO.ComfyNode):
                 IO.Combo.Input(
                     "image_type",
                     options=[i.value for i in KlingImageGenImageReferenceType],
+                    advanced=True,
                 ),
                 IO.Float.Input(
                     "image_fidelity",
@@ -2463,6 +2466,7 @@ class KlingImageGenerationNode(IO.ComfyNode):
                     step=0.01,
                     display_mode=IO.NumberDisplay.slider,
                     tooltip="Reference intensity for user-uploaded images",
+                    advanced=True,
                 ),
                 IO.Float.Input(
                     "human_fidelity",
@@ -2472,6 +2476,7 @@ class KlingImageGenerationNode(IO.ComfyNode):
                     step=0.01,
                     display_mode=IO.NumberDisplay.slider,
                     tooltip="Subject reference similarity",
+                    advanced=True,
                 ),
                 IO.Combo.Input("model_name", options=["kling-v3", "kling-v2", "kling-v1-5"]),
                 IO.Combo.Input(
@@ -2587,7 +2592,7 @@ class TextToVideoWithAudio(IO.ComfyNode):
                 IO.Combo.Input("mode", options=["pro"]),
                 IO.Combo.Input("aspect_ratio", options=["16:9", "9:16", "1:1"]),
                 IO.Combo.Input("duration", options=[5, 10]),
-                IO.Boolean.Input("generate_audio", default=True),
+                IO.Boolean.Input("generate_audio", default=True, advanced=True),
             ],
             outputs=[
                 IO.Video.Output(),
@@ -2655,7 +2660,7 @@ class ImageToVideoWithAudio(IO.ComfyNode):
                 IO.String.Input("prompt", multiline=True, tooltip="Positive text prompt."),
                 IO.Combo.Input("mode", options=["pro"]),
                 IO.Combo.Input("duration", options=[5, 10]),
-                IO.Boolean.Input("generate_audio", default=True),
+                IO.Boolean.Input("generate_audio", default=True, advanced=True),
             ],
             outputs=[
                 IO.Video.Output(),

comfy_api_nodes/nodes_ltxv.py

@@ -74,6 +74,7 @@ class TextToVideoNode(IO.ComfyNode):
                     default=False,
                     optional=True,
                     tooltip="When true, the generated video will include AI-generated audio matching the scene.",
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -151,6 +152,7 @@ class ImageToVideoNode(IO.ComfyNode):
                     default=False,
                     optional=True,
                     tooltip="When true, the generated video will include AI-generated audio matching the scene.",
+                    advanced=True,
                 ),
             ],
             outputs=[

comfy_api_nodes/nodes_magnific.py

@@ -110,11 +110,13 @@ class MagnificImageUpscalerCreativeNode(IO.ComfyNode):
                 IO.Combo.Input(
                     "engine",
                     options=["automatic", "magnific_illusio", "magnific_sharpy", "magnific_sparkle"],
+                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "auto_downscale",
                     default=False,
                     tooltip="Automatically downscale input image if output would exceed maximum pixel limit.",
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -280,6 +282,7 @@ class MagnificImageUpscalerPreciseV2Node(IO.ComfyNode):
                     "auto_downscale",
                     default=False,
                     tooltip="Automatically downscale input image if output would exceed maximum resolution.",
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -440,6 +443,7 @@ class MagnificImageStyleTransferNode(IO.ComfyNode):
                         "softy",
                     ],
                     tooltip="Processing engine selection.",
+                    advanced=True,
                 ),
                 IO.DynamicCombo.Input(
                     "portrait_mode",
@@ -468,6 +472,7 @@ class MagnificImageStyleTransferNode(IO.ComfyNode):
                     default=True,
                     tooltip="When disabled, expect each generation to introduce a degree of randomness, "
                     "leading to more diverse outcomes.",
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -582,16 +587,19 @@ class MagnificImageRelightNode(IO.ComfyNode):
                     "interpolate_from_original",
                     default=False,
                     tooltip="Restricts generation freedom to match original more closely.",
+                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "change_background",
                     default=True,
                     tooltip="Modifies background based on prompt/reference.",
+                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "preserve_details",
                     default=True,
                     tooltip="Maintains texture and fine details from original.",
+                    advanced=True,
                 ),
                 IO.DynamicCombo.Input(
                     "advanced_settings",

comfy_api_nodes/nodes_meshy.py

@@ -58,11 +58,12 @@ class MeshyTextToModelNode(IO.ComfyNode):
                     ],
                     tooltip="When set to false, returns an unprocessed triangular mesh.",
                 ),
-                IO.Combo.Input("symmetry_mode", options=["auto", "on", "off"]),
+                IO.Combo.Input("symmetry_mode", options=["auto", "on", "off"], advanced=True),
                 IO.Combo.Input(
                     "pose_mode",
                     options=["", "A-pose", "T-pose"],
                     tooltip="Specify the pose mode for the generated model.",
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -155,6 +156,7 @@ class MeshyRefineNode(IO.ComfyNode):
                     tooltip="Generate PBR Maps (metallic, roughness, normal) in addition to the base color. "
                     "Note: this should be set to false when using Sculpture style, "
                     "as Sculpture style generates its own set of PBR maps.",
+                    advanced=True,
                 ),
                 IO.String.Input(
                     "texture_prompt",
@@ -299,6 +301,7 @@ class MeshyImageToModelNode(IO.ComfyNode):
                     "pose_mode",
                     options=["", "A-pose", "T-pose"],
                     tooltip="Specify the pose mode for the generated model.",
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -429,7 +432,7 @@ class MeshyMultiImageToModelNode(IO.ComfyNode):
                     ],
                     tooltip="When set to false, returns an unprocessed triangular mesh.",
                 ),
-                IO.Combo.Input("symmetry_mode", options=["auto", "on", "off"]),
+                IO.Combo.Input("symmetry_mode", options=["auto", "on", "off"], advanced=True),
                 IO.DynamicCombo.Input(
                     "should_texture",
                     options=[
@@ -466,6 +469,7 @@ class MeshyMultiImageToModelNode(IO.ComfyNode):
                     "pose_mode",
                     options=["", "A-pose", "T-pose"],
                     tooltip="Specify the pose mode for the generated model.",
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -728,8 +732,9 @@ class MeshyTextureNode(IO.ComfyNode):
                     tooltip="Use the original UV of the model instead of generating new UVs. "
                     "When enabled, Meshy preserves existing textures from the uploaded model. "
                     "If the model has no original UV, the quality of the output might not be as good.",
+                    advanced=True,
                 ),
-                IO.Boolean.Input("pbr", default=False),
+                IO.Boolean.Input("pbr", default=False, advanced=True),
                 IO.String.Input(
                     "text_style_prompt",
                     default="",

comfy_api_nodes/nodes_openai.py

@@ -43,7 +43,6 @@ class SupportedOpenAIModel(str, Enum):
     o1 = "o1"
     o3 = "o3"
     o1_pro = "o1-pro"
-    gpt_4o = "gpt-4o"
     gpt_4_1 = "gpt-4.1"
     gpt_4_1_mini = "gpt-4.1-mini"
     gpt_4_1_nano = "gpt-4.1-nano"
@@ -576,6 +575,7 @@ class OpenAIChatNode(IO.ComfyNode):
             node_id="OpenAIChatNode",
             display_name="OpenAI ChatGPT",
             category="api node/text/OpenAI",
+            essentials_category="Text Generation",
             description="Generate text responses from an OpenAI model.",
             inputs=[
                 IO.String.Input(
@@ -588,6 +588,7 @@ class OpenAIChatNode(IO.ComfyNode):
                     "persist_context",
                     default=False,
                     tooltip="This parameter is deprecated and has no effect.",
+                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "model",
@@ -649,11 +650,6 @@ class OpenAIChatNode(IO.ComfyNode):
                     "usd": [0.01, 0.04],
                     "format": { "approximate": true, "separator": "-", "suffix": " per 1K tokens" }
                   }
-                  : $contains($m, "gpt-4o") ? {
-                    "type": "list_usd",
-                    "usd": [0.0025, 0.01],
-                    "format": { "approximate": true, "separator": "-", "suffix": " per 1K tokens" }
-                  }
                   : $contains($m, "gpt-4.1-nano") ? {
                     "type": "list_usd",
                     "usd": [0.0001, 0.0004],
@@ -862,6 +858,7 @@ class OpenAIChatConfig(IO.ComfyNode):
                     options=["auto", "disabled"],
                     default="auto",
                     tooltip="The truncation strategy to use for the model response. auto: If the context of this response and previous ones exceeds the model's context window size, the model will truncate the response to fit the context window by dropping input items in the middle of the conversation.disabled: If a model response will exceed the context window size for a model, the request will fail with a 400 error",
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "max_output_tokens",
@@ -870,6 +867,7 @@ class OpenAIChatConfig(IO.ComfyNode):
                     max=16384,
                     tooltip="An upper bound for the number of tokens that can be generated for a response, including visible output tokens",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.String.Input(
                     "instructions",

comfy_api_nodes/nodes_recraft.py

@@ -1,5 +1,4 @@
 from io import BytesIO
-from typing import Optional, Union
 
 import aiohttp
 import torch
@@ -9,6 +8,8 @@ from typing_extensions import override
 from comfy.utils import ProgressBar
 from comfy_api.latest import IO, ComfyExtension
 from comfy_api_nodes.apis.recraft import (
+    RECRAFT_V4_PRO_SIZES,
+    RECRAFT_V4_SIZES,
     RecraftColor,
     RecraftColorChain,
     RecraftControls,
@@ -18,7 +19,6 @@ from comfy_api_nodes.apis.recraft import (
     RecraftImageGenerationResponse,
     RecraftImageSize,
     RecraftIO,
-    RecraftModel,
     RecraftStyle,
     RecraftStyleV3,
     get_v3_substyles,
@@ -39,7 +39,7 @@ async def handle_recraft_file_request(
     cls: type[IO.ComfyNode],
     image: torch.Tensor,
     path: str,
-    mask: Optional[torch.Tensor] = None,
+    mask: torch.Tensor | None = None,
     total_pixels: int = 4096 * 4096,
     timeout: int = 1024,
     request=None,
@@ -73,11 +73,11 @@ async def handle_recraft_file_request(
 def recraft_multipart_parser(
     data,
     parent_key=None,
-    formatter: Optional[type[callable]] = None,
-    converted_to_check: Optional[list[list]] = None,
+    formatter: type[callable] | None = None,
+    converted_to_check: list[list] | None = None,
     is_list: bool = False,
     return_mode: str = "formdata",  # "dict" | "formdata"
-) -> Union[dict, aiohttp.FormData]:
+) -> dict | aiohttp.FormData:
     """
     Formats data such that multipart/form-data will work with aiohttp library when both files and data are present.
 
@@ -309,7 +309,7 @@ class RecraftStyleInfiniteStyleLibrary(IO.ComfyNode):
             node_id="RecraftStyleV3InfiniteStyleLibrary",
             display_name="Recraft Style - Infinite Style Library",
             category="api node/image/Recraft",
-            description="Select style based on preexisting UUID from Recraft's Infinite Style Library.",
+            description="Choose style based on preexisting UUID from Recraft's Infinite Style Library.",
             inputs=[
                 IO.String.Input("style_id", default="", tooltip="UUID of style from Infinite Style Library."),
             ],
@@ -485,7 +485,7 @@ class RecraftTextToImageNode(IO.ComfyNode):
             data=RecraftImageGenerationRequest(
                 prompt=prompt,
                 negative_prompt=negative_prompt,
-                model=RecraftModel.recraftv3,
+                model="recraftv3",
                 size=size,
                 n=n,
                 style=recraft_style.style,
@@ -598,7 +598,7 @@ class RecraftImageToImageNode(IO.ComfyNode):
         request = RecraftImageGenerationRequest(
             prompt=prompt,
             negative_prompt=negative_prompt,
-            model=RecraftModel.recraftv3,
+            model="recraftv3",
             n=n,
             strength=round(strength, 2),
             style=recraft_style.style,
@@ -698,7 +698,7 @@ class RecraftImageInpaintingNode(IO.ComfyNode):
         request = RecraftImageGenerationRequest(
             prompt=prompt,
             negative_prompt=negative_prompt,
-            model=RecraftModel.recraftv3,
+            model="recraftv3",
             n=n,
             style=recraft_style.style,
             substyle=recraft_style.substyle,
@@ -810,7 +810,7 @@ class RecraftTextToVectorNode(IO.ComfyNode):
             data=RecraftImageGenerationRequest(
                 prompt=prompt,
                 negative_prompt=negative_prompt,
-                model=RecraftModel.recraftv3,
+                model="recraftv3",
                 size=size,
                 n=n,
                 style=recraft_style.style,
@@ -933,7 +933,7 @@ class RecraftReplaceBackgroundNode(IO.ComfyNode):
         request = RecraftImageGenerationRequest(
             prompt=prompt,
             negative_prompt=negative_prompt,
-            model=RecraftModel.recraftv3,
+            model="recraftv3",
             n=n,
             style=recraft_style.style,
             substyle=recraft_style.substyle,
@@ -963,6 +963,7 @@ class RecraftRemoveBackgroundNode(IO.ComfyNode):
             node_id="RecraftRemoveBackgroundNode",
             display_name="Recraft Remove Background",
             category="api node/image/Recraft",
+            essentials_category="Image Tools",
             description="Remove background from image, and return processed image and mask.",
             inputs=[
                 IO.Image.Input("image"),
@@ -1078,6 +1079,252 @@ class RecraftCreativeUpscaleNode(RecraftCrispUpscaleNode):
         )
 
 
+class RecraftV4TextToImageNode(IO.ComfyNode):
+    @classmethod
+    def define_schema(cls):
+        return IO.Schema(
+            node_id="RecraftV4TextToImageNode",
+            display_name="Recraft V4 Text to Image",
+            category="api node/image/Recraft",
+            description="Generates images using Recraft V4 or V4 Pro models.",
+            inputs=[
+                IO.String.Input(
+                    "prompt",
+                    multiline=True,
+                    tooltip="Prompt for the image generation. Maximum 10,000 characters.",
+                ),
+                IO.String.Input(
+                    "negative_prompt",
+                    multiline=True,
+                    tooltip="An optional text description of undesired elements on an image.",
+                ),
+                IO.DynamicCombo.Input(
+                    "model",
+                    options=[
+                        IO.DynamicCombo.Option(
+                            "recraftv4",
+                            [
+                                IO.Combo.Input(
+                                    "size",
+                                    options=RECRAFT_V4_SIZES,
+                                    default="1024x1024",
+                                    tooltip="The size of the generated image.",
+                                ),
+                            ],
+                        ),
+                        IO.DynamicCombo.Option(
+                            "recraftv4_pro",
+                            [
+                                IO.Combo.Input(
+                                    "size",
+                                    options=RECRAFT_V4_PRO_SIZES,
+                                    default="2048x2048",
+                                    tooltip="The size of the generated image.",
+                                ),
+                            ],
+                        ),
+                    ],
+                    tooltip="The model to use for generation.",
+                ),
+                IO.Int.Input(
+                    "n",
+                    default=1,
+                    min=1,
+                    max=6,
+                    tooltip="The number of images to generate.",
+                ),
+                IO.Int.Input(
+                    "seed",
+                    default=0,
+                    min=0,
+                    max=0xFFFFFFFFFFFFFFFF,
+                    control_after_generate=True,
+                    tooltip="Seed to determine if node should re-run; "
+                    "actual results are nondeterministic regardless of seed.",
+                ),
+                IO.Custom(RecraftIO.CONTROLS).Input(
+                    "recraft_controls",
+                    tooltip="Optional additional controls over the generation via the Recraft Controls node.",
+                    optional=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", "n"]),
+                expr="""
+                (
+                    $prices := {"recraftv4": 0.04, "recraftv4_pro": 0.25};
+                    {"type":"usd","usd": $lookup($prices, widgets.model) * widgets.n}
+                )
+                """,
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        prompt: str,
+        negative_prompt: str,
+        model: dict,
+        n: int,
+        seed: int,
+        recraft_controls: RecraftControls | None = None,
+    ) -> IO.NodeOutput:
+        validate_string(prompt, strip_whitespace=False, min_length=1, max_length=10000)
+        response = await sync_op(
+            cls,
+            ApiEndpoint(path="/proxy/recraft/image_generation", method="POST"),
+            response_model=RecraftImageGenerationResponse,
+            data=RecraftImageGenerationRequest(
+                prompt=prompt,
+                negative_prompt=negative_prompt if negative_prompt else None,
+                model=model["model"],
+                size=model["size"],
+                n=n,
+                controls=recraft_controls.create_api_model() if recraft_controls else None,
+            ),
+            max_retries=1,
+        )
+        images = []
+        for data in response.data:
+            with handle_recraft_image_output():
+                image = bytesio_to_image_tensor(await download_url_as_bytesio(data.url, timeout=1024))
+            if len(image.shape) < 4:
+                image = image.unsqueeze(0)
+            images.append(image)
+        return IO.NodeOutput(torch.cat(images, dim=0))
+
+
+class RecraftV4TextToVectorNode(IO.ComfyNode):
+    @classmethod
+    def define_schema(cls):
+        return IO.Schema(
+            node_id="RecraftV4TextToVectorNode",
+            display_name="Recraft V4 Text to Vector",
+            category="api node/image/Recraft",
+            description="Generates SVG using Recraft V4 or V4 Pro models.",
+            inputs=[
+                IO.String.Input(
+                    "prompt",
+                    multiline=True,
+                    tooltip="Prompt for the image generation. Maximum 10,000 characters.",
+                ),
+                IO.String.Input(
+                    "negative_prompt",
+                    multiline=True,
+                    tooltip="An optional text description of undesired elements on an image.",
+                ),
+                IO.DynamicCombo.Input(
+                    "model",
+                    options=[
+                        IO.DynamicCombo.Option(
+                            "recraftv4",
+                            [
+                                IO.Combo.Input(
+                                    "size",
+                                    options=RECRAFT_V4_SIZES,
+                                    default="1024x1024",
+                                    tooltip="The size of the generated image.",
+                                ),
+                            ],
+                        ),
+                        IO.DynamicCombo.Option(
+                            "recraftv4_pro",
+                            [
+                                IO.Combo.Input(
+                                    "size",
+                                    options=RECRAFT_V4_PRO_SIZES,
+                                    default="2048x2048",
+                                    tooltip="The size of the generated image.",
+                                ),
+                            ],
+                        ),
+                    ],
+                    tooltip="The model to use for generation.",
+                ),
+                IO.Int.Input(
+                    "n",
+                    default=1,
+                    min=1,
+                    max=6,
+                    tooltip="The number of images to generate.",
+                ),
+                IO.Int.Input(
+                    "seed",
+                    default=0,
+                    min=0,
+                    max=0xFFFFFFFFFFFFFFFF,
+                    control_after_generate=True,
+                    tooltip="Seed to determine if node should re-run; "
+                    "actual results are nondeterministic regardless of seed.",
+                ),
+                IO.Custom(RecraftIO.CONTROLS).Input(
+                    "recraft_controls",
+                    tooltip="Optional additional controls over the generation via the Recraft Controls node.",
+                    optional=True,
+                ),
+            ],
+            outputs=[
+                IO.SVG.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", "n"]),
+                expr="""
+                (
+                    $prices := {"recraftv4": 0.08, "recraftv4_pro": 0.30};
+                    {"type":"usd","usd": $lookup($prices, widgets.model) * widgets.n}
+                )
+                """,
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        prompt: str,
+        negative_prompt: str,
+        model: dict,
+        n: int,
+        seed: int,
+        recraft_controls: RecraftControls | None = None,
+    ) -> IO.NodeOutput:
+        validate_string(prompt, strip_whitespace=False, min_length=1, max_length=10000)
+        response = await sync_op(
+            cls,
+            ApiEndpoint(path="/proxy/recraft/image_generation", method="POST"),
+            response_model=RecraftImageGenerationResponse,
+            data=RecraftImageGenerationRequest(
+                prompt=prompt,
+                negative_prompt=negative_prompt if negative_prompt else None,
+                model=model["model"],
+                size=model["size"],
+                n=n,
+                style="vector_illustration",
+                substyle=None,
+                controls=recraft_controls.create_api_model() if recraft_controls else None,
+            ),
+            max_retries=1,
+        )
+        svg_data = []
+        for data in response.data:
+            svg_data.append(await download_url_as_bytesio(data.url, timeout=1024))
+        return IO.NodeOutput(SVG(svg_data))
+
+
 class RecraftExtension(ComfyExtension):
     @override
     async def get_node_list(self) -> list[type[IO.ComfyNode]]:
@@ -1098,6 +1345,8 @@ class RecraftExtension(ComfyExtension):
             RecraftCreateStyleNode,
             RecraftColorRGBNode,
             RecraftControlsNode,
+            RecraftV4TextToImageNode,
+            RecraftV4TextToVectorNode,
         ]
 
 

comfy_api_nodes/nodes_rodin.py

@@ -493,7 +493,7 @@ class Rodin3D_Gen2(IO.ComfyNode):
                     default="500K-Triangle",
                     optional=True,
                 ),
-                IO.Boolean.Input("TAPose", default=False),
+                IO.Boolean.Input("TAPose", default=False, advanced=True),
             ],
             outputs=[
                 IO.String.Output(display_name="3D Model Path"),  # for backward compatibility only
@@ -505,6 +505,9 @@ class Rodin3D_Gen2(IO.ComfyNode):
                 IO.Hidden.unique_id,
             ],
             is_api_node=True,
+            price_badge=IO.PriceBadge(
+                expr="""{"type":"usd","usd":0.4}""",
+            ),
         )
 
     @classmethod

comfy_api_nodes/nodes_stability.py

@@ -86,6 +86,7 @@ class StabilityStableImageUltraNode(IO.ComfyNode):
                     "style_preset",
                     options=get_stability_style_presets(),
                     tooltip="Optional desired style of generated image.",
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -107,6 +108,7 @@ class StabilityStableImageUltraNode(IO.ComfyNode):
                     tooltip="A blurb of text describing what you do not wish to see in the output image. This is an advanced feature.",
                     force_input=True,
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Float.Input(
                     "image_denoise",
@@ -218,6 +220,7 @@ class StabilityStableImageSD_3_5Node(IO.ComfyNode):
                     "style_preset",
                     options=get_stability_style_presets(),
                     tooltip="Optional desired style of generated image.",
+                    advanced=True,
                 ),
                 IO.Float.Input(
                     "cfg_scale",
@@ -247,6 +250,7 @@ class StabilityStableImageSD_3_5Node(IO.ComfyNode):
                     tooltip="Keywords of what you do not wish to see in the output image. This is an advanced feature.",
                     force_input=True,
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Float.Input(
                     "image_denoise",
@@ -384,6 +388,7 @@ class StabilityUpscaleConservativeNode(IO.ComfyNode):
                     tooltip="Keywords of what you do not wish to see in the output image. This is an advanced feature.",
                     force_input=True,
                     optional=True,
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -474,6 +479,7 @@ class StabilityUpscaleCreativeNode(IO.ComfyNode):
                     "style_preset",
                     options=get_stability_style_presets(),
                     tooltip="Optional desired style of generated image.",
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -491,6 +497,7 @@ class StabilityUpscaleCreativeNode(IO.ComfyNode):
                     tooltip="Keywords of what you do not wish to see in the output image. This is an advanced feature.",
                     force_input=True,
                     optional=True,
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -624,6 +631,7 @@ class StabilityTextToAudio(IO.ComfyNode):
             node_id="StabilityTextToAudio",
             display_name="Stability AI Text To Audio",
             category="api node/audio/Stability AI",
+            essentials_category="Audio",
             description=cleandoc(cls.__doc__ or ""),
             inputs=[
                 IO.Combo.Input(
@@ -659,6 +667,7 @@ class StabilityTextToAudio(IO.ComfyNode):
                     step=1,
                     tooltip="Controls the number of sampling steps.",
                     optional=True,
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -736,6 +745,7 @@ class StabilityAudioToAudio(IO.ComfyNode):
                     step=1,
                     tooltip="Controls the number of sampling steps.",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Float.Input(
                     "strength",
@@ -829,6 +839,7 @@ class StabilityAudioInpaint(IO.ComfyNode):
                     step=1,
                     tooltip="Controls the number of sampling steps.",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "mask_start",
@@ -837,6 +848,7 @@ class StabilityAudioInpaint(IO.ComfyNode):
                     max=190,
                     step=1,
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "mask_end",
@@ -845,6 +857,7 @@ class StabilityAudioInpaint(IO.ComfyNode):
                     max=190,
                     step=1,
                     optional=True,
+                    advanced=True,
                 ),
             ],
             outputs=[

comfy_api_nodes/nodes_topaz.py

@@ -63,12 +63,14 @@ class TopazImageEnhance(IO.ComfyNode):
                     "subject_detection",
                     options=["All", "Foreground", "Background"],
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "face_enhancement",
                     default=True,
                     optional=True,
                     tooltip="Enhance faces (if present) during processing.",
+                    advanced=True,
                 ),
                 IO.Float.Input(
                     "face_enhancement_creativity",
@@ -79,6 +81,7 @@ class TopazImageEnhance(IO.ComfyNode):
                     display_mode=IO.NumberDisplay.number,
                     optional=True,
                     tooltip="Set the creativity level for face enhancement.",
+                    advanced=True,
                 ),
                 IO.Float.Input(
                     "face_enhancement_strength",
@@ -89,6 +92,7 @@ class TopazImageEnhance(IO.ComfyNode):
                     display_mode=IO.NumberDisplay.number,
                     optional=True,
                     tooltip="Controls how sharp enhanced faces are relative to the background.",
+                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "crop_to_fill",
@@ -96,6 +100,7 @@ class TopazImageEnhance(IO.ComfyNode):
                     optional=True,
                     tooltip="By default, the image is letterboxed when the output aspect ratio differs. "
                     "Enable to crop the image to fill the output dimensions.",
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "output_width",
@@ -106,6 +111,7 @@ class TopazImageEnhance(IO.ComfyNode):
                     display_mode=IO.NumberDisplay.number,
                     optional=True,
                     tooltip="Zero value means to calculate automatically (usually it will be original size or output_height if specified).",
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "output_height",
@@ -116,6 +122,7 @@ class TopazImageEnhance(IO.ComfyNode):
                     display_mode=IO.NumberDisplay.number,
                     optional=True,
                     tooltip="Zero value means to output in the same height as original or output width.",
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "creativity",
@@ -131,12 +138,14 @@ class TopazImageEnhance(IO.ComfyNode):
                     default=True,
                     optional=True,
                     tooltip="Preserve subjects' facial identity.",
+                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "color_preservation",
                     default=True,
                     optional=True,
                     tooltip="Preserve the original colors.",
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -234,9 +243,10 @@ class TopazVideoEnhance(IO.ComfyNode):
                     default="low",
                     tooltip="Creativity level (applies only to Starlight (Astra) Creative).",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Boolean.Input("interpolation_enabled", default=False, optional=True),
-                IO.Combo.Input("interpolation_model", options=["apo-8"], default="apo-8", optional=True),
+                IO.Combo.Input("interpolation_model", options=["apo-8"], default="apo-8", optional=True, advanced=True),
                 IO.Int.Input(
                     "interpolation_slowmo",
                     default=1,
@@ -246,6 +256,7 @@ class TopazVideoEnhance(IO.ComfyNode):
                     tooltip="Slow-motion factor applied to the input video. "
                     "For example, 2 makes the output twice as slow and doubles the duration.",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "interpolation_frame_rate",
@@ -261,6 +272,7 @@ class TopazVideoEnhance(IO.ComfyNode):
                     default=False,
                     tooltip="Analyze the input for duplicate frames and remove them.",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Float.Input(
                     "interpolation_duplicate_threshold",
@@ -271,6 +283,7 @@ class TopazVideoEnhance(IO.ComfyNode):
                     display_mode=IO.NumberDisplay.number,
                     tooltip="Detection sensitivity for duplicate frames.",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "dynamic_compression_level",
@@ -278,6 +291,7 @@ class TopazVideoEnhance(IO.ComfyNode):
                     default="Low",
                     tooltip="CQP level.",
                     optional=True,
+                    advanced=True,
                 ),
             ],
             outputs=[

comfy_api_nodes/nodes_tripo.py

@@ -89,13 +89,13 @@ class TripoTextToModelNode(IO.ComfyNode):
                 IO.Combo.Input("style", options=TripoStyle, default="None", optional=True),
                 IO.Boolean.Input("texture", default=True, optional=True),
                 IO.Boolean.Input("pbr", default=True, optional=True),
-                IO.Int.Input("image_seed", default=42, optional=True),
-                IO.Int.Input("model_seed", default=42, optional=True),
-                IO.Int.Input("texture_seed", default=42, optional=True),
-                IO.Combo.Input("texture_quality", default="standard", options=["standard", "detailed"], optional=True),
-                IO.Int.Input("face_limit", default=-1, min=-1, max=2000000, optional=True),
-                IO.Boolean.Input("quad", default=False, optional=True),
-                IO.Combo.Input("geometry_quality", default="standard", options=["standard", "detailed"], optional=True),
+                IO.Int.Input("image_seed", default=42, optional=True, advanced=True),
+                IO.Int.Input("model_seed", default=42, optional=True, advanced=True),
+                IO.Int.Input("texture_seed", default=42, optional=True, advanced=True),
+                IO.Combo.Input("texture_quality", default="standard", options=["standard", "detailed"], optional=True, advanced=True),
+                IO.Int.Input("face_limit", default=-1, min=-1, max=2000000, optional=True, advanced=True),
+                IO.Boolean.Input("quad", default=False, optional=True, advanced=True),
+                IO.Combo.Input("geometry_quality", default="standard", options=["standard", "detailed"], optional=True, advanced=True),
             ],
             outputs=[
                 IO.String.Output(display_name="model_file"),  # for backward compatibility only
@@ -210,18 +210,18 @@ class TripoImageToModelNode(IO.ComfyNode):
                 IO.Combo.Input("style", options=TripoStyle, default="None", optional=True),
                 IO.Boolean.Input("texture", default=True, optional=True),
                 IO.Boolean.Input("pbr", default=True, optional=True),
-                IO.Int.Input("model_seed", default=42, optional=True),
+                IO.Int.Input("model_seed", default=42, optional=True, advanced=True),
                 IO.Combo.Input(
-                    "orientation", options=TripoOrientation, default=TripoOrientation.DEFAULT, optional=True
+                    "orientation", options=TripoOrientation, default=TripoOrientation.DEFAULT, optional=True, advanced=True
                 ),
-                IO.Int.Input("texture_seed", default=42, optional=True),
-                IO.Combo.Input("texture_quality", default="standard", options=["standard", "detailed"], optional=True),
+                IO.Int.Input("texture_seed", default=42, optional=True, advanced=True),
+                IO.Combo.Input("texture_quality", default="standard", options=["standard", "detailed"], optional=True, advanced=True),
                 IO.Combo.Input(
-                    "texture_alignment", default="original_image", options=["original_image", "geometry"], optional=True
+                    "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),
-                IO.Boolean.Input("quad", default=False, optional=True),
-                IO.Combo.Input("geometry_quality", default="standard", options=["standard", "detailed"], optional=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.Combo.Input("geometry_quality", default="standard", options=["standard", "detailed"], optional=True, advanced=True),
             ],
             outputs=[
                 IO.String.Output(display_name="model_file"),  # for backward compatibility only
@@ -347,18 +347,19 @@ class TripoMultiviewToModelNode(IO.ComfyNode):
                     options=TripoOrientation,
                     default=TripoOrientation.DEFAULT,
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Boolean.Input("texture", default=True, optional=True),
                 IO.Boolean.Input("pbr", default=True, optional=True),
-                IO.Int.Input("model_seed", default=42, optional=True),
-                IO.Int.Input("texture_seed", default=42, optional=True),
-                IO.Combo.Input("texture_quality", default="standard", options=["standard", "detailed"], optional=True),
+                IO.Int.Input("model_seed", default=42, optional=True, advanced=True),
+                IO.Int.Input("texture_seed", default=42, optional=True, advanced=True),
+                IO.Combo.Input("texture_quality", default="standard", options=["standard", "detailed"], optional=True, advanced=True),
                 IO.Combo.Input(
-                    "texture_alignment", default="original_image", options=["original_image", "geometry"], optional=True
+                    "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),
-                IO.Boolean.Input("quad", default=False, optional=True),
-                IO.Combo.Input("geometry_quality", default="standard", options=["standard", "detailed"], optional=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.Combo.Input("geometry_quality", default="standard", options=["standard", "detailed"], optional=True, advanced=True),
             ],
             outputs=[
                 IO.String.Output(display_name="model_file"),  # for backward compatibility only
@@ -474,10 +475,10 @@ class TripoTextureNode(IO.ComfyNode):
                 IO.Custom("MODEL_TASK_ID").Input("model_task_id"),
                 IO.Boolean.Input("texture", default=True, optional=True),
                 IO.Boolean.Input("pbr", default=True, optional=True),
-                IO.Int.Input("texture_seed", default=42, optional=True),
-                IO.Combo.Input("texture_quality", default="standard", options=["standard", "detailed"], optional=True),
+                IO.Int.Input("texture_seed", default=42, optional=True, advanced=True),
+                IO.Combo.Input("texture_quality", default="standard", options=["standard", "detailed"], optional=True, advanced=True),
                 IO.Combo.Input(
-                    "texture_alignment", default="original_image", options=["original_image", "geometry"], optional=True
+                    "texture_alignment", default="original_image", options=["original_image", "geometry"], optional=True, advanced=True
                 ),
             ],
             outputs=[
@@ -682,13 +683,14 @@ class TripoConversionNode(IO.ComfyNode):
             inputs=[
                 IO.Custom("MODEL_TASK_ID,RIG_TASK_ID,RETARGET_TASK_ID").Input("original_model_task_id"),
                 IO.Combo.Input("format", options=["GLTF", "USDZ", "FBX", "OBJ", "STL", "3MF"]),
-                IO.Boolean.Input("quad", default=False, optional=True),
+                IO.Boolean.Input("quad", default=False, optional=True, advanced=True),
                 IO.Int.Input(
                     "face_limit",
                     default=-1,
                     min=-1,
                     max=2000000,
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "texture_size",
@@ -696,47 +698,53 @@ class TripoConversionNode(IO.ComfyNode):
                     min=128,
                     max=4096,
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "texture_format",
                     options=["BMP", "DPX", "HDR", "JPEG", "OPEN_EXR", "PNG", "TARGA", "TIFF", "WEBP"],
                     default="JPEG",
                     optional=True,
+                    advanced=True,
                 ),
-                IO.Boolean.Input("force_symmetry", default=False, optional=True),
-                IO.Boolean.Input("flatten_bottom", default=False, optional=True),
+                IO.Boolean.Input("force_symmetry", default=False, optional=True, advanced=True),
+                IO.Boolean.Input("flatten_bottom", default=False, optional=True, advanced=True),
                 IO.Float.Input(
                     "flatten_bottom_threshold",
                     default=0.0,
                     min=0.0,
                     max=1.0,
                     optional=True,
+                    advanced=True,
                 ),
-                IO.Boolean.Input("pivot_to_center_bottom", default=False, optional=True),
+                IO.Boolean.Input("pivot_to_center_bottom", default=False, optional=True, advanced=True),
                 IO.Float.Input(
                     "scale_factor",
                     default=1.0,
                     min=0.0,
                     optional=True,
+                    advanced=True,
                 ),
-                IO.Boolean.Input("with_animation", default=False, optional=True),
-                IO.Boolean.Input("pack_uv", default=False, optional=True),
-                IO.Boolean.Input("bake", default=False, optional=True),
-                IO.String.Input("part_names", default="", optional=True),  # comma-separated list
+                IO.Boolean.Input("with_animation", default=False, optional=True, advanced=True),
+                IO.Boolean.Input("pack_uv", default=False, optional=True, advanced=True),
+                IO.Boolean.Input("bake", default=False, optional=True, advanced=True),
+                IO.String.Input("part_names", default="", optional=True, advanced=True),  # comma-separated list
                 IO.Combo.Input(
                     "fbx_preset",
                     options=["blender", "mixamo", "3dsmax"],
                     default="blender",
                     optional=True,
+                    advanced=True,
                 ),
-                IO.Boolean.Input("export_vertex_colors", default=False, optional=True),
+                IO.Boolean.Input("export_vertex_colors", default=False, optional=True, advanced=True),
                 IO.Combo.Input(
                     "export_orientation",
                     options=["align_image", "default"],
                     default="default",
                     optional=True,
+                    advanced=True,
                 ),
-                IO.Boolean.Input("animate_in_place", default=False, optional=True),
+                IO.Boolean.Input("animate_in_place", default=False, optional=True, advanced=True),
             ],
             outputs=[],
             hidden=[

comfy_api_nodes/nodes_veo2.py

@@ -81,6 +81,7 @@ class VeoVideoGenerationNode(IO.ComfyNode):
                     default=True,
                     tooltip="Whether to enhance the prompt with AI assistance",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "person_generation",
@@ -88,6 +89,7 @@ class VeoVideoGenerationNode(IO.ComfyNode):
                     default="ALLOW",
                     tooltip="Whether to allow generating people in the video",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",
@@ -299,6 +301,7 @@ class Veo3VideoGenerationNode(VeoVideoGenerationNode):
                     default=True,
                     tooltip="This parameter is deprecated and ignored.",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "person_generation",
@@ -306,6 +309,7 @@ class Veo3VideoGenerationNode(VeoVideoGenerationNode):
                     default="ALLOW",
                     tooltip="Whether to allow generating people in the video",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "seed",

comfy_api_nodes/nodes_vidu.py

@@ -54,6 +54,7 @@ async def execute_task(
         response_model=TaskStatusResponse,
         status_extractor=lambda r: r.state,
         progress_extractor=lambda r: r.progress,
+        price_extractor=lambda r: r.credits * 0.005 if r.credits is not None else None,
         max_poll_attempts=max_poll_attempts,
     )
     if not response.creations:
@@ -111,12 +112,14 @@ class ViduTextToVideoNode(IO.ComfyNode):
                     options=["1080p"],
                     tooltip="Supported values may vary by model & duration",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "movement_amplitude",
                     options=["auto", "small", "medium", "large"],
                     tooltip="The movement amplitude of objects in the frame",
                     optional=True,
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -207,12 +210,14 @@ class ViduImageToVideoNode(IO.ComfyNode):
                     options=["1080p"],
                     tooltip="Supported values may vary by model & duration",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "movement_amplitude",
                     options=["auto", "small", "medium", "large"],
                     tooltip="The movement amplitude of objects in the frame",
                     optional=True,
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -313,12 +318,14 @@ class ViduReferenceVideoNode(IO.ComfyNode):
                     options=["1080p"],
                     tooltip="Supported values may vary by model & duration",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "movement_amplitude",
                     options=["auto", "small", "medium", "large"],
                     tooltip="The movement amplitude of objects in the frame",
                     optional=True,
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -425,12 +432,14 @@ class ViduStartEndToVideoNode(IO.ComfyNode):
                     options=["1080p"],
                     tooltip="Supported values may vary by model & duration",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "movement_amplitude",
                     options=["auto", "small", "medium", "large"],
                     tooltip="The movement amplitude of objects in the frame",
                     optional=True,
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -510,11 +519,12 @@ class Vidu2TextToVideoNode(IO.ComfyNode):
                     control_after_generate=True,
                 ),
                 IO.Combo.Input("aspect_ratio", options=["16:9", "9:16", "3:4", "4:3", "1:1"]),
-                IO.Combo.Input("resolution", options=["720p", "1080p"]),
+                IO.Combo.Input("resolution", options=["720p", "1080p"], advanced=True),
                 IO.Boolean.Input(
                     "background_music",
                     default=False,
                     tooltip="Whether to add background music to the generated video.",
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -608,11 +618,13 @@ class Vidu2ImageToVideoNode(IO.ComfyNode):
                 IO.Combo.Input(
                     "resolution",
                     options=["720p", "1080p"],
+                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "movement_amplitude",
                     options=["auto", "small", "medium", "large"],
                     tooltip="The movement amplitude of objects in the frame.",
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -726,6 +738,7 @@ class Vidu2ReferenceVideoNode(IO.ComfyNode):
                     "audio",
                     default=False,
                     tooltip="When enabled video will contain generated speech and background music based on the prompt.",
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "duration",
@@ -745,11 +758,12 @@ class Vidu2ReferenceVideoNode(IO.ComfyNode):
                     control_after_generate=True,
                 ),
                 IO.Combo.Input("aspect_ratio", options=["16:9", "9:16", "4:3", "3:4", "1:1"]),
-                IO.Combo.Input("resolution", options=["720p", "1080p"]),
+                IO.Combo.Input("resolution", options=["720p", "1080p"], advanced=True),
                 IO.Combo.Input(
                     "movement_amplitude",
                     options=["auto", "small", "medium", "large"],
                     tooltip="The movement amplitude of objects in the frame.",
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -863,11 +877,12 @@ class Vidu2StartEndToVideoNode(IO.ComfyNode):
                     display_mode=IO.NumberDisplay.number,
                     control_after_generate=True,
                 ),
-                IO.Combo.Input("resolution", options=["720p", "1080p"]),
+                IO.Combo.Input("resolution", options=["720p", "1080p"], advanced=True),
                 IO.Combo.Input(
                     "movement_amplitude",
                     options=["auto", "small", "medium", "large"],
                     tooltip="The movement amplitude of objects in the frame.",
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -1306,6 +1321,36 @@ class Vidu3TextToVideoNode(IO.ComfyNode):
                                 ),
                             ],
                         ),
+                        IO.DynamicCombo.Option(
+                            "viduq3-turbo",
+                            [
+                                IO.Combo.Input(
+                                    "aspect_ratio",
+                                    options=["16:9", "9:16", "3:4", "4:3", "1:1"],
+                                    tooltip="The aspect ratio of the output video.",
+                                ),
+                                IO.Combo.Input(
+                                    "resolution",
+                                    options=["720p", "1080p"],
+                                    tooltip="Resolution of the output video.",
+                                ),
+                                IO.Int.Input(
+                                    "duration",
+                                    default=5,
+                                    min=1,
+                                    max=16,
+                                    step=1,
+                                    display_mode=IO.NumberDisplay.slider,
+                                    tooltip="Duration of the output video in seconds.",
+                                ),
+                                IO.Boolean.Input(
+                                    "audio",
+                                    default=False,
+                                    tooltip="When enabled, outputs video with sound "
+                                    "(including dialogue and sound effects).",
+                                ),
+                            ],
+                        ),
                     ],
                     tooltip="Model to use for video generation.",
                 ),
@@ -1334,13 +1379,20 @@ class Vidu3TextToVideoNode(IO.ComfyNode):
             ],
             is_api_node=True,
             price_badge=IO.PriceBadge(
-                depends_on=IO.PriceBadgeDepends(widgets=["model.duration", "model.resolution"]),
+                depends_on=IO.PriceBadgeDepends(widgets=["model", "model.duration", "model.resolution"]),
                 expr="""
                 (
                   $res := $lookup(widgets, "model.resolution");
-                  $base := $lookup({"720p": 0.075, "1080p": 0.1}, $res);
-                  $perSec := $lookup({"720p": 0.025, "1080p": 0.05}, $res);
-                  {"type":"usd","usd": $base + $perSec * ($lookup(widgets, "model.duration") - 1)}
+                  $d := $lookup(widgets, "model.duration");
+                  $contains(widgets.model, "turbo")
+                    ? (
+                        $rate := $lookup({"720p": 0.06, "1080p": 0.08}, $res);
+                        {"type":"usd","usd": $rate * $d}
+                      )
+                    : (
+                        $rate := $lookup({"720p": 0.15, "1080p": 0.16}, $res);
+                        {"type":"usd","usd": $rate * $d}
+                      )
                 )
                 """,
             ),
@@ -1409,6 +1461,31 @@ class Vidu3ImageToVideoNode(IO.ComfyNode):
                                 ),
                             ],
                         ),
+                        IO.DynamicCombo.Option(
+                            "viduq3-turbo",
+                            [
+                                IO.Combo.Input(
+                                    "resolution",
+                                    options=["720p", "1080p"],
+                                    tooltip="Resolution of the output video.",
+                                ),
+                                IO.Int.Input(
+                                    "duration",
+                                    default=5,
+                                    min=1,
+                                    max=16,
+                                    step=1,
+                                    display_mode=IO.NumberDisplay.slider,
+                                    tooltip="Duration of the output video in seconds.",
+                                ),
+                                IO.Boolean.Input(
+                                    "audio",
+                                    default=False,
+                                    tooltip="When enabled, outputs video with sound "
+                                    "(including dialogue and sound effects).",
+                                ),
+                            ],
+                        ),
                     ],
                     tooltip="Model to use for video generation.",
                 ),
@@ -1442,13 +1519,20 @@ class Vidu3ImageToVideoNode(IO.ComfyNode):
             ],
             is_api_node=True,
             price_badge=IO.PriceBadge(
-                depends_on=IO.PriceBadgeDepends(widgets=["model.duration", "model.resolution"]),
+                depends_on=IO.PriceBadgeDepends(widgets=["model", "model.duration", "model.resolution"]),
                 expr="""
                 (
                   $res := $lookup(widgets, "model.resolution");
-                  $base := $lookup({"720p": 0.075, "1080p": 0.275, "2k": 0.35}, $res);
-                  $perSec := $lookup({"720p": 0.05, "1080p": 0.075, "2k": 0.075}, $res);
-                  {"type":"usd","usd": $base + $perSec * ($lookup(widgets, "model.duration") - 1)}
+                  $d := $lookup(widgets, "model.duration");
+                  $contains(widgets.model, "turbo")
+                    ? (
+                        $rate := $lookup({"720p": 0.06, "1080p": 0.08}, $res);
+                        {"type":"usd","usd": $rate * $d}
+                      )
+                    : (
+                        $rate := $lookup({"720p": 0.15, "1080p": 0.16, "2k": 0.2}, $res);
+                        {"type":"usd","usd": $rate * $d}
+                      )
                 )
                 """,
             ),
@@ -1481,6 +1565,145 @@ class Vidu3ImageToVideoNode(IO.ComfyNode):
         return IO.NodeOutput(await download_url_to_video_output(results[0].url))
 
 
+class Vidu3StartEndToVideoNode(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls):
+        return IO.Schema(
+            node_id="Vidu3StartEndToVideoNode",
+            display_name="Vidu Q3 Start/End Frame-to-Video Generation",
+            category="api node/video/Vidu",
+            description="Generate a video from a start frame, an end frame, and a prompt.",
+            inputs=[
+                IO.DynamicCombo.Input(
+                    "model",
+                    options=[
+                        IO.DynamicCombo.Option(
+                            "viduq3-pro",
+                            [
+                                IO.Combo.Input(
+                                    "resolution",
+                                    options=["720p", "1080p"],
+                                    tooltip="Resolution of the output video.",
+                                ),
+                                IO.Int.Input(
+                                    "duration",
+                                    default=5,
+                                    min=1,
+                                    max=16,
+                                    step=1,
+                                    display_mode=IO.NumberDisplay.slider,
+                                    tooltip="Duration of the output video in seconds.",
+                                ),
+                                IO.Boolean.Input(
+                                    "audio",
+                                    default=False,
+                                    tooltip="When enabled, outputs video with sound "
+                                    "(including dialogue and sound effects).",
+                                ),
+                            ],
+                        ),
+                        IO.DynamicCombo.Option(
+                            "viduq3-turbo",
+                            [
+                                IO.Combo.Input(
+                                    "resolution",
+                                    options=["720p", "1080p"],
+                                    tooltip="Resolution of the output video.",
+                                ),
+                                IO.Int.Input(
+                                    "duration",
+                                    default=5,
+                                    min=1,
+                                    max=16,
+                                    step=1,
+                                    display_mode=IO.NumberDisplay.slider,
+                                    tooltip="Duration of the output video in seconds.",
+                                ),
+                                IO.Boolean.Input(
+                                    "audio",
+                                    default=False,
+                                    tooltip="When enabled, outputs video with sound "
+                                    "(including dialogue and sound effects).",
+                                ),
+                            ],
+                        ),
+                    ],
+                    tooltip="Model to use for video generation.",
+                ),
+                IO.Image.Input("first_frame"),
+                IO.Image.Input("end_frame"),
+                IO.String.Input(
+                    "prompt",
+                    multiline=True,
+                    tooltip="Prompt description (max 2000 characters).",
+                ),
+                IO.Int.Input(
+                    "seed",
+                    default=1,
+                    min=0,
+                    max=2147483647,
+                    step=1,
+                    display_mode=IO.NumberDisplay.number,
+                    control_after_generate=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.duration", "model.resolution"]),
+                expr="""
+                (
+                  $res := $lookup(widgets, "model.resolution");
+                  $d := $lookup(widgets, "model.duration");
+                  $contains(widgets.model, "turbo")
+                    ? (
+                        $rate := $lookup({"720p": 0.06, "1080p": 0.08}, $res);
+                        {"type":"usd","usd": $rate * $d}
+                      )
+                    : (
+                        $rate := $lookup({"720p": 0.15, "1080p": 0.16}, $res);
+                        {"type":"usd","usd": $rate * $d}
+                      )
+                )
+                """,
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        model: dict,
+        first_frame: Input.Image,
+        end_frame: Input.Image,
+        prompt: str,
+        seed: int,
+    ) -> IO.NodeOutput:
+        validate_string(prompt, max_length=2000)
+        validate_images_aspect_ratio_closeness(first_frame, end_frame, min_rel=0.8, max_rel=1.25, strict=False)
+        payload = TaskCreationRequest(
+            model=model["model"],
+            prompt=prompt,
+            duration=model["duration"],
+            seed=seed,
+            resolution=model["resolution"],
+            audio=model["audio"],
+            images=[
+                (await upload_images_to_comfyapi(cls, frame, max_images=1, mime_type="image/png"))[0]
+                for frame in (first_frame, end_frame)
+            ],
+        )
+        results = await execute_task(cls, VIDU_START_END_VIDEO, payload)
+        return IO.NodeOutput(await download_url_to_video_output(results[0].url))
+
+
 class ViduExtension(ComfyExtension):
     @override
     async def get_node_list(self) -> list[type[IO.ComfyNode]]:
@@ -1497,6 +1720,7 @@ class ViduExtension(ComfyExtension):
             ViduMultiFrameVideoNode,
             Vidu3TextToVideoNode,
             Vidu3ImageToVideoNode,
+            Vidu3StartEndToVideoNode,
         ]
 
 

comfy_api_nodes/nodes_wan.py

@@ -227,12 +227,14 @@ class WanTextToImageApi(IO.ComfyNode):
                     default=True,
                     tooltip="Whether to enhance the prompt with AI assistance.",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "watermark",
                     default=False,
                     tooltip="Whether to add an AI-generated watermark to the result.",
                     optional=True,
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -355,6 +357,7 @@ class WanImageToImageApi(IO.ComfyNode):
                     default=False,
                     tooltip="Whether to add an AI-generated watermark to the result.",
                     optional=True,
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -495,18 +498,21 @@ class WanTextToVideoApi(IO.ComfyNode):
                     default=False,
                     optional=True,
                     tooltip="If no audio input is provided, generate audio automatically.",
+                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "prompt_extend",
                     default=True,
                     tooltip="Whether to enhance the prompt with AI assistance.",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "watermark",
                     default=False,
                     tooltip="Whether to add an AI-generated watermark to the result.",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "shot_type",
@@ -515,6 +521,7 @@ class WanTextToVideoApi(IO.ComfyNode):
                     "single continuous shot or multiple shots with cuts. "
                     "This parameter takes effect only when prompt_extend is True.",
                     optional=True,
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -667,18 +674,21 @@ class WanImageToVideoApi(IO.ComfyNode):
                     default=False,
                     optional=True,
                     tooltip="If no audio input is provided, generate audio automatically.",
+                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "prompt_extend",
                     default=True,
                     tooltip="Whether to enhance the prompt with AI assistance.",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "watermark",
                     default=False,
                     tooltip="Whether to add an AI-generated watermark to the result.",
                     optional=True,
+                    advanced=True,
                 ),
                 IO.Combo.Input(
                     "shot_type",
@@ -687,6 +697,7 @@ class WanImageToVideoApi(IO.ComfyNode):
                     "single continuous shot or multiple shots with cuts. "
                     "This parameter takes effect only when prompt_extend is True.",
                     optional=True,
+                    advanced=True,
                 ),
             ],
             outputs=[
@@ -839,11 +850,13 @@ class WanReferenceVideoApi(IO.ComfyNode):
                     options=["single", "multi"],
                     tooltip="Specifies the shot type for the generated video, that is, whether the video is a "
                     "single continuous shot or multiple shots with cuts.",
+                    advanced=True,
                 ),
                 IO.Boolean.Input(
                     "watermark",
                     default=False,
                     tooltip="Whether to add an AI-generated watermark to the result.",
+                    advanced=True,
                 ),
             ],
             outputs=[

comfy_api_nodes/util/__init__.py

@@ -9,6 +9,8 @@ from .client import (
 from .conversions import (
     audio_bytes_to_audio_input,
     audio_input_to_mp3,
+    audio_ndarray_to_bytesio,
+    audio_tensor_to_contiguous_ndarray,
     audio_to_base64_string,
     bytesio_to_image_tensor,
     convert_mask_to_image,
@@ -33,6 +35,7 @@ from .download_helpers import (
     download_url_to_video_output,
 )
 from .upload_helpers import (
+    upload_3d_model_to_comfyapi,
     upload_audio_to_comfyapi,
     upload_file_to_comfyapi,
     upload_image_to_comfyapi,
@@ -62,6 +65,7 @@ __all__ = [
     "sync_op",
     "sync_op_raw",
     # Upload helpers
+    "upload_3d_model_to_comfyapi",
     "upload_audio_to_comfyapi",
     "upload_file_to_comfyapi",
     "upload_image_to_comfyapi",
@@ -76,6 +80,8 @@ __all__ = [
     # Conversions
     "audio_bytes_to_audio_input",
     "audio_input_to_mp3",
+    "audio_ndarray_to_bytesio",
+    "audio_tensor_to_contiguous_ndarray",
     "audio_to_base64_string",
     "bytesio_to_image_tensor",
     "convert_mask_to_image",

comfy_api_nodes/util/conversions.py

@@ -57,7 +57,7 @@ def tensor_to_bytesio(
     image: torch.Tensor,
     *,
     total_pixels: int | None = 2048 * 2048,
-    mime_type: str = "image/png",
+    mime_type: str | None = "image/png",
 ) -> BytesIO:
     """Converts a torch.Tensor image to a named BytesIO object.
 

comfy_api_nodes/util/upload_helpers.py

@@ -164,6 +164,27 @@ async def upload_video_to_comfyapi(
     return await upload_file_to_comfyapi(cls, video_bytes_io, filename, upload_mime_type, wait_label)
 
 
+_3D_MIME_TYPES = {
+    "glb": "model/gltf-binary",
+    "obj": "model/obj",
+    "fbx": "application/octet-stream",
+}
+
+
+async def upload_3d_model_to_comfyapi(
+    cls: type[IO.ComfyNode],
+    model_3d: Types.File3D,
+    file_format: str,
+) -> str:
+    """Uploads a 3D model file to ComfyUI API and returns its download URL."""
+    return await upload_file_to_comfyapi(
+        cls,
+        model_3d.get_data(),
+        f"{uuid.uuid4()}.{file_format}",
+        _3D_MIME_TYPES.get(file_format, "application/octet-stream"),
+    )
+
+
 async def upload_file_to_comfyapi(
     cls: type[IO.ComfyNode],
     file_bytes_io: BytesIO,

comfy_extras/nodes_advanced_samplers.py

@@ -47,8 +47,8 @@ class SamplerLCMUpscale(io.ComfyNode):
             node_id="SamplerLCMUpscale",
             category="sampling/custom_sampling/samplers",
             inputs=[
-                io.Float.Input("scale_ratio", default=1.0, min=0.1, max=20.0, step=0.01),
-                io.Int.Input("scale_steps", default=-1, min=-1, max=1000, step=1),
+                io.Float.Input("scale_ratio", default=1.0, min=0.1, max=20.0, step=0.01, advanced=True),
+                io.Int.Input("scale_steps", default=-1, min=-1, max=1000, step=1, advanced=True),
                 io.Combo.Input("upscale_method", options=cls.UPSCALE_METHODS),
             ],
             outputs=[io.Sampler.Output()],
@@ -94,7 +94,7 @@ class SamplerEulerCFGpp(io.ComfyNode):
             display_name="SamplerEulerCFG++",
             category="_for_testing",  # "sampling/custom_sampling/samplers"
             inputs=[
-                io.Combo.Input("version", options=["regular", "alternative"]),
+                io.Combo.Input("version", options=["regular", "alternative"], advanced=True),
             ],
             outputs=[io.Sampler.Output()],
             is_experimental=True,

comfy_extras/nodes_apg.py

@@ -26,6 +26,7 @@ class APG(io.ComfyNode):
                     max=10.0,
                     step=0.01,
                     tooltip="Controls the scale of the parallel guidance vector. Default CFG behavior at a setting of 1.",
+                    advanced=True,
                 ),
                 io.Float.Input(
                     "norm_threshold",
@@ -34,6 +35,7 @@ class APG(io.ComfyNode):
                     max=50.0,
                     step=0.1,
                     tooltip="Normalize guidance vector to this value, normalization disable at a setting of 0.",
+                    advanced=True,
                 ),
                 io.Float.Input(
                     "momentum",
@@ -42,6 +44,7 @@ class APG(io.ComfyNode):
                     max=1.0,
                     step=0.01,
                     tooltip="Controls a running average of guidance during diffusion, disabled at a setting of 0.",
+                    advanced=True,
                 ),
             ],
             outputs=[io.Model.Output()],

comfy_extras/nodes_attention_multiply.py

@@ -28,10 +28,10 @@ class UNetSelfAttentionMultiply(io.ComfyNode):
             category="_for_testing/attention_experiments",
             inputs=[
                 io.Model.Input("model"),
-                io.Float.Input("q", default=1.0, min=0.0, max=10.0, step=0.01),
-                io.Float.Input("k", default=1.0, min=0.0, max=10.0, step=0.01),
-                io.Float.Input("v", default=1.0, min=0.0, max=10.0, step=0.01),
-                io.Float.Input("out", default=1.0, min=0.0, max=10.0, step=0.01),
+                io.Float.Input("q", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
+                io.Float.Input("k", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
+                io.Float.Input("v", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
+                io.Float.Input("out", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
             ],
             outputs=[io.Model.Output()],
             is_experimental=True,
@@ -51,10 +51,10 @@ class UNetCrossAttentionMultiply(io.ComfyNode):
             category="_for_testing/attention_experiments",
             inputs=[
                 io.Model.Input("model"),
-                io.Float.Input("q", default=1.0, min=0.0, max=10.0, step=0.01),
-                io.Float.Input("k", default=1.0, min=0.0, max=10.0, step=0.01),
-                io.Float.Input("v", default=1.0, min=0.0, max=10.0, step=0.01),
-                io.Float.Input("out", default=1.0, min=0.0, max=10.0, step=0.01),
+                io.Float.Input("q", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
+                io.Float.Input("k", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
+                io.Float.Input("v", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
+                io.Float.Input("out", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
             ],
             outputs=[io.Model.Output()],
             is_experimental=True,
@@ -75,10 +75,10 @@ class CLIPAttentionMultiply(io.ComfyNode):
             category="_for_testing/attention_experiments",
             inputs=[
                 io.Clip.Input("clip"),
-                io.Float.Input("q", default=1.0, min=0.0, max=10.0, step=0.01),
-                io.Float.Input("k", default=1.0, min=0.0, max=10.0, step=0.01),
-                io.Float.Input("v", default=1.0, min=0.0, max=10.0, step=0.01),
-                io.Float.Input("out", default=1.0, min=0.0, max=10.0, step=0.01),
+                io.Float.Input("q", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
+                io.Float.Input("k", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
+                io.Float.Input("v", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
+                io.Float.Input("out", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
             ],
             outputs=[io.Clip.Output()],
             is_experimental=True,
@@ -109,10 +109,10 @@ class UNetTemporalAttentionMultiply(io.ComfyNode):
             category="_for_testing/attention_experiments",
             inputs=[
                 io.Model.Input("model"),
-                io.Float.Input("self_structural", default=1.0, min=0.0, max=10.0, step=0.01),
-                io.Float.Input("self_temporal", default=1.0, min=0.0, max=10.0, step=0.01),
-                io.Float.Input("cross_structural", default=1.0, min=0.0, max=10.0, step=0.01),
-                io.Float.Input("cross_temporal", default=1.0, min=0.0, max=10.0, step=0.01),
+                io.Float.Input("self_structural", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
+                io.Float.Input("self_temporal", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
+                io.Float.Input("cross_structural", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
+                io.Float.Input("cross_temporal", default=1.0, min=0.0, max=10.0, step=0.01, advanced=True),
             ],
             outputs=[io.Model.Output()],
             is_experimental=True,

comfy_extras/nodes_audio.py

@@ -22,7 +22,7 @@ class EmptyLatentAudio(IO.ComfyNode):
             inputs=[
                 IO.Float.Input("seconds", default=47.6, 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."
+                    "batch_size", default=1, min=1, max=4096, tooltip="The number of latent images in the batch.",
                 ),
             ],
             outputs=[IO.Latent.Output()],
@@ -159,6 +159,7 @@ class SaveAudio(IO.ComfyNode):
             search_aliases=["export flac"],
             display_name="Save Audio (FLAC)",
             category="audio",
+            essentials_category="Audio",
             inputs=[
                 IO.Audio.Input("audio"),
                 IO.String.Input("filename_prefix", default="audio/ComfyUI"),
@@ -300,6 +301,7 @@ class LoadAudio(IO.ComfyNode):
             search_aliases=["import audio", "open audio", "audio file"],
             display_name="Load Audio",
             category="audio",
+            essentials_category="Audio",
             inputs=[
                 IO.Combo.Input("audio", upload=IO.UploadType.audio, options=sorted(files)),
             ],
@@ -677,6 +679,7 @@ class EmptyAudio(IO.ComfyNode):
                     tooltip="Sample rate of the empty audio clip.",
                     min=1,
                     max=192000,
+                    advanced=True,
                 ),
                 IO.Int.Input(
                     "channels",
@@ -684,6 +687,7 @@ class EmptyAudio(IO.ComfyNode):
                     min=1,
                     max=2,
                     tooltip="Number of audio channels (1 for mono, 2 for stereo).",
+                    advanced=True,
                 ),
             ],
             outputs=[IO.Audio.Output()],
@@ -698,6 +702,67 @@ class EmptyAudio(IO.ComfyNode):
     create_empty_audio = execute  # TODO: remove
 
 
+class AudioEqualizer3Band(IO.ComfyNode):
+    @classmethod
+    def define_schema(cls):
+        return IO.Schema(
+            node_id="AudioEqualizer3Band",
+            search_aliases=["eq", "bass boost", "treble boost", "equalizer"],
+            display_name="Audio Equalizer (3-Band)",
+            category="audio",
+            is_experimental=True,
+            inputs=[
+                IO.Audio.Input("audio"),
+                IO.Float.Input("low_gain_dB", default=0.0, min=-24.0, max=24.0, step=0.1, tooltip="Gain for Low frequencies (Bass)"),
+                IO.Int.Input("low_freq", default=100, min=20, max=500, tooltip="Cutoff frequency for Low shelf"),
+                IO.Float.Input("mid_gain_dB", default=0.0, min=-24.0, max=24.0, step=0.1, tooltip="Gain for Mid frequencies"),
+                IO.Int.Input("mid_freq", default=1000, min=200, max=4000, tooltip="Center frequency for Mids"),
+                IO.Float.Input("mid_q", default=0.707, min=0.1, max=10.0, step=0.1, tooltip="Q factor (bandwidth) for Mids"),
+                IO.Float.Input("high_gain_dB", default=0.0, min=-24.0, max=24.0, step=0.1, tooltip="Gain for High frequencies (Treble)"),
+                IO.Int.Input("high_freq", default=5000, min=1000, max=15000, tooltip="Cutoff frequency for High shelf"),
+            ],
+            outputs=[IO.Audio.Output()],
+        )
+
+    @classmethod
+    def execute(cls, audio, low_gain_dB, low_freq, mid_gain_dB, mid_freq, mid_q, high_gain_dB, high_freq) -> IO.NodeOutput:
+        waveform = audio["waveform"]
+        sample_rate = audio["sample_rate"]
+        eq_waveform = waveform.clone()
+
+        # 1. Apply Low Shelf (Bass)
+        if low_gain_dB != 0:
+            eq_waveform = torchaudio.functional.bass_biquad(
+                eq_waveform,
+                sample_rate,
+                gain=low_gain_dB,
+                central_freq=float(low_freq),
+                Q=0.707
+            )
+
+        # 2. Apply Peaking EQ (Mids)
+        if mid_gain_dB != 0:
+            eq_waveform = torchaudio.functional.equalizer_biquad(
+                eq_waveform,
+                sample_rate,
+                center_freq=float(mid_freq),
+                gain=mid_gain_dB,
+                Q=mid_q
+            )
+
+        # 3. Apply High Shelf (Treble)
+        if high_gain_dB != 0:
+            eq_waveform = torchaudio.functional.treble_biquad(
+                eq_waveform,
+                sample_rate,
+                gain=high_gain_dB,
+                central_freq=float(high_freq),
+                Q=0.707
+            )
+
+        return IO.NodeOutput({"waveform": eq_waveform, "sample_rate": sample_rate})
+
+
 class AudioExtension(ComfyExtension):
     @override
     async def get_node_list(self) -> list[type[IO.ComfyNode]]:
@@ -720,6 +785,7 @@ class AudioExtension(ComfyExtension):
             AudioMerge,
             AudioAdjustVolume,
             EmptyAudio,
+            AudioEqualizer3Band,
         ]
 
 async def comfy_entrypoint() -> AudioExtension:

comfy_extras/nodes_camera_trajectory.py

@@ -174,10 +174,10 @@ class WanCameraEmbedding(io.ComfyNode):
                 io.Int.Input("height", default=480, min=16, max=nodes.MAX_RESOLUTION, step=16),
                 io.Int.Input("length", default=81, min=1, max=nodes.MAX_RESOLUTION, step=4),
                 io.Float.Input("speed", default=1.0, min=0, max=10.0, step=0.1, optional=True),
-                io.Float.Input("fx", default=0.5, min=0, max=1, step=0.000000001, optional=True),
-                io.Float.Input("fy", default=0.5, min=0, max=1, step=0.000000001, optional=True),
-                io.Float.Input("cx", default=0.5, min=0, max=1, step=0.01, optional=True),
-                io.Float.Input("cy", default=0.5, min=0, max=1, step=0.01, optional=True),
+                io.Float.Input("fx", default=0.5, min=0, max=1, step=0.000000001, optional=True, advanced=True),
+                io.Float.Input("fy", default=0.5, min=0, max=1, step=0.000000001, optional=True, advanced=True),
+                io.Float.Input("cx", default=0.5, min=0, max=1, step=0.01, optional=True, advanced=True),
+                io.Float.Input("cy", default=0.5, min=0, max=1, step=0.01, optional=True, advanced=True),
             ],
             outputs=[
                 io.WanCameraEmbedding.Output(display_name="camera_embedding"),