Merge branch 'master' into ListInput

New Lora nodes using DynamicGroup widget (CORE-186) (#14655 )
feat: allow --comfy-api-base target ephemeral testenvs (#14569 )
2026-07-21 23:41:28 +08:00 · 2026-07-09 21:30:01 +08:00 · 2026-07-09 21:29:06 +08:00 · 2026-07-08 23:20:10 -07:00 · 2026-07-09 01:50:25 -04:00 · 2026-07-08 22:00:08 -07:00
328 changed files with 53084 additions and 15943 deletions
@@ -1,5 +1,4 @@
-As of the time of writing this you need this driver for best results:
-https://www.amd.com/en/resources/support-articles/release-notes/RN-AMDGPU-WINDOWS-PYTORCH-7-1-1.html
+As of the time of writing this you need a recent driver. Updating to the latest driver is recommended.

 HOW TO RUN:

@@ -7,9 +6,9 @@ If you have a AMD gpu:

 run_amd_gpu.bat

-If you have memory issues you can try disabling the smart memory management by running comfyui with:
+If you have memory issues you can try enabling the new dynamic memory management by running comfyui with:

-run_amd_gpu_disable_smart_memory.bat
+run_amd_gpu_enable_dynamic_vram.bat

 IF YOU GET A RED ERROR IN THE UI MAKE SURE YOU HAVE A MODEL/CHECKPOINT IN: ComfyUI\models\checkpoints

@@ -4,12 +4,12 @@ early_access: false
 tone_instructions: "Only comment on issues introduced by this PR's changes. Do not flag pre-existing problems in moved, re-indented, or reformatted code."

 reviews:
-  profile: "chill"
-  request_changes_workflow: false
+  profile: "assertive"
+  request_changes_workflow: true
  high_level_summary: false
  poem: false
  review_status: false
-  review_details: false
+  review_details: true
  commit_status: true
  collapse_walkthrough: true
  changed_files_summary: false
@@ -39,6 +39,14 @@ reviews:
    - path: "**"
      instructions: |
        IMPORTANT: Only comment on issues directly introduced by this PR's code changes.
+        Treat AGENTS.md as mandatory repository policy, not optional style guidance.
+        Flag PR changes that violate AGENTS.md even when the code is otherwise functional.
+        In particular, enforce architecture boundaries, dtype/device/memory rules,
+        interface contracts, import style, no unnecessary try/except blocks, no inline
+        imports, no outbound internet paths in core ComfyUI, and narrow scoped fixes.
+        Prefer direct findings over suggestions when a rule is violated. Only ignore
+        AGENTS.md when it clearly conflicts with a newer explicit maintainer instruction
+        in the PR.
        Do NOT flag pre-existing issues in code that was merely moved, re-indented,
        de-indented, or reformatted without logic changes. If code appears in the diff
        only due to whitespace or structural reformatting (e.g., removing a `with:` block),
@@ -123,5 +131,10 @@ chat:

 knowledge_base:
  opt_out: false
+  code_guidelines:
+    enabled: true
+    filePatterns:
+      - files: "AGENTS.md"
+        applyTo: "**"
  learnings:
    scope: "auto"
@@ -17,7 +17,7 @@ jobs:
      - name: Check for Windows line endings (CRLF)
        run: |
          # Get the list of changed files in the PR
-          CHANGED_FILES=$(git diff --name-only ${{ github.event.pull_request.base.sha }}..${{ github.event.pull_request.head.sha }})
+          CHANGED_FILES=$(git diff --name-only ${{ github.event.pull_request.base.sha }}..${{ github.event.pull_request.head.sha }} -- ':!.ci')

          # Flag to track if CRLF is found
          CRLF_FOUND=false
@@ -0,0 +1,38 @@
+name: CI - Cursor Review
+
+# Thin caller for the shared reusable cursor-review workflow in
+# Comfy-Org/github-workflows. The review logic (panel matrix, judge
+# consolidation, prompts, extract/post/notify scripts) lives there as the
+# single source of truth, so this repo only carries the repo-specific diff
+# excludes.
+
+on:
+  pull_request:
+    types: [labeled, unlabeled]
+
+concurrency:
+  group: cursor-review-pr-${{ github.event.pull_request.number }}-${{ github.event.label.name }}
+  cancel-in-progress: true
+
+jobs:
+  cursor-review:
+    if: github.event.label.name == 'cursor-review'
+    permissions:
+      contents: read
+      pull-requests: write
+    # SHA-pinned per zizmor `unpinned-uses: hash-pin`. Bump this SHA to pick up
+    # upstream changes; keep `workflows_ref` matching so prompts/scripts load
+    # from the same commit as the workflow definition.
+    uses: Comfy-Org/github-workflows/.github/workflows/cursor-review.yml@047ca48febe3a6647608ed2e0c4331b491cb9d6a # github-workflows#9
+    with:
+      workflows_ref: 047ca48febe3a6647608ed2e0c4331b491cb9d6a
+      diff_excludes: >-
+        :!**/.claude/**
+        :!**/dist/**
+        :!**/vendor/**
+        :!**/*.generated.*
+        :!**/*.min.js
+        :!**/*.min.css
+    secrets:
+      CURSOR_API_KEY: ${{ secrets.CURSOR_API_KEY }}
+      SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}
@@ -0,0 +1,91 @@
+name: CLA Assistant
+
+on:
+  issue_comment:
+    types: [created]
+  pull_request_target:
+    types: [opened, synchronize, closed]
+
+permissions:
+  actions: write
+  contents: read # 'read' is enough because signatures live in a REMOTE repo
+  pull-requests: write
+  statuses: write
+
+jobs:
+  cla-assistant:
+    runs-on: ubuntu-latest
+    steps:
+      # The CLA action normally requires every commit author in a PR to sign.
+      # We only want the PR author to sign, so we allowlist all other committers
+      # by computing them from the PR's commits and excluding the PR author.
+      - name: Build author-only allowlist
+        id: allowlist
+        if: >
+          github.event_name == 'pull_request_target' ||
+          (github.event_name == 'issue_comment' && github.event.issue.pull_request && (
+            github.event.comment.body == 'recheck' ||
+            github.event.comment.body == 'I have read and agree to the Contributor License Agreement'
+          ))
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          PR_NUMBER: ${{ github.event.pull_request.number || github.event.issue.number }}
+          PR_AUTHOR: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
+          BASE_ALLOWLIST: action@github.com,actions-user,ampagent,claude,comfy-pr-bot,GitHub Action,github-actions,github-actions[bot],Glary Bot,Glary-Bot,*[bot]
+        run: |
+          others=$(gh api "repos/${{ github.repository }}/pulls/${PR_NUMBER}/commits" --paginate \
+            --jq '.[] | (.author.login // empty), (.committer.login // empty)' \
+            | sort -u | grep -vix "${PR_AUTHOR}" | paste -sd, -)
+          if [ -n "$others" ]; then
+            echo "allowlist=${BASE_ALLOWLIST},${others}" >> "$GITHUB_OUTPUT"
+          else
+            echo "allowlist=${BASE_ALLOWLIST}" >> "$GITHUB_OUTPUT"
+          fi
+
+      - name: CLA Assistant
+        # Run on PR events, on "recheck" comment, or when someone posts the exact signing phrase.
+        # IMPORTANT: this phrase must match `custom-pr-sign-comment` below.
+        if: >
+          github.event_name == 'pull_request_target' ||
+          (github.event_name == 'issue_comment' && github.event.issue.pull_request && (
+            github.event.comment.body == 'recheck' ||
+            github.event.comment.body == 'I have read and agree to the Contributor License Agreement'
+          ))
+        uses: contributor-assistant/github-action@ca4a40a7d1004f18d9960b404b97e5f30a505a08 # v2.6.1
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          # PAT required to write to the centralized signatures repo.
+          PERSONAL_ACCESS_TOKEN: ${{ secrets.PERSONAL_ACCESS_TOKEN }}
+        with:
+          # Where the CLA document lives (shown to contributors)
+          path-to-document: https://github.com/Comfy-Org/comfy-cla/blob/main/comfyui_icla.md
+
+          # Centralized signature storage
+          remote-organization-name: comfy-org
+          remote-repository-name: comfy-cla
+          path-to-signatures: signatures/cla.json
+          branch: main
+
+          # Only the PR author must sign: bots plus every non-author committer
+          # are allowlisted via the "Build author-only allowlist" step above.
+          # *[bot] is a catch-all for any GitHub App bot account.
+          allowlist: ${{ steps.allowlist.outputs.allowlist }}
+
+          # Custom PR comment messages
+          custom-notsigned-prcomment: |
+            🎉 Thank you for your contribution, we really appreciate it! 🎉
+
+            Like many open source projects, we require contributors to sign our [Contributor License Agreement (CLA)](https://github.com/Comfy-Org/comfy-cla/blob/main/comfyui_icla.md). A CLA makes the ownership of contributions explicit, so contributors and the project share a clear understanding of how the code can be used. By signing, you:
+
+            - Confirm that you own your contribution.
+            - Keep the right to reuse your own code.
+            - Grant us a copyright license to include and share it within our projects.
+
+            CLAs are standard practice across major open source projects including those under the Apache Software Foundation and the Linux Foundation. Ours is based on the Apache Software Foundation's CLA. Most importantly, it would enable us to relicense the project under a more permissive license in the future, giving the project and its community greater flexibility.
+
+            ✍ **To sign, please post a new comment on this PR with exactly the following text:** ✍
+
+          custom-pr-sign-comment: I have read and agree to the Contributor License Agreement
+
+          custom-allsigned-prcomment: |
+            ✅ All contributors have signed the CLA. Thank you! This PR is ready to be merged.
@@ -0,0 +1,24 @@
+name: Detect Unreviewed Merge
+
+# SOC 2 compliance — reusable workflow lives in Comfy-Org/github-workflows,
+# tracking issues are filed in Comfy-Org/unreviewed-merges.
+
+on:
+  push:
+    branches: [master]
+
+concurrency:
+  group: detect-unreviewed-merge-${{ github.sha }}
+  cancel-in-progress: false
+
+permissions:
+  contents: read
+  pull-requests: read
+
+jobs:
+  detect:
+    uses: Comfy-Org/github-workflows/.github/workflows/detect-unreviewed-merge.yml@4d9cb6b87f953bb7cd69954280e1465fb9bd2040 # v1
+    with:
+      approval-mode: latest-per-reviewer
+    secrets:
+      UNREVIEWED_MERGES_TOKEN: ${{ secrets.UNREVIEWED_MERGES_TOKEN }}
@@ -0,0 +1,296 @@
+## Engineering Style
+
+- Keep changes small and direct. Most fixes should touch the narrowest code path
+  that explains the bug, performance issue, dtype issue, model-format issue, or
+  user-facing behavior.
+- Change the least amount of files possible. A change that touches many files is
+  more likely to be a bad change than a good one unless the broader scope is
+  directly required.
+- Prefer practical fixes over broad architecture work. Add abstractions only
+  when they remove real repeated logic or match an existing ComfyUI pattern.
+- Prefer fewer dependencies. Do not add new dependencies to ComfyUI unless they
+  are absolutely necessary.
+- Delete obsolete code aggressively when newer infrastructure makes it useless.
+  Remove dead fallbacks, migration paths, unused options, debug prints, and
+  compatibility branches that are no longer needed. Do not leave dead branches,
+  unreachable code, or functions that are never called. If code is not
+  necessary for the current behavior, remove it.
+- Revert or disable problematic behavior quickly when it breaks users. It is
+  better to remove a broken feature path than keep a complicated partial fix.
+- Preserve existing APIs, node names, model-loading behavior, file layout, and
+  workflow compatibility unless the change is explicitly about replacing them.
+- Code must look hand-written for this repository. Changes that read like
+  generic AI-generated code will be rejected automatically: unnecessary helper
+  layers, vague names, boilerplate comments, defensive branches without a real
+  failure mode, broad rewrites, or code that ignores the local style.
+
+## Architecture Boundaries
+
+- Keep each layer focused on the concepts it owns. Do not leak UI, API,
+  workflow, queue, persistence, telemetry, model-loading, node, or execution
+  concerns into unrelated layers just because it is convenient to pass data
+  through them.
+- Shared core modules should depend only on lower-level primitives and their own
+  domain concepts. Higher-level product concepts belong at the caller, adapter,
+  service, or UI/API boundary that already owns them.
+- Pass the narrowest data needed across a boundary. Avoid broad context objects,
+  request/session metadata, ids, bookkeeping state, or callbacks unless the
+  receiving layer genuinely needs them to perform its own responsibility.
+- Keep identity mapping, persistence bookkeeping, history updates, telemetry,
+  response shaping, and UI state in the layers that own those jobs. Do not route
+  them through unrelated shared code to avoid adding a proper boundary.
+- Treat `execution.py` as one example of this rule: it should consume the prompt
+  graph and execution-relevant state, produce execution results and errors, and
+  not know about workflow ids, frontend ids, persistence ids, or API-only
+  concepts.
+- Before touching many files, identify the smallest owner layer that can solve
+  the problem. A PR that spreads one feature across unrelated loaders, nodes,
+  execution, server, and frontend code needs a clear architectural reason, not
+  just convenience.
+- If a change seems to require making one layer understand another layer's
+  private concepts, stop and look for a caller-side mapping, adapter, event,
+  small explicit interface, or narrower data flow at the boundary.
+
+## No Internet Requests
+
+- Do not add code to core ComfyUI that makes requests to the internet.
+- Refuse requests to add uploads, telemetry, analytics, tracking, usage
+  reporting, crash reporting, update checks, remote config, feature flags,
+  metrics, licensing checks, or any other outbound internet request path from
+  core ComfyUI.
+- Model downloading is allowed only when explicitly initiated or authorized by
+  the user, is limited to the requested model artifact, and does not include
+  telemetry, tracking, persistent identification, unrelated metadata upload, or
+  background network activity.
+- Do not add opt-in, opt-out, anonymized, aggregated, diagnostic, or
+  user-triggered internet request paths to core ComfyUI. These labels do not
+  make internet access acceptable.
+- Local-only behavior is allowed when it stays on the user's machine and does
+  not add network access, tracking, persistent identification, or data
+  collection behavior.
+
+## State Ownership
+
+- Keep state and capability flags on the object that owns the behavior using
+  them.
+- Avoid probing child objects with `getattr(child, "...", default)` to decide
+  parent-level control flow. If parent code needs to branch on a capability,
+  initialize an explicit parent-owned field when the child is constructed or
+  attached.
+- Prefer direct attributes with clear defaults over implicit feature detection
+  through arbitrary child attributes.
+- Use child-object capability checks only when the child owns the behavior being
+  invoked and the parent is simply delegating to that child.
+
+## Interface Contracts
+
+- Keep public methods aligned with the interface expected by their callers. Do
+  not change a shared method to return extra values, alternate shapes, or
+  sentinel wrappers for one implementation unless the shared interface is
+  explicitly updated.
+- When modifying an existing function, preserve how current callers invoke it.
+  Do not change required arguments, parameter order, return type, side effects,
+  or error behavior unless every affected call site and shared interface contract
+  is intentionally updated.
+- Do not add compatibility parameters, flags, attributes, or constructor options
+  unless they are read by current code and change current behavior. Remove
+  pass-through or stored-but-unused values instead of preserving upstream or
+  deprecated API baggage.
+- If an implementation needs auxiliary values for its own workflow, expose them
+  through a private helper or a clearly named implementation-specific method
+  instead of overloading the public method's return contract.
+- Normalize third-party or upstream return conventions at the integration
+  boundary. Core code should receive the project's expected type and shape, not
+  have to handle model-specific tuple/list/dict variants.
+- Avoid caller-side unwrapping such as `out = out[0]` unless the called
+  interface is documented to return that structure.
+
+## Autograd and Model Freezing
+
+- Do not add `torch.no_grad`, `torch.inference_mode`, or inference-mode helper
+  wrappers in ComfyUI code. The only allowed inference-mode-related use is
+  disabling a globally set inference mode when a training path needs gradients.
+- Do not add freeze, unfreeze, or trainability toggles to model classes. ComfyUI
+  models are always treated as frozen for inference, so explicit freeze
+  functionality is redundant and should not be added.
+- Remove training-only behavior such as dropout from inference model code, but
+  preserve checkpoint and state-dict compatibility when doing so. If deleting a
+  module would change state-dict keys, module ordering, or checkpoint loading
+  behavior, replace it with a no-op such as `nn.Identity` instead of removing the
+  slot outright.
+
+## Python Style
+
+- Keep imports at module scope. Avoid inline imports unless they are already part
+  of an established optional-backend probe or are needed to avoid an import
+  cycle.
+- Do not add unnecessary `try`/`except` blocks. Use them for optional dependency,
+  platform, or backend capability detection only when the program has a useful
+  fallback. Prefer specific exception types when changing new code.
+- If a library version is pinned in `requirements.txt`, do not add code to
+  ComfyUI to handle older versions of that library.
+- Remove any workarounds for PyTorch versions that ComfyUI no longer officially
+  supports. Deprecated workarounds include catching an exception and rerunning
+  the same op with the input cast to float. If a workaround does not have a
+  comment naming the exact PyTorch version or versions that still need it,
+  remove it.
+- Let unsupported model formats, invalid quantization metadata, and bad states
+  fail with clear errors instead of silently producing lower quality output.
+- Match the existing local style in the file you edit. This codebase tolerates
+  long lines, simple helper functions, module-level state, and direct tensor
+  operations when they make the code easier to follow.
+- Keep comments sparse and useful. Strip useless comments that restate the code
+  or describe obvious behavior. Short TODOs are fine when they name the concrete
+  missing follow-up.
+
+## Model, Device, and Memory Behavior
+
+- Treat dtype, device placement, VRAM usage, and offloading behavior as core
+  correctness concerns. Check CPU, CUDA, ROCm, MPS, DirectML, XPU, NPU, and low
+  VRAM implications when touching shared execution or loading code.
+- Prefer native ComfyUI formats and existing quantization/offload helpers over
+  adding parallel code paths. Use `comfy.quant_ops`, `comfy.model_management`,
+  `comfy.memory_management`, `comfy.pinned_memory`, `comfy_aimdo`, and
+  `comfy-kitchen` helpers where they already solve the problem.
+- Use optimized comfy-kitchen ops in places where they improve performance
+  without changing the expected dtype, device, memory, or interface behavior.
+- All models should use the optimized attention function selected by ComfyUI.
+  Treat optimized backend functions, dispatch helpers, and capability-selected
+  callables as opaque. Higher-level code must not inspect function identity,
+  names, modules, or implementation details to decide behavior.
+- Apply the same opacity rule to similar patterns beyond attention: callers
+  should depend on the documented interface and result contract, not on which
+  backend implementation was selected underneath.
+- Do not use custom inference ops that only duplicate an existing op while
+  upcasting to float32, such as custom RMSNorm variants. Use the generic ComfyUI
+  ops and/or native torch ops instead.
+- If a model class `__init__` has an `operations` parameter, assume
+  `operations` is never `None`. Do not add fallback branches or default torch
+  ops for a missing `operations` object.
+- Do not add unnecessary parameters to model, model block, or model ops related
+  classes. Constructor and forward signatures should carry only values that are
+  actually needed by that object for inference.
+- Reuse existing model classes, blocks, ops, and helper modules when appropriate.
+  Before implementing a new version of a model component, search the existing
+  model code for a class or helper that already provides the behavior.
+- Model detection code that inspects linear weight shapes should only use the
+  first dimension. The second dimension may be half the original size for
+  NVFP4 or other 4-bit quantized models.
+- Avoid adding `einops` usage in core inference code. Use native torch tensor
+  ops such as `reshape`, `view`, `permute`, `transpose`, `flatten`, `unflatten`,
+  `unsqueeze`, and `squeeze` instead.
+- Do not use tensors as general-purpose Python data structures. Keep metadata,
+  bookkeeping, counters, flags, shape math, padding math, index planning, memory
+  estimates, and control-flow decisions in plain Python values unless the data
+  must participate directly in tensor computation. Do not create tensors for
+  structural metadata that is only used for Python-side control flow. Sequence
+  lengths, cumulative offsets, split indices, window counts, slice boundaries,
+  and repeat counts should be kept as Python ints/lists from the point they are
+  computed. Do not build them as CPU/GPU tensors and then cast, move, validate,
+  or convert them back to Python for `split`, `tensor_split`, indexing plans,
+  loops, or cache keys. Avoid creating temporary tensors just to use tensor
+  methods for scalar or structural calculations.
+- Avoid unnecessary casts and transfers. Preserve the intended compute dtype,
+  storage dtype, bias dtype, and original tensor shape metadata.
+- Keep model-native latent layout handling inside the model or latent-format
+  owner, not in helper nodes. Do not collapse, expand, pack, or unpack latent
+  dimensions in nodes or other caller-side adapters just to satisfy a model
+  forward; the model path should consume and return the native latent shape for
+  that model family.
+- Assume inputs to the main model forward are already in the compute dtype by
+  default, except integer inputs such as some model timestep tensors. Do not add
+  defensive or convenience casts in model code; it is better for invalid dtype
+  plumbing to error clearly than to hide it with unnecessary casts.
+- Raw model parameters that are not owned by an op and may be initialized in a
+  dtype different from the compute dtype should be cast at use in forward or
+  inference code with `comfy.ops.cast_to_input` or
+  `comfy.model_management.cast_to` to avoid dtype mismatches.
+- Model code should not care what dtype it is initialized in, and model
+  `__init__` methods should not contain workarounds for specific dtypes. Dtype
+  workaround code, such as making a model work with fp16 compute, belongs in the
+  execution or model-management layer that owns compute policy.
+- Model code should not perform unnecessary device-to-CPU or CPU-to-device
+  transfers. New allocations must be created on the correct device and dtype;
+  never allocate on CPU and then move to GPU, or allocate in one dtype and then
+  convert to another.
+- Model code itself should not perform memory management. Loading, unloading,
+  offloading, device movement, VRAM policy, cache lifetime, and cleanup belong
+  in the relevant model-management and execution layers, not inside model
+  implementations.
+- Do not add global, module-level, class-level, singleton, or model-owned stores
+  for tensors or other large memory that persist across executions. Temporary
+  caches must be scoped to a single execution or forward/encode/decode call:
+  allocate them in the owning top-level call, pass them explicitly through the
+  call stack, and let them be discarded when that call returns.
+- Follow the Wan VAE temporal cache pattern for temporary caches: create a local
+  cache such as `feat_map` for the encode/decode operation, pass it into the
+  blocks that need it, and do not retain it on the model or in global state.
+- In model init code, prefer `torch.empty` for parameter/buffer placeholders
+  that are populated from the model state dict instead of zero-initializing with
+  `torch.zeros` or similar. If an allocation is not loaded from the state dict
+  and is useless for inference, do not include it.
+- `nn.Parameter` tensors that are stored in and populated from the model state
+  dict should be initialized with `torch.empty`, not with zero, random, or
+  otherwise meaningful initialization.
+- Model initialization should describe module structure, not fabricate
+  checkpoint-owned tensor contents. Parameters and buffers that are loaded from
+  the state dict must not be manually initialized, reassigned, or filled with
+  fallback values unless that value is actually used when no checkpoint key
+  exists.
+- When slicing large tensors, copy the slice if the sliced tensor's lifetime
+  exceeds the current function scope. Do not keep a long-lived view into a large
+  backing tensor when a smaller copy would release memory sooner.
+- Use fused or compound torch operations such as `addcmul` when they naturally
+  match the math. Reducing Python and torch dispatch overhead is a valid
+  optimization when it does not obscure the code or change dtype/device
+  behavior.
+- Avoid caches that persist across different executions as much as possible.
+  Persistent caches are acceptable only when they use a very minimal amount of
+  memory and have a clear ownership and invalidation story.
+- When optimizing, favor small measurable changes: fewer allocations, fewer
+  device transfers, less peak memory, better batching, or use of a faster
+  existing backend op.
+
+## Nodes and User-Facing Behavior
+
+- Follow existing node conventions: `INPUT_TYPES`, `RETURN_TYPES`, `FUNCTION`,
+  `CATEGORY`, and registration through the local mapping used by that file.
+- Keep node changes backward compatible by default. Add inputs with sensible
+  defaults and avoid changing output types unless the request requires it.
+- Model implementations should add the minimal number of ComfyUI nodes required
+  to run the model. Reuse existing nodes as much as possible; adapting the model
+  to work with existing nodes is strongly preferred over creating new nodes.
+- Nodes should output only values they own. Do not add pass-through outputs for
+  workflow convenience unless the node is explicitly an output node. Existing
+  models, latents, conditioning, or other inputs should flow directly to the
+  next consumer instead of being re-emitted unchanged.
+- Nodes should expose only inputs they actually read to produce current
+  behavior. Do not add placeholder, pass-through, compatibility, or
+  workflow-shaping inputs that are ignored or could flow directly to another
+  node.
+- Node-level code must not patch model code directly. Any node behavior that
+  modifies, wraps, hooks, or changes model behavior must go through the model
+  patcher class instead of reaching into model internals.
+- The official mascot of ComfyUI is a very cute anime girl with massive fennec
+  ears, a big fluffy tail, long blonde wavy hair, and blue eyes. Feel free to
+  use her in ComfyUI materials, UI text, examples, tests, generated assets, or
+  comments, but do not disrespect her.
+- Warning and info messages should be short and actionable. Remove noisy or
+  misleading messages rather than adding more logging.
+- Documentation and README edits should be concise, factual, and tied to the
+  changed behavior.
+
+## Commit and Review Habits
+
+- If asked to write commit messages, use short direct subjects like the existing
+  history: `Fix ...`, `Add ...`, `Support ...`, `Remove ...`, `Update ...`,
+  `Make ...`, `Use ...`, `Disable ...`, `Bump ...`, or `Revert ...`.
+- Keep PR descriptions short and reviewable. State the problem, the behavioral
+  change, and the tests run; avoid long narrative explanations, implementation
+  diaries, or exhaustive file-by-file summaries unless the reviewer explicitly
+  needs that context.
+- Prefer one coherent behavioral change per commit. Dependency pins, tests, and
+  the code that needs them may be in the same commit when they are inseparable.
+- In reviews, prioritize real user impact: crashes, wrong dtype/device behavior,
+  memory regressions, broken model loading, workflow incompatibility, and noisy
+  or misleading user-facing output.
@@ -140,7 +140,7 @@ ComfyUI follows a weekly release cycle targeting Monday but this regularly chang
   - Commits outside of the stable release tags may be very unstable and break many custom nodes.
   - Serves as the foundation for the desktop release

-2. **[ComfyUI Desktop](https://github.com/Comfy-Org/desktop)**
+2. **[Comfy Desktop](https://github.com/Comfy-Org/Comfy-Desktop)**
   - Builds a new release using the latest stable core version

 3. **[ComfyUI Frontend](https://github.com/Comfy-Org/ComfyUI_frontend)**
@@ -229,7 +229,7 @@ Python 3.14 works but some custom nodes may have issues. The free threaded varia

 Python 3.13 is very well supported. If you have trouble with some custom node dependencies on 3.13 you can try 3.12

-torch 2.4 and above is supported but some features and optimizations might only work on newer versions. We generally recommend using the latest major version of pytorch with the latest cuda version unless it is less than 2 weeks old.
+torch 2.5 is minimally supported but using a newer version is extremely recommended. Some features and optimizations might only work on newer versions. We generally recommend using the latest major version of pytorch with the latest cuda version unless it is less than 2 weeks old. If your pytorch is more than 6 months old, please update it.

 ### Instructions:

@@ -309,7 +309,7 @@ After this you should have everything installed and can proceed to running Comfy

 #### Apple Mac silicon

-You can install ComfyUI in Apple Mac silicon (M1 or M2) with any recent macOS version.
+You can install ComfyUI in Apple Mac silicon (M1, M2, M3 or M4) with any recent macOS version.

 1. Install pytorch nightly. For instructions, read the [Accelerated PyTorch training on Mac](https://developer.apple.com/metal/pytorch/) Apple Developer guide (make sure to install the latest pytorch nightly).
 1. Follow the [ComfyUI manual installation](#manual-install-windows-linux) instructions for Windows and Linux.
@@ -364,7 +364,7 @@ For models compatible with Iluvatar Extension for PyTorch. Here's a step-by-step
 | Flag | Description |
 |------|-------------|
 | `--enable-manager` | Enable ComfyUI-Manager |
-| `--enable-manager-legacy-ui` | Use the legacy manager UI instead of the new UI (requires `--enable-manager`) |
+| `--enable-manager-legacy-ui` | Use the legacy manager UI instead of the new UI (implies `--enable-manager`) |
 | `--disable-manager-ui` | Disable the manager UI and endpoints while keeping background features like security checks and scheduled installation completion (requires `--enable-manager`) |


@@ -382,11 +382,7 @@ For AMD 7600 and maybe other RDNA3 cards: ```HSA_OVERRIDE_GFX_VERSION=11.0.0 pyt

 ### AMD ROCm Tips

-You can enable experimental memory efficient attention on recent pytorch in ComfyUI on some AMD GPUs using this command, it should already be enabled by default on RDNA3. If this improves speed for you on latest pytorch on your GPU please report it so that I can enable it by default.
-
-```TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL=1 python main.py --use-pytorch-cross-attention```
-
-You can also try setting this env variable `PYTORCH_TUNABLEOP_ENABLED=1` which might speed things up at the cost of a very slow initial run.
+You can try setting this env variable `PYTORCH_TUNABLEOP_ENABLED=1` which might speed things up at the cost of a very slow initial run.

 # Notes

@@ -462,16 +458,6 @@ To use the most up-to-date frontend version:

 This approach allows you to easily switch between the stable fortnightly release and the cutting-edge daily updates, or even specific versions for testing purposes.

-### Accessing the Legacy Frontend
-
-If you need to use the legacy frontend for any reason, you can access it using the following command line argument:
-
-```
--front-end-version Comfy-Org/ComfyUI_legacy_frontend@latest
-```
-
-This will use a snapshot of the legacy frontend preserved in the [ComfyUI Legacy Frontend repository](https://github.com/Comfy-Org/ComfyUI_legacy_frontend).
-
 # QA

 ### Which GPU should I buy for this?
@@ -0,0 +1,39 @@
+"""
+Drop the vestigial tags.tag_type column.
+
+tag_type was always "user" in practice — no code path ever set it to anything
+else (no system/seeded classification was ever wired up) and nothing queried it.
+The column, its index (ix_tags_tag_type), and the corresponding API field were
+dead weight, so they are removed.
+
+Revision ID: 0004_drop_tag_type
+Revises: 0003_add_metadata_job_id
+Create Date: 2026-06-03
+"""
+
+from alembic import op
+import sqlalchemy as sa
+
+revision = "0004_drop_tag_type"
+down_revision = "0003_add_metadata_job_id"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    with op.batch_alter_table("tags") as batch_op:
+        batch_op.drop_index("ix_tags_tag_type")
+        batch_op.drop_column("tag_type")
+
+
+def downgrade() -> None:
+    with op.batch_alter_table("tags") as batch_op:
+        batch_op.add_column(
+            sa.Column(
+                "tag_type",
+                sa.String(length=32),
+                nullable=False,
+                server_default="user",
+            )
+        )
+        batch_op.create_index("ix_tags_tag_type", ["tag_type"])
@@ -0,0 +1,107 @@
+"""
+Allow case-sensitive tag names.
+
+Revision ID: 0005_allow_case_sensitive_tags
+Revises: 0004_drop_tag_type
+Create Date: 2026-06-16
+"""
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = "0005_allow_case_sensitive_tags"
+down_revision = "0004_drop_tag_type"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    bind = op.get_bind()
+    if bind.dialect.name == "sqlite":
+        # SQLite cannot ALTER/DROP CHECK constraints. Recreate the small tag
+        # vocabulary table without the lowercase constraint while preserving
+        # existing tag names.
+        op.execute("PRAGMA foreign_keys=OFF")
+        try:
+            op.execute(
+                "CREATE TABLE tags_new ("
+                "name VARCHAR(512) NOT NULL, "
+                "CONSTRAINT pk_tags PRIMARY KEY (name)"
+                ")"
+            )
+            op.execute("INSERT INTO tags_new(name) SELECT name FROM tags")
+            op.execute("DROP TABLE tags")
+            op.execute("ALTER TABLE tags_new RENAME TO tags")
+        finally:
+            op.execute("PRAGMA foreign_keys=ON")
+        return
+
+    op.drop_constraint("ck_tags_ck_tags_lowercase", "tags", type_="check")
+
+
+def downgrade() -> None:
+    # Existing mixed-case tags cannot satisfy the old constraint. Lowercase them
+    # before restoring it, merging duplicate vocabulary/link rows that collide.
+    bind = op.get_bind()
+
+    tag_names = [row[0] for row in bind.execute(sa.text("SELECT name FROM tags"))]
+    existing_names = set(tag_names)
+    lowercase_names = sorted({name.lower() for name in tag_names})
+    missing_lowercase_rows = [
+        {"name": name} for name in lowercase_names if name not in existing_names
+    ]
+    if missing_lowercase_rows:
+        bind.execute(sa.text("INSERT INTO tags(name) VALUES (:name)"), missing_lowercase_rows)
+
+    link_rows = bind.execute(
+        sa.text(
+            "SELECT asset_reference_id, tag_name, origin, added_at "
+            "FROM asset_reference_tags "
+            "ORDER BY asset_reference_id, tag_name"
+        )
+    ).mappings()
+    deduped_links = {}
+    for row in link_rows:
+        key = (row["asset_reference_id"], row["tag_name"].lower())
+        deduped_links.setdefault(
+            key,
+            {
+                "asset_reference_id": row["asset_reference_id"],
+                "tag_name": row["tag_name"].lower(),
+                "origin": row["origin"],
+                "added_at": row["added_at"],
+            },
+        )
+
+    op.execute("DELETE FROM asset_reference_tags")
+    if deduped_links:
+        bind.execute(
+            sa.text(
+                "INSERT INTO asset_reference_tags "
+                "(asset_reference_id, tag_name, origin, added_at) "
+                "VALUES (:asset_reference_id, :tag_name, :origin, :added_at)"
+            ),
+            list(deduped_links.values()),
+        )
+    op.execute("DELETE FROM tags WHERE name != lower(name)")
+
+    if bind.dialect.name == "sqlite":
+        op.execute("PRAGMA foreign_keys=OFF")
+        try:
+            op.execute(
+                "CREATE TABLE tags_new ("
+                "name VARCHAR(512) NOT NULL, "
+                "CONSTRAINT pk_tags PRIMARY KEY (name), "
+                "CONSTRAINT ck_tags_lowercase CHECK (name = lower(name))"
+                ")"
+            )
+            op.execute("INSERT INTO tags_new(name) SELECT name FROM tags")
+            op.execute("DROP TABLE tags")
+            op.execute("ALTER TABLE tags_new RENAME TO tags")
+        finally:
+            op.execute("PRAGMA foreign_keys=ON")
+        return
+
+    op.create_check_constraint(
+        "ck_tags_ck_tags_lowercase", "tags", "name = lower(name)"
+    )
@@ -0,0 +1,30 @@
+"""
+Add loader_path column to asset_references.
+
+Stores the in-root loader path (path relative to the storage root with the
+top-level model category dropped) derived from file_path at scan/ingest time,
+so the assets API can return it without re-resolving against every registered
+model-folder base on every request.
+
+Revision ID: 0006_add_loader_path
+Revises: 0005_allow_case_sensitive_tags
+Create Date: 2026-07-02
+"""
+
+from alembic import op
+import sqlalchemy as sa
+
+revision = "0006_add_loader_path"
+down_revision = "0005_allow_case_sensitive_tags"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    with op.batch_alter_table("asset_references") as batch_op:
+        batch_op.add_column(sa.Column("loader_path", sa.Text(), nullable=True))
+
+
+def downgrade() -> None:
+    with op.batch_alter_table("asset_references") as batch_op:
+        batch_op.drop_column("loader_path")
@@ -39,6 +39,8 @@ from app.assets.services import (
    update_asset_metadata,
    upload_from_temp_path,
 )
+from app.assets.services.cursor import InvalidCursorError
+from app.assets.services.path_utils import compute_display_name
 from app.assets.services.tagging import list_tag_histogram

 ROUTES = web.RouteTableDef()
@@ -160,11 +162,19 @@ def _build_asset_response(result: schemas.AssetDetailResult | schemas.UploadResu
            preview_url = None
    else:
        preview_url = _build_preview_url_from_view(result.tags, result.ref.user_metadata)
+    if result.ref.file_path:
+        display_name = compute_display_name(result.ref.file_path)
+        # In-root loader path (model category dropped): what model loaders consume.
+        loader_path = result.ref.loader_path
+    else:
+        display_name, loader_path = None, None
    asset_content_hash = result.asset.hash if result.asset else None
    return schemas_out.Asset(
        id=result.ref.id,
        name=result.ref.name,
        hash=asset_content_hash,
+        loader_path=loader_path,
+        display_name=display_name,
        asset_hash=asset_content_hash,
        size=int(result.asset.size_bytes) if result.asset else None,
        mime_type=result.asset.mime_type if result.asset else None,
@@ -174,7 +184,7 @@ def _build_asset_response(result: schemas.AssetDetailResult | schemas.UploadResu
        user_metadata=result.ref.user_metadata or {},
        metadata=result.ref.system_metadata,
        job_id=result.ref.job_id,
-        prompt_id=result.ref.job_id,  # deprecated: mirrors job_id for cloud compat
+        prompt_id=result.ref.job_id,  # deprecated alias of job_id, kept for compatibility
        created_at=result.ref.created_at,
        updated_at=result.ref.updated_at,
        last_access_time=result.ref.last_access_time,
@@ -211,24 +221,37 @@ async def list_assets_route(request: web.Request) -> web.Response:
    order_candidate = (q.order or "desc").lower()
    order = order_candidate if order_candidate in {"asc", "desc"} else "desc"

-    result = list_assets_page(
-        owner_id=USER_MANAGER.get_request_user_id(request),
-        include_tags=q.include_tags,
-        exclude_tags=q.exclude_tags,
-        name_contains=q.name_contains,
-        metadata_filter=q.metadata_filter,
-        limit=q.limit,
-        offset=q.offset,
-        sort=sort,
-        order=order,
-    )
+    try:
+        result = list_assets_page(
+            owner_id=USER_MANAGER.get_request_user_id(request),
+            include_tags=q.include_tags,
+            exclude_tags=q.exclude_tags,
+            name_contains=q.name_contains,
+            metadata_filter=q.metadata_filter,
+            limit=q.limit,
+            offset=q.offset,
+            sort=sort,
+            order=order,
+            after=q.after,
+        )
+    except InvalidCursorError as e:
+        return _build_error_response(400, "INVALID_CURSOR", str(e))

    summaries = [_build_asset_response(item) for item in result.items]

+    # has_more semantics differ by mode:
+    #   - cursor mode: a non-empty next_cursor means there are more results.
+    #   - offset mode: derived from total - (offset + page size).
+    if q.after is not None:
+        has_more = result.next_cursor is not None
+    else:
+        has_more = (q.offset + len(summaries)) < result.total
+
    payload = schemas_out.AssetsList(
        assets=summaries,
        total=result.total,
-        has_more=(q.offset + len(summaries)) < result.total,
+        has_more=has_more,
+        next_cursor=result.next_cursor,
    )
    return web.json_response(payload.model_dump(mode="json", exclude_none=True))

@@ -292,12 +315,15 @@ async def download_asset_content(request: web.Request) -> web.Response:
            404, "FILE_NOT_FOUND", "Underlying file not found on disk."
        )

-    _DANGEROUS_MIME_TYPES = {
-        "text/html", "text/html-sandboxed", "application/xhtml+xml",
-        "text/javascript", "text/css",
-    }
-    if content_type in _DANGEROUS_MIME_TYPES:
+    # User-controlled asset content must never render inline in the app origin
+    # (stored XSS via SVG/HTML/XML). Force dangerous types to download and
+    # override any requested inline disposition. Centralised through
+    # folder_paths.is_dangerous_content_type so this can't drift from /view and
+    # /userdata (the previous inline set here omitted image/svg+xml and missed
+    # the charset/casing/+xml-dialect bypasses).
+    if folder_paths.is_dangerous_content_type(content_type):
        content_type = "application/octet-stream"
+        disposition = "attachment"

    safe_name = (filename or "").replace("\r", "").replace("\n", "")
    encoded = urllib.parse.quote(safe_name)
@@ -402,17 +428,6 @@ async def upload_asset(request: web.Request) -> web.Response:
            400, "INVALID_BODY", f"Validation failed: {ve.json()}"
        )

-    if spec.tags and spec.tags[0] == "models":
-        if (
-            len(spec.tags) < 2
-            or spec.tags[1] not in folder_paths.folder_names_and_paths
-        ):
-            delete_temp_file_if_exists(parsed.tmp_path)
-            category = spec.tags[1] if len(spec.tags) >= 2 else ""
-            return _build_error_response(
-                400, "INVALID_BODY", f"unknown models category '{category}'"
-            )
-
    try:
        # Fast path: hash exists, create AssetReference without writing anything
        if spec.hash and parsed.provided_hash_exists is True:
@@ -456,7 +471,7 @@ async def upload_asset(request: web.Request) -> web.Response:
        return _build_error_response(400, e.code, str(e))
    except ValueError as e:
        delete_temp_file_if_exists(parsed.tmp_path)
-        return _build_error_response(400, "BAD_REQUEST", str(e))
+        return _build_error_response(400, "INVALID_BODY", str(e))
    except HashMismatchError as e:
        delete_temp_file_if_exists(parsed.tmp_path)
        return _build_error_response(400, "HASH_MISMATCH", str(e))
@@ -519,18 +534,14 @@ async def update_asset_route(request: web.Request) -> web.Response:
@_require_assets_feature_enabled
 async def delete_asset_route(request: web.Request) -> web.Response:
    reference_id = str(uuid.UUID(request.match_info["id"]))
-    delete_content_param = request.query.get("delete_content")
-    delete_content = (
-        False
-        if delete_content_param is None
-        else delete_content_param.lower() not in {"0", "false", "no"}
-    )

    try:
+        # Deleting an asset is a soft delete of the reference; the underlying
+        # content is preserved (it may be shared with other references).
        deleted = delete_asset_reference(
            reference_id=reference_id,
            owner_id=USER_MANAGER.get_request_user_id(request),
-            delete_content_if_orphan=delete_content,
+            delete_content_if_orphan=False,
        )
    except Exception:
        logging.exception(
@@ -575,8 +586,8 @@ async def get_tags(request: web.Request) -> web.Response:
    )

    tags = [
-        schemas_out.TagUsage(name=name, count=count, type=tag_type)
-        for (name, tag_type, count) in rows
+        schemas_out.TagUsage(name=name, count=count)
+        for (name, count) in rows
    ]
    payload = schemas_out.TagsList(
        tags=tags, total=total, has_more=(query.offset + len(tags)) < total
@@ -59,6 +59,11 @@ class ListAssetsQuery(BaseModel):

    limit: conint(ge=1, le=500) = 20
    offset: conint(ge=0) = 0
+    # Opaque keyset cursor. When supplied, `offset` is ignored. Cursor pagination
+    # is supported for sort values `created_at`, `updated_at`, `name`, `size`.
+    # Supplying `after` together with `sort=last_access_time` returns
+    # 400 INVALID_CURSOR; that sort only supports offset/limit.
+    after: str | None = None

    sort: Literal["name", "created_at", "updated_at", "size", "last_access_time"] = (
        "created_at"
@@ -135,7 +140,7 @@ class CreateFromHashBody(BaseModel):
        if v is None:
            return []
        if isinstance(v, list):
-            out = [str(t).strip().lower() for t in v if str(t).strip()]
+            out = [str(t).strip() for t in v if str(t).strip()]
            seen = set()
            dedup = []
            for t in out:
@@ -144,7 +149,7 @@ class CreateFromHashBody(BaseModel):
                    dedup.append(t)
            return dedup
        if isinstance(v, str):
-            return [t.strip().lower() for t in v.split(",") if t.strip()]
+            return list(dict.fromkeys(t.strip() for t in v.split(",") if t.strip()))
        return []


@@ -201,7 +206,7 @@ class TagsListQuery(BaseModel):
        if v is None:
            return v
        v = v.strip()
-        return v.lower() or None
+        return v or None


 class TagsAdd(BaseModel):
@@ -215,7 +220,7 @@ class TagsAdd(BaseModel):
        for t in v:
            if not isinstance(t, str):
                raise TypeError("tags must be strings")
-            tnorm = t.strip().lower()
+            tnorm = t.strip()
            if tnorm:
                out.append(tnorm)
        seen = set()
@@ -234,8 +239,8 @@ class TagsRemove(TagsAdd):
 class UploadAssetSpec(BaseModel):
    """Upload Asset operation.

-    - tags: optional list; if provided, first is root ('models'|'input'|'output');
-            if root == 'models', second must be a valid category
+    - tags: labels plus one destination role ('models'|'input'|'output') for new bytes;
+            if role == 'models', exactly one model_type:<folder_name> tag is required
    - name: display name
    - user_metadata: arbitrary JSON object (optional)
    - hash: optional canonical 'blake3:<hex>' for validation / fast-path
@@ -304,7 +309,7 @@ class UploadAssetSpec(BaseModel):
        norm = []
        seen = set()
        for t in items:
-            tnorm = str(t).strip().lower()
+            tnorm = str(t).strip()
            if tnorm and tnorm not in seen:
                seen.add(tnorm)
                norm.append(tnorm)
@@ -330,14 +335,4 @@ class UploadAssetSpec(BaseModel):

    @model_validator(mode="after")
    def _validate_order(self):
-        if not self.tags:
-            raise ValueError("at least one tag is required for uploads")
-        root = self.tags[0]
-        if root not in {"models", "input", "output"}:
-            raise ValueError("first tag must be one of: models, input, output")
-        if root == "models":
-            if len(self.tags) < 2:
-                raise ValueError(
-                    "models uploads require a category tag as the second tag"
-                )
        return self
@@ -9,8 +9,20 @@ class Asset(BaseModel):
    ``id`` here is the AssetReference id, not the content-addressed Asset id."""

    id: str
-    name: str
+    name: str = Field(
+        ...,
+        deprecated=True,
+        description="Reference label, often caller-provided or derived from the filename. Deprecated for storage path/display semantics; use `loader_path` and `display_name` when present.",
+    )
    hash: str | None = None
+    loader_path: str | None = Field(
+        default=None,
+        description="The value a loader consumes to load this asset. `None` when no loader can resolve the file.",
+    )
+    display_name: str | None = Field(
+        default=None,
+        description="Human-facing label for the asset. Not unique.",
+    )
    asset_hash: str | None = None
    size: int | None = None
    mime_type: str | None = None
@@ -41,12 +53,13 @@ class AssetsList(BaseModel):
    assets: list[Asset]
    total: int
    has_more: bool
+    # Opaque cursor for the next page. Omitted when there are no more results.
+    next_cursor: str | None = None


 class TagUsage(BaseModel):
    name: str
    count: int
-    type: str


 class TagsList(BaseModel):
@@ -140,7 +140,6 @@ async def parse_multipart_upload(
            provided_mime_type = ((await field.text()) or "").strip() or None
        elif fname == "preview_id":
            provided_preview_id = ((await field.text()) or "").strip() or None
-
    if not file_present and not (provided_hash and provided_hash_exists):
        raise UploadError(
            400, "MISSING_FILE", "Form must include a 'file' part or a known 'hash'."
@@ -76,6 +76,8 @@ class AssetReference(Base):

    # Cache state fields (from former AssetCacheState)
    file_path: Mapped[str | None] = mapped_column(Text, nullable=True)
+    # In-root loader path derived from file_path at scan/ingest time.
+    loader_path: Mapped[str | None] = mapped_column(Text, nullable=True)
    mtime_ns: Mapped[int | None] = mapped_column(BigInteger, nullable=True)
    needs_verify: Mapped[bool] = mapped_column(Boolean, nullable=False, default=False)
    is_missing: Mapped[bool] = mapped_column(Boolean, nullable=False, default=False)
@@ -227,7 +229,6 @@ class Tag(Base):
    __tablename__ = "tags"

    name: Mapped[str] = mapped_column(String(512), primary_key=True)
-    tag_type: Mapped[str] = mapped_column(String(32), nullable=False, default="user")

    asset_reference_links: Mapped[list[AssetReferenceTag]] = relationship(
        back_populates="tag",
@@ -240,7 +241,5 @@ class Tag(Base):
        overlaps="asset_reference_links,tag_links,tags,asset_reference",
    )

-    __table_args__ = (Index("ix_tags_tag_type", "tag_type"),)
-
    def __repr__(self) -> str:
        return f"<Tag {self.name}>"
@@ -266,9 +266,18 @@ def list_references_page(
    metadata_filter: dict | None = None,
    sort: str | None = None,
    order: str | None = None,
+    after_cursor_value: object | None = None,
+    after_cursor_id: str | None = None,
 ) -> tuple[list[AssetReference], dict[str, list[str]], int]:
    """List references with pagination, filtering, and sorting.

+    When ``after_cursor_value``/``after_cursor_id`` are supplied the query uses
+    keyset pagination — ``offset`` is ignored and a WHERE clause selects rows
+    strictly after the given ``(sort_col, id)`` position in the active sort
+    direction. The cursor value must already be typed for the column
+    (datetime for time sorts, int for size, str for name); the caller decodes
+    the opaque cursor string and resolves to the typed value.
+
    Returns (references, tag_map, total_count).
    """
    base = (
@@ -297,9 +306,31 @@ def list_references_page(
        "size": Asset.size_bytes,
    }
    sort_col = sort_map.get(sort, AssetReference.created_at)
-    sort_exp = sort_col.desc() if order == "desc" else sort_col.asc()
+    descending = order == "desc"

-    base = base.order_by(sort_exp).limit(limit).offset(offset)
+    # Keyset WHERE: (sort_col, id) strictly less-than / greater-than the cursor.
+    # Equivalent to: sort_col <op> v  OR  (sort_col = v AND id <op> cursor_id).
+    if after_cursor_value is not None and after_cursor_id is not None:
+        if descending:
+            keyset = sa.or_(
+                sort_col < after_cursor_value,
+                sa.and_(sort_col == after_cursor_value, AssetReference.id < after_cursor_id),
+            )
+        else:
+            keyset = sa.or_(
+                sort_col > after_cursor_value,
+                sa.and_(sort_col == after_cursor_value, AssetReference.id > after_cursor_id),
+            )
+        base = base.where(keyset)
+
+    # Secondary ORDER BY id (matching the primary direction) gives the keyset
+    # comparison a deterministic tiebreaker on duplicate sort_col values.
+    id_exp = AssetReference.id.desc() if descending else AssetReference.id.asc()
+    sort_exp = sort_col.desc() if descending else sort_col.asc()
+
+    base = base.order_by(sort_exp, id_exp).limit(limit)
+    if after_cursor_id is None:
+        base = base.offset(offset)

    count_stmt = (
        select(sa.func.count())
@@ -619,6 +650,7 @@ def upsert_reference(
    name: str,
    mtime_ns: int,
    owner_id: str = "",
+    loader_path: str | None = None,
 ) -> tuple[bool, bool]:
    """Upsert a reference by file_path. Returns (created, updated).

@@ -628,6 +660,7 @@ def upsert_reference(
    vals = {
        "asset_id": asset_id,
        "file_path": file_path,
+        "loader_path": loader_path,
        "name": name,
        "owner_id": owner_id,
        "mtime_ns": int(mtime_ns),
@@ -655,13 +688,14 @@ def upsert_reference(
                AssetReference.asset_id != asset_id,
                AssetReference.mtime_ns.is_(None),
                AssetReference.mtime_ns != int(mtime_ns),
+                AssetReference.loader_path.is_distinct_from(loader_path),
                AssetReference.is_missing == True,  # noqa: E712
                AssetReference.deleted_at.isnot(None),
            )
        )
        .values(
-            asset_id=asset_id, mtime_ns=int(mtime_ns), is_missing=False,
-            deleted_at=None, updated_at=now,
+            asset_id=asset_id, mtime_ns=int(mtime_ns), loader_path=loader_path,
+            is_missing=False, deleted_at=None, updated_at=now,
        )
    )
    res2 = session.execute(upd)
@@ -55,13 +55,11 @@ def validate_tags_exist(session: Session, tags: list[str]) -> None:
        raise ValueError(f"Unknown tags: {missing}")


-def ensure_tags_exist(
-    session: Session, names: Iterable[str], tag_type: str = "user"
-) -> None:
+def ensure_tags_exist(session: Session, names: Iterable[str]) -> None:
    wanted = normalize_tags(list(names))
    if not wanted:
        return
-    rows = [{"name": n, "tag_type": tag_type} for n in list(dict.fromkeys(wanted))]
+    rows = [{"name": n} for n in list(dict.fromkeys(wanted))]
    ins = (
        sqlite.insert(Tag)
        .values(rows)
@@ -97,7 +95,7 @@ def set_reference_tags(
    to_remove = [t for t in current if t not in desired]

    if to_add:
-        ensure_tags_exist(session, to_add, tag_type="user")
+        ensure_tags_exist(session, to_add)
        session.add_all(
            [
                AssetReferenceTag(
@@ -142,7 +140,7 @@ def add_tags_to_reference(
        return AddTagsResult(added=[], already_present=[], total_tags=total)

    if create_if_missing:
-        ensure_tags_exist(session, norm, tag_type="user")
+        ensure_tags_exist(session, norm)

    current = set(get_reference_tags(session, reference_id))

@@ -267,6 +265,8 @@ def list_tags_with_usage(
    order: str = "count_desc",
    owner_id: str = "",
 ) -> tuple[list[tuple[str, str, int]], int]:
+    prefix_filter = prefix.strip() if prefix else ""
+
    counts_sq = (
        select(
            AssetReferenceTag.tag_name.label("tag_name"),
@@ -289,16 +289,14 @@ def list_tags_with_usage(
    q = (
        select(
            Tag.name,
-            Tag.tag_type,
            func.coalesce(counts_sq.c.cnt, 0).label("count"),
        )
        .select_from(Tag)
        .join(counts_sq, counts_sq.c.tag_name == Tag.name, isouter=True)
    )

-    if prefix:
-        escaped, esc = escape_sql_like_string(prefix.strip().lower())
-        q = q.where(Tag.name.like(escaped + "%", escape=esc))
+    if prefix_filter:
+        q = q.where(func.substr(Tag.name, 1, len(prefix_filter)) == prefix_filter)

    if not include_zero:
        q = q.where(func.coalesce(counts_sq.c.cnt, 0) > 0)
@@ -309,9 +307,8 @@ def list_tags_with_usage(
        q = q.order_by(func.coalesce(counts_sq.c.cnt, 0).desc(), Tag.name.asc())

    total_q = select(func.count()).select_from(Tag)
-    if prefix:
-        escaped, esc = escape_sql_like_string(prefix.strip().lower())
-        total_q = total_q.where(Tag.name.like(escaped + "%", escape=esc))
+    if prefix_filter:
+        total_q = total_q.where(func.substr(Tag.name, 1, len(prefix_filter)) == prefix_filter)
    if not include_zero:
        visible_tags_sq = (
            select(AssetReferenceTag.tag_name)
@@ -331,7 +328,7 @@ def list_tags_with_usage(
    rows = (session.execute(q.limit(limit).offset(offset))).all()
    total = (session.execute(total_q)).scalar_one()

-    rows_norm = [(name, ttype, int(count or 0)) for (name, ttype, count) in rows]
+    rows_norm = [(name, int(count or 0)) for (name, count) in rows]
    return rows_norm, int(total or 0)


@@ -41,10 +41,10 @@ def get_utc_now() -> datetime:
 def normalize_tags(tags: list[str] | None) -> list[str]:
    """
    Normalize a list of tags by:
-      - Stripping whitespace and converting to lowercase.
-      - Removing duplicates.
+      - Stripping whitespace.
+      - Removing exact duplicates while preserving order and case.
    """
-    return list(dict.fromkeys(t.strip().lower() for t in (tags or []) if (t or "").strip()))
+    return list(dict.fromkeys(t.strip() for t in (tags or []) if (t or "").strip()))


 def validate_blake3_hash(s: str) -> str:
@@ -33,9 +33,10 @@ from app.assets.services.file_utils import (
    verify_file_unchanged,
 )
 from app.assets.services.hashing import HashCheckpoint, compute_blake3_hash
+from app.assets.services.image_dimensions import extract_image_dimensions
 from app.assets.services.metadata_extract import extract_file_metadata
 from app.assets.services.path_utils import (
-    compute_relative_filename,
+    compute_loader_path,
    get_comfy_models_folders,
    get_name_and_tags_from_asset_path,
 )
@@ -62,7 +63,7 @@ RootType = Literal["models", "input", "output"]
 def get_prefixes_for_root(root: RootType) -> list[str]:
    if root == "models":
        bases: list[str] = []
-        for _bucket, paths in get_comfy_models_folders():
+        for _bucket, paths, _exts in get_comfy_models_folders():
            bases.extend(paths)
        return [os.path.abspath(p) for p in bases]
    if root == "input":
@@ -80,7 +81,7 @@ def get_all_known_prefixes() -> list[str]:

 def collect_models_files() -> list[str]:
    out: list[str] = []
-    for folder_name, bases in get_comfy_models_folders():
+    for folder_name, bases, _exts in get_comfy_models_folders():
        rel_files = folder_paths.get_filename_list(folder_name) or []
        for rel_path in rel_files:
            if not all(is_visible(part) for part in Path(rel_path).parts):
@@ -307,7 +308,7 @@ def build_asset_specs(
        if not stat_p.st_size:
            continue
        name, tags = get_name_and_tags_from_asset_path(abs_p)
-        rel_fname = compute_relative_filename(abs_p)
+        rel_fname = compute_loader_path(abs_p)

        # Extract metadata (tier 1: filesystem, tier 2: safetensors header)
        metadata = None
@@ -354,7 +355,7 @@ def insert_asset_specs(specs: list[SeedAssetSpec], tag_pool: set[str]) -> int:
        return 0
    with create_session() as sess:
        if tag_pool:
-            ensure_tags_exist(sess, tag_pool, tag_type="user")
+            ensure_tags_exist(sess, tag_pool)
        result = batch_insert_seed_assets(sess, specs=specs, owner_id="")
        sess.commit()
        return result.inserted_refs
@@ -429,7 +430,7 @@ def enrich_asset(
        return new_level

    initial_mtime_ns = get_mtime_ns(stat_p)
-    rel_fname = compute_relative_filename(file_path)
+    rel_fname = compute_loader_path(file_path)
    mime_type: str | None = None
    metadata = None

@@ -506,6 +507,10 @@ def enrich_asset(

    if extract_metadata and metadata:
        system_metadata = metadata.to_user_metadata()
+        if mime_type and mime_type.startswith("image/"):
+            dims = extract_image_dimensions(file_path, mime_type=mime_type)
+            if dims:
+                system_metadata.update(dims)
        set_reference_system_metadata(session, reference_id, system_metadata)

    if full_hash:
@@ -1,8 +1,19 @@
 import contextlib
 import mimetypes
 import os
+from datetime import timezone
 from typing import Sequence

+from app.assets.services.cursor import (
+    CursorPayload,
+    InvalidCursorError,
+    decode_cursor,
+    decode_cursor_int,
+    decode_cursor_time,
+    encode_cursor,
+    encode_cursor_from_time,
+)
+

 from app.assets.database.models import Asset
 from app.assets.database.queries import (
@@ -27,7 +38,7 @@ from app.assets.database.queries import (
    update_reference_updated_at,
 )
 from app.assets.helpers import select_best_live_path
-from app.assets.services.path_utils import compute_relative_filename
+from app.assets.services.path_utils import compute_loader_path
 from app.assets.services.schemas import (
    AssetData,
    AssetDetailResult,
@@ -80,7 +91,7 @@ def update_asset_metadata(
            update_reference_name(session, reference_id=reference_id, name=name)
            touched = True

-        computed_filename = compute_relative_filename(ref.file_path) if ref.file_path else None
+        computed_filename = compute_loader_path(ref.file_path) if ref.file_path else None

        new_meta: dict | None = None
        if user_metadata is not None:
@@ -149,6 +160,16 @@ def delete_asset_reference(
    owner_id: str,
    delete_content_if_orphan: bool = True,
 ) -> bool:
+    """Delete an asset reference.
+
+    With ``delete_content_if_orphan=False`` (a soft delete), the reference is
+    hidden and the underlying content is preserved. With ``True``, the content
+    is also removed once it becomes orphaned.
+
+    Note: the public DELETE /api/assets/{id} endpoint always soft-deletes
+    (passes ``False``); the orphan-reclamation path is intentionally
+    internal-only, retained for a future GC/admin caller.
+    """
    with create_session() as session:
        if not delete_content_if_orphan:
            # Soft delete: mark the reference as deleted but keep everything
@@ -242,6 +263,11 @@ def get_asset_by_hash(asset_hash: str) -> AssetData | None:
        return extract_asset_data(asset)


+# Sort fields that support cursor pagination. `last_access_time` is not
+# in this list — it falls back to offset/limit.
+_CURSOR_SORT_FIELDS = ("created_at", "updated_at", "name", "size")
+
+
 def list_assets_page(
    owner_id: str = "",
    include_tags: Sequence[str] | None = None,
@@ -252,7 +278,39 @@ def list_assets_page(
    offset: int = 0,
    sort: str = "created_at",
    order: str = "desc",
+    after: str | None = None,
 ) -> ListAssetsResult:
+    """List assets with optional cursor pagination.
+
+    When ``after`` is supplied it overrides ``offset``. The cursor's sort field
+    must match ``sort`` and be in the cursor-supported allowlist; mismatches
+    raise InvalidCursorError so the handler can map to 400 INVALID_CURSOR.
+    """
+    cursor_value: object | None = None
+    cursor_id: str | None = None
+    # Mint next_cursor on every page where the sort is cursor-supported, not
+    # only when the request itself arrived with a cursor. Otherwise a first
+    # request (no `after`) returns next_cursor=None and the client can never
+    # enter cursor mode.
+    mint_cursor = sort in _CURSOR_SORT_FIELDS
+
+    if after is not None:
+        if sort not in _CURSOR_SORT_FIELDS:
+            raise InvalidCursorError(
+                f"cursor pagination is not supported for sort={sort!r}"
+            )
+        payload = decode_cursor(after, _CURSOR_SORT_FIELDS, expected_order=order)
+        if payload.sort_field != sort:
+            raise InvalidCursorError(
+                f"cursor sort field {payload.sort_field!r} does not match request sort {sort!r}"
+            )
+        cursor_value, cursor_id = _resolve_cursor_value(payload), payload.id
+
+    # Over-fetch by one row so we can distinguish "exactly `limit` rows total
+    # remaining" from "more rows past this page" without a second query. Drop
+    # the sentinel before returning.
+    fetch_limit = limit + 1 if mint_cursor else limit
+
    with create_session() as session:
        refs, tag_map, total = list_references_page(
            session,
@@ -261,12 +319,22 @@ def list_assets_page(
            exclude_tags=exclude_tags,
            name_contains=name_contains,
            metadata_filter=metadata_filter,
-            limit=limit,
+            limit=fetch_limit,
            offset=offset,
            sort=sort,
            order=order,
+            after_cursor_value=cursor_value,
+            after_cursor_id=cursor_id,
        )

+        next_cursor: str | None = None
+        if mint_cursor and len(refs) > limit:
+            # There's at least one more row past this page — mint a cursor from
+            # the last row of the page (i.e. index `limit - 1`, since we
+            # over-fetched), and drop the sentinel.
+            next_cursor = _encode_next_cursor(refs[limit - 1], sort, order)
+            refs = refs[:limit]
+
        items: list[AssetSummaryData] = []
        for ref in refs:
            items.append(
@@ -277,7 +345,39 @@ def list_assets_page(
                )
            )

-        return ListAssetsResult(items=items, total=total)
+        return ListAssetsResult(items=items, total=total, next_cursor=next_cursor)
+
+
+def _resolve_cursor_value(payload: CursorPayload) -> object:
+    """Map a decoded cursor payload to a column-typed Python value."""
+    if payload.sort_field in ("created_at", "updated_at"):
+        # DB stores naive UTC; strip tzinfo so the comparison binds against a
+        # `TIMESTAMP WITHOUT TIME ZONE` column without an offset shift.
+        return decode_cursor_time(payload).replace(tzinfo=None)
+    if payload.sort_field == "size":
+        return decode_cursor_int(payload)
+    return payload.value  # name, str-typed
+
+
+def _encode_next_cursor(ref, sort: str, order: str) -> str | None:
+    """Mint a cursor pointing at *ref* for the given sort dimension.
+
+    Returns None when the boundary row carries a NULL sort value (e.g. an asset
+    record whose size_bytes hasn't been backfilled). Continuing pagination
+    across a NULL boundary is undefined under keyset ordering — better to
+    truncate cleanly here than to mint a cursor that mis-positions.
+    """
+    if sort == "name":
+        return encode_cursor("name", ref.name, ref.id, order=order)
+    if sort == "size":
+        if ref.asset is None or ref.asset.size_bytes is None:
+            return None
+        return encode_cursor("size", str(ref.asset.size_bytes), ref.id, order=order)
+    # created_at / updated_at — DB datetimes are naive UTC; attach tz before encoding.
+    value = ref.created_at if sort == "created_at" else ref.updated_at
+    if value is None:
+        return None
+    return encode_cursor_from_time(sort, value.replace(tzinfo=timezone.utc), ref.id, order=order)


 def resolve_hash_to_path(
@@ -56,6 +56,7 @@ class ReferenceRow(TypedDict):
    id: str
    asset_id: str
    file_path: str
+    loader_path: str | None
    mtime_ns: int
    owner_id: str
    name: str
@@ -134,6 +135,14 @@ def batch_insert_seed_assets(

    for spec in specs:
        absolute_path = os.path.abspath(spec["abs_path"])
+        existing_asset_id = path_to_asset_id.get(absolute_path)
+        if existing_asset_id is not None:
+            existing_tags = asset_id_to_ref_data[existing_asset_id]["tags"]
+            asset_id_to_ref_data[existing_asset_id]["tags"] = list(
+                dict.fromkeys([*existing_tags, *spec["tags"]])
+            )
+            continue
+
        asset_id = str(uuid.uuid4())
        reference_id = str(uuid.uuid4())
        absolute_path_list.append(absolute_path)
@@ -164,6 +173,8 @@ def batch_insert_seed_assets(
                "id": reference_id,
                "asset_id": asset_id,
                "file_path": absolute_path,
+                # spec["fname"] is compute_loader_path(abs_path) from build_asset_specs.
+                "loader_path": spec["fname"],
                "mtime_ns": spec["mtime_ns"],
                "owner_id": owner_id,
                "name": spec["info_name"],
@@ -0,0 +1,213 @@
+"""Opaque keyset-pagination cursor for /api/assets.
+
+Payload JSON uses short keys to keep the encoded length small:
+
+    {"s": <sort_field>, "v": <value>, "id": <id>, "o": <order>}
+
+The `o` key binds the cursor to the sort direction it was minted under,
+so replaying a `desc` cursor against an `asc` request fails with
+``INVALID_CURSOR`` rather than silently walking the wrong direction.
+`o` is mandatory on every payload — a cursor without it is rejected as
+malformed.
+
+Encoding is base64url with no padding. Cursors are opaque tokens: the
+payload format is internal to this server, and clients must treat a
+cursor as a black box handed back via `next_cursor`. No byte-level
+compatibility with any other implementation is required.
+
+Time values are serialized as Unix microseconds (UTC) — microsecond
+precision is sufficient to round-trip the timestamps stored by the
+database without rounding rows in the same millisecond bucket.
+"""
+from __future__ import annotations
+
+import base64
+import json
+from dataclasses import dataclass
+from datetime import datetime, timezone
+from typing import Iterable, Optional
+
+
+class InvalidCursorError(ValueError):
+    """Raised on a malformed, oversized, or unsupported-sort-field cursor.
+
+    Map to a 400 response with code ``INVALID_CURSOR`` at the handler.
+    """
+
+
+# Wire-format length caps. Cursors are user-controlled, so caps protect the
+# decode path from oversized allocations and downstream SQL predicates from
+# unbounded strings.
+#
+# MAX_CURSOR_VALUE_LENGTH is 512 to fit the `AssetReference.name` column max
+# (`String(512)`) — otherwise a long-named asset would mint a cursor the same
+# server then refuses on the next request.
+#
+# MAX_ENCODED_CURSOR_LENGTH is the decode-path guard, sized comfortably above
+# the largest cursor the per-field caps can produce. Worst case is value + id
+# at their caps with every character JSON-escaping to the six-byte `\uXXXX`
+# form (control characters), which is ~5.2 KB once base64url-encoded. At 8192
+# the encoder can never mint a cursor that exceeds it, so a freshly minted
+# cursor always decodes on the next request and there is no user-visible
+# "cursor too long" failure.
+MAX_ENCODED_CURSOR_LENGTH = 8192
+MAX_CURSOR_VALUE_LENGTH = 512
+MAX_CURSOR_ID_LENGTH = 128
+
+
+@dataclass(frozen=True)
+class CursorPayload:
+    sort_field: str
+    value: str
+    id: str
+    order: str
+
+
+_VALID_ORDERS = ("asc", "desc")
+
+
+def encode_cursor(sort_field: str, value: str, id: str, order: str = "desc") -> str:
+    """Encode a cursor payload as a base64url (no-padding) string.
+
+    `order` binds the cursor to the sort direction it was minted under so a
+    later request with a flipped `order` query parameter is rejected with
+    ``INVALID_CURSOR`` rather than silently walking the wrong direction.
+    """
+    if order not in _VALID_ORDERS:
+        raise InvalidCursorError(f"order must be one of {_VALID_ORDERS}, got {order!r}")
+    # Symmetric input validation: the encoder must reject anything the
+    # decoder rejects, or the same server will mint cursors it then 400s on
+    # the next request.
+    if not id:
+        raise InvalidCursorError("id must be non-empty")
+    if len(id) > MAX_CURSOR_ID_LENGTH:
+        raise InvalidCursorError("id exceeds maximum length")
+    if len(value) > MAX_CURSOR_VALUE_LENGTH:
+        raise InvalidCursorError("value exceeds maximum length")
+    payload = {"s": sort_field, "v": value, "id": id, "o": order}
+    raw = json.dumps(payload, separators=(",", ":"), ensure_ascii=False)
+    # No mint-time length guard is needed: the per-field caps above bound the
+    # encoded length well below MAX_ENCODED_CURSOR_LENGTH (see its definition),
+    # so the encoder can never produce a cursor the decode path would reject.
+    return base64.urlsafe_b64encode(raw.encode("utf-8")).rstrip(b"=").decode("ascii")
+
+
+def encode_cursor_from_time(sort_field: str, t: datetime, id: str, order: str = "desc") -> str:
+    """Encode a time-typed cursor at Unix microsecond precision.
+
+    Accepts an aware datetime (any timezone) and normalizes to UTC. Naive
+    datetimes are rejected so callers can't accidentally encode the local
+    wall-clock value of a UTC-stored timestamp.
+    """
+    if t.tzinfo is None:
+        raise ValueError("encode_cursor_from_time requires an aware datetime")
+    micros = _datetime_to_unix_micros(t.astimezone(timezone.utc))
+    return encode_cursor(sort_field, str(micros), id, order=order)
+
+
+def decode_cursor(
+    cursor: str,
+    allowed_sort_fields: Iterable[str],
+    expected_order: str | None = None,
+) -> CursorPayload:
+    """Parse an opaque cursor.
+
+    ``allowed_sort_fields`` is the endpoint's accepted sort-field list — a
+    cursor carrying a field outside this set is rejected so a cursor minted
+    for one column can't be replayed against another (e.g. a ``created_at``
+    timestamp string compared against a ``name`` column).
+
+    ``expected_order`` (``"asc"``/``"desc"``), when supplied, must match the
+    payload's ``o`` field. ``o`` is required on every payload; a cursor
+    missing it is rejected as malformed.
+
+    Passing no allowed fields rejects every cursor.
+    """
+    if len(cursor) > MAX_ENCODED_CURSOR_LENGTH:
+        raise InvalidCursorError("cursor exceeds maximum length")
+
+    try:
+        # urlsafe_b64decode requires correct padding; we strip on encode, so
+        # restore the trailing '=' pad here.
+        padding = "=" * (-len(cursor) % 4)
+        raw = base64.urlsafe_b64decode(cursor + padding)
+    except (ValueError, base64.binascii.Error) as e:
+        raise InvalidCursorError(f"encoding: {e}") from e
+
+    try:
+        decoded = json.loads(raw)
+    except (json.JSONDecodeError, UnicodeDecodeError) as e:
+        raise InvalidCursorError(f"payload: {e}") from e
+
+    if not isinstance(decoded, dict):
+        raise InvalidCursorError("payload: expected object")
+
+    sort_field = decoded.get("s")
+    value = decoded.get("v")
+    id = decoded.get("id")
+    order = decoded.get("o")
+
+    if not isinstance(sort_field, str) or not isinstance(value, str) or not isinstance(id, str):
+        raise InvalidCursorError("payload: missing or non-string s/v/id")
+
+    if id == "":
+        raise InvalidCursorError("missing id")
+    if len(id) > MAX_CURSOR_ID_LENGTH:
+        raise InvalidCursorError("id exceeds maximum length")
+    if len(value) > MAX_CURSOR_VALUE_LENGTH:
+        raise InvalidCursorError("value exceeds maximum length")
+
+    if sort_field not in allowed_sort_fields:
+        raise InvalidCursorError(f"unsupported sort field {sort_field!r}")
+
+    if not isinstance(order, str):
+        raise InvalidCursorError("missing or non-string o")
+    if order not in _VALID_ORDERS:
+        raise InvalidCursorError(f"unsupported order {order!r}")
+    if expected_order is not None and order != expected_order:
+        raise InvalidCursorError(
+            f"cursor order {order!r} does not match request order {expected_order!r}"
+        )
+
+    return CursorPayload(sort_field=sort_field, value=value, id=id, order=order)
+
+
+def decode_cursor_time(payload: Optional[CursorPayload]) -> datetime:
+    """Parse a time-typed cursor value as Unix microseconds, returning UTC."""
+    if payload is None:
+        raise InvalidCursorError("nil cursor payload")
+    try:
+        micros = int(payload.value)
+    except ValueError as e:
+        raise InvalidCursorError(f"value is not a valid timestamp: {e}") from e
+    try:
+        return _unix_micros_to_datetime(micros)
+    except (OverflowError, OSError, ValueError) as e:
+        # Crafted out-of-range microseconds (e.g. > datetime.MAX_YEAR) blow up
+        # in fromtimestamp / datetime construction. Map to 400, not 500.
+        raise InvalidCursorError(f"value is out of representable range: {e}") from e
+
+
+def decode_cursor_int(payload: Optional[CursorPayload]) -> int:
+    """Parse a cursor value as a base-10 integer."""
+    if payload is None:
+        raise InvalidCursorError("nil cursor payload")
+    try:
+        return int(payload.value)
+    except ValueError as e:
+        raise InvalidCursorError(f"value is not a valid integer: {e}") from e
+
+
+_EPOCH = datetime(1970, 1, 1, tzinfo=timezone.utc)
+
+
+def _datetime_to_unix_micros(t: datetime) -> int:
+    """Convert an aware UTC datetime to Unix microseconds (integer math)."""
+    delta = t - _EPOCH
+    return (delta.days * 86_400 + delta.seconds) * 1_000_000 + delta.microseconds
+
+
+def _unix_micros_to_datetime(micros: int) -> datetime:
+    """Convert Unix microseconds to a UTC datetime, preserving precision."""
+    seconds, micro_remainder = divmod(micros, 1_000_000)
+    return datetime.fromtimestamp(seconds, tz=timezone.utc).replace(microsecond=micro_remainder)
@@ -0,0 +1,63 @@
+"""Image dimension extraction for asset ingest.
+
+Reads only the image header via Pillow to capture width/height cheaply,
+without a full pixel decode. Returns a metadata dict suitable for merging
+into ``AssetReference.system_metadata``.
+"""
+from __future__ import annotations
+
+import logging
+from typing import Any
+
+logger = logging.getLogger(__name__)
+
+
+def extract_image_dimensions(
+    file_path: str, mime_type: str | None = None
+) -> dict[str, Any] | None:
+    """Extract image dimensions for the file at ``file_path``.
+
+    Args:
+        file_path: Absolute path to a file on disk.
+        mime_type: Optional MIME type hint. When provided and not prefixed
+            with ``image/``, extraction is skipped without touching the file.
+
+    Returns:
+        ``{"kind": "image", "width": W, "height": H}`` when the file is a
+        recognizable image with positive dimensions, otherwise ``None``.
+
+    The dict shape is intended to be merged into ``system_metadata`` so the
+    asset response surfaces ``metadata.kind`` plus dimension fields for image
+    assets. Forward-compatible: future media kinds (e.g. ``"video"`` with
+    duration/fps) can extend this shape without schema changes.
+    """
+    if mime_type is not None and not mime_type.startswith("image/"):
+        return None
+
+    try:
+        from PIL import Image, UnidentifiedImageError
+    except ImportError:
+        logger.debug(
+            "Pillow not available; skipping image dimension extraction for %s",
+            file_path,
+        )
+        return None
+
+    try:
+        with Image.open(file_path) as img:
+            width, height = img.size
+    except (OSError, UnidentifiedImageError, ValueError) as exc:
+        logger.debug(
+            "Failed to read image dimensions from %s: %s", file_path, exc
+        )
+        return None
+
+    if (
+        not isinstance(width, int)
+        or not isinstance(height, int)
+        or width <= 0
+        or height <= 0
+    ):
+        return None
+
+    return {"kind": "image", "width": width, "height": height}
@@ -17,9 +17,11 @@ from app.assets.database.queries import (
    get_reference_by_file_path,
    get_reference_tags,
    get_or_create_reference,
+    list_references_by_asset_id,
    reference_exists,
    remove_missing_tag_for_asset_id,
    set_reference_metadata,
+    set_reference_system_metadata,
    set_reference_tags,
    update_asset_hash_and_mime,
    upsert_asset,
@@ -29,9 +31,11 @@ from app.assets.database.queries import (
 from app.assets.helpers import get_utc_now, normalize_tags
 from app.assets.services.bulk_ingest import batch_insert_seed_assets
 from app.assets.services.file_utils import get_size_and_mtime_ns
+from app.assets.services.image_dimensions import extract_image_dimensions
 from app.assets.services.path_utils import (
-    compute_relative_filename,
+    compute_loader_path,
    get_name_and_tags_from_asset_path,
+    get_path_derived_tags_from_path,
    resolve_destination_from_tags,
    validate_path_within_base,
 )
@@ -88,6 +92,7 @@ def _ingest_file_from_path(
            name=info_name or os.path.basename(locator),
            mtime_ns=mtime_ns,
            owner_id=owner_id,
+            loader_path=compute_loader_path(locator),
        )

        # Get the reference we just created/updated
@@ -98,17 +103,32 @@ def _ingest_file_from_path(
            if preview_id and ref.preview_id != preview_id:
                ref.preview_id = preview_id

-            norm = normalize_tags(list(tags))
-            if norm:
+            try:
+                backend_tags = get_path_derived_tags_from_path(locator)
+            except ValueError:
+                backend_tags = []
+            caller_tags = normalize_tags(tags)
+            backend_tags = normalize_tags(backend_tags)
+            all_tags = normalize_tags([*caller_tags, *backend_tags])
+            if all_tags:
                if require_existing_tags:
-                    validate_tags_exist(session, norm)
-                add_tags_to_reference(
-                    session,
-                    reference_id=reference_id,
-                    tags=norm,
-                    origin=tag_origin,
-                    create_if_missing=not require_existing_tags,
-                )
+                    validate_tags_exist(session, all_tags)
+                if backend_tags:
+                    add_tags_to_reference(
+                        session,
+                        reference_id=reference_id,
+                        tags=backend_tags,
+                        origin="automatic",
+                        create_if_missing=not require_existing_tags,
+                    )
+                if caller_tags:
+                    add_tags_to_reference(
+                        session,
+                        reference_id=reference_id,
+                        tags=caller_tags,
+                        origin=tag_origin,
+                        create_if_missing=not require_existing_tags,
+                    )

            _update_metadata_with_filename(
                session,
@@ -118,6 +138,14 @@ def _ingest_file_from_path(
                user_metadata=user_metadata,
            )

+            _maybe_store_image_dimensions(
+                session,
+                reference_id=reference_id,
+                file_path=locator,
+                mime_type=mime_type,
+                current_system_metadata=ref.system_metadata,
+            )
+
        try:
            remove_missing_tag_for_asset_id(session, asset_id=asset.id)
        except Exception:
@@ -217,7 +245,7 @@ def ingest_existing_file(
            "mtime_ns": mtime_ns,
            "info_name": name,
            "tags": tags,
-            "fname": os.path.basename(abs_path),
+            "fname": compute_loader_path(abs_path),
            "metadata": None,
            "hash": None,
            "mime_type": mime_type,
@@ -277,7 +305,7 @@ def _register_existing_asset(
            return result

        new_meta = dict(user_metadata)
-        computed_filename = compute_relative_filename(ref.file_path) if ref.file_path else None
+        computed_filename = compute_loader_path(ref.file_path) if ref.file_path else None
        if computed_filename:
            new_meta["filename"] = computed_filename

@@ -288,6 +316,13 @@ def _register_existing_asset(
                user_metadata=new_meta,
            )

+        _backfill_image_dimensions_from_siblings(
+            session,
+            asset_id=asset.id,
+            new_reference_id=ref.id,
+            current_system_metadata=ref.system_metadata,
+        )
+
        if tags is not None:
            set_reference_tags(
                session,
@@ -317,7 +352,7 @@ def _update_metadata_with_filename(
    current_metadata: dict | None,
    user_metadata: dict[str, Any],
 ) -> None:
-    computed_filename = compute_relative_filename(file_path) if file_path else None
+    computed_filename = compute_loader_path(file_path) if file_path else None

    current_meta = current_metadata or {}
    new_meta = dict(current_meta)
@@ -334,6 +369,87 @@ def _update_metadata_with_filename(
        )


+_IMAGE_DIMENSION_KEYS = ("kind", "width", "height")
+
+
+def _maybe_store_image_dimensions(
+    session: Session,
+    reference_id: str,
+    file_path: str,
+    mime_type: str | None,
+    current_system_metadata: dict | None,
+) -> None:
+    """Populate ``kind``/``width``/``height`` on system_metadata for image refs.
+
+    Non-image MIME types are a no-op. Pre-existing keys (e.g. enricher-written
+    safetensors metadata, download provenance) are preserved by merge.
+    """
+    if not mime_type or not mime_type.startswith("image/"):
+        return
+
+    dims = extract_image_dimensions(file_path, mime_type=mime_type)
+    if not dims:
+        return
+
+    current = current_system_metadata or {}
+    merged = dict(current)
+    merged.update(dims)
+    if merged != current:
+        set_reference_system_metadata(
+            session,
+            reference_id=reference_id,
+            system_metadata=merged,
+        )
+
+
+def _backfill_image_dimensions_from_siblings(
+    session: Session,
+    asset_id: str,
+    new_reference_id: str,
+    current_system_metadata: dict | None,
+) -> None:
+    """Copy image dimension keys from any sibling reference of the same asset.
+
+    The from-hash path doesn't read the file bytes, so dimensions can't be
+    extracted there directly. When another reference of the same asset already
+    carries image dimensions, copy them onto the new reference so consumers
+    see consistent metadata regardless of how the asset was registered.
+
+    Best-effort: missing siblings, non-image siblings, or absent dimension
+    keys leave the target reference unchanged.
+    """
+    current = current_system_metadata or {}
+    if current.get("kind") == "image" and "width" in current and "height" in current:
+        return
+
+    for sibling in list_references_by_asset_id(session, asset_id):
+        if sibling.id == new_reference_id:
+            continue
+        meta = sibling.system_metadata or {}
+        if meta.get("kind") != "image":
+            continue
+        width = meta.get("width")
+        height = meta.get("height")
+        if (
+            type(width) is not int
+            or type(height) is not int
+            or width <= 0
+            or height <= 0
+        ):
+            continue
+        merged = dict(current)
+        merged["kind"] = "image"
+        merged["width"] = width
+        merged["height"] = height
+        if merged != current:
+            set_reference_system_metadata(
+                session,
+                reference_id=new_reference_id,
+                system_metadata=merged,
+            )
+        return
+
+
 def _sanitize_filename(name: str | None, fallback: str) -> str:
    n = os.path.basename((name or "").strip() or fallback)
    return n if n else fallback
@@ -375,6 +491,10 @@ def upload_from_temp_path(
        existing = get_asset_by_hash(session, asset_hash=asset_hash)

    if existing is not None:
+        # Once content is already known, duplicate byte uploads are treated as
+        # reference-only creation. Request tags are labels only here: do not
+        # require upload destination tags, do not move bytes, and do not
+        # synthesize path-derived classification or uploaded provenance.
        with contextlib.suppress(Exception):
            if temp_path and os.path.exists(temp_path):
                os.remove(temp_path)
@@ -436,7 +556,7 @@ def upload_from_temp_path(
        owner_id=owner_id,
        preview_id=preview_id,
        user_metadata=user_metadata or {},
-        tags=tags,
+        tags=[*(tags or []), "uploaded"],
        tag_origin="manual",
        require_existing_tags=False,
    )
@@ -470,15 +590,19 @@ def register_file_in_place(
 ) -> UploadResult:
    """Register an already-saved file in the asset database without moving it.

-    Tags are derived from the filesystem path (root category + subfolder names),
-    merged with any caller-provided tags, matching the behavior of the scanner.
+    This helper is used by upload paths that have already written bytes before
+    registering the file, so it records the same ``uploaded`` tag as the
+    multipart byte-upload path.
+
+    Tags are derived from trusted filesystem classification and merged with any
+    caller-provided tags, matching the behavior of the scanner.
    If the path is not under a known root, only the caller-provided tags are used.
    """
    try:
        _, path_tags = get_name_and_tags_from_asset_path(abs_path)
    except ValueError:
        path_tags = []
-    merged_tags = normalize_tags([*path_tags, *tags])
+    merged_tags = normalize_tags([*path_tags, *tags, "uploaded"])

    try:
        digest, _ = hashing.compute_blake3_hash(abs_path)
@@ -3,59 +3,66 @@ from pathlib import Path
 from typing import Literal

 import folder_paths
-from app.assets.helpers import normalize_tags


-_NON_MODEL_FOLDER_NAMES = frozenset({"custom_nodes"})
+_NON_MODEL_FOLDER_NAMES = frozenset({"configs", "custom_nodes"})
+_KNOWN_SUBFOLDER_TAGS = frozenset({"3d", "pasted", "painter", "threed", "webcam"})


-def get_comfy_models_folders() -> list[tuple[str, list[str]]]:
-    """Build list of (folder_name, base_paths[]) for all model locations.
+def get_comfy_models_folders() -> list[tuple[str, list[str], set[str]]]:
+    """Build list of (folder_name, base_paths[], extensions) for all model locations.

    Includes every category registered in folder_names_and_paths,
    regardless of whether its paths are under the main models_dir,
-    but excludes non-model entries like custom_nodes.
+    but excludes non-model entries like configs and custom_nodes.
+
+    An empty extensions set means the category accepts any extension,
+    matching folder_paths.filter_files_extensions semantics.
    """
-    targets: list[tuple[str, list[str]]] = []
+    targets: list[tuple[str, list[str], set[str]]] = []
    for name, values in folder_paths.folder_names_and_paths.items():
        if name in _NON_MODEL_FOLDER_NAMES:
            continue
-        paths, _exts = values[0], values[1]
+        paths, exts = values[0], values[1]
        if paths:
-            targets.append((name, paths))
+            targets.append((name, paths, set(exts)))
    return targets


 def resolve_destination_from_tags(tags: list[str]) -> tuple[str, list[str]]:
-    """Validates and maps tags -> (base_dir, subdirs_for_fs)"""
-    if not tags:
-        raise ValueError("tags must not be empty")
-    root = tags[0].lower()
+    """Validates and maps upload routing tags -> (base_dir, subdirs_for_fs).
+
+    The request tags are only used to choose the write destination. Extra tags
+    remain labels; they do not become path components or trusted classification.
+    """
+    destination_roles = [t for t in tags if t in {"input", "models", "output"}]
+    if len(destination_roles) != 1:
+        raise ValueError("uploads require exactly one destination role: input, models, or output")
+
+    root = destination_roles[0]
    if root == "models":
-        if len(tags) < 2:
-            raise ValueError("at least two tags required for model asset")
+        model_type_tags = [t for t in tags if t.startswith("model_type:")]
+        if len(model_type_tags) != 1:
+            raise ValueError("models uploads require exactly one model_type:<folder_name> tag")
+        folder_name = model_type_tags[0].split(":", 1)[1]
+        if not folder_name:
+            raise ValueError("models uploads require exactly one model_type:<folder_name> tag")
+        model_folder_paths = {
+            name: paths for name, paths, _exts in get_comfy_models_folders()
+        }
        try:
-            bases = folder_paths.folder_names_and_paths[tags[1]][0]
+            bases = model_folder_paths[folder_name]
        except KeyError:
-            raise ValueError(f"unknown model category '{tags[1]}'")
+            raise ValueError(f"unknown model category '{folder_name}'")
        if not bases:
-            raise ValueError(f"no base path configured for category '{tags[1]}'")
+            raise ValueError(f"no base path configured for category '{folder_name}'")
        base_dir = os.path.abspath(bases[0])
-        raw_subdirs = tags[2:]
    elif root == "input":
        base_dir = os.path.abspath(folder_paths.get_input_directory())
-        raw_subdirs = tags[1:]
-    elif root == "output":
-        base_dir = os.path.abspath(folder_paths.get_output_directory())
-        raw_subdirs = tags[1:]
    else:
-        raise ValueError(f"unknown root tag '{tags[0]}'; expected 'models', 'input', or 'output'")
-    _sep_chars = frozenset(("/", "\\", os.sep))
-    for i in raw_subdirs:
-        if i in (".", "..") or _sep_chars & set(i):
-            raise ValueError("invalid path component in tags")
+        base_dir = os.path.abspath(folder_paths.get_output_directory())

-    return base_dir, raw_subdirs if raw_subdirs else []
+    return base_dir, []


 def validate_path_within_base(candidate: str, base: str) -> None:
@@ -65,14 +72,79 @@ def validate_path_within_base(candidate: str, base: str) -> None:
        raise ValueError("destination escapes base directory")


-def compute_relative_filename(file_path: str) -> str | None:
+def _compute_relative_path(child: str, parent: str) -> str:
+    rel = os.path.relpath(os.path.abspath(child), os.path.abspath(parent))
+    if rel == ".":
+        return ""
+    return rel.replace(os.sep, "/")
+
+
+def _is_relative_to(child: str, parent: str) -> bool:
+    return Path(os.path.abspath(child)).is_relative_to(os.path.abspath(parent))
+
+
+def compute_asset_response_paths(file_path: str) -> tuple[str, str | None] | None:
+    """Return (logical_path, display_name) for a file path.
+
+    ``logical_path`` is the internal namespaced storage locator (e.g.
+    ``models/checkpoints/foo/bar.safetensors``); ``display_name`` is the
+    human-facing label below that namespace, served on Asset responses. These
+    are storage locators, not model-loader namespaces. Registered model-folder
+    membership is represented by backend tags such as
+    ``model_type:<folder_name>``; these paths only use known storage roots.
    """
-    Return the model's path relative to the last well-known folder (the model category),
-    using forward slashes, eg:
+    fp_abs = os.path.abspath(file_path)
+    candidates: list[tuple[int, int, str, str]] = []
+
+    for order, (namespace, base) in enumerate(
+        (
+            ("input", folder_paths.get_input_directory()),
+            ("output", folder_paths.get_output_directory()),
+            ("temp", folder_paths.get_temp_directory()),
+            ("models", getattr(folder_paths, "models_dir", "")),
+        )
+    ):
+        if not base:
+            continue
+        base_abs = os.path.abspath(base)
+        if _is_relative_to(fp_abs, base_abs):
+            candidates.append((len(base_abs), -order, namespace, base_abs))
+
+    if not candidates:
+        return None
+
+    _base_len, _order, namespace, base = max(candidates)
+    rel = _compute_relative_path(fp_abs, base)
+    public_path = f"{namespace}/{rel}" if rel else namespace
+    return public_path, rel or None
+
+
+def compute_display_name(file_path: str) -> str | None:
+    """Return the asset's `display_name`, or None for unknown paths."""
+    result = compute_asset_response_paths(file_path)
+    return result[1] if result else None
+
+
+def compute_logical_path(file_path: str) -> str | None:
+    """Return the internal namespaced storage locator, or None for unknown paths."""
+    result = compute_asset_response_paths(file_path)
+    return result[0] if result else None
+
+
+def compute_loader_path(file_path: str) -> str | None:
+    """
+    Return the asset's in-root loader path: the path relative to the last
+    well-known folder (the model category), using forward slashes, eg:
      /.../models/checkpoints/flux/123/flux.safetensors -> "flux/123/flux.safetensors"
      /.../models/text_encoders/clip_g.safetensors -> "clip_g.safetensors"

-    For non-model paths, returns None.
+    This is the value model loaders consume (the model category is dropped). It
+    is persisted as ``AssetReference.loader_path`` and served as the public
+    Asset response `loader_path` field. The human-facing `display_name` comes
+    from compute_asset_response_paths().
+
+    For input/output/temp paths the full path relative to that root is returned.
+    For paths outside any known root, returns None.
    """
    try:
        root_category, rel_path = get_asset_category_and_relative_path(file_path)
@@ -116,9 +188,10 @@ def get_asset_category_and_relative_path(
    def _compute_relative(child: str, parent: str) -> str:
        # Normalize relative path, stripping any leading ".." components
        # by anchoring to root (os.sep) then computing relpath back from it.
-        return os.path.relpath(
+        rel = os.path.relpath(
            os.path.join(os.sep, os.path.relpath(child, parent)), os.sep
        )
+        return "" if rel == "." else rel.replace(os.sep, "/")

    # 1) input
    input_base = os.path.abspath(folder_paths.get_input_directory())
@@ -136,8 +209,14 @@ def get_asset_category_and_relative_path(
        return "temp", _compute_relative(fp_abs, temp_base)

    # 4) models (check deepest matching base to avoid ambiguity)
+    ext = os.path.splitext(fp_abs)[1].lower()
    best: tuple[int, str, str] | None = None  # (base_len, bucket, rel_inside_bucket)
-    for bucket, bases in get_comfy_models_folders():
+    for bucket, bases, extensions in get_comfy_models_folders():
+        # A bucket only lists files within its extension set (empty set
+        # accepts any extension), so a bucket that cannot load the file
+        # must not contribute a loader path.
+        if extensions and ext not in extensions:
+            continue
        for b in bases:
            base_abs = os.path.abspath(b)
            if not _check_is_within(fp_abs, base_abs):
@@ -149,25 +228,111 @@ def get_asset_category_and_relative_path(
    if best is not None:
        _, bucket, rel_inside = best
        combined = os.path.join(bucket, rel_inside)
-        return "models", os.path.relpath(os.path.join(os.sep, combined), os.sep)
+        normalized = os.path.relpath(os.path.join(os.sep, combined), os.sep)
+        return "models", normalized.replace(os.sep, "/")

    raise ValueError(
        f"Path is not within input, output, temp, or configured model bases: {file_path}"
    )


+def get_backend_system_tags_from_path(path: str) -> list[str]:
+    """Return trusted backend tags derived from current filesystem facts.
+
+    The returned tags are only the backend-generated system tags: ``models``,
+    ``model_type:<folder_name>``, ``input``, ``output``, and ``temp``. Model
+    type tags are based on registered folder names, not path components.
+
+    A ``model_type:<folder_name>`` tag is only emitted when the file's
+    extension is accepted by that folder's registered extension set, so
+    categories sharing a base directory tag only the files they can
+    actually load. Files under a model base whose extension matches no
+    category still get the ``models`` tag.
+    """
+    fp_abs = os.path.abspath(path)
+    fp_path = Path(fp_abs)
+    tags: list[str] = []
+
+    def _add(tag: str) -> None:
+        if tag not in tags:
+            tags.append(tag)
+
+    for role, base in (
+        ("input", folder_paths.get_input_directory()),
+        ("output", folder_paths.get_output_directory()),
+        ("temp", folder_paths.get_temp_directory()),
+    ):
+        if fp_path.is_relative_to(os.path.abspath(base)):
+            _add(role)
+
+    ext = os.path.splitext(fp_abs)[1].lower()
+    model_types: list[str] = []
+    under_models_base = False
+    for folder_name, bases, extensions in get_comfy_models_folders():
+        for base in bases:
+            if fp_path.is_relative_to(os.path.abspath(base)):
+                under_models_base = True
+                # Empty set accepts any extension, matching
+                # folder_paths.filter_files_extensions semantics.
+                if not extensions or ext in extensions:
+                    model_types.append(folder_name)
+                break
+
+    if under_models_base:
+        _add("models")
+    for folder_name in model_types:
+        _add(f"model_type:{folder_name}")
+
+    if not tags:
+        raise ValueError(
+            f"Path is not within input, output, temp, or configured model bases: {path}"
+        )
+    return tags
+
+
+def get_known_subfolder_tags(subfolder: str | None) -> list[str]:
+    """Return tags for known UI/input subfolder names."""
+    if subfolder in _KNOWN_SUBFOLDER_TAGS:
+        return [subfolder]
+    return []
+
+
+def get_known_input_subfolder_tags_from_path(path: str) -> list[str]:
+    """Return known input-layout tags for files in canonical input subfolders.
+
+    These are compatibility tags for current UI-origin input directories such as
+    ``pasted`` and ``webcam``. They are intentionally narrow: only files directly
+    inside a known top-level input directory receive the matching tag.
+    """
+    fp_abs = os.path.abspath(path)
+    input_base = os.path.abspath(folder_paths.get_input_directory())
+    if not Path(fp_abs).is_relative_to(input_base):
+        return []
+
+    rel = os.path.relpath(fp_abs, input_base)
+    parts = Path(rel).parts
+    if len(parts) == 2:
+        return get_known_subfolder_tags(parts[0])
+    return []
+
+
+def get_path_derived_tags_from_path(path: str) -> list[str]:
+    """Return all backend-derived tags for an asset path."""
+    tags = get_backend_system_tags_from_path(path)
+    for tag in get_known_input_subfolder_tags_from_path(path):
+        if tag not in tags:
+            tags.append(tag)
+    return tags
+
+
 def get_name_and_tags_from_asset_path(file_path: str) -> tuple[str, list[str]]:
    """Return (name, tags) derived from a filesystem path.

    - name: base filename with extension
-    - tags: [root_category] + parent folder names in order
+    - tags: backend-derived tags from root/model classification and known input
+      subfolder layout conventions

    Raises:
        ValueError: path does not belong to any known root.
    """
-    root_category, some_path = get_asset_category_and_relative_path(file_path)
-    p = Path(some_path)
-    parent_parts = [
-        part for part in p.parent.parts if part not in (".", "..", p.anchor)
-    ]
-    return p.name, list(dict.fromkeys(normalize_tags([root_category, *parent_parts])))
+    return Path(file_path).name, get_path_derived_tags_from_path(file_path)
@@ -25,6 +25,7 @@ class ReferenceData:
    preview_id: str | None
    created_at: datetime
    updated_at: datetime
+    loader_path: str | None = None
    system_metadata: dict[str, Any] | None = None
    job_id: str | None = None
    last_access_time: datetime | None = None
@@ -56,7 +57,6 @@ class IngestResult:

 class TagUsage(NamedTuple):
    name: str
-    tag_type: str
    count: int


@@ -71,6 +71,7 @@ class AssetSummaryData:
 class ListAssetsResult:
    items: list[AssetSummaryData]
    total: int
+    next_cursor: str | None = None


@dataclass(frozen=True)
@@ -93,6 +94,7 @@ def extract_reference_data(ref: AssetReference) -> ReferenceData:
        id=ref.id,
        name=ref.name,
        file_path=ref.file_path,
+        loader_path=ref.loader_path,
        user_metadata=ref.user_metadata,
        preview_id=ref.preview_id,
        system_metadata=ref.system_metadata,
@@ -75,7 +75,7 @@ def list_tags(
            owner_id=owner_id,
        )

-    return [TagUsage(name, tag_type, count) for name, tag_type, count in rows], total
+    return [TagUsage(name, count) for name, count in rows], total


 def list_tag_histogram(
@@ -50,21 +50,45 @@ class ModelFileManager:
        @routes.get("/experiment/models/preview/{folder}/{path_index}/{filename:.*}")
        async def get_model_preview(request):
            folder_name = request.match_info.get("folder", None)
-            path_index = int(request.match_info.get("path_index", None))
            filename = request.match_info.get("filename", None)

            if folder_name not in folder_paths.folder_names_and_paths:
                return web.Response(status=404)

+            # The "{filename:.*}" capture also matches the empty string, which
+            # would resolve to the folder itself; reject it explicitly.
+            if not filename:
+                return web.Response(status=400)
+
+            try:
+                path_index = int(request.match_info.get("path_index", None))
+            except (TypeError, ValueError):
+                return web.Response(status=400)
+
            folders = folder_paths.folder_names_and_paths[folder_name]
+            if path_index < 0 or path_index >= len(folders[0]):
+                return web.Response(status=404)
            folder = folders[0][path_index]
-            full_filename = os.path.join(folder, filename)
+            full_filename = os.path.normpath(os.path.join(folder, filename))
+
+            # Prevent path traversal: the requested file must stay within the
+            # configured model folder. `filename` is an unrestricted ".*" capture,
+            # so values like "../../../../etc/passwd" would otherwise escape it.
+            if not folder_paths.is_within_directory(folder, full_filename):
+                return web.Response(status=403)

            previews = self.get_model_previews(full_filename)
            default_preview = previews[0] if len(previews) > 0 else None
            if default_preview is None or (isinstance(default_preview, str) and not os.path.isfile(default_preview)):
                return web.Response(status=404)

+            # The preview is selected by a glob inside get_model_previews, so a
+            # companion file (e.g. "model.preview.png") could itself be a symlink
+            # resolving outside the model folder. Re-validate the file actually
+            # opened: is_within_directory realpaths it, catching symlink escape.
+            if isinstance(default_preview, str) and not folder_paths.is_within_directory(folder, default_preview):
+                return web.Response(status=403)
+
            try:
                with Image.open(default_preview) as img:
                    img_bytes = BytesIO()
@@ -6,6 +6,7 @@ import glob
 import shutil
 import logging
 import tempfile
+import mimetypes
 from aiohttp import web
 from urllib import parse
 from comfy.cli_args import args
@@ -336,7 +337,20 @@ class UserManager():
            if not isinstance(path, str):
                return path

-            return web.FileResponse(path)
+            # User data files are arbitrary user-supplied content and are never
+            # meant to render inline. Disable MIME sniffing and force a download
+            # so uploaded markup/scripts can't execute in the app origin (stored
+            # XSS). Content-Disposition: attachment is the load-bearing guard;
+            # the content-type override and nosniff are defence in depth.
+            content_type = mimetypes.guess_type(path)[0] or 'application/octet-stream'
+            if folder_paths.is_dangerous_content_type(content_type):
+                content_type = 'application/octet-stream'
+
+            return web.FileResponse(path, headers={
+                "Content-Type": content_type,
+                "X-Content-Type-Options": "nosniff",
+                "Content-Disposition": "attachment",
+            })

        @routes.post("/userdata/{file}")
        async def post_userdata(request):
@@ -0,0 +1,569 @@
+{
+  "revision": 0,
+  "last_node_id": 89,
+  "last_link_id": 0,
+  "nodes": [
+    {
+      "id": 89,
+      "type": "85e595bd-af9e-40ee-85c5-b98bb15da47a",
+      "pos": [
+        320,
+        520
+      ],
+      "size": [
+        400,
+        360
+      ],
+      "flags": {},
+      "order": 3,
+      "mode": 0,
+      "inputs": [
+        {
+          "localized_name": "image",
+          "name": "image",
+          "type": "IMAGE",
+          "link": null
+        },
+        {
+          "name": "resolution",
+          "type": "INT",
+          "widget": {
+            "name": "resolution"
+          },
+          "link": null
+        },
+        {
+          "name": "resize_method",
+          "type": "COMBO",
+          "widget": {
+            "name": "resize_method"
+          },
+          "link": null
+        },
+        {
+          "label": "output_type",
+          "name": "output",
+          "type": "COMFY_DYNAMICCOMBO_V3",
+          "widget": {
+            "name": "output"
+          },
+          "link": null
+        },
+        {
+          "label": "output_normalization",
+          "name": "output.normalization",
+          "type": "COMBO",
+          "widget": {
+            "name": "output.normalization"
+          },
+          "link": null
+        },
+        {
+          "label": "apply_sky_clip",
+          "name": "output.apply_sky_clip",
+          "type": "BOOLEAN",
+          "widget": {
+            "name": "output.apply_sky_clip"
+          },
+          "link": null
+        },
+        {
+          "name": "model_name",
+          "type": "COMBO",
+          "widget": {
+            "name": "model_name"
+          },
+          "link": null
+        }
+      ],
+      "outputs": [
+        {
+          "localized_name": "IMAGE",
+          "name": "IMAGE",
+          "type": "IMAGE",
+          "links": []
+        }
+      ],
+      "properties": {
+        "proxyWidgets": [
+          [
+            "87",
+            "resolution"
+          ],
+          [
+            "87",
+            "resize_method"
+          ],
+          [
+            "86",
+            "output"
+          ],
+          [
+            "86",
+            "output.normalization"
+          ],
+          [
+            "86",
+            "output.apply_sky_clip"
+          ],
+          [
+            "88",
+            "model_name"
+          ]
+        ],
+        "cnr_id": "comfy-core",
+        "ver": "0.24.0"
+      },
+      "widgets_values": [],
+      "title": "Image Depth Estimation (Depth Anything 3)"
+    }
+  ],
+  "links": [],
+  "version": 0.4,
+  "definitions": {
+    "subgraphs": [
+      {
+        "id": "85e595bd-af9e-40ee-85c5-b98bb15da47a",
+        "version": 1,
+        "state": {
+          "lastGroupId": 4,
+          "lastNodeId": 89,
+          "lastLinkId": 109,
+          "lastRerouteId": 0
+        },
+        "revision": 2,
+        "config": {},
+        "name": "Image Depth Estimation (Depth Anything 3)",
+        "inputNode": {
+          "id": -10,
+          "bounding": [
+            400,
+            90,
+            166.998046875,
+            188
+          ]
+        },
+        "outputNode": {
+          "id": -20,
+          "bounding": [
+            1250,
+            146,
+            128,
+            68
+          ]
+        },
+        "inputs": [
+          {
+            "id": "43cf3118-495a-487d-8eb3-a17c7e92f64f",
+            "name": "image",
+            "type": "IMAGE",
+            "linkIds": [
+              19
+            ],
+            "localized_name": "image",
+            "pos": [
+              542.998046875,
+              114
+            ]
+          },
+          {
+            "id": "1089a0a1-6db1-45a8-84b0-0bfdc2ed920a",
+            "name": "resolution",
+            "type": "INT",
+            "linkIds": [
+              22
+            ],
+            "pos": [
+              542.998046875,
+              134
+            ]
+          },
+          {
+            "id": "25fb64ac-26d5-466d-995b-6d51b9afa2c4",
+            "name": "resize_method",
+            "type": "COMBO",
+            "linkIds": [
+              23
+            ],
+            "pos": [
+              542.998046875,
+              154
+            ]
+          },
+          {
+            "id": "8acafb7c-6c8b-46b3-9d74-c563498a3af1",
+            "name": "output",
+            "type": "COMFY_DYNAMICCOMBO_V3",
+            "linkIds": [
+              24
+            ],
+            "label": "output_type",
+            "pos": [
+              542.998046875,
+              174
+            ]
+          },
+          {
+            "id": "1da5009b-4648-43e8-a257-16426630cf22",
+            "name": "output.normalization",
+            "type": "COMBO",
+            "linkIds": [
+              25
+            ],
+            "label": "output_normalization",
+            "pos": [
+              542.998046875,
+              194
+            ]
+          },
+          {
+            "id": "fd7edb33-5fb1-4538-a411-26e5039a9321",
+            "name": "output.apply_sky_clip",
+            "type": "BOOLEAN",
+            "linkIds": [
+              26
+            ],
+            "label": "apply_sky_clip",
+            "pos": [
+              542.998046875,
+              214
+            ]
+          },
+          {
+            "id": "b5be4c8a-b833-4f1e-8c94-3ed1dd722190",
+            "name": "model_name",
+            "type": "COMBO",
+            "linkIds": [
+              106
+            ],
+            "pos": [
+              542.998046875,
+              234
+            ]
+          }
+        ],
+        "outputs": [
+          {
+            "id": "478ab537-63bc-4d74-a9f0-c975f550880f",
+            "name": "IMAGE",
+            "type": "IMAGE",
+            "linkIds": [
+              7
+            ],
+            "localized_name": "IMAGE",
+            "pos": [
+              1274,
+              170
+            ]
+          }
+        ],
+        "widgets": [],
+        "nodes": [
+          {
+            "id": 86,
+            "type": "DA3Render",
+            "pos": [
+              800,
+              310
+            ],
+            "size": [
+              380,
+              130
+            ],
+            "flags": {},
+            "order": 0,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "da3_geometry",
+                "name": "da3_geometry",
+                "type": "DA3_GEOMETRY",
+                "link": 12
+              },
+              {
+                "localized_name": "output",
+                "name": "output",
+                "type": "COMFY_DYNAMICCOMBO_V3",
+                "widget": {
+                  "name": "output"
+                },
+                "link": 24
+              },
+              {
+                "localized_name": "output.normalization",
+                "name": "output.normalization",
+                "type": "COMBO",
+                "widget": {
+                  "name": "output.normalization"
+                },
+                "link": 25
+              },
+              {
+                "localized_name": "output.apply_sky_clip",
+                "name": "output.apply_sky_clip",
+                "type": "BOOLEAN",
+                "widget": {
+                  "name": "output.apply_sky_clip"
+                },
+                "link": 26
+              },
+              {
+                "name": "geometry",
+                "type": "DA3_GEOMETRY",
+                "link": null
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "IMAGE",
+                "name": "IMAGE",
+                "type": "IMAGE",
+                "slot_index": 0,
+                "links": [
+                  7
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "DA3Render",
+              "cnr_id": "comfy-core",
+              "ver": "0.19.0"
+            },
+            "widgets_values": [
+              "depth",
+              "v2_style",
+              false
+            ]
+          },
+          {
+            "id": 87,
+            "type": "DA3Inference",
+            "pos": [
+              800,
+              50
+            ],
+            "size": [
+              390,
+              130
+            ],
+            "flags": {},
+            "order": 1,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "da3_model",
+                "name": "da3_model",
+                "type": "DA3_MODEL",
+                "link": 107
+              },
+              {
+                "localized_name": "image",
+                "name": "image",
+                "type": "IMAGE",
+                "link": 19
+              },
+              {
+                "localized_name": "resolution",
+                "name": "resolution",
+                "type": "INT",
+                "widget": {
+                  "name": "resolution"
+                },
+                "link": 22
+              },
+              {
+                "localized_name": "resize_method",
+                "name": "resize_method",
+                "type": "COMBO",
+                "widget": {
+                  "name": "resize_method"
+                },
+                "link": 23
+              },
+              {
+                "localized_name": "mode",
+                "name": "mode",
+                "type": "COMFY_DYNAMICCOMBO_V3",
+                "widget": {
+                  "name": "mode"
+                },
+                "link": null
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "da3_geometry",
+                "name": "da3_geometry",
+                "type": "DA3_GEOMETRY",
+                "slot_index": 0,
+                "links": [
+                  12
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "DA3Inference",
+              "cnr_id": "comfy-core",
+              "ver": "0.19.0"
+            },
+            "widgets_values": [
+              504,
+              "upper_bound_resize",
+              "mono"
+            ]
+          },
+          {
+            "id": 88,
+            "type": "LoadDA3Model",
+            "pos": [
+              810,
+              -160
+            ],
+            "size": [
+              400,
+              140
+            ],
+            "flags": {},
+            "order": 2,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "model_name",
+                "name": "model_name",
+                "type": "COMBO",
+                "widget": {
+                  "name": "model_name"
+                },
+                "link": 106
+              },
+              {
+                "localized_name": "weight_dtype",
+                "name": "weight_dtype",
+                "type": "COMBO",
+                "widget": {
+                  "name": "weight_dtype"
+                },
+                "link": null
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "DA3_MODEL",
+                "name": "DA3_MODEL",
+                "type": "DA3_MODEL",
+                "links": [
+                  107
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "LoadDA3Model",
+              "cnr_id": "comfy-core",
+              "ver": "0.24.0",
+              "models": [
+                {
+                  "name": "depth_anything_3_mono_large.safetensors",
+                  "url": "https://huggingface.co/Comfy-Org/Depth-Anything-3/resolve/main/geometry_estimation/depth_anything_3_mono_large.safetensors",
+                  "directory": "geometry_estimation"
+                }
+              ]
+            },
+            "widgets_values": [
+              "depth_anything_3_mono_large.safetensors",
+              "default"
+            ]
+          }
+        ],
+        "groups": [],
+        "links": [
+          {
+            "id": 12,
+            "origin_id": 87,
+            "origin_slot": 0,
+            "target_id": 86,
+            "target_slot": 0,
+            "type": "DA3_GEOMETRY"
+          },
+          {
+            "id": 19,
+            "origin_id": -10,
+            "origin_slot": 0,
+            "target_id": 87,
+            "target_slot": 1,
+            "type": "IMAGE"
+          },
+          {
+            "id": 7,
+            "origin_id": 86,
+            "origin_slot": 0,
+            "target_id": -20,
+            "target_slot": 0,
+            "type": "IMAGE"
+          },
+          {
+            "id": 22,
+            "origin_id": -10,
+            "origin_slot": 1,
+            "target_id": 87,
+            "target_slot": 2,
+            "type": "INT"
+          },
+          {
+            "id": 23,
+            "origin_id": -10,
+            "origin_slot": 2,
+            "target_id": 87,
+            "target_slot": 3,
+            "type": "COMBO"
+          },
+          {
+            "id": 24,
+            "origin_id": -10,
+            "origin_slot": 3,
+            "target_id": 86,
+            "target_slot": 1,
+            "type": "COMFY_DYNAMICCOMBO_V3"
+          },
+          {
+            "id": 25,
+            "origin_id": -10,
+            "origin_slot": 4,
+            "target_id": 86,
+            "target_slot": 2,
+            "type": "COMBO"
+          },
+          {
+            "id": 26,
+            "origin_id": -10,
+            "origin_slot": 5,
+            "target_id": 86,
+            "target_slot": 3,
+            "type": "BOOLEAN"
+          },
+          {
+            "id": 106,
+            "origin_id": -10,
+            "origin_slot": 6,
+            "target_id": 88,
+            "target_slot": 0,
+            "type": "COMBO"
+          },
+          {
+            "id": 107,
+            "origin_id": 88,
+            "origin_slot": 0,
+            "target_id": 87,
+            "target_slot": 0,
+            "type": "DA3_MODEL"
+          }
+        ],
+        "extra": {},
+        "category": "Conditioning & Preprocessors/Depth",
+        "description": "This subgraph takes an input image and produces a depth map using the Depth Anything 3 model, which recovers spatially consistent geometry from any number of views. It is ideal for single or multi-view images, videos, and 3D scenes where accurate depth estimation is needed for tasks like SLAM, novel view synthesis, or spatial perception. The model uses a plain transformer backbone and supports both monocular and multi-view inputs without."
+      }
+    ]
+  },
+  "extra": {
+    "BlueprintDescription": "This subgraph takes an input image and produces a depth map using the Depth Anything 3 model, which recovers spatially consistent geometry from any number of views. It is ideal for single or multi-view images, videos, and 3D scenes where accurate depth estimation is needed for tasks like SLAM, novel view synthesis, or spatial perception. The model uses a plain transformer backbone and supports both monocular and multi-view inputs without."
+  }
+}
@@ -1077,9 +1077,12 @@
          }
        ],
        "extra": {},
-        "category": "Image generation and editing/Text to image"
+        "category": "Image generation and editing/Text to image",
+        "description": "This subgraph converts text prompts into non-photorealistic illustrations using a 2-billion-parameter model optimized for anime and artistic styles. It is ideal for generating concept art, character designs, or stylized illustrations where photorealism is not required. The model excels with anime and artistic content but performs poorly on realistic subjects."
      }
    ]
  },
-  "extra": {}
+  "extra": {
+    "BlueprintDescription": "This subgraph converts text prompts into non-photorealistic illustrations using a 2-billion-parameter model optimized for anime and artistic styles. It is ideal for generating concept art, character designs, or stylized illustrations where photorealism is not required. The model excels with anime and artistic content but performs poorly on realistic subjects."
+  }
 }
@@ -0,0 +1,825 @@
+{
+  "revision": 0,
+  "last_node_id": 97,
+  "last_link_id": 0,
+  "nodes": [
+    {
+      "id": 97,
+      "type": "253ec5ca-8333-4ddf-a036-9fc0923651b9",
+      "pos": [
+        410,
+        500
+      ],
+      "size": [
+        400,
+        400
+      ],
+      "flags": {},
+      "order": 3,
+      "mode": 0,
+      "inputs": [
+        {
+          "name": "video",
+          "type": "VIDEO",
+          "link": null
+        },
+        {
+          "name": "start_time",
+          "type": "FLOAT",
+          "widget": {
+            "name": "start_time"
+          },
+          "link": null
+        },
+        {
+          "name": "duration",
+          "type": "FLOAT",
+          "widget": {
+            "name": "duration"
+          },
+          "link": null
+        },
+        {
+          "name": "resolution",
+          "type": "INT",
+          "widget": {
+            "name": "resolution"
+          },
+          "link": null
+        },
+        {
+          "name": "resize_method",
+          "type": "COMBO",
+          "widget": {
+            "name": "resize_method"
+          },
+          "link": null
+        },
+        {
+          "label": "output_type",
+          "name": "output",
+          "type": "COMFY_DYNAMICCOMBO_V3",
+          "widget": {
+            "name": "output"
+          },
+          "link": null
+        },
+        {
+          "label": "normalization",
+          "name": "output.normalization",
+          "type": "COMBO",
+          "widget": {
+            "name": "output.normalization"
+          },
+          "link": null
+        },
+        {
+          "name": "output.apply_sky_clip",
+          "type": "BOOLEAN",
+          "widget": {
+            "name": "output.apply_sky_clip"
+          },
+          "link": null
+        },
+        {
+          "name": "model_name",
+          "type": "COMBO",
+          "widget": {
+            "name": "model_name"
+          },
+          "link": null
+        }
+      ],
+      "outputs": [
+        {
+          "localized_name": "IMAGE",
+          "name": "IMAGE",
+          "type": "IMAGE",
+          "links": []
+        },
+        {
+          "name": "audio",
+          "type": "AUDIO",
+          "links": []
+        },
+        {
+          "name": "fps",
+          "type": "FLOAT",
+          "links": []
+        }
+      ],
+      "properties": {
+        "proxyWidgets": [
+          [
+            "96",
+            "start_time"
+          ],
+          [
+            "96",
+            "duration"
+          ],
+          [
+            "93",
+            "resolution"
+          ],
+          [
+            "93",
+            "resize_method"
+          ],
+          [
+            "92",
+            "output"
+          ],
+          [
+            "92",
+            "output.normalization"
+          ],
+          [
+            "92",
+            "output.apply_sky_clip"
+          ],
+          [
+            "94",
+            "model_name"
+          ]
+        ],
+        "cnr_id": "comfy-core",
+        "ver": "0.24.0"
+      },
+      "widgets_values": [],
+      "title": "Video Depth Estimation (Depth Anything 3)"
+    }
+  ],
+  "links": [],
+  "version": 0.4,
+  "definitions": {
+    "subgraphs": [
+      {
+        "id": "253ec5ca-8333-4ddf-a036-9fc0923651b9",
+        "version": 1,
+        "state": {
+          "lastGroupId": 4,
+          "lastNodeId": 97,
+          "lastLinkId": 129,
+          "lastRerouteId": 0
+        },
+        "revision": 2,
+        "config": {},
+        "name": "Video Depth Estimation (Depth Anything 3)",
+        "inputNode": {
+          "id": -10,
+          "bounding": [
+            -230,
+            130,
+            167.912109375,
+            228
+          ]
+        },
+        "outputNode": {
+          "id": -20,
+          "bounding": [
+            1520,
+            140,
+            128,
+            108
+          ]
+        },
+        "inputs": [
+          {
+            "id": "698c28c6-cf92-4039-8b39-f3062868ea7c",
+            "name": "video",
+            "type": "VIDEO",
+            "linkIds": [
+              119
+            ],
+            "pos": [
+              -86.087890625,
+              154
+            ]
+          },
+          {
+            "id": "97a1f63e-1585-4a40-9dec-e2700120d84a",
+            "name": "start_time",
+            "type": "FLOAT",
+            "linkIds": [
+              121
+            ],
+            "pos": [
+              -86.087890625,
+              174
+            ]
+          },
+          {
+            "id": "4dbbd3b3-c5ee-4a56-a0d3-3268d3b2fd64",
+            "name": "duration",
+            "type": "FLOAT",
+            "linkIds": [
+              122
+            ],
+            "pos": [
+              -86.087890625,
+              194
+            ]
+          },
+          {
+            "id": "16f55101-f99d-4c0c-bebf-c3b31c54f13e",
+            "name": "resolution",
+            "type": "INT",
+            "linkIds": [
+              124
+            ],
+            "pos": [
+              -86.087890625,
+              214
+            ]
+          },
+          {
+            "id": "d9cd7693-4bb3-4ed7-9a75-276b997abcd9",
+            "name": "resize_method",
+            "type": "COMBO",
+            "linkIds": [
+              125
+            ],
+            "pos": [
+              -86.087890625,
+              234
+            ]
+          },
+          {
+            "id": "a6e90532-323b-462e-ba9c-1672384d5b31",
+            "name": "output",
+            "type": "COMFY_DYNAMICCOMBO_V3",
+            "linkIds": [
+              126
+            ],
+            "label": "output_type",
+            "pos": [
+              -86.087890625,
+              254
+            ]
+          },
+          {
+            "id": "69e6aeef-437d-4fde-b2fc-d5ab9369238d",
+            "name": "output.normalization",
+            "type": "COMBO",
+            "linkIds": [
+              127
+            ],
+            "label": "normalization",
+            "pos": [
+              -86.087890625,
+              274
+            ]
+          },
+          {
+            "id": "73206f72-f89a-4698-885e-5d9277df2998",
+            "name": "output.apply_sky_clip",
+            "type": "BOOLEAN",
+            "linkIds": [
+              128
+            ],
+            "pos": [
+              -86.087890625,
+              294
+            ]
+          },
+          {
+            "id": "dddbc7fc-9431-448a-9ed3-9aa62404288b",
+            "name": "model_name",
+            "type": "COMBO",
+            "linkIds": [
+              129
+            ],
+            "pos": [
+              -86.087890625,
+              314
+            ]
+          }
+        ],
+        "outputs": [
+          {
+            "id": "478ab537-63bc-4d74-a9f0-c975f550880f",
+            "name": "IMAGE",
+            "type": "IMAGE",
+            "linkIds": [
+              7
+            ],
+            "localized_name": "IMAGE",
+            "pos": [
+              1544,
+              164
+            ]
+          },
+          {
+            "id": "cdaf037e-79bc-4a94-b06c-0fd32e76f615",
+            "name": "audio",
+            "type": "AUDIO",
+            "linkIds": [
+              112
+            ],
+            "pos": [
+              1544,
+              184
+            ]
+          },
+          {
+            "id": "4c0e5484-d193-49c7-b107-92619628880a",
+            "name": "fps",
+            "type": "FLOAT",
+            "linkIds": [
+              113
+            ],
+            "pos": [
+              1544,
+              204
+            ]
+          }
+        ],
+        "widgets": [],
+        "nodes": [
+          {
+            "id": 92,
+            "type": "DA3Render",
+            "pos": [
+              740,
+              230
+            ],
+            "size": [
+              380,
+              130
+            ],
+            "flags": {},
+            "order": 0,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "da3_geometry",
+                "name": "da3_geometry",
+                "type": "DA3_GEOMETRY",
+                "link": 12
+              },
+              {
+                "localized_name": "output",
+                "name": "output",
+                "type": "COMFY_DYNAMICCOMBO_V3",
+                "widget": {
+                  "name": "output"
+                },
+                "link": 126
+              },
+              {
+                "localized_name": "output.normalization",
+                "name": "output.normalization",
+                "type": "COMBO",
+                "widget": {
+                  "name": "output.normalization"
+                },
+                "link": 127
+              },
+              {
+                "localized_name": "output.apply_sky_clip",
+                "name": "output.apply_sky_clip",
+                "type": "BOOLEAN",
+                "widget": {
+                  "name": "output.apply_sky_clip"
+                },
+                "link": 128
+              },
+              {
+                "name": "geometry",
+                "type": "DA3_GEOMETRY",
+                "link": null
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "IMAGE",
+                "name": "IMAGE",
+                "type": "IMAGE",
+                "slot_index": 0,
+                "links": [
+                  7
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "DA3Render",
+              "cnr_id": "comfy-core",
+              "ver": "0.19.0"
+            },
+            "widgets_values": [
+              "depth",
+              "v2_style",
+              false
+            ]
+          },
+          {
+            "id": 93,
+            "type": "DA3Inference",
+            "pos": [
+              740,
+              -30
+            ],
+            "size": [
+              390,
+              130
+            ],
+            "flags": {},
+            "order": 1,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "da3_model",
+                "name": "da3_model",
+                "type": "DA3_MODEL",
+                "link": 107
+              },
+              {
+                "localized_name": "image",
+                "name": "image",
+                "type": "IMAGE",
+                "link": 111
+              },
+              {
+                "localized_name": "resolution",
+                "name": "resolution",
+                "type": "INT",
+                "widget": {
+                  "name": "resolution"
+                },
+                "link": 124
+              },
+              {
+                "localized_name": "resize_method",
+                "name": "resize_method",
+                "type": "COMBO",
+                "widget": {
+                  "name": "resize_method"
+                },
+                "link": 125
+              },
+              {
+                "localized_name": "mode",
+                "name": "mode",
+                "type": "COMFY_DYNAMICCOMBO_V3",
+                "widget": {
+                  "name": "mode"
+                },
+                "link": null
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "da3_geometry",
+                "name": "da3_geometry",
+                "type": "DA3_GEOMETRY",
+                "slot_index": 0,
+                "links": [
+                  12
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "DA3Inference",
+              "cnr_id": "comfy-core",
+              "ver": "0.19.0"
+            },
+            "widgets_values": [
+              504,
+              "lower_bound_resize",
+              "mono"
+            ]
+          },
+          {
+            "id": 94,
+            "type": "LoadDA3Model",
+            "pos": [
+              50,
+              410
+            ],
+            "size": [
+              400,
+              140
+            ],
+            "flags": {},
+            "order": 2,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "model_name",
+                "name": "model_name",
+                "type": "COMBO",
+                "widget": {
+                  "name": "model_name"
+                },
+                "link": 129
+              },
+              {
+                "localized_name": "weight_dtype",
+                "name": "weight_dtype",
+                "type": "COMBO",
+                "widget": {
+                  "name": "weight_dtype"
+                },
+                "link": null
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "DA3_MODEL",
+                "name": "DA3_MODEL",
+                "type": "DA3_MODEL",
+                "links": [
+                  107
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "LoadDA3Model",
+              "cnr_id": "comfy-core",
+              "ver": "0.24.0",
+              "models": [
+                {
+                  "name": "depth_anything_3_mono_large.safetensors",
+                  "url": "https://huggingface.co/Comfy-Org/Depth-Anything-3/resolve/main/geometry_estimation/depth_anything_3_mono_large.safetensors",
+                  "directory": "geometry_estimation"
+                }
+              ]
+            },
+            "widgets_values": [
+              "depth_anything_3_mono_large.safetensors",
+              "default"
+            ]
+          },
+          {
+            "id": 95,
+            "type": "GetVideoComponents",
+            "pos": [
+              70,
+              -140
+            ],
+            "size": [
+              260,
+              120
+            ],
+            "flags": {},
+            "order": 3,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "video",
+                "name": "video",
+                "type": "VIDEO",
+                "link": 120
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "images",
+                "name": "images",
+                "type": "IMAGE",
+                "links": [
+                  111
+                ]
+              },
+              {
+                "localized_name": "audio",
+                "name": "audio",
+                "type": "AUDIO",
+                "links": [
+                  112
+                ]
+              },
+              {
+                "localized_name": "fps",
+                "name": "fps",
+                "type": "FLOAT",
+                "links": [
+                  113
+                ]
+              },
+              {
+                "localized_name": "bit_depth",
+                "name": "bit_depth",
+                "type": "INT",
+                "links": null
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "GetVideoComponents",
+              "cnr_id": "comfy-core",
+              "ver": "0.24.0"
+            }
+          },
+          {
+            "id": 96,
+            "type": "Video Slice",
+            "pos": [
+              70,
+              -360
+            ],
+            "size": [
+              270,
+              170
+            ],
+            "flags": {},
+            "order": 4,
+            "mode": 0,
+            "inputs": [
+              {
+                "localized_name": "video",
+                "name": "video",
+                "type": "VIDEO",
+                "link": 119
+              },
+              {
+                "localized_name": "start_time",
+                "name": "start_time",
+                "type": "FLOAT",
+                "widget": {
+                  "name": "start_time"
+                },
+                "link": 121
+              },
+              {
+                "localized_name": "duration",
+                "name": "duration",
+                "type": "FLOAT",
+                "widget": {
+                  "name": "duration"
+                },
+                "link": 122
+              },
+              {
+                "localized_name": "strict_duration",
+                "name": "strict_duration",
+                "type": "BOOLEAN",
+                "widget": {
+                  "name": "strict_duration"
+                },
+                "link": null
+              }
+            ],
+            "outputs": [
+              {
+                "localized_name": "VIDEO",
+                "name": "VIDEO",
+                "type": "VIDEO",
+                "links": [
+                  120
+                ]
+              }
+            ],
+            "properties": {
+              "Node name for S&R": "Video Slice",
+              "cnr_id": "comfy-core",
+              "ver": "0.24.0"
+            },
+            "widgets_values": [
+              0,
+              5,
+              false
+            ]
+          }
+        ],
+        "groups": [],
+        "links": [
+          {
+            "id": 12,
+            "origin_id": 93,
+            "origin_slot": 0,
+            "target_id": 92,
+            "target_slot": 0,
+            "type": "DA3_GEOMETRY"
+          },
+          {
+            "id": 7,
+            "origin_id": 92,
+            "origin_slot": 0,
+            "target_id": -20,
+            "target_slot": 0,
+            "type": "IMAGE"
+          },
+          {
+            "id": 107,
+            "origin_id": 94,
+            "origin_slot": 0,
+            "target_id": 93,
+            "target_slot": 0,
+            "type": "DA3_MODEL"
+          },
+          {
+            "id": 111,
+            "origin_id": 95,
+            "origin_slot": 0,
+            "target_id": 93,
+            "target_slot": 1,
+            "type": "IMAGE"
+          },
+          {
+            "id": 112,
+            "origin_id": 95,
+            "origin_slot": 1,
+            "target_id": -20,
+            "target_slot": 1,
+            "type": "AUDIO"
+          },
+          {
+            "id": 113,
+            "origin_id": 95,
+            "origin_slot": 2,
+            "target_id": -20,
+            "target_slot": 2,
+            "type": "FLOAT"
+          },
+          {
+            "id": 119,
+            "origin_id": -10,
+            "origin_slot": 0,
+            "target_id": 96,
+            "target_slot": 0,
+            "type": "VIDEO"
+          },
+          {
+            "id": 120,
+            "origin_id": 96,
+            "origin_slot": 0,
+            "target_id": 95,
+            "target_slot": 0,
+            "type": "VIDEO"
+          },
+          {
+            "id": 121,
+            "origin_id": -10,
+            "origin_slot": 1,
+            "target_id": 96,
+            "target_slot": 1,
+            "type": "FLOAT"
+          },
+          {
+            "id": 122,
+            "origin_id": -10,
+            "origin_slot": 2,
+            "target_id": 96,
+            "target_slot": 2,
+            "type": "FLOAT"
+          },
+          {
+            "id": 124,
+            "origin_id": -10,
+            "origin_slot": 3,
+            "target_id": 93,
+            "target_slot": 2,
+            "type": "INT"
+          },
+          {
+            "id": 125,
+            "origin_id": -10,
+            "origin_slot": 4,
+            "target_id": 93,
+            "target_slot": 3,
+            "type": "COMBO"
+          },
+          {
+            "id": 126,
+            "origin_id": -10,
+            "origin_slot": 5,
+            "target_id": 92,
+            "target_slot": 1,
+            "type": "COMFY_DYNAMICCOMBO_V3"
+          },
+          {
+            "id": 127,
+            "origin_id": -10,
+            "origin_slot": 6,
+            "target_id": 92,
+            "target_slot": 2,
+            "type": "COMBO"
+          },
+          {
+            "id": 128,
+            "origin_id": -10,
+            "origin_slot": 7,
+            "target_id": 92,
+            "target_slot": 3,
+            "type": "BOOLEAN"
+          },
+          {
+            "id": 129,
+            "origin_id": -10,
+            "origin_slot": 8,
+            "target_id": 94,
+            "target_slot": 0,
+            "type": "COMBO"
+          }
+        ],
+        "extra": {},
+        "category": "Conditioning & Preprocessors/Depth",
+        "description": "This subgraph processes a video input through Depth Anything 3 to produce temporally consistent depth maps for each frame, outputting a depth video. It is ideal for video content requiring spatial geometry estimation, such as 3D reconstruction, SLAM, or novel view synthesis from moving cameras. The model uses a plain transformer backbone trained with a depth-ray representation, supporting any number of views without requiring known camera poses."
+      }
+    ]
+  },
+  "extra": {
+    "BlueprintDescription": "This subgraph processes a video input through Depth Anything 3 to produce temporally consistent depth maps for each frame, outputting a depth video. It is ideal for video content requiring spatial geometry estimation, such as 3D reconstruction, SLAM, or novel view synthesis from moving cameras. The model uses a plain transformer backbone trained with a depth-ray representation, supporting any number of views without requiring known camera poses."
+  }
+}
@@ -105,7 +105,7 @@ class WindowAttention(nn.Module):

        relative_position_bias = self.relative_position_bias_table[self.relative_position_index.long().view(-1)].view(
            self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1)  # Wh*Ww,Wh*Ww,nH
-        relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
+        relative_position_bias = comfy.ops.cast_to_input(relative_position_bias.permute(2, 0, 1).contiguous(), attn)  # nH, Wh*Ww, Wh*Ww
        attn = attn + relative_position_bias.unsqueeze(0)

        if mask is not None:
@@ -55,12 +55,7 @@ class BackgroundRemovalModel():
        out = torch.nn.functional.interpolate(out, size=(H, W), mode="bicubic", antialias=False)

        mask = out.sigmoid().to(device=comfy.model_management.intermediate_device(), dtype=comfy.model_management.intermediate_dtype())
-        if mask.ndim == 3:
-            mask = mask.unsqueeze(0)
-        if mask.shape[1] != 1:
-            mask = mask.movedim(-1, 1)
-
-        return mask
+        return mask.squeeze(1)  # (B, 1, H, W) -> (B, H, W)


 def load_background_removal_model(sd):
@@ -115,6 +115,7 @@ cache_group.add_argument("--cache-ram", nargs='*', type=float, default=[], metav
 cache_group.add_argument("--cache-classic", action="store_true", help="Use the old style (aggressive) caching.")
 cache_group.add_argument("--cache-lru", type=int, default=0, help="Use LRU caching with a maximum of N node results cached. May use more RAM/VRAM.")
 cache_group.add_argument("--cache-none", action="store_true", help="Reduced RAM/VRAM usage at the expense of executing every node for each run.")
+cache_group.add_argument("--high-ram", action="store_true", help="Can improve performance slightly on high RAM or on systems where pagefile use is preferred over model loading.")

 attn_group = parser.add_mutually_exclusive_group()
 attn_group.add_argument("--use-split-cross-attention", action="store_true", help="Use the split cross attention optimization. Ignored when xformers is used.")
@@ -133,7 +134,7 @@ upcast.add_argument("--dont-upcast-attention", action="store_true", help="Disabl
 parser.add_argument("--enable-manager", action="store_true", help="Enable the ComfyUI-Manager feature.")
 manager_group = parser.add_mutually_exclusive_group()
 manager_group.add_argument("--disable-manager-ui", action="store_true", help="Disables only the ComfyUI-Manager UI and endpoints. Scheduled installations and similar background tasks will still operate.")
-manager_group.add_argument("--enable-manager-legacy-ui", action="store_true", help="Enables the legacy UI of ComfyUI-Manager")
+manager_group.add_argument("--enable-manager-legacy-ui", action="store_true", help="Enables the legacy UI of ComfyUI-Manager. Implies --enable-manager.")


 vram_group = parser.add_mutually_exclusive_group()
@@ -144,11 +145,13 @@ vram_group.add_argument("--novram", action="store_true", help="When lowvram isn'
 vram_group.add_argument("--cpu", action="store_true", help="To use the CPU for everything (slow).")

 parser.add_argument("--reserve-vram", type=float, default=None, help="Set the amount of vram in GB you want to reserve for use by your OS/other software. By default some amount is reserved depending on your OS.")
+parser.add_argument("--vram-headroom", type=float, default=0, help="Set the amount of vram in GB for DynamicVRAM to maintain as extra headroom above default. ComfyUI will try and keep this much VRAM completely free and unused, even counting VRAM from other apps.")

 parser.add_argument("--async-offload", nargs='?', const=2, type=int, default=None, metavar="NUM_STREAMS", help="Use async weight offloading. An optional argument controls the amount of offload streams. Default is 2. Enabled by default on Nvidia.")
 parser.add_argument("--disable-async-offload", action="store_true", help="Disable async weight offloading.")
 parser.add_argument("--disable-dynamic-vram", action="store_true", help="Disable dynamic VRAM and use estimate based model loading.")
 parser.add_argument("--enable-dynamic-vram", action="store_true", help="Enable dynamic VRAM on systems where it's not enabled by default.")
+parser.add_argument("--fast-disk", action="store_true", help="Prefer disk-backed dynamic loading and offload over unpinned RAM. Can be faster for users with fast NVME disks.")

 parser.add_argument("--force-non-blocking", action="store_true", help="Force ComfyUI to use non-blocking operations for all applicable tensors. This may improve performance on some non-Nvidia systems but can cause issues with some workflows.")

@@ -165,6 +168,8 @@ class PerformanceFeature(enum.Enum):

 parser.add_argument("--fast", nargs="*", type=PerformanceFeature, help="Enable some untested and potentially quality deteriorating optimizations. This is used to test new features so using it might crash your comfyui. --fast with no arguments enables everything. You can pass a list specific optimizations if you only want to enable specific ones. Current valid optimizations: {}".format(" ".join(map(lambda c: c.value, PerformanceFeature))))

+parser.add_argument("--debug-hang", action="store_true", help="Enable stack trace dumps on Ctrl-C for debugging hangs.")
+
 parser.add_argument("--disable-pinned-memory", action="store_true", help="Disable pinned memory use.")

 parser.add_argument("--mmap-torch-files", action="store_true", help="Use mmap when loading ckpt/pt files.")
@@ -220,6 +225,7 @@ parser.add_argument(
 )

 parser.add_argument("--user-directory", type=is_valid_directory, default=None, help="Set the ComfyUI user directory with an absolute path. Overrides --base-directory.")
+parser.add_argument("--models-directory", type=is_valid_directory, default=None, help="Set the ComfyUI models directory. Overrides the models folder in --base-directory.")

 parser.add_argument("--enable-compress-response-body", action="store_true", help="Enable compressing response body.")

@@ -235,6 +241,7 @@ database_default_path = os.path.abspath(
 )
 parser.add_argument("--database-url", type=str, default=f"sqlite:///{database_default_path}", help="Specify the database URL, e.g. for an in-memory database you can use 'sqlite:///:memory:'.")
 parser.add_argument("--enable-assets", action="store_true", help="Enable the assets system (API routes, database synchronization, and background scanning).")
+parser.add_argument("--enable-asset-hashing", action="store_true", help="Compute blake3 content hashes when scanning assets. Hashing enables future asset-portability features (deduplication, cross-machine model resolution) but adds startup cost and per-output cost on large models directories. Off by default; enable to opt in.")
 parser.add_argument("--feature-flag", type=str, action='append', default=[], metavar="KEY[=VALUE]", help="Set a server feature flag. Use KEY=VALUE to set an explicit value, or bare KEY to set it to true. Can be specified multiple times. Boolean values (true/false) and numbers are auto-converted. Examples: --feature-flag show_signin_button=true  or  --feature-flag show_signin_button")
 parser.add_argument("--list-feature-flags", action="store_true", help="Print the registry of known CLI-settable feature flags as JSON and exit.")

@@ -246,6 +253,9 @@ else:
 if args.cache_ram is not None and len(args.cache_ram) > 2:
    parser.error("--cache-ram accepts at most two values: active GB and inactive GB")

+if args.high_ram:
+    args.cache_classic = True
+
 if args.windows_standalone_build:
    args.auto_launch = True

@@ -255,6 +265,10 @@ if args.disable_auto_launch:
 if args.force_fp16:
    args.fp16_unet = True

+# '--enable-manager-legacy-ui' is meaningless unless the manager is enabled, so imply '--enable-manager'.
+if args.enable_manager_legacy_ui:
+    args.enable_manager = True
+

 # '--fast' is not provided, use an empty set
 if args.fast is None:
@@ -9,6 +9,7 @@ import comfy.model_management
 import comfy.utils
 import comfy.clip_model
 import comfy.image_encoders.dino2
+import comfy.image_encoders.dino3

 class Output:
    def __getitem__(self, key):
@@ -23,12 +24,16 @@ IMAGE_ENCODERS = {
    "siglip_vision_model": comfy.clip_model.CLIPVisionModelProjection,
    "siglip2_vision_model": comfy.clip_model.CLIPVisionModelProjection,
    "dinov2": comfy.image_encoders.dino2.Dinov2Model,
+    "dinov3": comfy.image_encoders.dino3.DINOv3ViTModel,
 }

 class ClipVisionModel():
    def __init__(self, json_config):
-        with open(json_config) as f:
-            config = json.load(f)
+        if isinstance(json_config, dict):
+            config = json_config
+        else:
+            with open(json_config) as f:
+                config = json.load(f)

        self.image_size = config.get("image_size", 224)
        self.image_mean = config.get("image_mean", [0.48145466, 0.4578275, 0.40821073])
@@ -134,6 +139,8 @@ def load_clipvision_from_sd(sd, prefix="", convert_keys=False):
        json_config = os.path.join(os.path.join(os.path.dirname(os.path.realpath(__file__)), "image_encoders"), "dino2_giant.json")
    elif 'encoder.layer.23.layer_scale2.lambda1' in sd:
        json_config = os.path.join(os.path.join(os.path.dirname(os.path.realpath(__file__)), "image_encoders"), "dino2_large.json")
+    elif 'layer.0.mlp.gate_proj.weight' in sd and 'layer.31.norm1.weight' in sd: # Dinov3 ViT-H/16+ (SwiGLU gated MLP, 32 layers)
+        json_config = comfy.image_encoders.dino3.DINOV3_VITH_CONFIG
    else:
        return None

@@ -0,0 +1,46 @@
+"""Runtime config the frontend reads from /features to follow --comfy-api-base.
+
+For a non-prod comfy.org backend (staging or an ephemeral preview env), "/features" exposes the api and
+platform base so the frontend talks to it without a rebuild, plus the Firebase environment it should use.
+Prod bases are left alone and keep their build-time defaults.
+"""
+
+from typing import Any
+from urllib.parse import urlparse
+
+from comfy.cli_args import args
+
+_STAGING_API_HOST = "stagingapi.comfy.org"
+_TESTENV_HOST_SUFFIX = ".testenvs.comfy.org"
+_STAGING_PLATFORM_BASE_URL = "https://stagingplatform.comfy.org"
+
+
+def _is_staging_tier(host: str) -> bool:
+    return host == _STAGING_API_HOST or host.endswith(_TESTENV_HOST_SUFFIX)
+
+
+def normalize_comfy_api_base(url: str) -> str:
+    """Rewrite a testenv's friendly main host to its comfy-api '-registry' sibling."""
+    parsed = urlparse(url)
+    host = parsed.hostname or ""
+    if not host.endswith(_TESTENV_HOST_SUFFIX):
+        return url
+    label = host[: -len(_TESTENV_HOST_SUFFIX)]
+    if label.endswith("-registry"):
+        return url
+    return f"{parsed.scheme or 'https'}://{label}-registry{_TESTENV_HOST_SUFFIX}"
+
+
+def environment_overrides_for_base(base_url: str) -> dict[str, Any] | None:
+    """The /features overrides for a staging-tier base, or None for prod."""
+    if not _is_staging_tier(urlparse(base_url).hostname or ""):
+        return None
+    return {
+        "comfy_api_base_url": normalize_comfy_api_base(base_url).rstrip("/"),
+        "comfy_platform_base_url": _STAGING_PLATFORM_BASE_URL,
+        "firebase_env": "dev",
+    }
+
+
+def get_environment_overrides() -> dict[str, Any] | None:
+    return environment_overrides_for_base(getattr(args, "comfy_api_base", "") or "")
@@ -8,6 +8,8 @@ from abc import ABC, abstractmethod
 import logging
 import comfy.model_management
 import comfy.patcher_extension
+import comfy.utils
+import comfy.conds
 if TYPE_CHECKING:
    from comfy.model_base import BaseModel
    from comfy.model_patcher import ModelPatcher
@@ -51,12 +53,18 @@ class ContextHandlerABC(ABC):


 class IndexListContextWindow(ContextWindowABC):
-    def __init__(self, index_list: list[int], dim: int=0, total_frames: int=0):
+    def __init__(self, index_list: list[int], dim: int=0, total_frames: int=0, modality_windows: dict=None, context_overlap: int=0):
        self.index_list = index_list
        self.context_length = len(index_list)
+        self.context_overlap = context_overlap
        self.dim = dim
        self.total_frames = total_frames
        self.center_ratio = (min(index_list) + max(index_list)) / (2 * total_frames)
+        self.modality_windows = modality_windows  # dict of {mod_idx: IndexListContextWindow}
+        self.guide_frames_indices: list[int] = []
+        self.guide_overlap_info: list[tuple[int, int]] = []
+        self.guide_kf_local_positions: list[int] = []
+        self.guide_downscale_factors: list[int] = []

    def get_tensor(self, full: torch.Tensor, device=None, dim=None, retain_index_list=[]) -> torch.Tensor:
        if dim is None:
@@ -85,6 +93,11 @@ class IndexListContextWindow(ContextWindowABC):
        region_idx = int(self.center_ratio * num_regions)
        return min(max(region_idx, 0), num_regions - 1)

+    def get_window_for_modality(self, modality_idx: int) -> 'IndexListContextWindow':
+        if modality_idx == 0:
+            return self
+        return self.modality_windows[modality_idx]
+

 class IndexListCallbacks:
    EVALUATE_CONTEXT_WINDOWS = "evaluate_context_windows"
@@ -148,6 +161,172 @@ def slice_cond(cond_value, window: IndexListContextWindow, x_in: torch.Tensor, d
    return cond_value._copy_with(sliced)


+def compute_guide_overlap(guide_entries: list[dict], keyframe_idxs: torch.Tensor, temporal_downscale_ratio: int, window_index_list: list[int]):
+    """Compute which concatenated guide frames overlap with a context window.
+
+    Each guide's latent-space start is derived from its first token's pixel-t-start
+    in keyframe_idxs (shape (B, [t,h,w], num_tokens, [start, end])), divided by the
+    model's temporal_downscale_ratio.
+
+    Args:
+        guide_entries: list of guide_attention_entry dicts
+        keyframe_idxs: per-token pixel coords cond tensor for the modality
+        temporal_downscale_ratio: model's pixel-to-latent temporal compression ratio
+        window_index_list: the window's frame indices into the video portion
+
+    Returns:
+        suffix_indices: indices into the guide_frames tensor for frame selection
+        overlap_info: list of (entry_idx, overlap_count) for guide_attention_entries adjustment
+        kf_local_positions: window-local frame positions for keyframe_idxs regeneration
+        total_overlap: total number of overlapping guide frames
+    """
+    window_set = set(window_index_list)
+    window_list = list(window_index_list)
+    suffix_indices = []
+    overlap_info = []
+    kf_local_positions = []
+    suffix_base = 0
+    token_offset = 0
+
+    for entry_idx, entry in enumerate(guide_entries):
+        first_t_pixel = int(keyframe_idxs[0, 0, token_offset, 0].item())
+        latent_start = (first_t_pixel + temporal_downscale_ratio - 1) // temporal_downscale_ratio
+        guide_len = entry["latent_shape"][0]
+        entry_overlap = 0
+
+        for local_offset in range(guide_len):
+            video_pos = latent_start + local_offset
+            if video_pos in window_set:
+                suffix_indices.append(suffix_base + local_offset)
+                kf_local_positions.append(window_list.index(video_pos))
+                entry_overlap += 1
+
+        if entry_overlap > 0:
+            overlap_info.append((entry_idx, entry_overlap))
+        suffix_base += guide_len
+        token_offset += entry["pre_filter_count"]
+
+    return suffix_indices, overlap_info, kf_local_positions, len(suffix_indices)
+
+
+@dataclass
+class WindowingState:
+    """Per-modality context windowing state for each step,
+    built using IndexListContextHandler._build_window_state().
+    For non-multimodal models the lists are length 1
+    """
+    latents: list[torch.Tensor]                  # per-modality working latents (guide frames stripped)
+    guide_latents: list[torch.Tensor | None]     # per-modality guide frames stripped from latents
+    guide_entries: list[list[dict] | None]       # per-modality guide_attention_entry metadata
+    keyframe_idxs: list[torch.Tensor | None]     # per-modality keyframe_idxs tensor for guide latent_start derivation
+    latent_shapes: list | None                   # original packed shapes for unpack/pack (None if not multimodal)
+    dim: int = 0                                 # primary modality temporal dim for context windowing
+    is_multimodal: bool = False
+    temporal_downscale_ratio: int = 1            # model's pixel-to-latent temporal compression ratio
+
+    def prepare_window(self, window: IndexListContextWindow, model) -> IndexListContextWindow:
+        """Reformat window for multimodal contexts by deriving per-modality index lists.
+        Non-multimodal contexts return the input window unchanged.
+        """
+        if not self.is_multimodal:
+            return window
+
+        x = self.latents[0]
+        primary_total = self.latent_shapes[0][self.dim]
+        primary_overlap = window.context_overlap
+        map_shapes = self.latent_shapes
+        if x.size(self.dim) != primary_total:
+            map_shapes = list(self.latent_shapes)
+            video_shape = list(self.latent_shapes[0])
+            video_shape[self.dim] = x.size(self.dim)
+            map_shapes[0] = torch.Size(video_shape)
+        try:
+            per_modality_indices = model.map_context_window_to_modalities(
+                window.index_list, map_shapes, self.dim)
+        except AttributeError:
+            raise NotImplementedError(
+                f"{type(model).__name__} must implement map_context_window_to_modalities for multimodal context windows.")
+        modality_windows = {}
+        for mod_idx in range(1, len(self.latents)):
+            modality_total_frames = self.latents[mod_idx].shape[self.dim]
+            ratio = modality_total_frames / primary_total if primary_total > 0 else 1
+            modality_overlap = max(round(primary_overlap * ratio), 0)
+            modality_windows[mod_idx] = IndexListContextWindow(
+                per_modality_indices[mod_idx], dim=self.dim,
+                total_frames=modality_total_frames,
+                context_overlap=modality_overlap)
+        return IndexListContextWindow(
+            window.index_list, dim=self.dim, total_frames=x.shape[self.dim],
+            modality_windows=modality_windows, context_overlap=primary_overlap)
+
+    def slice_for_window(self, window: IndexListContextWindow, retain_index_list: list[int], device=None) -> tuple[list[torch.Tensor], list[int]]:
+        """Slice latents for a context window, injecting guide frames where applicable.
+        For multimodal contexts, uses the modality-specific windows derived in prepare_window().
+        """
+        sliced = []
+        guide_frame_counts = []
+        for idx in range(len(self.latents)):
+            modality_window = window.get_window_for_modality(idx)
+            retain = retain_index_list if idx == 0 else []
+            s = modality_window.get_tensor(self.latents[idx], device, retain_index_list=retain)
+            if self.guide_entries[idx] is not None:
+                s, ng = self._inject_guide_frames(s, modality_window, modality_idx=idx)
+            else:
+                ng = 0
+            sliced.append(s)
+            guide_frame_counts.append(ng)
+        return sliced, guide_frame_counts
+
+    def strip_guide_frames(self, out_per_modality: list[list[torch.Tensor]], guide_frame_counts: list[int], window: IndexListContextWindow):
+        """Strip injected guide frames from per-cond, per-modality outputs in place."""
+        for idx in range(len(self.latents)):
+            if guide_frame_counts[idx] > 0:
+                window_len = len(window.get_window_for_modality(idx).index_list)
+                for ci in range(len(out_per_modality)):
+                    out_per_modality[ci][idx] = out_per_modality[ci][idx].narrow(self.dim, 0, window_len)
+
+    def _inject_guide_frames(self, latent_slice: torch.Tensor, window: IndexListContextWindow, modality_idx: int = 0) -> tuple[torch.Tensor, int]:
+        guide_entries = self.guide_entries[modality_idx]
+        guide_frames = self.guide_latents[modality_idx]
+        keyframe_idxs = self.keyframe_idxs[modality_idx]
+        suffix_idx, overlap_info, kf_local_pos, guide_frame_count = compute_guide_overlap(
+            guide_entries, keyframe_idxs, self.temporal_downscale_ratio, window.index_list)
+        # Shift keyframe positions to account for causal_window_fix anchor occupying sub-pos 0.
+        anchor_idx = getattr(window, 'causal_anchor_index', None)
+        if anchor_idx is not None and anchor_idx >= 0:
+            kf_local_pos = [p + 1 for p in kf_local_pos]
+        window.guide_frames_indices = suffix_idx
+        window.guide_overlap_info = overlap_info
+        window.guide_kf_local_positions = kf_local_pos
+
+        # Derive per-overlap-entry latent_downscale_factor from guide entry latent_shape vs guide frame spatial dims.
+        # guide_frames has full (post-dilation) spatial dims; entry["latent_shape"] has pre-dilation dims.
+        guide_downscale_factors = []
+        if guide_frame_count > 0:
+            full_H = guide_frames.shape[3]
+            for entry_idx, _ in overlap_info:
+                entry_H = guide_entries[entry_idx]["latent_shape"][1]
+                guide_downscale_factors.append(full_H // entry_H)
+        window.guide_downscale_factors = guide_downscale_factors
+
+        if guide_frame_count > 0:
+            idx = tuple([slice(None)] * self.dim + [suffix_idx])
+            return torch.cat([latent_slice, guide_frames[idx]], dim=self.dim), guide_frame_count
+        return latent_slice, 0
+
+    def patch_latent_shapes(self, sub_conds, new_shapes):
+        if not self.is_multimodal:
+            return
+
+        for cond_list in sub_conds:
+            if cond_list is None:
+                continue
+            for cond_dict in cond_list:
+                model_conds = cond_dict.get('model_conds', {})
+                if 'latent_shapes' in model_conds:
+                    model_conds['latent_shapes'] = comfy.conds.CONDConstant(new_shapes)
+
+
@dataclass
 class ContextSchedule:
    name: str
@@ -162,7 +341,7 @@ ContextResults = collections.namedtuple("ContextResults", ['window_idx', 'sub_co
 class IndexListContextHandler(ContextHandlerABC):
    def __init__(self, context_schedule: ContextSchedule, fuse_method: ContextFuseMethod, context_length: int=1, context_overlap: int=0, context_stride: int=1,
                 closed_loop: bool=False, dim:int=0, freenoise: bool=False, cond_retain_index_list: list[int]=[], split_conds_to_windows: bool=False,
-                 causal_window_fix: bool=True):
+                 latent_retain_index_list: list[int]=[], causal_window_fix: bool=True):
        self.context_schedule = context_schedule
        self.fuse_method = fuse_method
        self.context_length = context_length
@@ -174,17 +353,118 @@ class IndexListContextHandler(ContextHandlerABC):
        self.freenoise = freenoise
        self.cond_retain_index_list = [int(x.strip()) for x in cond_retain_index_list.split(",")] if cond_retain_index_list else []
        self.split_conds_to_windows = split_conds_to_windows
+        self.latent_retain_index_list = [int(x.strip()) for x in latent_retain_index_list.split(",")] if latent_retain_index_list else []
        self.causal_window_fix = causal_window_fix

        self.callbacks = {}

+    @staticmethod
+    def _get_latent_shapes(conds):
+        for cond_list in conds:
+            if cond_list is None:
+                continue
+            for cond_dict in cond_list:
+                model_conds = cond_dict.get('model_conds', {})
+                if 'latent_shapes' in model_conds:
+                    return model_conds['latent_shapes'].cond
+        return None
+
+    @staticmethod
+    def _get_guide_entries(conds):
+        for cond_list in conds:
+            if cond_list is None:
+                continue
+            for cond_dict in cond_list:
+                model_conds = cond_dict.get('model_conds', {})
+                entries = model_conds.get('guide_attention_entries')
+                if entries is not None and hasattr(entries, 'cond') and entries.cond:
+                    return entries.cond
+        return None
+
+    @staticmethod
+    def _get_keyframe_idxs(conds):
+        for cond_list in conds:
+            if cond_list is None:
+                continue
+            for cond_dict in cond_list:
+                model_conds = cond_dict.get('model_conds', {})
+                kf = model_conds.get('keyframe_idxs')
+                if kf is not None and hasattr(kf, 'cond') and kf.cond is not None:
+                    return kf.cond
+        return None
+
+    def _apply_freenoise(self, noise: torch.Tensor, conds: list[list[dict]], seed: int) -> torch.Tensor:
+        """Apply FreeNoise shuffling, scaling context length/overlap per-modality by frame ratio.
+        If guide frames are present on the primary modality, only the video portion is shuffled.
+        """
+        guide_entries = self._get_guide_entries(conds)
+        guide_count = sum(e["latent_shape"][0] for e in guide_entries) if guide_entries else 0
+
+        latent_shapes = self._get_latent_shapes(conds)
+        if latent_shapes is not None and len(latent_shapes) > 1:
+            modalities = comfy.utils.unpack_latents(noise, latent_shapes)
+            primary_total = latent_shapes[0][self.dim]
+            primary_video_count = modalities[0].size(self.dim) - guide_count
+            apply_freenoise(modalities[0].narrow(self.dim, 0, primary_video_count), self.dim, self.context_length, self.context_overlap, seed)
+            for i in range(1, len(modalities)):
+                mod_total = latent_shapes[i][self.dim]
+                ratio = mod_total / primary_total if primary_total > 0 else 1
+                mod_ctx_len = max(round(self.context_length * ratio), 1)
+                mod_ctx_overlap = max(round(self.context_overlap * ratio), 0)
+                modalities[i] = apply_freenoise(modalities[i], self.dim, mod_ctx_len, mod_ctx_overlap, seed)
+            noise, _ = comfy.utils.pack_latents(modalities)
+            return noise
+        video_count = noise.size(self.dim) - guide_count
+        apply_freenoise(noise.narrow(self.dim, 0, video_count), self.dim, self.context_length, self.context_overlap, seed)
+        return noise
+
+    def _build_window_state(self, x_in: torch.Tensor, conds: list[list[dict]], model: BaseModel) -> WindowingState:
+        """Build windowing state for the current step, including unpacking latents and extracting guide frame info from conds."""
+        latent_shapes = self._get_latent_shapes(conds)
+        is_multimodal = latent_shapes is not None and len(latent_shapes) > 1
+        unpacked_latents = comfy.utils.unpack_latents(x_in, latent_shapes) if is_multimodal else [x_in]
+
+        unpacked_latents_list = list(unpacked_latents)
+        guide_latents_list = [None] * len(unpacked_latents)
+        guide_entries_list = [None] * len(unpacked_latents)
+        keyframe_idxs_list = [None] * len(unpacked_latents)
+
+        extracted_guide_entries = self._get_guide_entries(conds)
+        extracted_keyframe_idxs = self._get_keyframe_idxs(conds)
+
+        # Strip guide frames (only from first modality for now)
+        if extracted_guide_entries is not None:
+            guide_count = sum(e["latent_shape"][0] for e in extracted_guide_entries)
+            if guide_count > 0:
+                x = unpacked_latents[0]
+                latent_count = x.size(self.dim) - guide_count
+                unpacked_latents_list[0] = x.narrow(self.dim, 0, latent_count)
+                guide_latents_list[0] = x.narrow(self.dim, latent_count, guide_count)
+                guide_entries_list[0] = extracted_guide_entries
+                keyframe_idxs_list[0] = extracted_keyframe_idxs
+
+
+        return WindowingState(
+            latents=unpacked_latents_list,
+            guide_latents=guide_latents_list,
+            guide_entries=guide_entries_list,
+            keyframe_idxs=keyframe_idxs_list,
+            latent_shapes=latent_shapes,
+            dim=self.dim,
+            is_multimodal=is_multimodal,
+            temporal_downscale_ratio=model.latent_format.temporal_downscale_ratio)
+
    def should_use_context(self, model: BaseModel, conds: list[list[dict]], x_in: torch.Tensor, timestep: torch.Tensor, model_options: dict[str]) -> bool:
-        # for now, assume first dim is batch - should have stored on BaseModel in actual implementation
-        if x_in.size(self.dim) > self.context_length:
-            logging.info(f"Using context windows {self.context_length} with overlap {self.context_overlap} for {x_in.size(self.dim)} frames.")
+        window_state = self._build_window_state(x_in, conds, model) # build window_state to check frame counts, will be built again in execute
+        total_frame_count = window_state.latents[0].size(self.dim)
+        if total_frame_count > self.context_length:
+            logging.info(f"\nUsing context windows: Context length {self.context_length} with overlap {self.context_overlap} for {total_frame_count} frames.")
            if self.cond_retain_index_list:
                logging.info(f"Retaining original cond for indexes: {self.cond_retain_index_list}")
+            if self.latent_retain_index_list:
+                logging.info(f"Retaining original latent for indexes: {self.latent_retain_index_list}")
            return True
+        logging.info(f"\nNot using context windows since context length ({self.context_length}) exceeds input frames ({total_frame_count}).")
        return False

    def prepare_control_objects(self, control: ControlBase, device=None) -> ControlBase:
@@ -275,7 +555,9 @@ class IndexListContextHandler(ContextHandlerABC):
        return resized_cond

    def set_step(self, timestep: torch.Tensor, model_options: dict[str]):
-        mask = torch.isclose(model_options["transformer_options"]["sample_sigmas"], timestep[0], rtol=0.0001)
+        sample_sigmas = model_options["transformer_options"]["sample_sigmas"]
+        current_timestep = timestep[0].to(sample_sigmas.dtype)
+        mask = torch.isclose(sample_sigmas, current_timestep, rtol=0.0001)
        matches = torch.nonzero(mask)
        if torch.numel(matches) == 0:
            return  # substep from multi-step sampler: keep self._step from the last full step
@@ -284,54 +566,98 @@ class IndexListContextHandler(ContextHandlerABC):
    def get_context_windows(self, model: BaseModel, x_in: torch.Tensor, model_options: dict[str]) -> list[IndexListContextWindow]:
        full_length = x_in.size(self.dim) # TODO: choose dim based on model
        context_windows = self.context_schedule.func(full_length, self, model_options)
-        context_windows = [IndexListContextWindow(window, dim=self.dim, total_frames=full_length) for window in context_windows]
+        context_windows = [IndexListContextWindow(window, dim=self.dim, total_frames=full_length, context_overlap=self.context_overlap) for window in context_windows]
        return context_windows

    def execute(self, calc_cond_batch: Callable, model: BaseModel, conds: list[list[dict]], x_in: torch.Tensor, timestep: torch.Tensor, model_options: dict[str]):
        self._model = model
        self.set_step(timestep, model_options)
-        context_windows = self.get_context_windows(model, x_in, model_options)
-        enumerated_context_windows = list(enumerate(context_windows))

-        conds_final = [torch.zeros_like(x_in) for _ in conds]
+        window_state = self._build_window_state(x_in, conds, model)
+        num_modalities = len(window_state.latents)
+
+        context_windows = self.get_context_windows(model, window_state.latents[0], model_options)
+        enumerated_context_windows = list(enumerate(context_windows))
+        total_windows = len(enumerated_context_windows)
+
+        # Initialize per-modality accumulators (length 1 for single-modality)
+        accum = [[torch.zeros_like(m) for _ in conds] for m in window_state.latents]
        if self.fuse_method.name == ContextFuseMethods.RELATIVE:
-            counts_final = [torch.ones(get_shape_for_dim(x_in, self.dim), device=x_in.device) for _ in conds]
+            counts = [[torch.ones(get_shape_for_dim(m, self.dim), device=m.device) for _ in conds] for m in window_state.latents]
        else:
-            counts_final = [torch.zeros(get_shape_for_dim(x_in, self.dim), device=x_in.device) for _ in conds]
-        biases_final = [([0.0] * x_in.shape[self.dim]) for _ in conds]
+            counts = [[torch.zeros(get_shape_for_dim(m, self.dim), device=m.device) for _ in conds] for m in window_state.latents]
+        biases = [[([0.0] * m.shape[self.dim]) for _ in conds] for m in window_state.latents]

        for callback in comfy.patcher_extension.get_all_callbacks(IndexListCallbacks.EXECUTE_START, self.callbacks):
            callback(self, model, x_in, conds, timestep, model_options)

+        # accumulate results from each context window
        for enum_window in enumerated_context_windows:
-            results = self.evaluate_context_windows(calc_cond_batch, model, x_in, conds, timestep, [enum_window], model_options)
+            results = self.evaluate_context_windows(
+                calc_cond_batch, model, x_in, conds, timestep, [enum_window],
+                model_options, window_state=window_state, total_windows=total_windows)
            for result in results:
-                self.combine_context_window_results(x_in, result.sub_conds_out, result.sub_conds, result.window, result.window_idx, len(enumerated_context_windows), timestep,
-                                            conds_final, counts_final, biases_final)
+                # result.sub_conds_out is per-cond, per-modality: list[list[Tensor]]
+                for mod_idx in range(num_modalities):
+                    mod_out = [result.sub_conds_out[ci][mod_idx] for ci in range(len(conds))]
+                    modality_window = result.window.get_window_for_modality(mod_idx)
+                    self.combine_context_window_results(
+                        window_state.latents[mod_idx], mod_out, result.sub_conds, modality_window,
+                        result.window_idx, total_windows, timestep,
+                        accum[mod_idx], counts[mod_idx], biases[mod_idx])
+
+        # fuse accumulated results into final conds
        try:
-            # finalize conds
-            if self.fuse_method.name == ContextFuseMethods.RELATIVE:
-                # relative is already normalized, so return as is
-                del counts_final
-                return conds_final
-            else:
-                # normalize conds via division by context usage counts
-                for i in range(len(conds_final)):
-                    conds_final[i] /= counts_final[i]
-                del counts_final
-                return conds_final
+            result_out = []
+            for ci in range(len(conds)):
+                finalized = []
+                for mod_idx in range(num_modalities):
+                    if self.fuse_method.name != ContextFuseMethods.RELATIVE:
+                        accum[mod_idx][ci] /= counts[mod_idx][ci]
+                    f = accum[mod_idx][ci]
+
+                    # if guide frames were injected, append them to the end of the fused latents for the next step
+                    if window_state.guide_latents[mod_idx] is not None:
+                        f = torch.cat([f, window_state.guide_latents[mod_idx]], dim=self.dim)
+                    finalized.append(f)
+
+                # pack modalities together if needed
+                if window_state.is_multimodal and len(finalized) > 1:
+                    packed, _ = comfy.utils.pack_latents(finalized)
+                else:
+                    packed = finalized[0]
+
+                result_out.append(packed)
+            return result_out
        finally:
            for callback in comfy.patcher_extension.get_all_callbacks(IndexListCallbacks.EXECUTE_CLEANUP, self.callbacks):
                callback(self, model, x_in, conds, timestep, model_options)

-    def evaluate_context_windows(self, calc_cond_batch: Callable, model: BaseModel, x_in: torch.Tensor, conds, timestep: torch.Tensor, enumerated_context_windows: list[tuple[int, IndexListContextWindow]],
-                                model_options, device=None, first_device=None):
+    def evaluate_context_windows(self, calc_cond_batch: Callable, model: BaseModel, x_in: torch.Tensor, conds,
+                                timestep: torch.Tensor, enumerated_context_windows: list[tuple[int, IndexListContextWindow]],
+                                model_options, window_state: WindowingState, total_windows: int = None,
+                                device=None, first_device=None):
+        """Evaluate context windows and return per-cond, per-modality outputs in ContextResults.sub_conds_out
+
+        For each window:
+        1. Builds windows (for each modality if multimodal)
+        2. Slices window for each modality
+        3. Injects concatenated latent guide frames where present
+        4. Packs together if needed and calls model
+        5. Unpacks and strips any guides from outputs
+        """
+        x = window_state.latents[0]
+
        results: list[ContextResults] = []
        for window_idx, window in enumerated_context_windows:
            # allow processing to end between context window executions for faster Cancel
            comfy.model_management.throw_exception_if_processing_interrupted()

-            # causal_window_fix: prepend a pre-window frame that will be stripped post-forward
+            # prepare the window accounting for multimodal windows
+            window = window_state.prepare_window(window, model)
+
+            # causal_window_fix: prepend a pre-window frame that will be stripped post-forward.
+            # Set anchor before slice_for_window so the latent slice and downstream cond slices both pick it up.
            anchor_applied = False
            if self.causal_window_fix:
                anchor_idx = window.index_list[0] - 1
@@ -339,27 +665,46 @@ class IndexListContextHandler(ContextHandlerABC):
                    window.causal_anchor_index = anchor_idx
                    anchor_applied = True

+            # slice the window for each modality, injecting guide frames where applicable
+            sliced, guide_frame_counts_per_modality = window_state.slice_for_window(window, self.latent_retain_index_list, device)
+
            for callback in comfy.patcher_extension.get_all_callbacks(IndexListCallbacks.EVALUATE_CONTEXT_WINDOWS, self.callbacks):
                callback(self, model, x_in, conds, timestep, model_options, window_idx, window, model_options, device, first_device)

-            # update exposed params
+            logging.info(f"Context window {window_idx + 1}/{total_windows or len(enumerated_context_windows)}: frames {window.index_list[0]}-{window.index_list[-1]} of {x.shape[self.dim]}"
+                         + (f" (+{guide_frame_counts_per_modality[0]} guide frames)" if guide_frame_counts_per_modality[0] > 0 else "")
+                         )
+
+            # if multimodal, pack modalities together
+            if window_state.is_multimodal and len(sliced) > 1:
+                sub_x, sub_shapes = comfy.utils.pack_latents(sliced)
+            else:
+                sub_x, sub_shapes = sliced[0], [sliced[0].shape]
+
+            # get resized conds for window
            model_options["transformer_options"]["context_window"] = window
-            # get subsections of x, timestep, conds
-            sub_x = window.get_tensor(x_in, device)
-            sub_timestep = window.get_tensor(timestep, device, dim=0)
-            sub_conds = [self.get_resized_cond(cond, x_in, window, device) for cond in conds]
+            sub_timestep = window.get_tensor(timestep, dim=0)
+            sub_conds = [self.get_resized_cond(cond, x, window) for cond in conds]

+            # if multimodal, patch latent_shapes in conds for correct unpacking in model
+            window_state.patch_latent_shapes(sub_conds, sub_shapes)
+
+            # call model on window
            sub_conds_out = calc_cond_batch(model, sub_conds, sub_x, sub_timestep, model_options)
-            if device is not None:
-                for i in range(len(sub_conds_out)):
-                    sub_conds_out[i] = sub_conds_out[i].to(x_in.device)

-            # strip causal_window_fix anchor if applied
+            # unpack outputs
+            out_per_modality = [comfy.utils.unpack_latents(sub_conds_out[i], sub_shapes) for i in range(len(sub_conds_out))]
+
+            # strip causal_window_fix anchor from primary modality before guide strip so window_len math stays correct
            if anchor_applied:
-                for i in range(len(sub_conds_out)):
-                    sub_conds_out[i] = sub_conds_out[i].narrow(self.dim, 1, sub_conds_out[i].shape[self.dim] - 1)
+                for ci in range(len(out_per_modality)):
+                    t = out_per_modality[ci][0]
+                    out_per_modality[ci][0] = t.narrow(self.dim, 1, t.shape[self.dim] - 1)

-            results.append(ContextResults(window_idx, sub_conds_out, sub_conds, window))
+            # strip injected guide frames
+            window_state.strip_guide_frames(out_per_modality, guide_frame_counts_per_modality, window)
+
+            results.append(ContextResults(window_idx, out_per_modality, sub_conds, window))
        return results


@@ -383,7 +728,7 @@ class IndexListContextHandler(ContextHandlerABC):
                    biases_final[i][idx] = bias_total + bias
        else:
            # add conds and counts based on weights of fuse method
-            weights = get_context_weights(window.context_length, x_in.shape[self.dim], window.index_list, self, sigma=timestep)
+            weights = get_context_weights(window.context_length, x_in.shape[self.dim], window.index_list, self, sigma=timestep, context_overlap=window.context_overlap)
            weights_tensor = match_weights_to_dim(weights, x_in, self.dim, device=x_in.device)
            for i in range(len(sub_conds_out)):
                window.add_window(conds_final[i], sub_conds_out[i] * weights_tensor)
@@ -393,16 +738,22 @@ class IndexListContextHandler(ContextHandlerABC):
            callback(self, x_in, sub_conds_out, sub_conds, window, window_idx, total_windows, timestep, conds_final, counts_final, biases_final)


-def _prepare_sampling_wrapper(executor, model, noise_shape: torch.Tensor, *args, **kwargs):
-    # limit noise_shape length to context_length for more accurate vram use estimation
+def _prepare_sampling_wrapper(executor, model, noise_shape: torch.Tensor, conds, *args, **kwargs):
+    # Scale noise_shape to a single context window so VRAM estimation budgets per-window.
    model_options = kwargs.get("model_options", None)
    if model_options is None:
        raise Exception("model_options not found in prepare_sampling_wrapper; this should never happen, something went wrong.")
    handler: IndexListContextHandler = model_options.get("context_handler", None)
    if handler is not None:
        noise_shape = list(noise_shape)
-        noise_shape[handler.dim] = min(noise_shape[handler.dim], handler.context_length)
-    return executor(model, noise_shape, *args, **kwargs)
+        is_packed = len(noise_shape) == 3 and noise_shape[1] == 1
+        if is_packed:
+            # TODO: latent_shapes cond isn't attached yet at this point, so we can't compute a
+            # per-window flat latent here. Skipping the clamp over-estimates but prevents immediate OOM.
+            pass
+        elif handler.dim < len(noise_shape) and noise_shape[handler.dim] > handler.context_length:
+            noise_shape[handler.dim] = min(noise_shape[handler.dim], handler.context_length)
+    return executor(model, noise_shape, conds, *args, **kwargs)


 def create_prepare_sampling_wrapper(model: ModelPatcher):
@@ -422,11 +773,12 @@ def _sampler_sample_wrapper(executor, guider, sigmas, extra_args, callback, nois
        raise Exception("context_handler not found in sampler_sample_wrapper; this should never happen, something went wrong.")
    if not handler.freenoise:
        return executor(guider, sigmas, extra_args, callback, noise, *args, **kwargs)
-    noise = apply_freenoise(noise, handler.dim, handler.context_length, handler.context_overlap, extra_args["seed"])
+
+    conds = [guider.conds.get('positive', guider.conds.get('negative', []))]
+    noise = handler._apply_freenoise(noise, conds, extra_args["seed"])

    return executor(guider, sigmas, extra_args, callback, noise, *args, **kwargs)

-
 def create_sampler_sample_wrapper(model: ModelPatcher):
    model.add_wrapper_with_key(
        comfy.patcher_extension.WrappersMP.SAMPLER_SAMPLE,
@@ -434,7 +786,6 @@ def create_sampler_sample_wrapper(model: ModelPatcher):
        _sampler_sample_wrapper
    )

-
 def match_weights_to_dim(weights: list[float], x_in: torch.Tensor, dim: int, device=None) -> torch.Tensor:
    total_dims = len(x_in.shape)
    weights_tensor = torch.Tensor(weights).to(device=device)
@@ -580,8 +931,9 @@ def get_matching_context_schedule(context_schedule: str) -> ContextSchedule:
    return ContextSchedule(context_schedule, func)


-def get_context_weights(length: int, full_length: int, idxs: list[int], handler: IndexListContextHandler, sigma: torch.Tensor=None):
-    return handler.fuse_method.func(length, sigma=sigma, handler=handler, full_length=full_length, idxs=idxs)
+def get_context_weights(length: int, full_length: int, idxs: list[int], handler: IndexListContextHandler, sigma: torch.Tensor=None, context_overlap: int=None):
+    context_overlap = handler.context_overlap if context_overlap is None else context_overlap
+    return handler.fuse_method.func(length, sigma=sigma, handler=handler, full_length=full_length, idxs=idxs, context_overlap=context_overlap)


 def create_weights_flat(length: int, **kwargs) -> list[float]:
@@ -599,18 +951,18 @@ def create_weights_pyramid(length: int, **kwargs) -> list[float]:
        weight_sequence = list(range(1, max_weight, 1)) + [max_weight] + list(range(max_weight - 1, 0, -1))
    return weight_sequence

-def create_weights_overlap_linear(length: int, full_length: int, idxs: list[int], handler: IndexListContextHandler, **kwargs):
+def create_weights_overlap_linear(length: int, full_length: int, idxs: list[int], context_overlap: int, **kwargs):
    # based on code in Kijai's WanVideoWrapper: https://github.com/kijai/ComfyUI-WanVideoWrapper/blob/dbb2523b37e4ccdf45127e5ae33e31362f755c8e/nodes.py#L1302
    # only expected overlap is given different weights
    weights_torch = torch.ones((length))
    # blend left-side on all except first window
    if min(idxs) > 0:
-        ramp_up = torch.linspace(1e-37, 1, handler.context_overlap)
-        weights_torch[:handler.context_overlap] = ramp_up
+        ramp_up = torch.linspace(1e-37, 1, context_overlap)
+        weights_torch[:context_overlap] = ramp_up
    # blend right-side on all except last window
    if max(idxs) < full_length-1:
-        ramp_down = torch.linspace(1, 1e-37, handler.context_overlap)
-        weights_torch[-handler.context_overlap:] = ramp_down
+        ramp_down = torch.linspace(1, 1e-37, context_overlap)
+        weights_torch[-context_overlap:] = ramp_down
    return weights_torch

 class ContextFuseMethods:
@@ -1,5 +1,20 @@
+import logging
+
 import torch

+_CK_STOCHASTIC_ROUNDING_AVAILABLE = False
+try:
+    import comfy_kitchen as ck
+    _ck_stochastic_rounding_fp8 = ck.stochastic_rounding_fp8
+    _CK_STOCHASTIC_ROUNDING_AVAILABLE = True
+except (AttributeError, ImportError):
+    logging.warning("comfy_kitchen does not support stochastic FP8 rounding, please update comfy_kitchen.")
+
+if not _CK_STOCHASTIC_ROUNDING_AVAILABLE:
+    def _ck_stochastic_rounding_fp8(value, rng, dtype):
+        raise NotImplementedError("comfy_kitchen does not support stochastic FP8 rounding")
+
+
 def calc_mantissa(abs_x, exponent, normal_mask, MANTISSA_BITS, EXPONENT_BIAS, generator=None):
    mantissa_scaled = torch.where(
        normal_mask,
@@ -57,6 +72,10 @@ def stochastic_rounding(value, dtype, seed=0):
    if dtype == torch.float8_e4m3fn or dtype == torch.float8_e5m2:
        generator = torch.Generator(device=value.device)
        generator.manual_seed(seed)
+        if _CK_STOCHASTIC_ROUNDING_AVAILABLE:
+            rng = torch.randint(0, 256, value.size(), dtype=torch.uint8, layout=value.layout, device=value.device, generator=generator)
+            return _ck_stochastic_rounding_fp8(value, rng, dtype)
+
        output = torch.empty_like(value, dtype=dtype)
        num_slices = max(1, (value.numel() / (4096 * 4096)))
        slice_size = max(1, round(value.shape[0] / num_slices))
@@ -1,7 +1,13 @@
 import torch
+import torch.nn.functional as F
+
 from comfy.text_encoders.bert import BertAttention
 import comfy.model_management
 from comfy.ldm.modules.attention import optimized_attention_for_device
+from comfy.ldm.depth_anything_3.reference_view_selector import (
+    select_reference_view, reorder_by_reference, restore_original_order,
+    THRESH_FOR_REF_SELECTION,
+)


 class Dino2AttentionOutput(torch.nn.Module):
@@ -14,13 +20,41 @@ class Dino2AttentionOutput(torch.nn.Module):


 class Dino2AttentionBlock(torch.nn.Module):
-    def __init__(self, embed_dim, heads, layer_norm_eps, dtype, device, operations):
+    def __init__(self, embed_dim, heads, layer_norm_eps, dtype, device, operations,
+                 qk_norm=False):
        super().__init__()
+        self.heads = heads
+        self.head_dim = embed_dim // heads
        self.attention = BertAttention(embed_dim, heads, dtype, device, operations)
        self.output = Dino2AttentionOutput(embed_dim, embed_dim, layer_norm_eps, dtype, device, operations)
+        if qk_norm:
+            self.q_norm = operations.LayerNorm(self.head_dim, dtype=dtype, device=device)
+            self.k_norm = operations.LayerNorm(self.head_dim, dtype=dtype, device=device)
+        else:
+            self.q_norm = None
+            self.k_norm = None

-    def forward(self, x, mask, optimized_attention):
-        return self.output(self.attention(x, mask, optimized_attention))
+    def forward(self, x, mask, optimized_attention, pos=None, rope=None):
+        # Fast path used by the existing CLIP-vision DINOv2 (no DA3 extensions).
+        if self.q_norm is None and rope is None:
+            return self.output(self.attention(x, mask, optimized_attention))
+
+        # DA3 path: do QKV manually so we can apply per-head QK-norm and 2D RoPE.
+        attn = self.attention
+        B, N, C = x.shape
+        h = self.heads
+        d = self.head_dim
+        q = attn.query(x).view(B, N, h, d).transpose(1, 2)
+        k = attn.key(x).view(B, N, h, d).transpose(1, 2)
+        v = attn.value(x).view(B, N, h, d).transpose(1, 2)
+        if self.q_norm is not None:
+            q = self.q_norm(q)
+            k = self.k_norm(k)
+        if rope is not None and pos is not None:
+            q = rope(q, pos)
+            k = rope(k, pos)
+        out = optimized_attention(q, k, v, h, mask=mask, skip_reshape=True)
+        return self.output(out)


 class LayerScale(torch.nn.Module):
@@ -64,9 +98,11 @@ class SwiGLUFFN(torch.nn.Module):


 class Dino2Block(torch.nn.Module):
-    def __init__(self, dim, num_heads, layer_norm_eps, dtype, device, operations, use_swiglu_ffn):
+    def __init__(self, dim, num_heads, layer_norm_eps, dtype, device, operations, use_swiglu_ffn,
+                 qk_norm=False):
        super().__init__()
-        self.attention = Dino2AttentionBlock(dim, num_heads, layer_norm_eps, dtype, device, operations)
+        self.attention = Dino2AttentionBlock(dim, num_heads, layer_norm_eps, dtype, device, operations,
+                                             qk_norm=qk_norm)
        self.layer_scale1 = LayerScale(dim, dtype, device, operations)
        self.layer_scale2 = LayerScale(dim, dtype, device, operations)
        if use_swiglu_ffn:
@@ -76,19 +112,90 @@ class Dino2Block(torch.nn.Module):
        self.norm1 = operations.LayerNorm(dim, eps=layer_norm_eps, dtype=dtype, device=device)
        self.norm2 = operations.LayerNorm(dim, eps=layer_norm_eps, dtype=dtype, device=device)

-    def forward(self, x, optimized_attention):
-        x = x + self.layer_scale1(self.attention(self.norm1(x), None, optimized_attention))
+    def forward(self, x, optimized_attention, pos=None, rope=None, attn_mask=None):
+        x = x + self.layer_scale1(self.attention(self.norm1(x), attn_mask, optimized_attention,
+                                                 pos=pos, rope=rope))
        x = x + self.layer_scale2(self.mlp(self.norm2(x)))
        return x


-class Dino2Encoder(torch.nn.Module):
-    def __init__(self, dim, num_heads, layer_norm_eps, num_layers, dtype, device, operations, use_swiglu_ffn):
+# -----------------------------------------------------------------------------
+# 2D Rotary position embedding (DA3 extension)
+# -----------------------------------------------------------------------------
+
+
+class _PositionGetter:
+    """Cache (h, w) -> flat (y, x) position grid used to feed ``rope``."""
+
+    def __init__(self):
+        self._cache: dict = {}
+
+    def __call__(self, batch_size: int, height: int, width: int, device) -> torch.Tensor:
+        key = (height, width, device)
+        if key not in self._cache:
+            y = torch.arange(height, device=device)
+            x = torch.arange(width, device=device)
+            self._cache[key] = torch.cartesian_prod(y, x)
+        cached = self._cache[key]
+        return cached.view(1, height * width, 2).expand(batch_size, -1, -1).clone()
+
+
+class RotaryPositionEmbedding2D(torch.nn.Module):
+    """2D RoPE used by DA3-Small/Base. No learnable parameters."""
+
+    def __init__(self, frequency: float = 100.0):
        super().__init__()
-        self.layer = torch.nn.ModuleList([Dino2Block(dim, num_heads, layer_norm_eps, dtype, device, operations, use_swiglu_ffn = use_swiglu_ffn)
-                                          for _ in range(num_layers)])
+        self.base_frequency = frequency
+        self._freq_cache: dict = {}
+
+    def _components(self, dim: int, seq_len: int, device, dtype):
+        key = (dim, seq_len, device, dtype)
+        if key not in self._freq_cache:
+            exp = torch.arange(0, dim, 2, device=device).float() / dim
+            inv_freq = 1.0 / (self.base_frequency ** exp)
+            pos = torch.arange(seq_len, device=device, dtype=inv_freq.dtype)
+            ang = torch.einsum("i,j->ij", pos, inv_freq)
+            ang = ang.to(dtype)
+            ang = torch.cat((ang, ang), dim=-1)
+            self._freq_cache[key] = (ang.cos().to(dtype), ang.sin().to(dtype))
+        return self._freq_cache[key]
+
+    @staticmethod
+    def _rotate(x: torch.Tensor) -> torch.Tensor:
+        d = x.shape[-1]
+        x1, x2 = x[..., : d // 2], x[..., d // 2:]
+        return torch.cat((-x2, x1), dim=-1)
+
+    def _apply_1d(self, tokens, positions, cos_c, sin_c):
+        cos = F.embedding(positions, cos_c)[:, None, :, :]
+        sin = F.embedding(positions, sin_c)[:, None, :, :]
+        return (tokens * cos) + (self._rotate(tokens) * sin)
+
+    def forward(self, tokens: torch.Tensor, positions: torch.Tensor) -> torch.Tensor:
+        feature_dim = tokens.size(-1) // 2
+        max_pos = int(positions.max()) + 1
+        cos_c, sin_c = self._components(feature_dim, max_pos, tokens.device, tokens.dtype)
+        v, h = tokens.chunk(2, dim=-1)
+        v = self._apply_1d(v, positions[..., 0], cos_c, sin_c)
+        h = self._apply_1d(h, positions[..., 1], cos_c, sin_c)
+        return torch.cat((v, h), dim=-1)
+
+
+class Dino2Encoder(torch.nn.Module):
+    def __init__(self, dim, num_heads, layer_norm_eps, num_layers, dtype, device, operations, use_swiglu_ffn,
+                 qknorm_start: int = -1):
+        super().__init__()
+        self.layer = torch.nn.ModuleList([
+            Dino2Block(
+                dim, num_heads, layer_norm_eps, dtype, device, operations,
+                use_swiglu_ffn=use_swiglu_ffn,
+                qk_norm=(qknorm_start != -1 and i >= qknorm_start),
+            )
+            for i in range(num_layers)
+        ])

    def forward(self, x, intermediate_output=None):
+        # Backward-compat path used by ``ClipVisionModel`` (no DA3 extensions).
        optimized_attention = optimized_attention_for_device(x.device, False, small_input=True)

        if intermediate_output is not None:
@@ -122,16 +229,27 @@ class Dino2PatchEmbeddings(torch.nn.Module):


 class Dino2Embeddings(torch.nn.Module):
-    def __init__(self, dim, dtype, device, operations):
+    def __init__(self, dim, dtype, device, operations,
+                 patch_size: int = 14, image_size: int = 518,
+                 use_mask_token: bool = True,
+                 num_camera_tokens: int = 0):
        super().__init__()
-        patch_size = 14
-        image_size = 518
        self.patch_size = patch_size
+        self.image_size = image_size

        self.patch_embeddings = Dino2PatchEmbeddings(dim, patch_size=patch_size, image_size=image_size, dtype=dtype, device=device, operations=operations)
        self.position_embeddings = torch.nn.Parameter(torch.empty(1, (image_size // patch_size) ** 2 + 1, dim, dtype=dtype, device=device))
        self.cls_token = torch.nn.Parameter(torch.empty(1, 1, dim, dtype=dtype, device=device)) # mask_token is a pre-training param, kept only so strict loading accepts the key.
-        self.mask_token = torch.nn.Parameter(torch.empty(1, dim, dtype=dtype, device=device))
+        if use_mask_token:
+            self.mask_token = torch.nn.Parameter(torch.empty(1, dim, dtype=dtype, device=device))
+        else:
+            self.mask_token = None
+        if num_camera_tokens > 0:
+            # DA3 stores (ref_token, src_token) pairs that get injected at the
+            # alt-attn boundary; see ``Dinov2Model._inject_camera_token``.
+            self.camera_token = torch.nn.Parameter(torch.empty(1, num_camera_tokens, dim, dtype=dtype, device=device))
+        else:
+            self.camera_token = None

    def interpolate_pos_encoding(self, x, h_pixels, w_pixels):
        pos_embed = comfy.model_management.cast_to_device(self.position_embeddings, x.device, torch.float32)
@@ -140,12 +258,22 @@ class Dino2Embeddings(torch.nn.Module):
        patch_pos = pos_embed[:, 1:]
        N = patch_pos.shape[1]
        M = int(N ** 0.5)
+        assert N == M * M, f"DINOv2 position grid must be square, got N={N} patches (sqrt={M})"
        h0 = h_pixels // self.patch_size
        w0 = w_pixels // self.patch_size
-        scale_factor = ((h0 + 0.1) / M, (w0 + 0.1) / M)  # +0.1 matches upstream DINOv2's FP-rounding workaround so the interpolate output size lands on (h0, w0).
+        # +0.1 matches upstream DINOv2's FP-rounding workaround so the interpolate output size lands on (h0, w0).
+        # scale_factor is (height_scale, width_scale) -- height MUST come first;
+        # swapping these only happens to work for square inputs and breaks
+        # non-square paths like DA3-Small / DA3-Base multi-view.
+        scale_factor = ((h0 + 0.1) / M, (w0 + 0.1) / M)

        patch_pos = patch_pos.reshape(1, M, M, -1).permute(0, 3, 1, 2)
        patch_pos = torch.nn.functional.interpolate(patch_pos, scale_factor=scale_factor, mode="bicubic", antialias=False)
+        assert (h0, w0) == patch_pos.shape[-2:], (
+            f"Interpolated pos-embed grid {tuple(patch_pos.shape[-2:])} does not match "
+            f"target patch grid ({h0}, {w0}) for input {h_pixels}x{w_pixels} (patch_size={self.patch_size}); "
+            f"check scale_factor axis order and +0.1 rounding workaround"
+        )
        patch_pos = patch_pos.permute(0, 2, 3, 1).flatten(1, 2)
        return torch.cat((class_pos, patch_pos), dim=1).to(x.dtype)

@@ -168,12 +296,51 @@ class Dinov2Model(torch.nn.Module):
        heads = config_dict["num_attention_heads"]
        layer_norm_eps = config_dict["layer_norm_eps"]
        use_swiglu_ffn = config_dict["use_swiglu_ffn"]
+        patch_size = config_dict.get("patch_size", 14)
+        image_size = config_dict.get("image_size", 518)
+        use_mask_token = config_dict.get("use_mask_token", True)

-        self.embeddings = Dino2Embeddings(dim, dtype, device, operations)
-        self.encoder = Dino2Encoder(dim, heads, layer_norm_eps, num_layers, dtype, device, operations, use_swiglu_ffn = use_swiglu_ffn)
+        # DA3 extensions (all default to disabled).
+        self.alt_start = config_dict.get("alt_start", -1)
+        self.qknorm_start = config_dict.get("qknorm_start", -1)
+        self.rope_start = config_dict.get("rope_start", -1)
+        self.cat_token = config_dict.get("cat_token", False)
+        rope_freq = config_dict.get("rope_freq", 100.0)
+
+        self.embed_dim = dim
+        self.patch_size = patch_size
+        self.num_register_tokens = 0
+        self.patch_start_idx = 1
+
+        if self.rope_start != -1 and rope_freq > 0:
+            self.rope = RotaryPositionEmbedding2D(frequency=rope_freq)
+            self._position_getter = _PositionGetter()
+        else:
+            self.rope = None
+            self._position_getter = None
+
+        # camera_token shape: (1, 2, dim) -> (ref_token, src_token).
+        num_cam_tokens = 2 if self.alt_start != -1 else 0
+
+        self.embeddings = Dino2Embeddings(
+            dim, dtype, device, operations,
+            patch_size=patch_size, image_size=image_size,
+            use_mask_token=use_mask_token, num_camera_tokens=num_cam_tokens,
+        )
+        self.encoder = Dino2Encoder(
+            dim, heads, layer_norm_eps, num_layers, dtype, device, operations,
+            use_swiglu_ffn=use_swiglu_ffn,
+            qknorm_start=self.qknorm_start,
+        )
        self.layernorm = operations.LayerNorm(dim, eps=layer_norm_eps, dtype=dtype, device=device)

    def forward(self, pixel_values, attention_mask=None, intermediate_output=None):
+        if self.alt_start != -1:
+            raise RuntimeError(
+                "Dinov2Model.forward() is the backward-compatible CLIP-vision path and does not "
+                "apply DA3 extensions (RoPE, alternating attention, camera-token injection). "
+                "Use get_intermediate_layers_da3() for Depth Anything 3 models."
+            )
        x = self.embeddings(pixel_values)
        x, i = self.encoder(x, intermediate_output=intermediate_output)
        x = self.layernorm(x)
@@ -181,6 +348,7 @@ class Dinov2Model(torch.nn.Module):
        return x, i, pooled_output, None

    def get_intermediate_layers(self, pixel_values, indices, apply_norm=True):
+        """Single-view multi-layer feature extraction."""
        x = self.embeddings(pixel_values)
        optimized_attention = optimized_attention_for_device(x.device, False, small_input=True)
        n_layers = len(self.encoder.layer)
@@ -197,3 +365,132 @@ class Dinov2Model(torch.nn.Module):
            if i >= max_idx:
                break
        return [cache[i] for i in resolved]
+
+    # ------------------------------------------------------------------
+    # Depth Anything 3 forward
+    # ------------------------------------------------------------------
+    def _prepare_rope_positions(self, B, S, H, W, device):
+        if self.rope is None:
+            return None, None
+        ph, pw = H // self.patch_size, W // self.patch_size
+        pos = self._position_getter(B * S, ph, pw, device=device)
+        # Shift so the cls/cam token at position 0 is reserved for "no diff".
+        pos = pos + 1
+        cls_pos = torch.zeros(B * S, self.patch_start_idx, 2, device=device, dtype=pos.dtype)
+        # Per-view local: real grid positions for patches, 0 for cls token.
+        pos_local = torch.cat([cls_pos, pos], dim=1)
+        # Global (across views): same grid positions; cls token still at 0,
+        # but patches share the same positions in every view.
+        pos_global = torch.cat([cls_pos, torch.zeros_like(pos) + 1], dim=1)
+        return pos_local, pos_global
+
+    def _inject_camera_token(self, x: torch.Tensor, B: int, S: int, cam_token: "torch.Tensor | None") -> torch.Tensor:
+        # x: (B, S, N, C). Replace token at index 0 with the camera token.
+        if cam_token is not None:
+            inj = cam_token
+        else:
+            ct = comfy.model_management.cast_to_device(self.embeddings.camera_token, x.device, x.dtype)
+            ref_token = ct[:, :1].expand(B, -1, -1)
+            src_token = ct[:, 1:].expand(B, max(S - 1, 0), -1)
+            inj = torch.cat([ref_token, src_token], dim=1)
+        x = x.clone()
+        x[:, :, 0] = inj
+        return x
+
+    def get_intermediate_layers_da3(self, pixel_values, out_layers, cam_token=None, ref_view_strategy="saddle_balanced", export_feat_layers=None):
+        """Multi-view multi-layer feature extraction used by Depth Anything 3."""
+        if pixel_values.ndim == 4:
+            pixel_values = pixel_values.unsqueeze(1)
+        assert pixel_values.ndim == 5 and pixel_values.shape[2] == 3, \
+            f"expected (B,3,H,W) or (B,S,3,H,W); got {tuple(pixel_values.shape)}"
+        B, S, _, H, W = pixel_values.shape
+
+        # Patch + cls + (interpolated) pos embed for each view.
+        x = pixel_values.reshape(B * S, 3, H, W)
+        x = self.embeddings(x)                          # (B*S, 1+N, C)
+        x = x.reshape(B, S, x.shape[-2], x.shape[-1])    # (B, S, 1+N, C)
+
+        pos_local, pos_global = self._prepare_rope_positions(B, S, H, W, x.device)
+        # optimized_attention is only used by blocks without QK-norm/RoPE
+        # (vanilla DINOv2 path); enabling-aware blocks fall through to SDPA.
+        optimized_attention = optimized_attention_for_device(x.device, False, small_input=True)
+
+        out_set = set(out_layers)
+        export_set = set(export_feat_layers) if export_feat_layers else set()
+        outputs: list[torch.Tensor] = []
+        aux_outputs: list[torch.Tensor] = []
+        local_x = x
+        b_idx = None
+
+
+        for i, blk in enumerate(self.encoder.layer):
+            apply_rope = self.rope is not None and i >= self.rope_start
+            block_rope = self.rope if apply_rope else None
+            l_pos = pos_local if apply_rope else None
+            g_pos = pos_global if apply_rope else None
+
+            # Reference-view selection threshold: matches the upstream constant
+            # THRESH_FOR_REF_SELECTION = 3. Skipped when a user-supplied
+            # cam_token is provided (camera info already pins the geometry).
+            if (self.alt_start != -1 and i == self.alt_start - 1 and S >= THRESH_FOR_REF_SELECTION and cam_token is None):
+                b_idx = select_reference_view(x, strategy=ref_view_strategy)
+                x = reorder_by_reference(x, b_idx)
+                local_x = reorder_by_reference(local_x, b_idx)
+
+            if self.alt_start != -1 and i == self.alt_start:
+                x = self._inject_camera_token(x, B, S, cam_token)
+
+            if self.alt_start != -1 and i >= self.alt_start and (i % 2 == 1):
+                # Global attention across views: flatten S into the seq dim.
+                t = x.reshape(B, S * x.shape[-2], x.shape[-1])
+                p = g_pos.reshape(B, S * g_pos.shape[-2], g_pos.shape[-1]) if g_pos is not None else None
+                t = blk(t, optimized_attention=optimized_attention, pos=p, rope=block_rope)
+                x = t.reshape(B, S, x.shape[-2], x.shape[-1])
+            else:
+                # Per-view local attention.
+                t = x.reshape(B * S, x.shape[-2], x.shape[-1])
+                p = l_pos.reshape(B * S, l_pos.shape[-2], l_pos.shape[-1]) if l_pos is not None else None
+                t = blk(t, optimized_attention=optimized_attention, pos=p, rope=block_rope)
+                x = t.reshape(B, S, x.shape[-2], x.shape[-1])
+                local_x = x
+
+            if i in out_set:
+                if self.cat_token:
+                    out_x = torch.cat([local_x, x], dim=-1)
+                else:
+                    out_x = x
+                # Restore original view order on the way out so heads see views
+                # in the user's expected order.
+                if b_idx is not None and self.alt_start != -1:
+                    out_x = restore_original_order(out_x, b_idx)
+                outputs.append(out_x)
+
+            if i in export_set:
+                aux = x
+                if b_idx is not None and self.alt_start != -1:
+                    aux = restore_original_order(aux, b_idx)
+                aux_outputs.append(aux)
+
+        # Apply final norm. When cat_token is set, only the right half
+        # ("global" features) is normalised; the left half is left as-is to
+        # match the upstream DA3 head signature.
+        normed: list[torch.Tensor] = []
+        cls_tokens: list[torch.Tensor] = []
+        for out_x in outputs:
+            cls_tokens.append(out_x[:, :, 0])
+            if out_x.shape[-1] == self.embed_dim:
+                normed.append(self.layernorm(out_x))
+            elif out_x.shape[-1] == self.embed_dim * 2:
+                left = out_x[..., :self.embed_dim]
+                right = self.layernorm(out_x[..., self.embed_dim:])
+                normed.append(torch.cat([left, right], dim=-1))
+            else:
+                raise ValueError(f"Unexpected token width: {out_x.shape[-1]}")
+
+        # Drop cls/cam token from the patch sequence.
+        normed = [o[..., 1 + self.num_register_tokens:, :] for o in normed]
+
+        # Final layernorm + drop cls token from auxiliary features too.
+        aux_normed = [self.layernorm(o)[..., 1 + self.num_register_tokens:, :]
+                      for o in aux_outputs]
+        return list(zip(normed, cls_tokens)), aux_normed
@@ -0,0 +1,259 @@
+import math
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+import comfy.ops
+from comfy.ldm.modules.attention import optimized_attention_for_device
+from comfy.image_encoders.dino2 import LayerScale as DINOv3ViTLayerScale
+
+
+# DINOv3 ViT-H/16+ (SwiGLU)
+DINOV3_VITH_CONFIG = {
+    "model_type": "dinov3",
+    "num_hidden_layers": 32,
+    "hidden_size": 1280,
+    "num_attention_heads": 20,
+    "num_register_tokens": 4,
+    "intermediate_size": 5120,
+    "layer_norm_eps": 1e-5,
+    "num_channels": 3,
+    "patch_size": 16,
+    "rope_theta": 100.0,
+    "use_gated_mlp": True,
+    "gated_mlp_act": "silu",
+    "image_size": 1024,
+    "image_mean": [0.485, 0.456, 0.406],
+    "image_std": [0.229, 0.224, 0.225],
+}
+
+
+class DINOv3ViTMLP(nn.Module):
+    def __init__(self, hidden_size, intermediate_size, mlp_bias, device, dtype, operations):
+        super().__init__()
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.up_proj = operations.Linear(self.hidden_size, self.intermediate_size, bias=mlp_bias, device=device, dtype=dtype)
+        self.down_proj = operations.Linear(self.intermediate_size, self.hidden_size, bias=mlp_bias, device=device, dtype=dtype)
+        self.act_fn = torch.nn.GELU()
+
+    def forward(self, x):
+        return self.down_proj(self.act_fn(self.up_proj(x)))
+
+
+def rotate_half(x):
+    x1 = x[..., : x.shape[-1] // 2]
+    x2 = x[..., x.shape[-1] // 2 :]
+    return torch.cat((-x2, x1), dim=-1)
+
+
+def apply_rotary_pos_emb(q, k, cos, sin, **kwargs):
+    num_tokens = q.shape[-2]
+    num_patches = sin.shape[-2]
+    num_prefix_tokens = num_tokens - num_patches
+
+    q_prefix_tokens, q_patches = q.split((num_prefix_tokens, num_patches), dim=-2)
+    k_prefix_tokens, k_patches = k.split((num_prefix_tokens, num_patches), dim=-2)
+
+    q_patches = (q_patches * cos) + (rotate_half(q_patches) * sin)
+    k_patches = (k_patches * cos) + (rotate_half(k_patches) * sin)
+
+    q = torch.cat((q_prefix_tokens, q_patches), dim=-2)
+    k = torch.cat((k_prefix_tokens, k_patches), dim=-2)
+
+    return q, k
+
+
+class DINOv3ViTAttention(nn.Module):
+    def __init__(self, hidden_size, num_attention_heads, device, dtype, operations):
+        super().__init__()
+        self.embed_dim = hidden_size
+        self.num_heads = num_attention_heads
+        self.head_dim = self.embed_dim // self.num_heads
+
+        self.k_proj = operations.Linear(self.embed_dim, self.embed_dim, bias=False, device=device, dtype=dtype)  # key_bias = False
+        self.v_proj = operations.Linear(self.embed_dim, self.embed_dim, bias=True, device=device, dtype=dtype)
+        self.q_proj = operations.Linear(self.embed_dim, self.embed_dim, bias=True, device=device, dtype=dtype)
+        self.o_proj = operations.Linear(self.embed_dim, self.embed_dim, bias=True, device=device, dtype=dtype)
+
+    def forward(self, hidden_states, attention_mask=None, position_embeddings=None, **kwargs):
+        batch_size, patches, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states)
+        value_states = self.v_proj(hidden_states)
+
+        query_states = query_states.view(batch_size, patches, self.num_heads, self.head_dim).transpose(1, 2)
+        key_states = key_states.view(batch_size, patches, self.num_heads, self.head_dim).transpose(1, 2)
+        value_states = value_states.view(batch_size, patches, self.num_heads, self.head_dim).transpose(1, 2)
+
+        if position_embeddings is not None:
+            cos, sin = position_embeddings
+            query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
+
+        attn = optimized_attention_for_device(query_states.device, mask=False)
+        attn_output = attn(
+            query_states, key_states, value_states, self.num_heads, attention_mask,
+            skip_reshape=True, skip_output_reshape=True, low_precision_attention=False,
+        )
+
+        attn_output = attn_output.transpose(1, 2)
+        attn_output = attn_output.reshape(batch_size, patches, -1).contiguous()
+        attn_output = self.o_proj(attn_output)
+        return attn_output
+
+
+class DINOv3ViTGatedMLP(nn.Module):
+    def __init__(self, hidden_size, intermediate_size, mlp_bias, device, dtype, operations, act="silu"):
+        super().__init__()
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.gate_proj = operations.Linear(self.hidden_size, self.intermediate_size, bias=mlp_bias, device=device, dtype=dtype)
+        self.up_proj = operations.Linear(self.hidden_size, self.intermediate_size, bias=mlp_bias, device=device, dtype=dtype)
+        self.down_proj = operations.Linear(self.intermediate_size, self.hidden_size, bias=mlp_bias, device=device, dtype=dtype)
+        self.act_fn = torch.nn.SiLU() if act == "silu" else torch.nn.GELU()
+
+    def forward(self, x):
+        return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
+
+
+def get_patches_center_coordinates(num_patches_h, num_patches_w, dtype, device):
+    coords_h = torch.arange(0.5, num_patches_h, dtype=dtype, device=device)
+    coords_w = torch.arange(0.5, num_patches_w, dtype=dtype, device=device)
+    coords_h = coords_h / num_patches_h
+    coords_w = coords_w / num_patches_w
+    coords = torch.stack(torch.meshgrid(coords_h, coords_w, indexing="ij"), dim=-1)
+    coords = coords.flatten(0, 1)
+    coords = 2.0 * coords - 1.0
+    return coords
+
+
+class DINOv3ViTRopePositionEmbedding(nn.Module):
+    inv_freq: torch.Tensor
+
+    def __init__(self, rope_theta, hidden_size, num_attention_heads, patch_size, device, dtype):
+        super().__init__()
+        self.base = rope_theta
+        self.head_dim = hidden_size // num_attention_heads
+        self.patch_size = patch_size
+
+        inv_freq = 1 / self.base ** torch.arange(0, 1, 4 / self.head_dim, dtype=torch.float32, device=device)
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+
+    def forward(self, pixel_values):
+        _, _, height, width = pixel_values.shape
+        num_patches_h = height // self.patch_size
+        num_patches_w = width // self.patch_size
+
+        patch_coords = get_patches_center_coordinates(num_patches_h, num_patches_w, dtype=torch.float32, device=pixel_values.device)
+        self.inv_freq = self.inv_freq.to(pixel_values.device)
+        angles = 2 * math.pi * patch_coords[:, :, None] * self.inv_freq[None, None, :]
+        angles = angles.flatten(1, 2)
+        angles = angles.tile(2)
+        cos = torch.cos(angles).to(dtype=pixel_values.dtype)
+        sin = torch.sin(angles).to(dtype=pixel_values.dtype)
+        return cos, sin
+
+
+class DINOv3ViTEmbeddings(nn.Module):
+    def __init__(self, hidden_size, num_register_tokens, num_channels, patch_size, dtype, device, operations):
+        super().__init__()
+        self.cls_token = nn.Parameter(torch.empty(1, 1, hidden_size, device=device, dtype=dtype))
+        self.mask_token = nn.Parameter(torch.empty(1, 1, hidden_size, device=device, dtype=dtype))
+        self.register_tokens = nn.Parameter(torch.empty(1, num_register_tokens, hidden_size, device=device, dtype=dtype))
+        self.patch_embeddings = operations.Conv2d(
+            num_channels, hidden_size, kernel_size=patch_size, stride=patch_size, device=device, dtype=dtype
+        )
+
+    def forward(self, pixel_values, bool_masked_pos=None):
+        batch_size = pixel_values.shape[0]
+
+        patch_embeddings = self.patch_embeddings(pixel_values)
+        patch_embeddings = patch_embeddings.flatten(2).transpose(1, 2)
+
+        if bool_masked_pos is not None:
+            mask_token = comfy.ops.cast_to_input(self.mask_token, patch_embeddings)
+            patch_embeddings = torch.where(bool_masked_pos.unsqueeze(-1), mask_token, patch_embeddings)
+
+        cls_token = comfy.ops.cast_to_input(self.cls_token.expand(batch_size, -1, -1), patch_embeddings)
+        register_tokens = comfy.ops.cast_to_input(self.register_tokens.expand(batch_size, -1, -1), patch_embeddings)
+        embeddings = torch.cat([cls_token, register_tokens, patch_embeddings], dim=1)
+        return embeddings
+
+
+class DINOv3ViTLayer(nn.Module):
+    def __init__(self, hidden_size, layer_norm_eps, use_gated_mlp, mlp_bias, intermediate_size,
+                 num_attention_heads, device, dtype, operations, gated_mlp_act="silu"):
+        super().__init__()
+        self.norm1 = operations.LayerNorm(hidden_size, eps=layer_norm_eps, device=device, dtype=dtype)
+        self.attention = DINOv3ViTAttention(hidden_size, num_attention_heads, device=device, dtype=dtype, operations=operations)
+        self.layer_scale1 = DINOv3ViTLayerScale(hidden_size, device=device, dtype=dtype, operations=None)
+
+        self.norm2 = operations.LayerNorm(hidden_size, eps=layer_norm_eps, device=device, dtype=dtype)
+        if use_gated_mlp:
+            self.mlp = DINOv3ViTGatedMLP(hidden_size, intermediate_size, mlp_bias, device=device, dtype=dtype, operations=operations, act=gated_mlp_act)
+        else:
+            self.mlp = DINOv3ViTMLP(hidden_size, intermediate_size=intermediate_size, mlp_bias=mlp_bias, device=device, dtype=dtype, operations=operations)
+        self.layer_scale2 = DINOv3ViTLayerScale(hidden_size, device=device, dtype=dtype, operations=None)
+
+    def forward(self, hidden_states, attention_mask=None, position_embeddings=None):
+        residual = hidden_states
+        hidden_states = self.norm1(hidden_states)
+        hidden_states = self.attention(hidden_states, attention_mask=attention_mask, position_embeddings=position_embeddings)
+        hidden_states = self.layer_scale1(hidden_states)
+        hidden_states = hidden_states + residual
+
+        residual = hidden_states
+        hidden_states = self.norm2(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = self.layer_scale2(hidden_states)
+        hidden_states = hidden_states + residual
+        return hidden_states
+
+
+class DINOv3ViTModel(nn.Module):
+    def __init__(self, config, dtype, device, operations):
+        super().__init__()
+        num_hidden_layers = config["num_hidden_layers"]
+        hidden_size = config["hidden_size"]
+        num_attention_heads = config["num_attention_heads"]
+        num_register_tokens = config["num_register_tokens"]
+        intermediate_size = config["intermediate_size"]
+        layer_norm_eps = config["layer_norm_eps"]
+        num_channels = config["num_channels"]
+        patch_size = config["patch_size"]
+        rope_theta = config["rope_theta"]
+        use_gated_mlp = config.get("use_gated_mlp", False)
+        gated_mlp_act = config.get("gated_mlp_act", "silu")
+
+        self.embeddings = DINOv3ViTEmbeddings(
+            hidden_size, num_register_tokens, num_channels=num_channels, patch_size=patch_size,
+            dtype=dtype, device=device, operations=operations
+        )
+        self.rope_embeddings = DINOv3ViTRopePositionEmbedding(
+            rope_theta, hidden_size, num_attention_heads, patch_size=patch_size, dtype=dtype, device=device
+        )
+        self.layer = nn.ModuleList([
+            DINOv3ViTLayer(hidden_size, layer_norm_eps, use_gated_mlp=use_gated_mlp, mlp_bias=True,
+                           intermediate_size=intermediate_size, num_attention_heads=num_attention_heads,
+                           dtype=dtype, device=device, operations=operations, gated_mlp_act=gated_mlp_act)
+            for _ in range(num_hidden_layers)])
+        self.norm = operations.LayerNorm(hidden_size, eps=layer_norm_eps, dtype=dtype, device=device)
+
+    def get_input_embeddings(self):
+        return self.embeddings.patch_embeddings
+
+    def forward(self, pixel_values, bool_masked_pos=None, **kwargs):
+        hidden_states = self.embeddings(pixel_values, bool_masked_pos=bool_masked_pos)
+        position_embeddings = self.rope_embeddings(pixel_values)
+
+        for layer_module in self.layer:
+            hidden_states = layer_module(hidden_states, position_embeddings=position_embeddings)
+
+        if kwargs.get("skip_norm_elementwise", False):
+            sequence_output = F.layer_norm(hidden_states, hidden_states.shape[-1:])
+        else:
+            norm = self.norm.to(hidden_states.device)
+            sequence_output = norm(hidden_states)
+        pooled_output = sequence_output[:, 0, :]
+        return sequence_output, None, pooled_output, None
@@ -239,6 +239,16 @@ class Flux2(LatentFormat):
    def process_out(self, latent):
        return latent

+class TripoSplat(LatentFormat):
+    # Sequence latent (B, 8192, 16) the camera token rides alongside as a second nested latent
+    latent_channels = 16
+
+    def process_in(self, latent):
+        return latent
+
+    def process_out(self, latent):
+        return latent
+
 class Mochi(LatentFormat):
    latent_channels = 12
    latent_dimensions = 3
@@ -799,13 +809,15 @@ class ZImagePixelSpace(ChromaRadiance):
    """
    pass

-
 class HiDreamO1Pixel(ChromaRadiance):
    """Pixel-space latent format for HiDream-O1.
    No VAE — model patches/unpatches raw RGB internally with patch_size=32.
    """
    pass

+class PixelDiTPixel(ChromaRadiance):
+    pass
+
 class CogVideoX(LatentFormat):
    """Latent format for CogVideoX-2b (THUDM/CogVideoX-2b).

@@ -217,10 +217,7 @@ class AceStepAttention(nn.Module):
                cos, sin = position_embeddings
                query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)

-        n_rep = self.num_heads // self.num_kv_heads
-        if n_rep > 1:
-            key_states = key_states.repeat_interleave(n_rep, dim=1)
-            value_states = value_states.repeat_interleave(n_rep, dim=1)
+        gqa_kwargs = {"enable_gqa": True} if self.num_heads != self.num_kv_heads else {}

        attn_bias = None
        if self.sliding_window is not None and not self.is_cross_attention:
@@ -244,7 +241,7 @@ class AceStepAttention(nn.Module):
            else:
                attn_bias = window_bias

-        attn_output = optimized_attention(query_states, key_states, value_states, self.num_heads, attn_bias, skip_reshape=True, low_precision_attention=False)
+        attn_output = optimized_attention(query_states, key_states, value_states, self.num_heads, attn_bias, skip_reshape=True, low_precision_attention=False, **gqa_kwargs)
        attn_output = self.o_proj(attn_output)

        return attn_output
@@ -425,19 +425,16 @@ class Attention(nn.Module):
        if n == 1 and causal:
            causal = False

-        if h != kv_h:
-            # Repeat interleave kv_heads to match q_heads
-            heads_per_kv_head = h // kv_h
-            k, v = map(lambda t: t.repeat_interleave(heads_per_kv_head, dim = 1), (k, v))
+        gqa_kwargs = {"enable_gqa": True} if h != kv_h else {}

        if self.differential:
            q, q_diff = q.unbind(dim=1)
            k, k_diff = k.unbind(dim=1)
-            out      = optimized_attention(q,      k,      v, h, skip_reshape=True, transformer_options=transformer_options)
-            out_diff = optimized_attention(q_diff, k_diff, v, h, skip_reshape=True, transformer_options=transformer_options)
+            out      = optimized_attention(q,      k,      v, h, skip_reshape=True, low_precision_attention=False, transformer_options=transformer_options, **gqa_kwargs)
+            out_diff = optimized_attention(q_diff, k_diff, v, h, skip_reshape=True, low_precision_attention=False, transformer_options=transformer_options, **gqa_kwargs)
            out = out - out_diff
        else:
-            out = optimized_attention(q, k, v, h, skip_reshape=True, transformer_options=transformer_options)
+            out = optimized_attention(q, k, v, h, skip_reshape=True, low_precision_attention=False, transformer_options=transformer_options, **gqa_kwargs)

        out = self.to_out(out)

@@ -138,11 +138,11 @@ class Attention(nn.Module):
                k_diff = _apply_rotary_pos_emb(k_diff.float(), freqs).to(k_dtype)

        if self.differential:
-            out = (optimized_attention(q, k, v, h, mask=mask, skip_reshape=True)
-                   - optimized_attention(q_diff, k_diff, v, h, mask=mask, skip_reshape=True))
+            out = (optimized_attention(q, k, v, h, mask=mask, skip_reshape=True, low_precision_attention=False)
+                   - optimized_attention(q_diff, k_diff, v, h, mask=mask, skip_reshape=True, low_precision_attention=False))
            del q, k, v, q_diff, k_diff
        else:
-            out = optimized_attention(q, k, v, h, mask=mask, skip_reshape=True)
+            out = optimized_attention(q, k, v, h, mask=mask, skip_reshape=True, low_precision_attention=False)
            del q, k, v

        return self.to_out(out)
@@ -0,0 +1,318 @@
+# Boogu-Image-0.1 transformer
+# Architecture is an OmniGen2 derivative (see comfy/ldm/omnigen/omnigen2.py) with an
+# added dual-stream ("double_stream") stage before the single-stream layers, conditioned
+# by a Qwen3-VL multimodal LLM. Reuses the OmniGen2/Lumina building blocks and the Flux
+# RoPE core, the only new component is the double-stream block + the hybrid forward order.
+
+from typing import Optional, Tuple
+
+import torch
+import torch.nn as nn
+from einops import rearrange
+
+import comfy.ldm.common_dit
+import comfy.ldm.omnigen.omnigen2
+from comfy.ldm.modules.attention import optimized_attention_masked
+from comfy.ldm.omnigen.omnigen2 import (
+    OmniGen2RotaryPosEmbed,
+    Lumina2CombinedTimestepCaptionEmbedding,
+    LuminaRMSNormZero,
+    LuminaLayerNormContinuous,
+    LuminaFeedForward,
+    Attention,
+    OmniGen2TransformerBlock,
+    apply_rotary_emb,
+)
+
+class BooguDoubleStreamProcessor(nn.Module):
+    # Joint attention over [instruct ; img] with separate per-stream q/k/v and output projections.
+    def __init__(self, dim, head_dim, heads, kv_heads, dtype=None, device=None, operations=None):
+        super().__init__()
+        query_dim = head_dim * heads
+        kv_dim = head_dim * kv_heads
+
+        self.img_to_q = operations.Linear(query_dim, query_dim, bias=False, dtype=dtype, device=device)
+        self.img_to_k = operations.Linear(query_dim, kv_dim, bias=False, dtype=dtype, device=device)
+        self.img_to_v = operations.Linear(query_dim, kv_dim, bias=False, dtype=dtype, device=device)
+
+        self.instruct_to_q = operations.Linear(query_dim, query_dim, bias=False, dtype=dtype, device=device)
+        self.instruct_to_k = operations.Linear(query_dim, kv_dim, bias=False, dtype=dtype, device=device)
+        self.instruct_to_v = operations.Linear(query_dim, kv_dim, bias=False, dtype=dtype, device=device)
+
+        self.instruct_out = operations.Linear(query_dim, query_dim, bias=False, dtype=dtype, device=device)
+        self.img_out = operations.Linear(query_dim, query_dim, bias=False, dtype=dtype, device=device)
+
+    def forward(self, attn, img_hidden_states, instruct_hidden_states, rotary_emb, attention_mask=None, transformer_options={}):
+        batch_size = img_hidden_states.shape[0]
+        L_instruct = instruct_hidden_states.shape[1]
+
+        img_q = self.img_to_q(img_hidden_states)
+        img_k = self.img_to_k(img_hidden_states)
+        img_v = self.img_to_v(img_hidden_states)
+
+        instruct_q = self.instruct_to_q(instruct_hidden_states)
+        instruct_k = self.instruct_to_k(instruct_hidden_states)
+        instruct_v = self.instruct_to_v(instruct_hidden_states)
+
+        # Concatenate instruction first, then image (matches reference processor order).
+        query = torch.cat([instruct_q, img_q], dim=1)
+        key = torch.cat([instruct_k, img_k], dim=1)
+        value = torch.cat([instruct_v, img_v], dim=1)
+
+        query = query.view(batch_size, -1, attn.heads, attn.dim_head)
+        key = key.view(batch_size, -1, attn.kv_heads, attn.dim_head)
+        value = value.view(batch_size, -1, attn.kv_heads, attn.dim_head)
+
+        query = attn.norm_q(query)
+        key = attn.norm_k(key)
+
+        if rotary_emb is not None:
+            query = apply_rotary_emb(query, rotary_emb)
+            key = apply_rotary_emb(key, rotary_emb)
+
+        query = query.transpose(1, 2)
+        key = key.transpose(1, 2)
+        value = value.transpose(1, 2)
+
+        gqa_kwargs = {"enable_gqa": True} if attn.kv_heads < attn.heads else {}
+        hidden_states = optimized_attention_masked(query, key, value, attn.heads, attention_mask, skip_reshape=True, transformer_options=transformer_options, **gqa_kwargs)
+
+        # Split back to instruction/image, apply per-stream output projections, recombine.
+        instruct_hidden_states = self.instruct_out(hidden_states[:, :L_instruct])
+        img_hidden_states = self.img_out(hidden_states[:, L_instruct:])
+        hidden_states = torch.cat([instruct_hidden_states, img_hidden_states], dim=1)
+
+        hidden_states = attn.to_out[0](hidden_states)
+        return hidden_states
+
+
+class BooguJointAttention(nn.Module):
+    # Holds the shared q/k RMSNorm + final output projection
+    def __init__(self, dim, head_dim, heads, kv_heads, eps=1e-5, dtype=None, device=None, operations=None):
+        super().__init__()
+        self.heads = heads
+        self.kv_heads = kv_heads
+        self.dim_head = head_dim
+        self.scale = head_dim ** -0.5
+
+        self.norm_q = operations.RMSNorm(head_dim, eps=eps, dtype=dtype, device=device)
+        self.norm_k = operations.RMSNorm(head_dim, eps=eps, dtype=dtype, device=device)
+        self.to_out = nn.Sequential(
+            operations.Linear(heads * head_dim, dim, bias=False, dtype=dtype, device=device),
+            nn.Dropout(0.0),
+        )
+        self.processor = BooguDoubleStreamProcessor(dim, head_dim, heads, kv_heads, dtype=dtype, device=device, operations=operations)
+
+    def forward(self, img_hidden_states, instruct_hidden_states, rotary_emb, attention_mask=None, transformer_options={}):
+        return self.processor(self, img_hidden_states, instruct_hidden_states, rotary_emb, attention_mask, transformer_options=transformer_options)
+
+
+class BooguDoubleStreamBlock(nn.Module):
+    # Dual-stream block: joint attention over [instruct ; img] + image self-attention, each stream with its own modulation/MLP.
+    def __init__(self, dim, num_attention_heads, num_kv_heads, multiple_of, ffn_dim_multiplier, norm_eps, dtype=None, device=None, operations=None):
+        super().__init__()
+        head_dim = dim // num_attention_heads
+
+        self.img_instruct_attn = BooguJointAttention(dim, head_dim, num_attention_heads, num_kv_heads, eps=1e-5, dtype=dtype, device=device, operations=operations)
+        self.img_self_attn = Attention(
+            query_dim=dim, dim_head=head_dim, heads=num_attention_heads, kv_heads=num_kv_heads,
+            eps=1e-5, bias=False, dtype=dtype, device=device, operations=operations,
+        )
+
+        self.img_feed_forward = LuminaFeedForward(dim=dim, inner_dim=4 * dim, multiple_of=multiple_of, dtype=dtype, device=device, operations=operations)
+        self.instruct_feed_forward = LuminaFeedForward(dim=dim, inner_dim=4 * dim, multiple_of=multiple_of, dtype=dtype, device=device, operations=operations)
+
+        self.img_norm1 = LuminaRMSNormZero(embedding_dim=dim, norm_eps=norm_eps, dtype=dtype, device=device, operations=operations)
+        self.img_norm2 = LuminaRMSNormZero(embedding_dim=dim, norm_eps=norm_eps, dtype=dtype, device=device, operations=operations)
+        self.img_norm3 = LuminaRMSNormZero(embedding_dim=dim, norm_eps=norm_eps, dtype=dtype, device=device, operations=operations)
+        self.instruct_norm1 = LuminaRMSNormZero(embedding_dim=dim, norm_eps=norm_eps, dtype=dtype, device=device, operations=operations)
+        self.instruct_norm2 = LuminaRMSNormZero(embedding_dim=dim, norm_eps=norm_eps, dtype=dtype, device=device, operations=operations)
+
+        self.img_attn_norm = operations.RMSNorm(dim, eps=norm_eps, dtype=dtype, device=device)
+        self.img_self_attn_norm = operations.RMSNorm(dim, eps=norm_eps, dtype=dtype, device=device)
+        self.img_ffn_norm1 = operations.RMSNorm(dim, eps=norm_eps, dtype=dtype, device=device)
+        self.img_ffn_norm2 = operations.RMSNorm(dim, eps=norm_eps, dtype=dtype, device=device)
+
+        self.instruct_attn_norm = operations.RMSNorm(dim, eps=norm_eps, dtype=dtype, device=device)
+        self.instruct_ffn_norm1 = operations.RMSNorm(dim, eps=norm_eps, dtype=dtype, device=device)
+        self.instruct_ffn_norm2 = operations.RMSNorm(dim, eps=norm_eps, dtype=dtype, device=device)
+
+    def forward(self, img_hidden_states, instruct_hidden_states, joint_rotary_emb, img_rotary_emb, temb, joint_attention_mask=None, img_attention_mask=None, transformer_options={}):
+        L_instruct = instruct_hidden_states.shape[1]
+
+        img_norm1_out, img_gate_msa, img_scale_mlp, img_gate_mlp = self.img_norm1(img_hidden_states, temb)
+        img_norm2_out, img_shift_mlp, _, _ = self.img_norm2(img_hidden_states, temb)
+        img_norm3_out, img_gate_self, _, _ = self.img_norm3(img_hidden_states, temb)
+
+        instruct_norm1_out, instruct_gate_msa, instruct_scale_mlp, instruct_gate_mlp = self.instruct_norm1(instruct_hidden_states, temb)
+        instruct_norm2_out, instruct_shift_mlp, _, _ = self.instruct_norm2(instruct_hidden_states, temb)
+
+        joint_attn_out = self.img_instruct_attn(img_norm1_out, instruct_norm1_out, joint_rotary_emb, joint_attention_mask, transformer_options=transformer_options)
+        instruct_attn_out = joint_attn_out[:, :L_instruct]
+        img_attn_out = joint_attn_out[:, L_instruct:]
+
+        img_self_attn_out = self.img_self_attn(img_norm3_out, img_norm3_out, img_attention_mask, img_rotary_emb, transformer_options=transformer_options)
+
+        img_hidden_states = img_hidden_states + img_gate_msa.unsqueeze(1).tanh() * self.img_attn_norm(img_attn_out)
+        img_hidden_states = img_hidden_states + img_gate_self.unsqueeze(1).tanh() * self.img_self_attn_norm(img_self_attn_out)
+        img_mlp_input = (1 + img_scale_mlp.unsqueeze(1)) * img_norm2_out + img_shift_mlp.unsqueeze(1)
+        img_mlp_out = self.img_feed_forward(self.img_ffn_norm1(img_mlp_input))
+        img_hidden_states = img_hidden_states + img_gate_mlp.unsqueeze(1).tanh() * self.img_ffn_norm2(img_mlp_out)
+
+        instruct_hidden_states = instruct_hidden_states + instruct_gate_msa.unsqueeze(1).tanh() * self.instruct_attn_norm(instruct_attn_out)
+        instruct_mlp_input = (1 + instruct_scale_mlp.unsqueeze(1)) * instruct_norm2_out + instruct_shift_mlp.unsqueeze(1)
+        instruct_mlp_out = self.instruct_feed_forward(self.instruct_ffn_norm1(instruct_mlp_input))
+        instruct_hidden_states = instruct_hidden_states + instruct_gate_mlp.unsqueeze(1).tanh() * self.instruct_ffn_norm2(instruct_mlp_out)
+
+        return img_hidden_states, instruct_hidden_states
+
+
+class BooguTransformer2DModel(nn.Module):
+    def __init__(
+        self,
+        patch_size: int = 2,
+        in_channels: int = 16,
+        out_channels: Optional[int] = None,
+        hidden_size: int = 3360,
+        num_layers: int = 32,
+        num_double_stream_layers: int = 8,
+        num_refiner_layers: int = 2,
+        num_attention_heads: int = 28,
+        num_kv_heads: int = 7,
+        multiple_of: int = 256,
+        ffn_dim_multiplier: Optional[float] = None,
+        norm_eps: float = 1e-5,
+        axes_dim_rope: Tuple[int, int, int] = (40, 40, 40),
+        axes_lens: Tuple[int, int, int] = (2048, 1664, 1664),
+        instruction_feat_dim: int = 4096,
+        timestep_scale: float = 1000.0,
+        image_model=None,
+        device=None, dtype=None, operations=None,
+    ):
+        super().__init__()
+
+        self.patch_size = patch_size
+        self.out_channels = out_channels or in_channels
+        self.hidden_size = hidden_size
+        self.dtype = dtype
+
+        self.rope_embedder = OmniGen2RotaryPosEmbed(
+            theta=10000,
+            axes_dim=axes_dim_rope,
+            axes_lens=axes_lens,
+            patch_size=patch_size,
+        )
+
+        self.x_embedder = operations.Linear(patch_size * patch_size * in_channels, hidden_size, dtype=dtype, device=device)
+        self.ref_image_patch_embedder = operations.Linear(patch_size * patch_size * in_channels, hidden_size, dtype=dtype, device=device)
+
+        self.time_caption_embed = Lumina2CombinedTimestepCaptionEmbedding(
+            hidden_size=hidden_size,
+            text_feat_dim=instruction_feat_dim,
+            norm_eps=norm_eps,
+            timestep_scale=timestep_scale, dtype=dtype, device=device, operations=operations
+        )
+
+        self.noise_refiner = nn.ModuleList([
+            OmniGen2TransformerBlock(hidden_size, num_attention_heads, num_kv_heads, multiple_of, ffn_dim_multiplier, norm_eps, modulation=True, dtype=dtype, device=device, operations=operations)
+            for _ in range(num_refiner_layers)
+        ])
+
+        self.ref_image_refiner = nn.ModuleList([
+            OmniGen2TransformerBlock(hidden_size, num_attention_heads, num_kv_heads, multiple_of, ffn_dim_multiplier, norm_eps, modulation=True, dtype=dtype, device=device, operations=operations)
+            for _ in range(num_refiner_layers)
+        ])
+
+        self.context_refiner = nn.ModuleList([
+            OmniGen2TransformerBlock(hidden_size, num_attention_heads, num_kv_heads, multiple_of, ffn_dim_multiplier, norm_eps, modulation=False, dtype=dtype, device=device, operations=operations)
+            for _ in range(num_refiner_layers)
+        ])
+
+        self.double_stream_layers = nn.ModuleList([
+            BooguDoubleStreamBlock(hidden_size, num_attention_heads, num_kv_heads, multiple_of, ffn_dim_multiplier, norm_eps, dtype=dtype, device=device, operations=operations)
+            for _ in range(num_double_stream_layers)
+        ])
+
+        self.single_stream_layers = nn.ModuleList([
+            OmniGen2TransformerBlock(hidden_size, num_attention_heads, num_kv_heads, multiple_of, ffn_dim_multiplier, norm_eps, modulation=True, dtype=dtype, device=device, operations=operations)
+            for _ in range(num_layers)
+        ])
+
+        self.norm_out = LuminaLayerNormContinuous(
+            embedding_dim=hidden_size,
+            conditioning_embedding_dim=min(hidden_size, 1024),
+            elementwise_affine=False,
+            eps=1e-6,
+            out_dim=patch_size * patch_size * self.out_channels, dtype=dtype, device=device, operations=operations
+        )
+
+        self.image_index_embedding = nn.Parameter(torch.empty(5, hidden_size, device=device, dtype=dtype))
+
+    # Patchify/refine helpers are identical to OmniGen2; reuse via bound methods.
+    flat_and_pad_to_seq = comfy.ldm.omnigen.omnigen2.OmniGen2Transformer2DModel.flat_and_pad_to_seq
+    img_patch_embed_and_refine = comfy.ldm.omnigen.omnigen2.OmniGen2Transformer2DModel.img_patch_embed_and_refine
+
+    def forward(self, x, timesteps, context, num_tokens, ref_latents=None, attention_mask=None, transformer_options={}, **kwargs):
+        B, C, H, W = x.shape
+        hidden_states = comfy.ldm.common_dit.pad_to_patch_size(x, (self.patch_size, self.patch_size))
+        _, _, H_padded, W_padded = hidden_states.shape
+        timestep = 1.0 - timesteps
+        text_hidden_states = context
+        text_attention_mask = attention_mask
+        ref_image_hidden_states = ref_latents
+        device = hidden_states.device
+
+        temb, text_hidden_states = self.time_caption_embed(timestep, text_hidden_states, hidden_states[0].dtype)
+
+        (
+            hidden_states, ref_image_hidden_states,
+            img_mask, ref_img_mask,
+            l_effective_ref_img_len, l_effective_img_len,
+            ref_img_sizes, img_sizes,
+        ) = self.flat_and_pad_to_seq(hidden_states, ref_image_hidden_states)
+
+        (
+            context_rotary_emb, ref_img_rotary_emb, noise_rotary_emb,
+            rotary_emb, encoder_seq_lengths, seq_lengths,
+        ) = self.rope_embedder(
+            hidden_states.shape[0], text_hidden_states.shape[1], [num_tokens] * text_hidden_states.shape[0],
+            l_effective_ref_img_len, l_effective_img_len,
+            ref_img_sizes, img_sizes, device,
+        )
+
+        for layer in self.context_refiner:
+            text_hidden_states = layer(text_hidden_states, text_attention_mask, context_rotary_emb, transformer_options=transformer_options)
+
+        img_len = hidden_states.shape[1]
+        combined_img_hidden_states = self.img_patch_embed_and_refine(
+            hidden_states, ref_image_hidden_states,
+            img_mask, ref_img_mask,
+            noise_rotary_emb, ref_img_rotary_emb,
+            l_effective_ref_img_len, l_effective_img_len,
+            temb,
+            transformer_options=transformer_options,
+        )
+
+        # Double-stream stage: the image self-attention only sees the [ref ; noise] tokens,
+        # which sit after the instruction tokens in the joint rope.
+        L_instruct = text_hidden_states.shape[1]
+        combined_img_rotary_emb = rotary_emb[:, L_instruct:]
+        for layer in self.double_stream_layers:
+            combined_img_hidden_states, text_hidden_states = layer(
+                combined_img_hidden_states, text_hidden_states,
+                rotary_emb, combined_img_rotary_emb, temb,
+                joint_attention_mask=None, img_attention_mask=None,
+                transformer_options=transformer_options,
+            )
+
+        hidden_states = torch.cat([text_hidden_states, combined_img_hidden_states], dim=1)
+
+        for layer in self.single_stream_layers:
+            hidden_states = layer(hidden_states, None, rotary_emb, temb, transformer_options=transformer_options)
+
+        hidden_states = self.norm_out(hidden_states, temb)
+
+        p = self.patch_size
+        output = rearrange(hidden_states[:, -img_len:], 'b (h w) (p1 p2 c) -> b c (h p1) (w p2)', h=H_padded // p, w=W_padded // p, p1=p, p2=p)[:, :, :H, :W]
+
+        return -output
@@ -38,6 +38,8 @@ class ChromaRadianceParams(ChromaParams):
    # None means use the same dtype as the model.
    nerf_embedder_dtype: Optional[torch.dtype]
    use_x0: bool
+    # Use sequential txt_ids instead of zeros
+    use_sequential_txt_ids: bool

 class ChromaRadiance(Chroma):
    """
@@ -162,6 +164,9 @@ class ChromaRadiance(Chroma):
        if params.use_x0:
            self.register_buffer("__x0__", torch.tensor([]))

+        if params.use_sequential_txt_ids:
+            self.register_buffer("__sequential__", torch.tensor([]))
+
    @property
    def _nerf_final_layer(self) -> nn.Module:
        if self.params.nerf_final_head_type == "linear":
@@ -313,6 +318,9 @@ class ChromaRadiance(Chroma):
        img_ids[:, :, 2] = img_ids[:, :, 2] + torch.linspace(0, w_len - 1, steps=w_len, device=x.device, dtype=x.dtype).unsqueeze(0)
        img_ids = repeat(img_ids, "h w c -> b (h w) c", b=bs)
        txt_ids = torch.zeros((bs, context.shape[1], 3), device=x.device, dtype=x.dtype)
+        # Radiance after 2026-05-22 uses sequential txt_ids instead of zeros
+        if params.use_sequential_txt_ids:
+            txt_ids[:, :, 0] = torch.arange(context.shape[1], device=x.device, dtype=x.dtype).unsqueeze(0).expand(bs, -1)

        img_out = self.forward_orig(
            img,
@@ -0,0 +1,25 @@
+"""Colormap utilities for depth and geometry visualisation."""
+
+from __future__ import annotations
+
+import torch
+
+
+def turbo(x: torch.Tensor) -> torch.Tensor:
+    """Anton Mikhailov polynomial approximation of the Turbo colormap.
+
+    Args:
+        x: Float tensor with values in [0, 1].
+
+    Returns:
+        RGB tensor of the same shape as ``x`` with a trailing size-3 dimension.
+    """
+    x = x.clamp(0.0, 1.0)
+    x2 = x * x
+    x3 = x2 * x
+    x4 = x2 * x2
+    x5 = x4 * x
+    r = 0.13572138 + 4.61539260*x - 42.66032258*x2 + 132.13108234*x3 - 152.94239396*x4 + 59.28637943*x5
+    g = 0.09140261 + 2.19418839*x + 4.84296658*x2 - 14.18503333*x3 +   4.27729857*x4 +  2.82956604*x5
+    b = 0.10667330 + 12.64194608*x - 60.58204836*x2 + 110.36276771*x3 - 89.90310912*x4 + 27.34824973*x5
+    return torch.stack([r, g, b], dim=-1).clamp(0.0, 1.0)
@@ -14,15 +14,7 @@ from torchvision import transforms
 import comfy.patcher_extension
 from comfy.ldm.modules.attention import optimized_attention
 import comfy.ldm.common_dit
-
-def apply_rotary_pos_emb(
-    t: torch.Tensor,
-    freqs: torch.Tensor,
-) -> torch.Tensor:
-    t_ = t.reshape(*t.shape[:-1], 2, -1).movedim(-2, -1).unsqueeze(-2).float()
-    t_out = freqs[..., 0] * t_[..., 0] + freqs[..., 1] * t_[..., 1]
-    t_out = t_out.movedim(-1, -2).reshape(*t.shape).type_as(t)
-    return t_out
+import comfy.quant_ops


 # ---------------------- Feed Forward Network -----------------------
@@ -173,8 +165,7 @@ class Attention(nn.Module):
            k = self.k_norm(k)
            v = self.v_norm(v)
            if self.is_selfattn and rope_emb is not None:  # only apply to self-attention!
-                q = apply_rotary_pos_emb(q, rope_emb)
-                k = apply_rotary_pos_emb(k, rope_emb)
+                q, k = comfy.quant_ops.ck.apply_rope_split_half(q, k, rope_emb)
            return q, k, v

        q, k, v = apply_norm_and_rotary_pos_emb(q, k, v, rope_emb)
@@ -524,7 +515,7 @@ class Block(nn.Module):
            h=H,
            w=W,
        )
-        x_B_T_H_W_D = x_B_T_H_W_D + gate_self_attn_B_T_1_1_D.to(residual_dtype) * result_B_T_H_W_D.to(residual_dtype)
+        x_B_T_H_W_D = torch.addcmul(x_B_T_H_W_D, gate_self_attn_B_T_1_1_D.to(residual_dtype), result_B_T_H_W_D.to(residual_dtype))

        def _x_fn(
            _x_B_T_H_W_D: torch.Tensor,
@@ -557,7 +548,7 @@ class Block(nn.Module):
            shift_cross_attn_B_T_1_1_D,
            transformer_options=transformer_options,
        )
-        x_B_T_H_W_D = result_B_T_H_W_D.to(residual_dtype) * gate_cross_attn_B_T_1_1_D.to(residual_dtype) + x_B_T_H_W_D
+        x_B_T_H_W_D = torch.addcmul(x_B_T_H_W_D, gate_cross_attn_B_T_1_1_D.to(residual_dtype), result_B_T_H_W_D.to(residual_dtype))

        normalized_x_B_T_H_W_D = _fn(
            x_B_T_H_W_D,
@@ -566,7 +557,7 @@ class Block(nn.Module):
            shift_mlp_B_T_1_1_D,
        )
        result_B_T_H_W_D = self.mlp(normalized_x_B_T_H_W_D.to(compute_dtype))
-        x_B_T_H_W_D = x_B_T_H_W_D + gate_mlp_B_T_1_1_D.to(residual_dtype) * result_B_T_H_W_D.to(residual_dtype)
+        x_B_T_H_W_D = torch.addcmul(x_B_T_H_W_D, gate_mlp_B_T_1_1_D.to(residual_dtype), result_B_T_H_W_D.to(residual_dtype))
        return x_B_T_H_W_D


@@ -0,0 +1,177 @@
+"""Camera-token encoder and decoder for Depth Anything 3."""
+
+from __future__ import annotations
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from comfy.ldm.modules.attention import optimized_attention_for_device
+from .transform import affine_inverse, extri_intri_to_pose_encoding
+
+
+# -----------------------------------------------------------------------
+# Building blocks (mirror depth_anything_3.model.utils.{attention,block})
+# -----------------------------------------------------------------------
+
+
+class _Mlp(nn.Module):
+    """Standard 2-layer MLP with GELU. Matches upstream ``utils.attention.Mlp``."""
+
+    def __init__(self, in_features, hidden_features=None, out_features=None, *, device=None, dtype=None, operations=None):
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+        self.fc1 = operations.Linear(in_features, hidden_features, bias=True, device=device, dtype=dtype)
+        self.fc2 = operations.Linear(hidden_features, out_features, bias=True, device=device, dtype=dtype)
+
+    def forward(self, x):
+        return self.fc2(F.gelu(self.fc1(x)))
+
+
+class _LayerScale(nn.Module):
+    """Per-channel learnable scaling. Matches upstream LayerScale."""
+
+    def __init__(self, dim, *, device=None, dtype=None):
+        super().__init__()
+        self.gamma = nn.Parameter(torch.empty(dim, device=device, dtype=dtype))
+
+    def forward(self, x):
+        return x * self.gamma.to(dtype=x.dtype, device=x.device)
+
+
+class _Attention(nn.Module):
+    """ Self-attention with fused QKV projection. Mirrors upstream utils.attention.Attention;
+    Layout matches the HF safetensors (attn.qkv.{weight,bias} and attn.proj.{weight,bias})."""
+
+    def __init__(self, dim, num_heads, *, device=None, dtype=None, operations=None):
+        super().__init__()
+        assert dim % num_heads == 0
+        self.num_heads = num_heads
+        self.head_dim = dim // num_heads
+        self.qkv = operations.Linear(dim, dim * 3, bias=True, device=device, dtype=dtype)
+        self.proj = operations.Linear(dim, dim, bias=True, device=device, dtype=dtype)
+
+    def forward(self, x):
+        B, N, C = x.shape
+        qkv = self.qkv(x).reshape(B, N, 3, C)
+        q, k, v = qkv.unbind(2)                      # each (B, N, C)
+        attn_fn = optimized_attention_for_device(x.device, small_input=True)
+        out = attn_fn(q, k, v, heads=self.num_heads)
+        return self.proj(out)
+
+
+class _Block(nn.Module):
+    """Pre-norm transformer block with LayerScale. Used by :class:CameraEnc. Layout follows upstream utils.block.Block."""
+
+    def __init__(self, dim, num_heads, mlp_ratio=4, init_values=0.01, *, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.norm1 = operations.LayerNorm(dim, device=device, dtype=dtype)
+        self.attn = _Attention(dim, num_heads, device=device, dtype=dtype, operations=operations)
+        self.ls1 = _LayerScale(dim, device=device, dtype=dtype) if init_values else nn.Identity()
+        self.norm2 = operations.LayerNorm(dim, device=device, dtype=dtype)
+        self.mlp = _Mlp(in_features=dim, hidden_features=int(dim * mlp_ratio), device=device, dtype=dtype, operations=operations)
+        self.ls2 = _LayerScale(dim, device=device, dtype=dtype) if init_values else nn.Identity()
+
+    def forward(self, x):
+        x = x + self.ls1(self.attn(self.norm1(x)))
+        x = x + self.ls2(self.mlp(self.norm2(x)))
+        return x
+
+
+class CameraEnc(nn.Module):
+    """Encode per-view (extrinsics, intrinsics) into a camera token.
+
+    Maps a 9-D pose-encoding vector through a small MLP up to the backbone's
+    ``embed_dim``, then runs ``trunk_depth`` transformer blocks. The output
+    has shape ``(B, S, embed_dim)`` and is injected at block ``alt_start``
+    of the DINOv2 backbone in place of the cls token.
+
+    Parameters mirror the upstream ``cam_enc.py`` so HF weights load directly.
+    """
+
+    def __init__(
+        self,
+        dim_out: int = 1024,
+        dim_in: int = 9,
+        trunk_depth: int = 4,
+        target_dim: int = 9,
+        num_heads: int = 16,
+        mlp_ratio: int = 4,
+        init_values: float = 0.01,
+        *,
+        device=None, dtype=None, operations=None,
+        **_kwargs,
+    ):
+        super().__init__()
+        self.target_dim = target_dim
+        self.trunk_depth = trunk_depth
+        self.trunk = nn.Sequential(*[
+            _Block(dim_out, num_heads=num_heads, mlp_ratio=mlp_ratio,
+                   init_values=init_values,
+                   device=device, dtype=dtype, operations=operations)
+            for _ in range(trunk_depth)
+        ])
+        self.token_norm = operations.LayerNorm(dim_out, device=device, dtype=dtype)
+        self.trunk_norm = operations.LayerNorm(dim_out, device=device, dtype=dtype)
+        self.pose_branch = _Mlp(
+            in_features=dim_in,
+            hidden_features=dim_out // 2,
+            out_features=dim_out,
+            device=device, dtype=dtype, operations=operations,
+        )
+
+    def forward(self, extrinsics: torch.Tensor, intrinsics: torch.Tensor,
+                image_size_hw) -> torch.Tensor:
+        """Encode camera parameters into ``(B, S, dim_out)`` tokens."""
+        c2ws = affine_inverse(extrinsics)
+        pose_encoding = extri_intri_to_pose_encoding(c2ws, intrinsics, image_size_hw)
+        tokens = self.pose_branch(pose_encoding.to(self.pose_branch.fc1.weight.dtype))
+        tokens = self.token_norm(tokens)
+        tokens = self.trunk(tokens)
+        tokens = self.trunk_norm(tokens)
+        return tokens
+
+
+class CameraDec(nn.Module):
+    """Decode the final cam token into a 9-D pose encoding.
+
+    Output layout: ``[T(3), quat_xyzw(4), fov_h, fov_w]``. The translation is
+    always predicted by the network; the quaternion and FoV can either be
+    predicted or supplied via ``camera_encoding`` (used at training time
+    when GT cameras are available -- not exercised at inference here).
+
+    Parameters mirror the upstream ``cam_dec.py`` so HF weights load directly.
+    """
+
+    def __init__(self, dim_in: int = 1536,
+                 *, device=None, dtype=None, operations=None, **_kwargs):
+        super().__init__()
+        d = dim_in
+        self.backbone = nn.Sequential(
+            operations.Linear(d, d, device=device, dtype=dtype),
+            nn.ReLU(),
+            operations.Linear(d, d, device=device, dtype=dtype),
+            nn.ReLU(),
+        )
+        self.fc_t = operations.Linear(d, 3, device=device, dtype=dtype)
+        self.fc_qvec = operations.Linear(d, 4, device=device, dtype=dtype)
+        self.fc_fov = nn.Sequential(
+            operations.Linear(d, 2, device=device, dtype=dtype),
+            nn.ReLU(),
+        )
+
+    def forward(self, feat: torch.Tensor,
+                camera_encoding: "torch.Tensor | None" = None) -> torch.Tensor:
+        """Decode ``(B, N, dim_in)`` cam tokens into ``(B, N, 9)`` pose enc."""
+        B, N = feat.shape[:2]
+        feat = feat.reshape(B * N, -1)
+        feat = self.backbone(feat)
+        out_t = self.fc_t(feat.float()).reshape(B, N, 3)
+        if camera_encoding is None:
+            out_qvec = self.fc_qvec(feat.float()).reshape(B, N, 4)
+            out_fov = self.fc_fov(feat.float()).reshape(B, N, 2)
+        else:
+            out_qvec = camera_encoding[..., 3:7]
+            out_fov = camera_encoding[..., -2:]
+        return torch.cat([out_t, out_qvec, out_fov], dim=-1)
@@ -0,0 +1,489 @@
+"""DPT / DualDPT heads for Depth Anything 3."""
+
+from __future__ import annotations
+
+from typing import List, Optional, Sequence, Tuple
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+class Permute(nn.Module):
+    def __init__(self, dims: Tuple[int, ...]):
+        super().__init__()
+        self.dims = dims
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return x.permute(*self.dims)
+
+
+def _custom_interpolate(
+    x: torch.Tensor,
+    size: Optional[Tuple[int, int]] = None,
+    scale_factor: Optional[float] = None,
+    mode: str = "bilinear",
+    align_corners: bool = True,
+) -> torch.Tensor:
+    if size is None:
+        assert scale_factor is not None
+        size = (int(x.shape[-2] * scale_factor), int(x.shape[-1] * scale_factor))
+    INT_MAX = 1610612736
+    total = size[0] * size[1] * x.shape[0] * x.shape[1]
+    if total > INT_MAX:
+        chunks = torch.chunk(x, chunks=(total // INT_MAX) + 1, dim=0)
+        outs = [F.interpolate(c, size=size, mode=mode, align_corners=align_corners) for c in chunks]
+        return torch.cat(outs, dim=0).contiguous()
+    return F.interpolate(x, size=size, mode=mode, align_corners=align_corners)
+
+
+def _create_uv_grid(width: int, height: int, aspect_ratio: float, dtype, device) -> torch.Tensor:
+    """Normalised UV grid spanning (-x_span, -y_span)..(x_span, y_span)."""
+    diag_factor = (aspect_ratio ** 2 + 1.0) ** 0.5
+    span_x = aspect_ratio / diag_factor
+    span_y = 1.0 / diag_factor
+    left_x = -span_x * (width - 1) / width
+    right_x = span_x * (width - 1) / width
+    top_y = -span_y * (height - 1) / height
+    bottom_y = span_y * (height - 1) / height
+    x_coords = torch.linspace(left_x, right_x, steps=width, dtype=dtype, device=device)
+    y_coords = torch.linspace(top_y, bottom_y, steps=height, dtype=dtype, device=device)
+    uu, vv = torch.meshgrid(x_coords, y_coords, indexing="xy")
+    return torch.stack((uu, vv), dim=-1)  # (H, W, 2)
+
+
+def _make_sincos_pos_embed(embed_dim: int, pos: torch.Tensor, omega_0: float = 100.0) -> torch.Tensor:
+    omega = torch.arange(embed_dim // 2, dtype=torch.float32, device=pos.device)
+    omega = 1.0 / omega_0 ** (omega / (embed_dim / 2.0))
+    pos = pos.reshape(-1)
+    out = torch.einsum("m,d->md", pos, omega)
+    return torch.cat([out.sin(), out.cos()], dim=1).float()
+
+
+def _position_grid_to_embed(pos_grid: torch.Tensor, embed_dim: int, omega_0: float = 100.0) -> torch.Tensor:
+    H, W, _ = pos_grid.shape
+    pos_flat = pos_grid.reshape(-1, 2)
+    emb_x = _make_sincos_pos_embed(embed_dim // 2, pos_flat[:, 0], omega_0=omega_0)
+    emb_y = _make_sincos_pos_embed(embed_dim // 2, pos_flat[:, 1], omega_0=omega_0)
+    emb = torch.cat([emb_x, emb_y], dim=-1)
+    return emb.view(H, W, embed_dim)
+
+
+def _add_pos_embed(x: torch.Tensor, W: int, H: int, ratio: float = 0.1) -> torch.Tensor:
+    """Stateless UV positional embedding added to a feature map (B, C, h, w)."""
+    pw, ph = x.shape[-1], x.shape[-2]
+    pe = _create_uv_grid(pw, ph, aspect_ratio=W / H, dtype=x.dtype, device=x.device)
+    pe = _position_grid_to_embed(pe, x.shape[1]) * ratio
+    pe = pe.permute(2, 0, 1)[None].expand(x.shape[0], -1, -1, -1).to(dtype=x.dtype)
+    return x + pe
+
+
+def _apply_activation(x: torch.Tensor, activation: str) -> torch.Tensor:
+    act = (activation or "linear").lower()
+    if act == "exp":
+        return torch.exp(x)
+    if act == "expp1":
+        return torch.exp(x) + 1
+    if act == "expm1":
+        return torch.expm1(x)
+    if act == "relu":
+        return torch.relu(x)
+    if act == "sigmoid":
+        return torch.sigmoid(x)
+    if act == "softplus":
+        return F.softplus(x)
+    if act == "tanh":
+        return torch.tanh(x)
+    return x
+
+
+# -----------------------------------------------------------------------------
+# Fusion building blocks
+# -----------------------------------------------------------------------------
+
+
+class ResidualConvUnit(nn.Module):
+    def __init__(self, features: int, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.conv1 = operations.Conv2d(features, features, 3, 1, 1, bias=True, device=device, dtype=dtype)
+        self.conv2 = operations.Conv2d(features, features, 3, 1, 1, bias=True, device=device, dtype=dtype)
+        self.activation = nn.ReLU(inplace=False)
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        out = self.activation(x)
+        out = self.conv1(out)
+        out = self.activation(out)
+        out = self.conv2(out)
+        return out + x
+
+
+class FeatureFusionBlock(nn.Module):
+    def __init__(self, features: int, has_residual: bool = True, align_corners: bool = True, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.align_corners = align_corners
+        self.has_residual = has_residual
+        if has_residual:
+            self.resConfUnit1 = ResidualConvUnit(features, device=device, dtype=dtype, operations=operations)
+        else:
+            self.resConfUnit1 = None
+        self.resConfUnit2 = ResidualConvUnit(features, device=device, dtype=dtype, operations=operations)
+        self.out_conv = operations.Conv2d(features, features, 1, 1, 0, bias=True, device=device, dtype=dtype)
+
+    def forward(self, *xs: torch.Tensor, size: Optional[Tuple[int, int]] = None) -> torch.Tensor:
+        y = xs[0]
+        if self.has_residual and len(xs) > 1 and self.resConfUnit1 is not None:
+            y = y + self.resConfUnit1(xs[1])
+        y = self.resConfUnit2(y)
+        if size is None:
+            up_kwargs = {"scale_factor": 2.0}
+        else:
+            up_kwargs = {"size": size}
+        y = _custom_interpolate(y, **up_kwargs, mode="bilinear", align_corners=self.align_corners)
+        y = self.out_conv(y)
+        return y
+
+
+class _Scratch(nn.Module):
+    """Container that mirrors upstream ``scratch`` attribute layout."""
+
+
+def _make_scratch(in_shape: List[int], out_shape: int, device=None, dtype=None, operations=None) -> _Scratch:
+    scratch = _Scratch()
+    scratch.layer1_rn = operations.Conv2d(in_shape[0], out_shape, 3, 1, 1, bias=False, device=device, dtype=dtype)
+    scratch.layer2_rn = operations.Conv2d(in_shape[1], out_shape, 3, 1, 1, bias=False, device=device, dtype=dtype)
+    scratch.layer3_rn = operations.Conv2d(in_shape[2], out_shape, 3, 1, 1, bias=False, device=device, dtype=dtype)
+    scratch.layer4_rn = operations.Conv2d(in_shape[3], out_shape, 3, 1, 1, bias=False, device=device, dtype=dtype)
+    return scratch
+
+
+def _make_fusion_block(features: int, has_residual: bool = True, device=None, dtype=None, operations=None) -> FeatureFusionBlock:
+    return FeatureFusionBlock(features, has_residual=has_residual, align_corners=True, device=device, dtype=dtype, operations=operations)
+
+
+# -----------------------------------------------------------------------------
+# DPT (single head + optional sky head) -- used by DA3Mono/Metric
+# -----------------------------------------------------------------------------
+
+
+class DPT(nn.Module):
+    """Single-head DPT used by DA3Mono-Large and DA3Metric-Large."""
+
+    def __init__(
+        self,
+        dim_in: int,
+        patch_size: int = 14,
+        output_dim: int = 1,
+        activation: str = "exp",
+        conf_activation: str = "expp1",
+        features: int = 256,
+        out_channels: Sequence[int] = (256, 512, 1024, 1024),
+        pos_embed: bool = False,
+        down_ratio: int = 1,
+        head_name: str = "depth",
+        use_sky_head: bool = True,
+        sky_name: str = "sky",
+        sky_activation: str = "relu",
+        norm_type: str = "idt",
+        device=None, dtype=None, operations=None,
+    ):
+        super().__init__()
+        self.patch_size = patch_size
+        self.activation = activation
+        self.conf_activation = conf_activation
+        self.pos_embed = pos_embed
+        self.down_ratio = down_ratio
+        self.head_main = head_name
+        self.sky_name = sky_name
+        self.out_dim = output_dim
+        self.has_conf = output_dim > 1
+        self.use_sky_head = use_sky_head
+        self.sky_activation = sky_activation
+        self.intermediate_layer_idx: Tuple[int, int, int, int] = (0, 1, 2, 3)
+
+        if norm_type == "layer":
+            self.norm = operations.LayerNorm(dim_in, device=device, dtype=dtype)
+        else:
+            self.norm = nn.Identity()
+
+        out_channels = list(out_channels)
+        self.projects = nn.ModuleList([
+            operations.Conv2d(dim_in, oc, kernel_size=1, stride=1, padding=0, device=device, dtype=dtype)
+            for oc in out_channels
+        ])
+        self.resize_layers = nn.ModuleList([
+            operations.ConvTranspose2d(out_channels[0], out_channels[0], kernel_size=4, stride=4, padding=0, device=device, dtype=dtype),
+            operations.ConvTranspose2d(out_channels[1], out_channels[1], kernel_size=2, stride=2, padding=0, device=device, dtype=dtype),
+            nn.Identity(),
+            operations.Conv2d(out_channels[3], out_channels[3], kernel_size=3, stride=2, padding=1, device=device, dtype=dtype),
+        ])
+
+        self.scratch = _make_scratch(out_channels, features, device=device, dtype=dtype, operations=operations)
+        self.scratch.refinenet1 = _make_fusion_block(features, device=device, dtype=dtype, operations=operations)
+        self.scratch.refinenet2 = _make_fusion_block(features, device=device, dtype=dtype, operations=operations)
+        self.scratch.refinenet3 = _make_fusion_block(features, device=device, dtype=dtype, operations=operations)
+        self.scratch.refinenet4 = _make_fusion_block(features, has_residual=False, device=device, dtype=dtype, operations=operations)
+
+        head_features_1 = features
+        head_features_2 = 32
+        self.scratch.output_conv1 = operations.Conv2d(
+            head_features_1, head_features_1 // 2, kernel_size=3, stride=1, padding=1,
+            device=device, dtype=dtype,
+        )
+        self.scratch.output_conv2 = nn.Sequential(
+            operations.Conv2d(head_features_1 // 2, head_features_2, kernel_size=3, stride=1, padding=1, device=device, dtype=dtype),
+            nn.ReLU(inplace=False),
+            operations.Conv2d(head_features_2, output_dim, kernel_size=1, stride=1, padding=0, device=device, dtype=dtype),
+        )
+
+        if self.use_sky_head:
+            self.scratch.sky_output_conv2 = nn.Sequential(
+                operations.Conv2d(head_features_1 // 2, head_features_2, kernel_size=3, stride=1, padding=1, device=device, dtype=dtype),
+                nn.ReLU(inplace=False),
+                operations.Conv2d(head_features_2, 1, kernel_size=1, stride=1, padding=0, device=device, dtype=dtype),
+            )
+
+    def forward(self, feats: List[torch.Tensor], H: int, W: int, patch_start_idx: int = 0, **_kwargs) -> dict:
+        # feats[i][0] is the patch-token tensor with shape (B, S, N_patch, C)
+        B, S, N, C = feats[0][0].shape
+        feats_flat = [feat[0].reshape(B * S, N, C) for feat in feats]
+
+        ph, pw = H // self.patch_size, W // self.patch_size
+        resized = []
+        for stage_idx, take_idx in enumerate(self.intermediate_layer_idx):
+            x = feats_flat[take_idx][:, patch_start_idx:]
+            x = self.norm(x)
+            x = x.permute(0, 2, 1).contiguous().reshape(B * S, C, ph, pw)
+            x = self.projects[stage_idx](x)
+            if self.pos_embed:
+                x = _add_pos_embed(x, W, H)
+            x = self.resize_layers[stage_idx](x)
+            resized.append(x)
+
+        l1_rn = self.scratch.layer1_rn(resized[0])
+        l2_rn = self.scratch.layer2_rn(resized[1])
+        l3_rn = self.scratch.layer3_rn(resized[2])
+        l4_rn = self.scratch.layer4_rn(resized[3])
+
+        out = self.scratch.refinenet4(l4_rn, size=l3_rn.shape[2:])
+        out = self.scratch.refinenet3(out, l3_rn, size=l2_rn.shape[2:])
+        out = self.scratch.refinenet2(out, l2_rn, size=l1_rn.shape[2:])
+        out = self.scratch.refinenet1(out, l1_rn)
+
+        h_out = int(ph * self.patch_size / self.down_ratio)
+        w_out = int(pw * self.patch_size / self.down_ratio)
+
+        fused = self.scratch.output_conv1(out)
+        fused = _custom_interpolate(fused, (h_out, w_out), mode="bilinear", align_corners=True)
+        if self.pos_embed:
+            fused = _add_pos_embed(fused, W, H)
+        feat = fused
+
+        main_logits = self.scratch.output_conv2(feat)
+        outs = {}
+        if self.has_conf:
+            fmap = main_logits.permute(0, 2, 3, 1)
+            pred = _apply_activation(fmap[..., :-1], self.activation)
+            conf = _apply_activation(fmap[..., -1], self.conf_activation)
+            outs[self.head_main] = pred.squeeze(-1).view(B, S, *pred.shape[1:-1])
+            outs[f"{self.head_main}_conf"] = conf.view(B, S, *conf.shape[1:])
+        else:
+            pred = _apply_activation(main_logits, self.activation)
+            outs[self.head_main] = pred.squeeze(1).view(B, S, *pred.shape[2:])
+
+        if self.use_sky_head:
+            sky_logits = self.scratch.sky_output_conv2(feat)
+            if self.sky_activation.lower() == "sigmoid":
+                sky = torch.sigmoid(sky_logits)
+            elif self.sky_activation.lower() == "relu":
+                sky = F.relu(sky_logits)
+            else:
+                sky = sky_logits
+            outs[self.sky_name] = sky.squeeze(1).view(B, S, *sky.shape[2:])
+
+        return outs
+
+
+# -----------------------------------------------------------------------------
+# DualDPT (depth + auxiliary "ray" head) -- used by DA3-Small / DA3-Base
+# -----------------------------------------------------------------------------
+
+
+class DualDPT(nn.Module):
+    """Two-head DPT used by DA3-Small / DA3-Base."""
+
+    def __init__(
+        self,
+        dim_in: int,
+        patch_size: int = 14,
+        output_dim: int = 2,
+        activation: str = "exp",
+        conf_activation: str = "expp1",
+        features: int = 256,
+        out_channels: Sequence[int] = (256, 512, 1024, 1024),
+        pos_embed: bool = True,
+        down_ratio: int = 1,
+        aux_pyramid_levels: int = 4,
+        aux_out1_conv_num: int = 5,
+        head_names: Tuple[str, str] = ("depth", "ray"),
+        device=None, dtype=None, operations=None,
+    ):
+        super().__init__()
+        self.patch_size = patch_size
+        self.activation = activation
+        self.conf_activation = conf_activation
+        self.pos_embed = pos_embed
+        self.down_ratio = down_ratio
+        self.aux_levels = aux_pyramid_levels
+        self.aux_out1_conv_num = aux_out1_conv_num
+        self.head_main, self.head_aux = head_names
+        self.intermediate_layer_idx: Tuple[int, int, int, int] = (0, 1, 2, 3)
+        # Toggle the auxiliary ray branch at runtime. Default off (mono path).
+        # DepthAnything3Net flips this on when running multi-view + ray-pose.
+        self.enable_aux: bool = False
+
+        self.norm = operations.LayerNorm(dim_in, device=device, dtype=dtype)
+        out_channels = list(out_channels)
+        self.projects = nn.ModuleList([
+            operations.Conv2d(dim_in, oc, kernel_size=1, stride=1, padding=0, device=device, dtype=dtype)
+            for oc in out_channels
+        ])
+        self.resize_layers = nn.ModuleList([
+            operations.ConvTranspose2d(out_channels[0], out_channels[0], kernel_size=4, stride=4, padding=0, device=device, dtype=dtype),
+            operations.ConvTranspose2d(out_channels[1], out_channels[1], kernel_size=2, stride=2, padding=0, device=device, dtype=dtype),
+            nn.Identity(),
+            operations.Conv2d(out_channels[3], out_channels[3], kernel_size=3, stride=2, padding=1, device=device, dtype=dtype),
+        ])
+
+        self.scratch = _make_scratch(out_channels, features, device=device, dtype=dtype, operations=operations)
+        # Main fusion chain
+        self.scratch.refinenet1 = _make_fusion_block(features, device=device, dtype=dtype, operations=operations)
+        self.scratch.refinenet2 = _make_fusion_block(features, device=device, dtype=dtype, operations=operations)
+        self.scratch.refinenet3 = _make_fusion_block(features, device=device, dtype=dtype, operations=operations)
+        self.scratch.refinenet4 = _make_fusion_block(features, has_residual=False, device=device, dtype=dtype, operations=operations)
+        # Auxiliary fusion chain (separate copies)
+        self.scratch.refinenet1_aux = _make_fusion_block(features, device=device, dtype=dtype, operations=operations)
+        self.scratch.refinenet2_aux = _make_fusion_block(features, device=device, dtype=dtype, operations=operations)
+        self.scratch.refinenet3_aux = _make_fusion_block(features, device=device, dtype=dtype, operations=operations)
+        self.scratch.refinenet4_aux = _make_fusion_block(features, has_residual=False, device=device, dtype=dtype, operations=operations)
+
+        head_features_1 = features
+        head_features_2 = 32
+
+        # Main head neck + final projection
+        self.scratch.output_conv1 = operations.Conv2d(
+            head_features_1, head_features_1 // 2, kernel_size=3, stride=1, padding=1,
+            device=device, dtype=dtype,
+        )
+        self.scratch.output_conv2 = nn.Sequential(
+            operations.Conv2d(head_features_1 // 2, head_features_2, kernel_size=3, stride=1, padding=1, device=device, dtype=dtype),
+            nn.ReLU(inplace=False),
+            operations.Conv2d(head_features_2, output_dim, kernel_size=1, stride=1, padding=0, device=device, dtype=dtype),
+        )
+
+        # Aux pre-head per level (multi-level pyramid)
+        self.scratch.output_conv1_aux = nn.ModuleList([
+            self._make_aux_out1_block(head_features_1, device=device, dtype=dtype, operations=operations)
+            for _ in range(self.aux_levels)
+        ])
+
+        # Aux final projection per level (includes LayerNorm permute path).
+        ln_seq = [Permute((0, 2, 3, 1)),
+                  operations.LayerNorm(head_features_2, device=device, dtype=dtype),
+                  Permute((0, 3, 1, 2))]
+        self.scratch.output_conv2_aux = nn.ModuleList([
+            nn.Sequential(
+                operations.Conv2d(head_features_1 // 2, head_features_2, kernel_size=3, stride=1, padding=1, device=device, dtype=dtype),
+                *ln_seq,
+                nn.ReLU(inplace=False),
+                operations.Conv2d(head_features_2, 7, kernel_size=1, stride=1, padding=0, device=device, dtype=dtype),
+            )
+            for _ in range(self.aux_levels)
+        ])
+
+    @staticmethod
+    def _make_aux_out1_block(in_ch: int, *, device=None, dtype=None, operations=None) -> nn.Sequential:
+        # aux_out1_conv_num=5 in all Apache-2.0 variants.
+        return nn.Sequential(
+            operations.Conv2d(in_ch, in_ch // 2, 3, 1, 1, device=device, dtype=dtype),
+            operations.Conv2d(in_ch // 2, in_ch, 3, 1, 1, device=device, dtype=dtype),
+            operations.Conv2d(in_ch, in_ch // 2, 3, 1, 1, device=device, dtype=dtype),
+            operations.Conv2d(in_ch // 2, in_ch, 3, 1, 1, device=device, dtype=dtype),
+            operations.Conv2d(in_ch, in_ch // 2, 3, 1, 1, device=device, dtype=dtype),
+        )
+
+    def forward(self, feats: List[torch.Tensor], H: int, W: int, patch_start_idx: int = 0, **_kwargs) -> dict:
+        B, S, N, C = feats[0][0].shape
+        feats_flat = [feat[0].reshape(B * S, N, C) for feat in feats]
+
+        ph, pw = H // self.patch_size, W // self.patch_size
+        resized = []
+        for stage_idx, take_idx in enumerate(self.intermediate_layer_idx):
+            x = feats_flat[take_idx][:, patch_start_idx:]
+            x = self.norm(x)
+            x = x.permute(0, 2, 1).contiguous().reshape(B * S, C, ph, pw)
+            x = self.projects[stage_idx](x)
+            if self.pos_embed:
+                x = _add_pos_embed(x, W, H)
+            x = self.resize_layers[stage_idx](x)
+            resized.append(x)
+
+        l1_rn = self.scratch.layer1_rn(resized[0])
+        l2_rn = self.scratch.layer2_rn(resized[1])
+        l3_rn = self.scratch.layer3_rn(resized[2])
+        l4_rn = self.scratch.layer4_rn(resized[3])
+
+        # Main pyramid (output_conv1 is applied inside the upstream `_fuse`,
+        # before interpolation -- replicate that order here).
+        m = self.scratch.refinenet4(l4_rn, size=l3_rn.shape[2:])
+        if self.enable_aux:
+            a4 = self.scratch.refinenet4_aux(l4_rn, size=l3_rn.shape[2:])
+            aux_pyr = [a4]
+        m = self.scratch.refinenet3(m, l3_rn, size=l2_rn.shape[2:])
+        if self.enable_aux:
+            aux_pyr.append(self.scratch.refinenet3_aux(aux_pyr[-1], l3_rn, size=l2_rn.shape[2:]))
+        m = self.scratch.refinenet2(m, l2_rn, size=l1_rn.shape[2:])
+        if self.enable_aux:
+            aux_pyr.append(self.scratch.refinenet2_aux(aux_pyr[-1], l2_rn, size=l1_rn.shape[2:]))
+        m = self.scratch.refinenet1(m, l1_rn)
+        if self.enable_aux:
+            aux_pyr.append(self.scratch.refinenet1_aux(aux_pyr[-1], l1_rn))
+        m = self.scratch.output_conv1(m)
+
+        h_out = int(ph * self.patch_size / self.down_ratio)
+        w_out = int(pw * self.patch_size / self.down_ratio)
+
+        m = _custom_interpolate(m, (h_out, w_out), mode="bilinear", align_corners=True)
+        if self.pos_embed:
+            m = _add_pos_embed(m, W, H)
+        main_logits = self.scratch.output_conv2(m)
+        fmap = main_logits.permute(0, 2, 3, 1)
+        depth_pred = _apply_activation(fmap[..., :-1], self.activation)
+        depth_conf = _apply_activation(fmap[..., -1], self.conf_activation)
+
+        outs = {
+            self.head_main: depth_pred.squeeze(-1).view(B, S, *depth_pred.shape[1:-1]),
+            f"{self.head_main}_conf": depth_conf.view(B, S, *depth_conf.shape[1:]),
+        }
+
+        if self.enable_aux:
+            # Auxiliary "ray" head (multi-level inside) -- only the last level
+            # is returned. Mirrors upstream ``DualDPT._fuse`` + ``_forward_impl``:
+            # each aux pyramid level goes through ``output_conv1_aux[i]``
+            # (5-layer conv stack that ends at ``features // 2`` channels),
+            # then the last level optionally gets a pos-embed and finally
+            # ``output_conv2_aux[-1]``.
+            aux_processed = [
+                self.scratch.output_conv1_aux[i](a) for i, a in enumerate(aux_pyr)
+            ]
+            last_aux = aux_processed[-1]
+            if self.pos_embed:
+                last_aux = _add_pos_embed(last_aux, W, H)
+            last_aux_logits = self.scratch.output_conv2_aux[-1](last_aux)
+            fmap_last = last_aux_logits.permute(0, 2, 3, 1)
+            # Channels: [ray(6), ray_conf(1)]; ray uses 'linear' activation.
+            aux_pred = fmap_last[..., :-1]
+            aux_conf = _apply_activation(fmap_last[..., -1], self.conf_activation)
+            outs[self.head_aux] = aux_pred.view(B, S, *aux_pred.shape[1:])
+            outs[f"{self.head_aux}_conf"] = aux_conf.view(B, S, *aux_conf.shape[1:])
+
+        return outs
@@ -0,0 +1,236 @@
+from __future__ import annotations
+
+from typing import Dict, Optional, Sequence
+
+import torch
+import torch.nn as nn
+
+from comfy.image_encoders.dino2 import Dinov2Model
+
+from .camera import CameraDec, CameraEnc
+from .dpt import DPT, DualDPT
+from .ray_pose import get_extrinsic_from_camray
+from .transform import affine_inverse, pose_encoding_to_extri_intri
+
+
+_HEAD_REGISTRY = {
+    "dpt": DPT,
+    "dualdpt": DualDPT,
+}
+
+
+# Backbone presets (mirror the upstream DINOv2 ViT variants).
+_BACKBONE_PRESETS = {
+    "vits": dict(hidden_size=384,  num_hidden_layers=12, num_attention_heads=6,  use_swiglu_ffn=False),
+    "vitb": dict(hidden_size=768,  num_hidden_layers=12, num_attention_heads=12, use_swiglu_ffn=False),
+    "vitl": dict(hidden_size=1024, num_hidden_layers=24, num_attention_heads=16, use_swiglu_ffn=False),
+    "vitg": dict(hidden_size=1536, num_hidden_layers=40, num_attention_heads=24, use_swiglu_ffn=True),
+}
+
+
+def _build_backbone_config(
+    backbone_name: str,
+    *,
+    alt_start: int,
+    qknorm_start: int,
+    rope_start: int,
+    cat_token: bool,
+) -> dict:
+    if backbone_name not in _BACKBONE_PRESETS:
+        raise ValueError(f"Unknown DINOv2 backbone variant: {backbone_name!r}")
+    cfg = dict(_BACKBONE_PRESETS[backbone_name])
+    cfg.update(dict(
+        layer_norm_eps=1e-6,
+        patch_size=14,
+        image_size=518,
+        # No mask_token in DA3 weights; omit param to avoid load warnings.
+        use_mask_token=False,
+        alt_start=alt_start,
+        qknorm_start=qknorm_start,
+        rope_start=rope_start,
+        cat_token=cat_token,
+        rope_freq=100.0,
+    ))
+    return cfg
+
+
+class DepthAnything3Net(nn.Module):
+
+    PATCH_SIZE = 14
+
+    def __init__(
+        self,
+        # --- Backbone ---
+        backbone_name: str = "vitl",
+        out_layers: Sequence[int] = (4, 11, 17, 23),
+        alt_start: int = -1,
+        qknorm_start: int = -1,
+        rope_start: int = -1,
+        cat_token: bool = False,
+        # --- Head ---
+        head_type: str = "dpt",  # dpt or dualdpt
+        head_dim_in: int = 1024,
+        head_output_dim: int = 1,  # 1 = depth only, 2 = depth+conf
+        head_features: int = 256,
+        head_out_channels: Sequence[int] = (256, 512, 1024, 1024),
+        head_use_sky_head: bool = True, # ignored by DualDPT
+        head_pos_embed: Optional[bool] = None,  # default: True for DualDPT, False for DPT
+        # --- Camera (multi-view) ---
+        has_cam_enc: bool = False,
+        has_cam_dec: bool = False,
+        cam_dim_out: Optional[int] = None,  # CameraEnc dim_out (defaults to embed_dim)
+        cam_dec_dim_in: Optional[int] = None,  # CameraDec dim_in  (defaults to 2*embed_dim with cat_token)
+        # ComfyUI plumbing
+        device=None, dtype=None, operations=None,
+        **_ignored,
+    ):
+        super().__init__()
+        head_cls = _HEAD_REGISTRY[head_type.lower()]
+        self.head_type = head_type.lower()
+        self.has_sky = (self.head_type == "dpt") and head_use_sky_head
+        self.has_conf = head_output_dim > 1
+        self.out_layers = list(out_layers)
+
+        backbone_cfg = _build_backbone_config(
+            backbone_name,
+            alt_start=alt_start,
+            qknorm_start=qknorm_start,
+            rope_start=rope_start,
+            cat_token=cat_token,
+        )
+        self.backbone = Dinov2Model(backbone_cfg, dtype, device, operations)
+
+        head_kwargs = dict(
+            dim_in=head_dim_in,
+            patch_size=self.PATCH_SIZE,
+            output_dim=head_output_dim,
+            features=head_features,
+            out_channels=tuple(head_out_channels),
+            device=device, dtype=dtype, operations=operations,
+        )
+        if self.head_type == "dpt":
+            head_kwargs.update(
+                use_sky_head=head_use_sky_head,
+                pos_embed=(False if head_pos_embed is None else head_pos_embed),
+            )
+        else:  # dualdpt
+            head_kwargs.update(
+                pos_embed=(True if head_pos_embed is None else head_pos_embed),
+            )
+        self.head = head_cls(**head_kwargs)
+
+        # Built only if checkpoint has weights; cam_enc output dim == embed_dim.
+        embed_dim = backbone_cfg["hidden_size"]
+        if has_cam_enc:
+            self.cam_enc = CameraEnc(
+                dim_out=cam_dim_out if cam_dim_out is not None else embed_dim,
+                num_heads=max(1, embed_dim // 64),
+                device=device, dtype=dtype, operations=operations,
+            )
+        else:
+            self.cam_enc = None
+        if has_cam_dec:
+            default_dim = embed_dim * (2 if cat_token else 1)
+            self.cam_dec = CameraDec(
+                dim_in=cam_dec_dim_in if cam_dec_dim_in is not None else default_dim,
+                device=device, dtype=dtype, operations=operations,
+            )
+        else:
+            self.cam_dec = None
+
+        self.dtype = dtype
+
+    def forward(
+        self,
+        image: torch.Tensor,
+        extrinsics: Optional[torch.Tensor] = None,
+        intrinsics: Optional[torch.Tensor] = None,
+        *,
+        use_ray_pose: bool = False,
+        ref_view_strategy: str = "saddle_balanced",
+        export_feat_layers: Optional[Sequence[int]] = None,
+        **_unused,
+    ) -> Dict[str, torch.Tensor]:
+        """Run depth and optionally pose prediction."""
+        if image.ndim == 4:
+            image = image.unsqueeze(1)  # (B, 1, 3, H, W)
+        assert image.ndim == 5 and image.shape[2] == 3, \
+            f"image must be (B,3,H,W) or (B,S,3,H,W); got {tuple(image.shape)}"
+
+        B, S, _, H, W = image.shape
+        assert H % self.PATCH_SIZE == 0 and W % self.PATCH_SIZE == 0, \
+            f"image H,W must be multiples of {self.PATCH_SIZE}; got {(H, W)}"
+
+        # Camera-token preparation (multi-view path).
+        cam_token = None
+        if extrinsics is not None and intrinsics is not None and self.cam_enc is not None:
+            cam_token = self.cam_enc(extrinsics, intrinsics, (H, W))
+
+        # Toggle aux ray output on/off depending on what the caller asked for.
+        if isinstance(self.head, DualDPT):
+            self.head.enable_aux = bool(use_ray_pose)
+
+        feats, aux_feats = self.backbone.get_intermediate_layers_da3(
+            image, self.out_layers, cam_token=cam_token,
+            ref_view_strategy=ref_view_strategy,
+            export_feat_layers=export_feat_layers,
+        )
+        head_out = self.head(feats, H=H, W=W, patch_start_idx=0)
+
+        # Pose prediction.
+        out: Dict[str, torch.Tensor] = {}
+        if use_ray_pose and "ray" in head_out and "ray_conf" in head_out:
+            ray = head_out["ray"]
+            ray_conf = head_out["ray_conf"]
+            extr_c2w, focal, pp = get_extrinsic_from_camray(
+                ray, ray_conf, ray.shape[-3], ray.shape[-2],
+            )
+            # Match the upstream output: w2c, drop the homogeneous row.
+            extr_w2c = affine_inverse(extr_c2w)[:, :, :3, :]
+            # Build pixel-space intrinsics from the normalised focal/pp output.
+            intr = torch.eye(3, device=ray.device, dtype=ray.dtype)
+            intr = intr[None, None].expand(extr_c2w.shape[0], extr_c2w.shape[1], 3, 3).clone()
+            intr[:, :, 0, 0] = focal[:, :, 0] / 2 * W
+            intr[:, :, 1, 1] = focal[:, :, 1] / 2 * H
+            intr[:, :, 0, 2] = pp[:, :, 0] * W * 0.5
+            intr[:, :, 1, 2] = pp[:, :, 1] * H * 0.5
+            out["extrinsics"] = extr_w2c
+            out["intrinsics"] = intr
+        elif self.cam_dec is not None and S > 1:
+            # Decode the cam-token of the final out_layer into a pose encoding.
+            cam_feat = feats[-1][1]  # (B, S, dim_in_to_cam_dec)
+            pose_enc = self.cam_dec(cam_feat)
+            c2w_3x4, intr = pose_encoding_to_extri_intri(pose_enc, (H, W))
+            # Match the upstream output convention: w2c (world->camera), 3x4.
+            c2w_4x4 = torch.cat([
+                c2w_3x4,
+                torch.tensor([0, 0, 0, 1], device=c2w_3x4.device, dtype=c2w_3x4.dtype)
+                    .view(1, 1, 1, 4).expand(B, S, 1, 4),
+            ], dim=-2)
+            out["extrinsics"] = affine_inverse(c2w_4x4)[:, :, :3, :]
+            out["intrinsics"] = intr
+
+        # Flatten the views axis for per-pixel outputs (depth/conf/sky) so the
+        # per-image consumer keeps its (B*S, H, W) interface.
+        for k, v in head_out.items():
+            if k in ("ray", "ray_conf"):
+                # Keep multi-view shape for downstream pose work.
+                out[k] = v
+            elif v.ndim >= 3 and v.shape[0] == B and v.shape[1] == S:
+                out[k] = v.reshape(B * S, *v.shape[2:])
+            else:
+                out[k] = v
+
+        if export_feat_layers:
+            out["aux_features"] = self._reshape_aux_features(aux_feats, H, W)
+        return out
+
+    def _reshape_aux_features(self, aux_feats, H: int, W: int):
+        """Reshape (B, S, N, C) aux features into (B, S, h_p, w_p, C)."""
+        ph, pw = H // self.PATCH_SIZE, W // self.PATCH_SIZE
+        out = []
+        for f in aux_feats:
+            B, S, N, C = f.shape
+            assert N == ph * pw, f"aux feature seq mismatch: {N} != {ph}*{pw}"
+            out.append(f.reshape(B, S, ph, pw, C))
+        return out
@@ -0,0 +1,128 @@
+"""Input/output preprocessing helpers for Depth Anything 3."""
+
+from __future__ import annotations
+
+from typing import Tuple
+
+import torch
+
+import comfy.utils
+
+PATCH_SIZE = 14
+
+# ImageNet normalization constants used during DA3 training.
+_IMAGENET_MEAN = torch.tensor([0.485, 0.456, 0.406])
+_IMAGENET_STD = torch.tensor([0.229, 0.224, 0.225])
+
+
+def _round_to_patch(x: int, patch: int = PATCH_SIZE) -> int:
+    down = (x // patch) * patch
+    up = down + patch
+    return up if abs(up - x) <= abs(x - down) else down
+
+
+def compute_target_size(orig_h: int, orig_w: int, process_res: int, method: str = "upper_bound_resize") -> Tuple[int, int]:
+    """Compute (target_h, target_w) for a single image.
+    upper_bound_resize: scale longest side to process_res, then round each dim to nearest multiple of 14 (default upstream method).
+    lower_bound_resize: scale shortest side to process_res, then round."""
+
+    if method == "upper_bound_resize":
+        longest = max(orig_h, orig_w)
+        scale = process_res / float(longest)
+    elif method == "lower_bound_resize":
+        shortest = min(orig_h, orig_w)
+        scale = process_res / float(shortest)
+    else:
+        raise ValueError(f"Unsupported process_res_method: {method}")
+
+    new_w = max(1, _round_to_patch(int(round(orig_w * scale))))
+    new_h = max(1, _round_to_patch(int(round(orig_h * scale))))
+    return new_h, new_w
+
+
+def preprocess_image(image: torch.Tensor, process_res: int = 504, method: str = "upper_bound_resize") -> torch.Tensor:
+    assert image.ndim == 4 and image.shape[-1] == 3, f"expected (B,H,W,3) IMAGE; got {tuple(image.shape)}"
+    B, H, W, _ = image.shape
+    target_h, target_w = compute_target_size(H, W, process_res, method)
+
+    # (B, H, W, 3) -> (B, 3, H, W)
+    x = image.movedim(-1, 1).contiguous()
+    if (target_h, target_w) != (H, W):
+        # Upstream uses cv2 INTER_CUBIC (upscale) / INTER_AREA (downscale).
+        # Lanczos in ``common_upscale`` is anti-aliased and produces the
+        # closest pixel-wise match in a sweep across {bilinear, bicubic,
+        # area, lanczos, bislerp}. Used in both directions for simplicity.
+        x = comfy.utils.common_upscale(x.float(), target_w, target_h, "lanczos", "disabled",)
+    x = x.clamp(0.0, 1.0)
+
+    mean = _IMAGENET_MEAN.to(device=x.device, dtype=x.dtype).view(1, 3, 1, 1)
+    std = _IMAGENET_STD.to(device=x.device, dtype=x.dtype).view(1, 3, 1, 1)
+    x = (x - mean) / std
+    return x
+
+
+# -----------------------------------------------------------------------------
+# Output post-processing (sky-aware clipping for Mono/Metric variants)
+# -----------------------------------------------------------------------------
+
+
+def compute_non_sky_mask(sky_prediction: torch.Tensor, threshold: float = 0.3) -> torch.Tensor:
+    """Boolean mask: True for non-sky pixels (sky probability < threshold)."""
+    return sky_prediction < threshold
+
+
+def apply_sky_aware_clip(depth: torch.Tensor, sky: torch.Tensor, threshold: float = 0.3, quantile: float = 0.99) -> torch.Tensor:
+    """Clips sky regions to the 99th percentile of non-sky depth. Returns a new depth tensor."""
+    non_sky = compute_non_sky_mask(sky, threshold=threshold)
+    if non_sky.sum() <= 10 or (~non_sky).sum() <= 10:
+        return depth.clone()
+
+    non_sky_depth = depth[non_sky]
+    if non_sky_depth.numel() > 100_000:
+        idx = torch.randint(0, non_sky_depth.numel(), (100_000,), device=non_sky_depth.device)
+        sampled = non_sky_depth[idx]
+    else:
+        sampled = non_sky_depth
+
+    max_depth = torch.quantile(sampled, quantile)
+    out = depth.clone()
+    out[~non_sky] = max_depth
+    return out
+
+
+def normalize_depth_v2_style(depth: torch.Tensor, sky: torch.Tensor | None = None, low_quantile: float = 0.01, high_quantile: float = 0.99) -> torch.Tensor:
+    """V2-style normalization computes percentile bounds over non-sky pixels (when available), then maps depth into [0, 1] with near = white (1.0)."""
+    if sky is not None:
+        mask = compute_non_sky_mask(sky)
+        if mask.any():
+            valid = depth[mask]
+        else:
+            valid = depth.flatten()
+    else:
+        valid = depth.flatten()
+
+    if valid.numel() > 100_000:
+        idx = torch.randint(0, valid.numel(), (100_000,), device=valid.device)
+        sample = valid[idx]
+    else:
+        sample = valid
+
+    lo = torch.quantile(sample, low_quantile)
+    hi = torch.quantile(sample, high_quantile)
+    rng = (hi - lo).clamp(min=1e-6)
+    norm = ((depth - lo) / rng).clamp(0.0, 1.0)
+    # Nearer pixels are brighter (1.0)
+    norm = 1.0 - norm
+    if sky is not None:
+        # Sky pixels become black (far / unknown)
+        sky_mask = ~compute_non_sky_mask(sky)
+        norm = torch.where(sky_mask, torch.zeros_like(norm), norm)
+    return norm
+
+
+def normalize_depth_min_max(depth: torch.Tensor) -> torch.Tensor:
+    """Simple per-frame min/max normalization with near=1.0 convention."""
+    lo = depth.amin(dim=(-2, -1), keepdim=True)
+    hi = depth.amax(dim=(-2, -1), keepdim=True)
+    rng = (hi - lo).clamp(min=1e-6)
+    return 1.0 - ((depth - lo) / rng).clamp(0.0, 1.0)
@@ -0,0 +1,272 @@
+"""Ray-to-pose conversion for the multi-view path of Depth Anything 3."""
+
+from __future__ import annotations
+
+from typing import Optional, Tuple
+
+import torch
+
+
+# qr/svd use fp32: CUDA often has no fp16/bf16 kernels for these ops.
+
+
+def _ql_decomposition(A: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:
+    """Decompose A = Q @ L with Q orthogonal and L lower-triangular.
+    Implemented in terms of QR by reversing the columns/rows; the standard
+    trick from the upstream reference. Inputs A are (3, 3)."""
+    P = torch.tensor([[0, 0, 1], [0, 1, 0], [1, 0, 0]], device=A.device, dtype=A.dtype)
+    A_tilde = A @ P
+    # CUDA QR is not implemented for fp16/bf16; upcast just for this call.
+    Q_tilde, R_tilde = torch.linalg.qr(A_tilde.float())
+    Q_tilde = Q_tilde.to(A.dtype)
+    R_tilde = R_tilde.to(A.dtype)
+    Q = Q_tilde @ P
+    L = P @ R_tilde @ P
+    d = torch.diag(L)
+    sign = torch.sign(d)
+    Q = Q * sign[None, :]  # scale columns of Q
+    L = L * sign[:, None]  # scale rows of L
+    return Q, L
+
+
+def _homogenize_points(points: torch.Tensor) -> torch.Tensor:
+    return torch.cat([points, torch.ones_like(points[..., :1])], dim=-1)
+
+
+# -----------------------------------------------------------------------------
+# Weighted-LSQ + RANSAC homography (batched)
+# -----------------------------------------------------------------------------
+
+
+def _find_homography_weighted_lsq(src_pts: torch.Tensor, dst_pts: torch.Tensor, confident_weight: torch.Tensor,) -> torch.Tensor:
+    """Solve a single H with weighted least-squares (DLT)."""
+    N = src_pts.shape[0]
+    if N < 4:
+        raise ValueError("At least 4 points are required to compute a homography.")
+    w = confident_weight.sqrt().unsqueeze(1)  # (N, 1)
+    x = src_pts[:, 0:1]
+    y = src_pts[:, 1:2]
+    u = dst_pts[:, 0:1]
+    v = dst_pts[:, 1:2]
+    zeros = torch.zeros_like(x)
+    A1 = torch.cat([-x * w, -y * w, -w, zeros, zeros, zeros, x * u * w, y * u * w, u * w], dim=1)
+    A2 = torch.cat([zeros, zeros, zeros, -x * w, -y * w, -w, x * v * w, y * v * w, v * w], dim=1)
+    A = torch.cat([A1, A2], dim=0)        # (2N, 9)
+    # CUDA SVD is not implemented for fp16/bf16; upcast just for this call.
+    _, _, Vh = torch.linalg.svd(A.float())
+    Vh = Vh.to(A.dtype)
+    H = Vh[-1].reshape(3, 3)
+    return H / H[-1, -1]
+
+
+def _find_homography_weighted_lsq_batched(src_pts_batch: torch.Tensor, dst_pts_batch: torch.Tensor, confident_weight_batch: torch.Tensor) -> torch.Tensor:
+    """Batched DLT solver. Inputs (B, K, 2) / (B, K); output (B, 3, 3)."""
+    B, K, _ = src_pts_batch.shape
+    w = confident_weight_batch.sqrt().unsqueeze(2)
+    x = src_pts_batch[:, :, 0:1]
+    y = src_pts_batch[:, :, 1:2]
+    u = dst_pts_batch[:, :, 0:1]
+    v = dst_pts_batch[:, :, 1:2]
+    zeros = torch.zeros_like(x)
+    A1 = torch.cat([-x * w, -y * w, -w, zeros, zeros, zeros, x * u * w, y * u * w, u * w], dim=2)
+    A2 = torch.cat([zeros, zeros, zeros, -x * w, -y * w, -w, x * v * w, y * v * w, v * w], dim=2)
+    A = torch.cat([A1, A2], dim=1)        # (B, 2K, 9)
+    # CUDA SVD is not implemented for fp16/bf16; upcast just for this call.
+    _, _, Vh = torch.linalg.svd(A.float())
+    Vh = Vh.to(A.dtype)
+    H = Vh[:, -1].reshape(B, 3, 3)
+    return H / H[:, 2:3, 2:3]
+
+
+def _ransac_find_homography_weighted_batched(
+    src_pts: torch.Tensor,                # (B, N, 2)
+    dst_pts: torch.Tensor,                # (B, N, 2)
+    confident_weight: torch.Tensor,       # (B, N)
+    n_sample: int,
+    n_iter: int = 100,
+    reproj_threshold: float = 3.0,
+    num_sample_for_ransac: int = 8,
+    max_inlier_num: int = 10000,
+    rand_sample_iters_idx: Optional[torch.Tensor] = None,
+) -> torch.Tensor:
+    """Batched weighted-RANSAC homography estimator. Returns (B, 3, 3) homography matrices."""
+    B, N, _ = src_pts.shape
+    assert N >= 4
+    device = src_pts.device
+
+    sorted_idx = torch.argsort(confident_weight, descending=True, dim=1)
+    candidate_idx = sorted_idx[:, :n_sample]                  # (B, n_sample)
+
+    if rand_sample_iters_idx is None:
+        rand_sample_iters_idx = torch.stack(
+            [torch.randperm(n_sample, device=device)[:num_sample_for_ransac]
+             for _ in range(n_iter)],
+            dim=0,
+        )
+
+    rand_idx = candidate_idx[:, rand_sample_iters_idx]        # (B, n_iter, k)
+    b_idx = (
+        torch.arange(B, device=device)
+        .view(B, 1, 1)
+        .expand(B, n_iter, num_sample_for_ransac)
+    )
+    src_b = src_pts[b_idx, rand_idx]
+    dst_b = dst_pts[b_idx, rand_idx]
+    w_b = confident_weight[b_idx, rand_idx]
+
+    cB, cN = src_b.shape[:2]
+    H_batch = _find_homography_weighted_lsq_batched(
+        src_b.flatten(0, 1), dst_b.flatten(0, 1), w_b.flatten(0, 1),
+    ).unflatten(0, (cB, cN))                                  # (B, n_iter, 3, 3)
+
+    src_homo = torch.cat([src_pts, torch.ones(B, N, 1, device=device, dtype=src_pts.dtype)], dim=2)
+    proj = torch.bmm(
+        src_homo.unsqueeze(1).expand(B, n_iter, N, 3).reshape(-1, N, 3),
+        H_batch.reshape(-1, 3, 3).transpose(1, 2),
+    )                                                          # (B*n_iter, N, 3)
+    proj_xy = (proj[:, :, :2] / proj[:, :, 2:3]).reshape(B, n_iter, N, 2)
+    err = ((proj_xy - dst_pts.unsqueeze(1)) ** 2).sum(-1).sqrt()  # (B, n_iter, N)
+    inlier_mask = err < reproj_threshold
+    score = (inlier_mask * confident_weight.unsqueeze(1)).sum(dim=2)
+    best_idx = torch.argmax(score, dim=1)
+    best_inlier_mask = inlier_mask[torch.arange(B, device=device), best_idx]
+
+    # Refit with the inlier set (per-batch, since the inlier counts vary).
+    H_inlier_list = []
+    for b in range(B):
+        mask = best_inlier_mask[b]
+        in_src = src_pts[b][mask]
+        in_dst = dst_pts[b][mask]
+        in_w = confident_weight[b][mask]
+        if in_src.shape[0] < 4:
+            # Fall back to identity when RANSAC fails to find enough inliers.
+            H_inlier_list.append(torch.eye(3, device=device, dtype=src_pts.dtype))
+            continue
+        sorted_w = torch.argsort(in_w, descending=True)
+        if len(sorted_w) > max_inlier_num:
+            keep = max(int(len(sorted_w) * 0.95), max_inlier_num)
+            sorted_w = sorted_w[:keep][torch.randperm(keep, device=device)[:max_inlier_num]]
+        H_inlier_list.append(
+            _find_homography_weighted_lsq(in_src[sorted_w], in_dst[sorted_w], in_w[sorted_w])
+        )
+    return torch.stack(H_inlier_list, dim=0)
+
+
+# -----------------------------------------------------------------------------
+# Camera-ray utilities
+# -----------------------------------------------------------------------------
+
+
+def _unproject_identity(num_y: int, num_x: int, B: int, S: int, device, dtype) -> torch.Tensor:
+    """Camera-space unit rays for an identity intrinsic on a 2x2 image plane."""
+    dx = 1.0 / num_x
+    dy = 1.0 / num_y
+    # Centered camera-space coords directly (skip the K^-1 step since it's
+    # just a translation by -1 on x and y when K is identity-with-center=1).
+    y = torch.linspace(-(1 - dy), (1 - dy), num_y, device=device, dtype=dtype)
+    x = torch.linspace(-(1 - dx), (1 - dx), num_x, device=device, dtype=dtype)
+    yy, xx = torch.meshgrid(y, x, indexing="ij")
+    grid = torch.stack((xx, yy), dim=-1)            # (h, w, 2)
+    grid = grid.unsqueeze(0).unsqueeze(0).expand(B, S, num_y, num_x, 2)
+    return torch.cat([grid, torch.ones_like(grid[..., :1])], dim=-1)
+
+
+def _camray_to_caminfo(
+    camray: torch.Tensor,  # (B, S, h, w, 6)
+    confidence: Optional[torch.Tensor] = None,  # (B, S, h, w)
+    reproj_threshold: float = 0.2,
+) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]:
+    """Convert per-pixel camera rays to per-view (R, T, focal, principal)."""
+    if confidence is None:
+        confidence = torch.ones_like(camray[..., 0])
+    B, S, h, w, _ = camray.shape
+    device = camray.device
+    dtype = camray.dtype
+
+    rays_target = camray[..., :3]                           # (B, S, h, w, 3)
+    rays_origin = _unproject_identity(h, w, B, S, device, dtype)
+
+    # Flatten (B*S, h*w, *) for the RANSAC routine.
+    rays_target = rays_target.flatten(0, 1).flatten(1, 2)
+    rays_origin = rays_origin.flatten(0, 1).flatten(1, 2)
+    weights = confidence.flatten(0, 1).flatten(1, 2).clone()
+
+    # Project to 2D in homogeneous form (the upstream calls this "perspective division").
+    z_thresh = 1e-4
+    mask = (rays_target[:, :, 2].abs() > z_thresh) & (rays_origin[:, :, 2].abs() > z_thresh)
+    weights = torch.where(mask, weights, torch.zeros_like(weights))
+    src = rays_origin.clone()
+    dst = rays_target.clone()
+    src[..., 0] = torch.where(mask, src[..., 0] / src[..., 2], src[..., 0])
+    src[..., 1] = torch.where(mask, src[..., 1] / src[..., 2], src[..., 1])
+    dst[..., 0] = torch.where(mask, dst[..., 0] / dst[..., 2], dst[..., 0])
+    dst[..., 1] = torch.where(mask, dst[..., 1] / dst[..., 2], dst[..., 1])
+    src = src[..., :2]
+    dst = dst[..., :2]
+
+    N = src.shape[1]
+    n_iter = 100
+    sample_ratio = 0.3
+    num_sample_for_ransac = 8
+    n_sample = max(num_sample_for_ransac, int(N * sample_ratio))
+    rand_idx = torch.stack(
+        [torch.randperm(n_sample, device=device)[:num_sample_for_ransac] for _ in range(n_iter)],
+        dim=0,
+    )
+
+    # Chunk along the view axis to keep peak memory predictable.
+    chunk = 2
+    A_list = []
+    for i in range(0, src.shape[0], chunk):
+        A = _ransac_find_homography_weighted_batched(
+            src[i:i + chunk], dst[i:i + chunk], weights[i:i + chunk],
+            n_sample=n_sample, n_iter=n_iter,
+            num_sample_for_ransac=num_sample_for_ransac,
+            reproj_threshold=reproj_threshold,
+            rand_sample_iters_idx=rand_idx,
+            max_inlier_num=8000,
+        )
+        # Flip sign on dets that come out < 0 (so that the QL produces a
+        # right-handed rotation). ``det`` lacks fp16/bf16 CUDA kernels, so
+        # do the comparison in fp32.
+        flip = torch.linalg.det(A.float()) < 0
+        A = torch.where(flip[:, None, None], -A, A)
+        A_list.append(A)
+    A = torch.cat(A_list, dim=0)                            # (B*S, 3, 3)
+
+    R_list, f_list, pp_list = [], [], []
+    for i in range(A.shape[0]):
+        R, L = _ql_decomposition(A[i])
+        L = L / L[2][2]
+        f_list.append(torch.stack((L[0][0], L[1][1])))
+        pp_list.append(torch.stack((L[2][0], L[2][1])))
+        R_list.append(R)
+    R = torch.stack(R_list).reshape(B, S, 3, 3)
+    focal = torch.stack(f_list).reshape(B, S, 2)
+    pp = torch.stack(pp_list).reshape(B, S, 2)
+
+    # Translation: confidence-weighted average of camray direction(s).
+    cf = confidence.flatten(0, 1).flatten(1, 2)
+    T = (camray.flatten(0, 1).flatten(1, 2)[..., 3:] * cf.unsqueeze(-1)).sum(dim=1)
+    T = T / cf.sum(dim=-1, keepdim=True)
+    T = T.reshape(B, S, 3)
+
+    # Match upstream output convention: focal -> 1/focal, pp + 1.
+    return R, T, 1.0 / focal, pp + 1.0
+
+
+def get_extrinsic_from_camray(
+    camray: torch.Tensor,  # (B, S, h, w, 6)
+    conf: torch.Tensor,  # (B, S, h, w, 1) or (B, S, h, w)
+    patch_size_y: int,
+    patch_size_x: int,
+) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+    """Wrap a 4x4 extrinsic + per-view focal + principal-point output."""
+    if conf.ndim == 5 and conf.shape[-1] == 1:
+        conf = conf.squeeze(-1)
+    R, T, focal, pp = _camray_to_caminfo(camray, confidence=conf)
+    extr = torch.cat([R, T.unsqueeze(-1)], dim=-1)           # (B, S, 3, 4)
+    homo_row = torch.tensor([0, 0, 0, 1], dtype=R.dtype, device=R.device)
+    homo_row = homo_row.view(1, 1, 1, 4).expand(R.shape[0], R.shape[1], 1, 4)
+    extr = torch.cat([extr, homo_row], dim=-2)               # (B, S, 4, 4)
+    return extr, focal, pp
@@ -0,0 +1,87 @@
+"""Reference-view selection for the multi-view path of Depth Anything 3."""
+
+from __future__ import annotations
+
+from typing import Literal
+
+import torch
+
+
+RefViewStrategy = Literal["first", "middle", "saddle_balanced", "saddle_sim_range"]
+
+
+# Per the upstream constants module: ``THRESH_FOR_REF_SELECTION = 3``.
+# Reference selection only runs when there are at least this many views.
+THRESH_FOR_REF_SELECTION: int = 3
+
+
+def select_reference_view(x: torch.Tensor, strategy: RefViewStrategy = "saddle_balanced") -> torch.Tensor:
+    """Pick a reference view index per batch element."""
+    B, S, _, _ = x.shape
+    if S <= 1:
+        return torch.zeros(B, dtype=torch.long, device=x.device)
+    if strategy == "first":
+        return torch.zeros(B, dtype=torch.long, device=x.device)
+    if strategy == "middle":
+        return torch.full((B,), S // 2, dtype=torch.long, device=x.device)
+
+    # Feature-based strategies: normalised cls/cam token per view.
+    img_class_feat = x[:, :, 0] / x[:, :, 0].norm(dim=-1, keepdim=True)  # (B,S,C)
+
+    if strategy == "saddle_balanced":
+        sim = torch.matmul(img_class_feat, img_class_feat.transpose(1, 2))  # (B,S,S)
+        sim_no_diag = sim - torch.eye(S, device=sim.device).unsqueeze(0)
+        sim_score = sim_no_diag.sum(dim=-1) / (S - 1)               # (B,S)
+        feat_norm = x[:, :, 0].norm(dim=-1)                          # (B,S)
+        feat_var = img_class_feat.var(dim=-1)                        # (B,S)
+
+        def _normalize(metric):
+            mn = metric.min(dim=1, keepdim=True).values
+            mx = metric.max(dim=1, keepdim=True).values
+            return (metric - mn) / (mx - mn + 1e-8)
+
+        sim_n, norm_n, var_n = _normalize(sim_score), _normalize(feat_norm), _normalize(feat_var)
+        balance = (sim_n - 0.5).abs() + (norm_n - 0.5).abs() + (var_n - 0.5).abs()
+        return balance.argmin(dim=1)
+
+    if strategy == "saddle_sim_range":
+        sim = torch.matmul(img_class_feat, img_class_feat.transpose(1, 2))
+        sim_no_diag = sim - torch.eye(S, device=sim.device).unsqueeze(0)
+        sim_max = sim_no_diag.max(dim=-1).values
+        sim_min = sim_no_diag.min(dim=-1).values
+        return (sim_max - sim_min).argmax(dim=1)
+
+    raise ValueError(
+        f"Unknown reference view selection strategy: {strategy!r}. "
+        f"Must be one of: 'first', 'middle', 'saddle_balanced', 'saddle_sim_range'"
+    )
+
+
+def reorder_by_reference(x: torch.Tensor, b_idx: torch.Tensor) -> torch.Tensor:
+    """Reorder x so the reference view is at position 0 in axis S."""
+    B, S = x.shape[0], x.shape[1]
+    if S <= 1:
+        return x
+    positions = torch.arange(S, device=x.device).unsqueeze(0).expand(B, -1)
+    b_idx_exp = b_idx.unsqueeze(1)
+    reorder = torch.where(
+        (positions > 0) & (positions <= b_idx_exp),
+        positions - 1,
+        positions,
+    )
+    reorder[:, 0] = b_idx
+    batch = torch.arange(B, device=x.device).unsqueeze(1)
+    return x[batch, reorder]
+
+
+def restore_original_order(x: torch.Tensor, b_idx: torch.Tensor) -> torch.Tensor:
+    """Inverse of reorder_by_reference."""
+    B, S = x.shape[0], x.shape[1]
+    if S <= 1:
+        return x
+    target_positions = torch.arange(S, device=x.device).unsqueeze(0).expand(B, -1)
+    b_idx_exp = b_idx.unsqueeze(1)
+    restore = torch.where(target_positions < b_idx_exp, target_positions + 1, target_positions)
+    restore = torch.scatter(restore, dim=1, index=b_idx_exp, src=torch.zeros_like(b_idx_exp))
+    batch = torch.arange(B, device=x.device).unsqueeze(1)
+    return x[batch, restore]
@@ -0,0 +1,160 @@
+"""Geometry / camera transform helpers for Depth Anything 3."""
+
+from __future__ import annotations
+
+from typing import Tuple
+
+import torch
+import torch.nn.functional as F
+
+
+# -----------------------------------------------------------------------------
+# Affine 4x4 helpers
+# -----------------------------------------------------------------------------
+
+
+def as_homogeneous(ext: torch.Tensor) -> torch.Tensor:
+    """Promote (...,3,4) extrinsics to (...,4,4) homogeneous form. No-op when the input is already ``(...,4,4)``."""
+    if ext.shape[-2:] == (4, 4):
+        return ext
+    if ext.shape[-2:] == (3, 4):
+        ones = torch.zeros_like(ext[..., :1, :4])
+        ones[..., 0, 3] = 1.0
+        return torch.cat([ext, ones], dim=-2)
+    raise ValueError(f"Invalid affine shape: {ext.shape}")
+
+
+def affine_inverse(A: torch.Tensor) -> torch.Tensor:
+    """Inverse of an affine matrix ``[R|T; 0 0 0 1]``."""
+    R = A[..., :3, :3]
+    T = A[..., :3, 3:]
+    P = A[..., 3:, :]
+    return torch.cat([torch.cat([R.mT, -R.mT @ T], dim=-1), P], dim=-2)
+
+
+# -----------------------------------------------------------------------------
+# Quaternion <-> rotation matrix (xyzw / scalar-last)
+# -----------------------------------------------------------------------------
+
+
+def _sqrt_positive_part(x: torch.Tensor) -> torch.Tensor:
+    """sqrt(max(0, x)) with a zero subgradient where x == 0."""
+    ret = torch.zeros_like(x)
+    positive_mask = x > 0
+    if torch.is_grad_enabled():
+        ret[positive_mask] = torch.sqrt(x[positive_mask])
+    else:
+        ret = torch.where(positive_mask, torch.sqrt(x), ret)
+    return ret
+
+
+def standardize_quaternion(quaternions: torch.Tensor) -> torch.Tensor:
+    """Force the real part of a unit quaternion (xyzw) to be non-negative."""
+    return torch.where(quaternions[..., 3:4] < 0, -quaternions, quaternions)
+
+
+def quat_to_mat(quaternions: torch.Tensor) -> torch.Tensor:
+    """Convert quaternions (xyzw) to (...,3,3) rotation matrices."""
+    i, j, k, r = torch.unbind(quaternions, -1)
+    two_s = 2.0 / (quaternions * quaternions).sum(-1)
+    o = torch.stack(
+        (
+            1 - two_s * (j * j + k * k),
+            two_s * (i * j - k * r),
+            two_s * (i * k + j * r),
+            two_s * (i * j + k * r),
+            1 - two_s * (i * i + k * k),
+            two_s * (j * k - i * r),
+            two_s * (i * k - j * r),
+            two_s * (j * k + i * r),
+            1 - two_s * (i * i + j * j),
+        ),
+        -1,
+    )
+    return o.reshape(quaternions.shape[:-1] + (3, 3))
+
+
+def mat_to_quat(matrix: torch.Tensor) -> torch.Tensor:
+    """Convert (...,3,3) rotation matrices to quaternions (xyzw)."""
+    if matrix.size(-1) != 3 or matrix.size(-2) != 3:
+        raise ValueError(f"Invalid rotation matrix shape {matrix.shape}.")
+
+    batch_dim = matrix.shape[:-2]
+    m00, m01, m02, m10, m11, m12, m20, m21, m22 = torch.unbind(
+        matrix.reshape(batch_dim + (9,)), dim=-1
+    )
+
+    q_abs = _sqrt_positive_part(
+        torch.stack(
+            [
+                1.0 + m00 + m11 + m22,
+                1.0 + m00 - m11 - m22,
+                1.0 - m00 + m11 - m22,
+                1.0 - m00 - m11 + m22,
+            ],
+            dim=-1,
+        )
+    )
+
+    quat_by_rijk = torch.stack(
+        [
+            torch.stack([q_abs[..., 0] ** 2, m21 - m12, m02 - m20, m10 - m01], dim=-1),
+            torch.stack([m21 - m12, q_abs[..., 1] ** 2, m10 + m01, m02 + m20], dim=-1),
+            torch.stack([m02 - m20, m10 + m01, q_abs[..., 2] ** 2, m12 + m21], dim=-1),
+            torch.stack([m10 - m01, m20 + m02, m21 + m12, q_abs[..., 3] ** 2], dim=-1),
+        ],
+        dim=-2,
+    )
+
+    flr = torch.tensor(0.1).to(dtype=q_abs.dtype, device=q_abs.device)
+    quat_candidates = quat_by_rijk / (2.0 * q_abs[..., None].max(flr))
+
+    out = quat_candidates[F.one_hot(q_abs.argmax(dim=-1), num_classes=4) > 0.5, :].reshape(
+        batch_dim + (4,)
+    )
+    # Reorder rijk -> xyzw (i.e. ijkr).
+    out = out[..., [1, 2, 3, 0]]
+    return standardize_quaternion(out)
+
+
+# -----------------------------------------------------------------------------
+# Pose-encoding <-> extrinsics + intrinsics
+# -----------------------------------------------------------------------------
+
+
+def extri_intri_to_pose_encoding(extrinsics: torch.Tensor, intrinsics: torch.Tensor, image_size_hw: Tuple[int, int]) -> torch.Tensor:
+    """Pack (extr, intr, image_size) into the 9-D pose-encoding vector.
+    extrinsics: camera-to-world (c2w) (B,S,4,4) matrices,
+    intrinsics: pixel-space (B,S,3,3) matrices,
+    image_size_hw: is a (H, W) pair.
+    """
+    R = extrinsics[..., :3, :3]
+    T = extrinsics[..., :3, 3]
+    quat = mat_to_quat(R)
+    H, W = image_size_hw
+    fov_h = 2 * torch.atan((H / 2) / intrinsics[..., 1, 1])
+    fov_w = 2 * torch.atan((W / 2) / intrinsics[..., 0, 0])
+    return torch.cat([T, quat, fov_h[..., None], fov_w[..., None]], dim=-1).float()
+
+
+def pose_encoding_to_extri_intri(pose_encoding: torch.Tensor, image_size_hw: Tuple[int, int]) -> Tuple[torch.Tensor, torch.Tensor]:
+    """Inverse of extri_intri_to_pose_encoding."""
+    T = pose_encoding[..., :3]
+    quat = pose_encoding[..., 3:7]
+    fov_h = pose_encoding[..., 7]
+    fov_w = pose_encoding[..., 8]
+    # Normalize to unit quaternion. CameraDec outputs raw values; a near-zero
+    # quaternion causes two_s = 2/norm² → inf in quat_to_mat → NaN extrinsics.
+    quat = quat / quat.norm(dim=-1, keepdim=True).clamp(min=1e-6)
+    R = quat_to_mat(quat)
+    extrinsics = torch.cat([R, T[..., None]], dim=-1)
+    H, W = image_size_hw
+    fy = (H / 2.0) / torch.clamp(torch.tan(fov_h / 2.0), 1e-6)
+    fx = (W / 2.0) / torch.clamp(torch.tan(fov_w / 2.0), 1e-6)
+    intrinsics = torch.zeros(pose_encoding.shape[:2] + (3, 3), device=pose_encoding.device, dtype=pose_encoding.dtype)
+    intrinsics[..., 0, 0] = fx
+    intrinsics[..., 1, 1] = fy
+    intrinsics[..., 0, 2] = W / 2
+    intrinsics[..., 1, 2] = H / 2
+    intrinsics[..., 2, 2] = 1.0
+    return extrinsics, intrinsics
@@ -5,6 +5,7 @@ import torch.nn.functional as F

 from comfy.ldm.modules.attention import optimized_attention
 import comfy.model_management
+import comfy.quant_ops

 def rope(pos: torch.Tensor, dim: int, theta: int) -> torch.Tensor:
    assert dim % 2 == 0
@@ -19,15 +20,6 @@ def rope(pos: torch.Tensor, dim: int, theta: int) -> torch.Tensor:
    out = torch.stack([torch.cos(out), torch.sin(out)], dim=0)
    return out.to(dtype=torch.float32, device=pos.device)

-def apply_rotary_emb(x_in: torch.Tensor, freqs_cis: torch.Tensor) -> torch.Tensor:
-    rot_dim = freqs_cis.shape[-1]
-    x, x_pass = x_in[..., :rot_dim], x_in[..., rot_dim:]
-    cos_ = freqs_cis[0]
-    sin_ = freqs_cis[1]
-    x1, x2 = x.chunk(2, dim=-1)
-    x_rotated = torch.cat((-x2, x1), dim=-1)
-    return torch.cat((x * cos_ + x_rotated * sin_, x_pass), dim=-1)
-
 class ErnieImageEmbedND3(nn.Module):
    def __init__(self, dim: int, theta: int, axes_dim: tuple):
        super().__init__()
@@ -37,8 +29,16 @@ class ErnieImageEmbedND3(nn.Module):

    def forward(self, ids: torch.Tensor) -> torch.Tensor:
        emb = torch.cat([rope(ids[..., i], self.axes_dim[i], self.theta) for i in range(3)], dim=-1)
-        emb = emb.unsqueeze(3)  # [2, B, S, 1, head_dim//2]
-        return torch.stack([emb, emb], dim=-1).reshape(*emb.shape[:-1], -1)  # [B, S, 1, head_dim]
+        cos_ = emb[0]
+        sin_ = emb[1]
+        N = cos_.shape[-1]
+        half = N // 2
+        cos_top = cos_[..., :half].repeat_interleave(2, dim=-1)
+        sin_top = sin_[..., :half].repeat_interleave(2, dim=-1)
+        cos_bot = cos_[..., half:].repeat_interleave(2, dim=-1)
+        sin_bot = sin_[..., half:].repeat_interleave(2, dim=-1)
+        rot = torch.stack([cos_top, -sin_top, sin_bot, cos_bot], dim=-1)
+        return rot.reshape(*rot.shape[:-1], 2, 2).unsqueeze(2)

 class ErnieImagePatchEmbedDynamic(nn.Module):
    def __init__(self, in_channels: int, embed_dim: int, patch_size: int, operations, device=None, dtype=None):
@@ -115,8 +115,7 @@ class ErnieImageAttention(nn.Module):
        key = self.norm_k(key)

        if image_rotary_emb is not None:
-            query = apply_rotary_emb(query, image_rotary_emb)
-            key = apply_rotary_emb(key, image_rotary_emb)
+            query, key = comfy.quant_ops.ck.apply_rope_split_half(query, key, image_rotary_emb)

        q_flat = query.reshape(B, S, -1)
        k_flat = key.reshape(B, S, -1)
@@ -274,7 +273,7 @@ class ErnieImageModel(nn.Module):

        image_ids = image_ids.view(1, N_img, 3).expand(B, -1, -1)

-        rotary_pos_emb = self.pos_embed(torch.cat([image_ids, text_ids], dim=1)).to(x.dtype)
+        rotary_pos_emb = self.pos_embed(torch.cat([image_ids, text_ids], dim=1))
        del image_ids, text_ids

        sample = self.time_proj(timesteps).to(dtype)
@@ -4,7 +4,7 @@ from torch import Tensor

 from comfy.ldm.modules.attention import optimized_attention
 import comfy.model_management
-import logging
+import comfy.quant_ops


 def attention(q: Tensor, k: Tensor, v: Tensor, pe: Tensor, mask=None, transformer_options={}) -> Tensor:
@@ -44,21 +44,15 @@ def _apply_rope(xq: Tensor, xk: Tensor, freqs_cis: Tensor):
    return apply_rope1(xq, freqs_cis), apply_rope1(xk, freqs_cis)


-try:
-    import comfy.quant_ops
-    q_apply_rope = comfy.quant_ops.ck.apply_rope
-    q_apply_rope1 = comfy.quant_ops.ck.apply_rope1
-    def apply_rope(xq, xk, freqs_cis):
-        if comfy.model_management.in_training:
-            return _apply_rope(xq, xk, freqs_cis)
-        else:
-            return apply_rope1(xq, freqs_cis), apply_rope1(xk, freqs_cis)
-    def apply_rope1(x, freqs_cis):
-        if comfy.model_management.in_training:
-            return _apply_rope1(x, freqs_cis)
-        else:
-            return q_apply_rope1(x, freqs_cis)
-except:
-    logging.warning("No comfy kitchen, using old apply_rope functions.")
-    apply_rope = _apply_rope
-    apply_rope1 = _apply_rope1
+def apply_rope(xq, xk, freqs_cis):
+    if comfy.model_management.in_training:
+        return _apply_rope(xq, xk, freqs_cis)
+    else:
+        return comfy.quant_ops.ck.apply_rope(xq, xk, freqs_cis)
+
+
+def apply_rope1(x, freqs_cis):
+    if comfy.model_management.in_training:
+        return _apply_rope1(x, freqs_cis)
+    else:
+        return comfy.quant_ops.ck.apply_rope1(x, freqs_cis)
@@ -0,0 +1,297 @@
+"""
+The Ideogram 4 transformer is a NextDiT/Lumina2-family single-stream model
+consumes Qwen3-VL hidden-state features (concatenated from 13 layers -> 53248 dims)
+packs ``[text tokens, image tokens]`` into one sequence with block-diagonal segment attention and 3D interleaved MRoPE.
+"""
+
+from __future__ import annotations
+
+import math
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+import comfy.patcher_extension
+from comfy.ldm.lumina.model import FeedForward
+from comfy.ldm.modules.attention import optimized_attention_masked
+from comfy.text_encoders.llama import apply_rope, precompute_freqs_cis
+
+# Per-token role indicators
+SEQUENCE_PADDING_INDICATOR = -1
+OUTPUT_IMAGE_INDICATOR = 2
+LLM_TOKEN_INDICATOR = 3
+# Image grid coordinates are offset so they never collide with text positions
+IMAGE_POSITION_OFFSET = 65536
+
+
+class Ideogram4Attention(nn.Module):
+    def __init__(self, hidden_size, num_heads, eps=1e-5, dtype=None, device=None, operations=None):
+        super().__init__()
+        self.num_heads = num_heads
+        self.head_dim = hidden_size // num_heads
+        self.hidden_size = hidden_size
+
+        self.qkv = operations.Linear(hidden_size, hidden_size * 3, bias=False, dtype=dtype, device=device)
+        self.norm_q = operations.RMSNorm(self.head_dim, eps=eps, elementwise_affine=True, dtype=dtype, device=device)
+        self.norm_k = operations.RMSNorm(self.head_dim, eps=eps, elementwise_affine=True, dtype=dtype, device=device)
+        self.o = operations.Linear(hidden_size, hidden_size, bias=False, dtype=dtype, device=device)
+
+    def forward(self, x, attn_mask, freqs_cis, transformer_options={}):
+        batch_size, seq_len, _ = x.shape
+        qkv = self.qkv(x).view(batch_size, seq_len, 3, self.num_heads, self.head_dim)
+        q, k, v = qkv.unbind(dim=2)
+
+        q = self.norm_q(q)
+        k = self.norm_k(k)
+
+        # (B, heads, L, head_dim)
+        q = q.transpose(1, 2)
+        k = k.transpose(1, 2)
+        v = v.transpose(1, 2)
+
+        q, k = apply_rope(q, k, freqs_cis)
+
+        out = optimized_attention_masked(q, k, v, self.num_heads, attn_mask, skip_reshape=True, transformer_options=transformer_options)
+        return self.o(out)
+
+
+class Ideogram4TransformerBlock(nn.Module):
+    def __init__(self, hidden_size, intermediate_size, num_heads, norm_eps, adaln_dim, dtype=None, device=None, operations=None):
+        super().__init__()
+        self.attention = Ideogram4Attention(hidden_size, num_heads, eps=1e-5, dtype=dtype, device=device, operations=operations)
+        self.feed_forward = FeedForward(
+            dim=hidden_size, hidden_dim=intermediate_size, multiple_of=1, ffn_dim_multiplier=None,
+            operation_settings={"operations": operations, "dtype": dtype, "device": device},
+        )
+
+        self.attention_norm1 = operations.RMSNorm(hidden_size, eps=norm_eps, elementwise_affine=True, dtype=dtype, device=device)
+        self.ffn_norm1 = operations.RMSNorm(hidden_size, eps=norm_eps, elementwise_affine=True, dtype=dtype, device=device)
+        self.attention_norm2 = operations.RMSNorm(hidden_size, eps=norm_eps, elementwise_affine=True, dtype=dtype, device=device)
+        self.ffn_norm2 = operations.RMSNorm(hidden_size, eps=norm_eps, elementwise_affine=True, dtype=dtype, device=device)
+
+        self.adaln_modulation = operations.Linear(adaln_dim, 4 * hidden_size, bias=True, dtype=dtype, device=device)
+
+    def forward(self, x, attn_mask, freqs_cis, adaln_input, transformer_options={}):
+        mod = self.adaln_modulation(adaln_input)
+        scale_msa, gate_msa, scale_mlp, gate_mlp = mod.chunk(4, dim=-1)
+        gate_msa = torch.tanh(gate_msa)
+        gate_mlp = torch.tanh(gate_mlp)
+        scale_msa = 1.0 + scale_msa
+        scale_mlp = 1.0 + scale_mlp
+
+        attn_out = self.attention(self.attention_norm1(x) * scale_msa, attn_mask, freqs_cis, transformer_options=transformer_options)
+        x = x + gate_msa * self.attention_norm2(attn_out)
+        x = x + gate_mlp * self.ffn_norm2(self.feed_forward(self.ffn_norm1(x) * scale_mlp))
+        return x
+
+
+def _sinusoidal_embedding(t, dim, scale=1e4):
+    t = t.to(torch.float32)
+    half = dim // 2
+    freq = math.log(scale) / (half - 1)
+    freq = torch.exp(torch.arange(half, dtype=torch.float32, device=t.device) * -freq)
+    emb = t.unsqueeze(-1) * freq
+    emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=-1)
+    if dim % 2 == 1:
+        emb = F.pad(emb, (0, 1))
+    return emb
+
+
+class Ideogram4EmbedScalar(nn.Module):
+    def __init__(self, dim, input_range=(0.0, 1.0), dtype=None, device=None, operations=None):
+        super().__init__()
+        self.dim = dim
+        self.range_min, self.range_max = input_range
+        self.mlp_in = operations.Linear(dim, dim, bias=True, dtype=dtype, device=device)
+        self.mlp_out = operations.Linear(dim, dim, bias=True, dtype=dtype, device=device)
+
+    def forward(self, x, dtype):
+        x = x.to(torch.float32)
+        scaled = 1e4 * (x - self.range_min) / (self.range_max - self.range_min)
+        emb = _sinusoidal_embedding(scaled, self.dim)
+        emb = emb.to(dtype)
+        emb = F.silu(self.mlp_in(emb))
+        return self.mlp_out(emb)
+
+
+class Ideogram4FinalLayer(nn.Module):
+    def __init__(self, hidden_size, out_channels, adaln_dim, dtype=None, device=None, operations=None):
+        super().__init__()
+        self.norm_final = operations.LayerNorm(hidden_size, eps=1e-6, elementwise_affine=False, dtype=dtype, device=device)
+        self.linear = operations.Linear(hidden_size, out_channels, bias=True, dtype=dtype, device=device)
+        self.adaln_modulation = operations.Linear(adaln_dim, hidden_size, bias=True, dtype=dtype, device=device)
+
+    def forward(self, x, c):
+        scale = 1.0 + self.adaln_modulation(F.silu(c))
+        return self.linear(self.norm_final(x) * scale)
+
+
+class Ideogram4Transformer(nn.Module):
+    """A single Ideogram 4 backbone operating on a packed token sequence."""
+
+    def __init__(self, emb_dim, num_layers, num_heads, intermediate_size, adaln_dim,
+                 in_channels, llm_features_dim, rope_theta, mrope_section, norm_eps,
+                 dtype=None, device=None, operations=None):
+        super().__init__()
+        self.head_dim = emb_dim // num_heads
+        self.rope_theta = rope_theta
+        self.mrope_section = tuple(mrope_section)
+
+        self.input_proj = operations.Linear(in_channels, emb_dim, bias=True, dtype=dtype, device=device)
+        self.llm_cond_norm = operations.RMSNorm(llm_features_dim, eps=1e-6, elementwise_affine=True, dtype=dtype, device=device)
+        self.llm_cond_proj = operations.Linear(llm_features_dim, emb_dim, bias=True, dtype=dtype, device=device)
+        self.t_embedding = Ideogram4EmbedScalar(emb_dim, input_range=(0.0, 1.0), dtype=dtype, device=device, operations=operations)
+        self.adaln_proj = operations.Linear(emb_dim, adaln_dim, bias=True, dtype=dtype, device=device)
+
+        self.embed_image_indicator = operations.Embedding(2, emb_dim, dtype=dtype, device=device)
+
+        self.layers = nn.ModuleList([
+            Ideogram4TransformerBlock(emb_dim, intermediate_size, num_heads, norm_eps, adaln_dim,
+                                      dtype=dtype, device=device, operations=operations)
+            for _ in range(num_layers)
+        ])
+
+        self.final_layer = Ideogram4FinalLayer(emb_dim, in_channels, adaln_dim, dtype=dtype, device=device, operations=operations)
+
+    def _backbone(self, llm_features, x, t, position_ids, attn_mask, indicator, transformer_options={}):
+        indicator = indicator.to(torch.long)
+        output_image_mask = (indicator == OUTPUT_IMAGE_INDICATOR).to(x.dtype).unsqueeze(-1)
+
+        x = x * output_image_mask
+        h = self.input_proj(x) * output_image_mask
+
+        t_cond = self.t_embedding(t, dtype=x.dtype)
+        if t.dim() == 1:
+            t_cond = t_cond.unsqueeze(1)
+        adaln_input = F.silu(self.adaln_proj(t_cond))
+
+        # h is zero on the text rows (content lives only on image rows), add writes the text features in place
+        if llm_features is not None:
+            L_text = llm_features.shape[1]
+            text_mask = (indicator[:, :L_text] == LLM_TOKEN_INDICATOR).to(x.dtype).unsqueeze(-1)
+            llm = self.llm_cond_norm(llm_features * text_mask)
+            llm = self.llm_cond_proj(llm) * text_mask
+            h[:, :L_text] = h[:, :L_text] + llm
+
+        h = h + self.embed_image_indicator((indicator == OUTPUT_IMAGE_INDICATOR).to(torch.long), out_dtype=h.dtype)
+
+        # Qwen3-VL interleaved MRoPE; position_ids (B, L, 3) -> (3, L) (same across batch).
+        freqs_cis = precompute_freqs_cis(
+            self.head_dim, position_ids[0].transpose(0, 1), self.rope_theta,
+            rope_dims=self.mrope_section, interleaved_mrope=True, device=position_ids.device,
+        )
+
+        if attn_mask is not None and attn_mask.dtype == torch.bool:
+            attn_mask = torch.zeros_like(attn_mask, dtype=h.dtype).masked_fill_(~attn_mask, -torch.finfo(h.dtype).max)
+
+        for layer in self.layers:
+            h = layer(h, attn_mask, freqs_cis, adaln_input, transformer_options=transformer_options)
+
+        return self.final_layer(h, adaln_input)
+
+
+class Ideogram4Transformer2DModel(Ideogram4Transformer):
+    """Ideogram 4 single-stream DiT.
+
+    Runs a packed ``[text, image]`` sequence when text context is supplied, or an image-only sequence when ``context is None``.
+    """
+
+    def __init__(self, image_model=None, in_channels=128, num_layers=34, num_attention_heads=18, attention_head_dim=256, intermediate_size=12288,
+                 adaln_dim=512, llm_features_dim=53248, rope_theta=5000000, mrope_section=(24, 20, 20), norm_eps=1e-5,
+                 dtype=None, device=None, operations=None, **kwargs):
+        emb_dim = num_attention_heads * attention_head_dim
+        super().__init__(
+            emb_dim=emb_dim, num_layers=num_layers, num_heads=num_attention_heads,
+            intermediate_size=intermediate_size, adaln_dim=adaln_dim, in_channels=in_channels,
+            llm_features_dim=llm_features_dim, rope_theta=rope_theta, mrope_section=mrope_section,
+            norm_eps=norm_eps, dtype=dtype, device=device, operations=operations)
+        self.dtype = dtype
+        self.in_channels = in_channels
+        self.out_channels = in_channels
+        # 128-dim token = patch (2x2) * ae_channels (32).
+        self.patch_size = 2
+        self.ae_channels = in_channels // (self.patch_size * self.patch_size)
+
+    def _img_to_tokens(self, x):
+        B, C, gh, gw = x.shape
+        x = x.view(B, self.ae_channels, self.patch_size, self.patch_size, gh, gw)
+        x = x.permute(0, 4, 5, 2, 3, 1)  # (B, gh, gw, pi, pj, c)
+        return x.reshape(B, gh * gw, C)
+
+    def _tokens_to_img(self, tokens, gh, gw):
+        B = tokens.shape[0]
+        C = tokens.shape[-1]
+        x = tokens.reshape(B, gh, gw, self.patch_size, self.patch_size, self.ae_channels)
+        x = x.permute(0, 5, 3, 4, 1, 2)  # (B, c, pi, pj, gh, gw)
+        return x.reshape(B, C, gh, gw)
+
+    def _image_position_ids(self, gh, gw, device):
+        h_idx = torch.arange(gh, device=device).view(-1, 1).expand(gh, gw).reshape(-1)
+        w_idx = torch.arange(gw, device=device).view(1, -1).expand(gh, gw).reshape(-1)
+        t_idx = torch.zeros_like(h_idx)
+        return torch.stack([t_idx, h_idx, w_idx], dim=1) + IMAGE_POSITION_OFFSET  # (L_img, 3)
+
+    def _run_conditional(self, x_chunk, context_chunk, attn_mask_chunk, t_chunk, gh, gw, transformer_options):
+        B = x_chunk.shape[0]
+        device = x_chunk.device
+        img_tokens = self._img_to_tokens(x_chunk)
+        L_img = img_tokens.shape[1]
+        L_text = context_chunk.shape[1]
+        L = L_text + L_img
+        latent_dim = img_tokens.shape[-1]
+
+        x_full = torch.zeros(B, L, latent_dim, dtype=img_tokens.dtype, device=device)
+        x_full[:, L_text:] = img_tokens
+
+        text_pos = torch.arange(L_text, device=device).view(-1, 1).expand(L_text, 3)
+        img_pos = self._image_position_ids(gh, gw, device)
+        position_ids = torch.cat([text_pos, img_pos], dim=0).unsqueeze(0).expand(B, L, 3)
+
+        indicator = torch.empty(B, L, dtype=torch.long, device=device)
+        indicator[:, :L_text] = LLM_TOKEN_INDICATOR
+        indicator[:, L_text:] = OUTPUT_IMAGE_INDICATOR
+
+        attn_mask = None
+        if attn_mask_chunk is not None:
+            segment_ids = torch.ones(B, L, dtype=torch.long, device=device)
+            pad = (attn_mask_chunk == 0)
+            segment_ids[:, :L_text][pad] = SEQUENCE_PADDING_INDICATOR
+            indicator[:, :L_text][pad] = 0
+            # Block-diagonal mask from segment ids: (B, 1, L, L), True = attend.
+            attn_mask = (segment_ids.unsqueeze(2) == segment_ids.unsqueeze(1)).unsqueeze(1)
+
+        out = self._backbone(context_chunk, x_full, t_chunk, position_ids, attn_mask, indicator,
+                             transformer_options=transformer_options)
+        return self._tokens_to_img(out[:, L_text:], gh, gw)
+
+    def _run_image_only(self, x_chunk, t_chunk, gh, gw, transformer_options):
+        B = x_chunk.shape[0]
+        device = x_chunk.device
+        img_tokens = self._img_to_tokens(x_chunk)
+        L_img = img_tokens.shape[1]
+
+        position_ids = self._image_position_ids(gh, gw, device).unsqueeze(0).expand(B, L_img, 3)
+        indicator = torch.full((B, L_img), OUTPUT_IMAGE_INDICATOR, dtype=torch.long, device=device)
+
+        # Image-only sequence is a single segment -> no mask, full attention, no LLM context.
+        out = self._backbone(None, img_tokens, t_chunk, position_ids, None, indicator, transformer_options=transformer_options)
+        return self._tokens_to_img(out, gh, gw)
+
+    def forward(self, x, timesteps, context=None, attention_mask=None, transformer_options={}, **kwargs):
+        return comfy.patcher_extension.WrapperExecutor.new_class_executor(
+            self._forward,
+            self,
+            comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.DIFFUSION_MODEL, transformer_options),
+        ).execute(x, timesteps, context, attention_mask, transformer_options, **kwargs)
+
+    def _forward(self, x, timesteps, context=None, attention_mask=None, transformer_options={}, **kwargs):
+        bs, c, gh, gw = x.shape
+
+        timesteps = 1.0 - timesteps
+
+        # unconditional pass
+        if context is None:
+            return -self._run_image_only(x, timesteps, gh, gw, transformer_options)
+
+        return -self._run_conditional(x, context, attention_mask, timesteps, gh, gw, transformer_options)
@@ -0,0 +1,290 @@
+"""Krea 2 (K2) — single-stream MMDiT.
+
+Text tokens produced by a Qwen3-VL-4B 12-layer ``txtfusion`` adapter and patchified image tokens are
+concatenated into one sequence and run through ``layers`` shared transformer blocks with
+AdaLN-single modulation, GQA + per-head QK-norm + sigmoid-gated attention, SwiGLU MLP, and 3-axis RoPE.
+"""
+
+from typing import Optional
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from einops import rearrange
+
+import comfy.model_management
+import comfy.patcher_extension
+import comfy.ldm.common_dit
+from comfy.ldm.flux.layers import EmbedND, timestep_embedding
+from comfy.ldm.flux.math import apply_rope
+from comfy.ldm.modules.attention import optimized_attention_masked
+
+
+class RMSNorm(nn.Module):
+    """RMSNorm with the reference ``(1 + scale)`` weight convention (scale stored zero-centered)."""
+
+    def __init__(self, features: int, eps: float = 1e-5, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.eps = eps
+        self.scale = nn.Parameter(torch.empty(features, device=device, dtype=dtype))
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        dtype = x.dtype
+        weight = comfy.model_management.cast_to(self.scale, dtype=torch.float32, device=x.device) + 1.0
+        return F.rms_norm(x.float(), (x.shape[-1],), weight=weight, eps=self.eps).to(dtype)
+
+
+class QKNorm(nn.Module):
+    def __init__(self, dim: int, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.qnorm = RMSNorm(dim, device=device, dtype=dtype, operations=operations)
+        self.knorm = RMSNorm(dim, device=device, dtype=dtype, operations=operations)
+
+    def forward(self, q, k):
+        return self.qnorm(q), self.knorm(k)
+
+
+class SwiGLU(nn.Module):
+    def __init__(self, features: int, multiplier: int, bias: bool = False, multiple: int = 128,
+                 device=None, dtype=None, operations=None):
+        super().__init__()
+        mlpdim = int(2 * features / 3) * multiplier
+        mlpdim = multiple * ((mlpdim + multiple - 1) // multiple)
+        self.gate = operations.Linear(features, mlpdim, bias=bias, device=device, dtype=dtype)
+        self.up = operations.Linear(features, mlpdim, bias=bias, device=device, dtype=dtype)
+        self.down = operations.Linear(mlpdim, features, bias=bias, device=device, dtype=dtype)
+
+    def forward(self, x):
+        return self.down(F.silu(self.gate(x)).mul_(self.up(x)))
+
+
+class Attention(nn.Module):
+    def __init__(self, dim: int, heads: int, kvheads: Optional[int] = None, bias: bool = False,
+                 device=None, dtype=None, operations=None):
+        super().__init__()
+        self.heads = heads
+        self.kvheads = kvheads if kvheads is not None else heads
+        self.headdim = dim // self.heads
+        self.wq = operations.Linear(dim, self.headdim * self.heads, bias=bias, device=device, dtype=dtype)
+        self.wk = operations.Linear(dim, self.headdim * self.kvheads, bias=bias, device=device, dtype=dtype)
+        self.wv = operations.Linear(dim, self.headdim * self.kvheads, bias=bias, device=device, dtype=dtype)
+        self.gate = operations.Linear(dim, dim, bias=bias, device=device, dtype=dtype)
+        self.qknorm = QKNorm(self.headdim, device=device, dtype=dtype, operations=operations)
+        self.wo = operations.Linear(dim, dim, bias=bias, device=device, dtype=dtype)
+
+    def forward(self, x, freqs=None, mask=None, transformer_options={}):
+        q, k, v, gate = self.wq(x), self.wk(x), self.wv(x), self.gate(x)
+        q = rearrange(q, "B L (H D) -> B H L D", H=self.heads)
+        k = rearrange(k, "B L (H D) -> B H L D", H=self.kvheads)
+        v = rearrange(v, "B L (H D) -> B H L D", H=self.kvheads)
+        q, k = self.qknorm(q, k)
+        if freqs is not None:
+            q, k = apply_rope(q, k, freqs)
+        if self.kvheads != self.heads:
+            rep = self.heads // self.kvheads
+            k = k.repeat_interleave(rep, dim=1)
+            v = v.repeat_interleave(rep, dim=1)
+        out = optimized_attention_masked(q, k, v, self.heads, mask=mask, skip_reshape=True,
+                                         transformer_options=transformer_options)
+        return self.wo(out * F.sigmoid(gate))
+
+
+class SimpleModulation(nn.Module):
+    def __init__(self, dim: int, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.lin = nn.Parameter(torch.empty(2, dim, device=device, dtype=dtype))
+
+    def forward(self, vec):
+        out = vec + comfy.model_management.cast_to(self.lin, dtype=vec.dtype, device=vec.device).unsqueeze(0)
+        scale, shift = out.chunk(2, dim=1)
+        return scale, shift
+
+
+class DoubleSharedModulation(nn.Module):
+    def __init__(self, dim: int, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.lin = nn.Parameter(torch.empty(6 * dim, device=device, dtype=dtype))
+
+    def forward(self, vec):
+        out = vec + comfy.model_management.cast_to(self.lin, dtype=vec.dtype, device=vec.device)
+        return out.chunk(6, dim=-1)
+
+
+class TextFusionBlock(nn.Module):
+    def __init__(self, features, heads, multiplier, bias=False, kvheads=None, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.prenorm = RMSNorm(features, device=device, dtype=dtype, operations=operations)
+        self.postnorm = RMSNorm(features, device=device, dtype=dtype, operations=operations)
+        self.attn = Attention(features, heads, kvheads=kvheads, bias=bias, device=device, dtype=dtype, operations=operations)
+        self.mlp = SwiGLU(features, multiplier, bias, device=device, dtype=dtype, operations=operations)
+
+    def forward(self, x, mask=None, transformer_options={}):
+        x = x + self.attn(self.prenorm(x), mask=mask, transformer_options=transformer_options)
+        x = x + self.mlp(self.postnorm(x))
+        return x
+
+
+class TextFusionTransformer(nn.Module):
+    def __init__(self, num_txt_layers, txt_dim, heads, multiplier, bias=False, kvheads=None, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.layerwise_blocks = nn.ModuleList([
+            TextFusionBlock(txt_dim, heads, multiplier, bias, kvheads, device=device, dtype=dtype, operations=operations)
+            for _ in range(2)
+        ])
+        self.projector = operations.Linear(num_txt_layers, 1, bias=False, device=device, dtype=dtype)
+        self.refiner_blocks = nn.ModuleList([
+            TextFusionBlock(txt_dim, heads, multiplier, bias, kvheads, device=device, dtype=dtype, operations=operations)
+            for _ in range(2)
+        ])
+
+    def forward(self, x, mask=None, transformer_options={}):
+        b, l, n, d = x.shape
+        x = x.reshape(b * l, n, d)
+        for block in self.layerwise_blocks:
+            x = block(x.contiguous(), mask=None, transformer_options=transformer_options)
+        x = rearrange(x, "(b l) n d -> b l d n", b=b, l=l)
+        x = self.projector(x).squeeze(-1)
+        for block in self.refiner_blocks:
+            x = block(x, mask=mask, transformer_options=transformer_options)
+        return x
+
+
+class SingleStreamBlock(nn.Module):
+    def __init__(self, features, heads, multiplier, bias=False, kvheads=None, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.mod = DoubleSharedModulation(features, device=device, dtype=dtype, operations=operations)
+        self.prenorm = RMSNorm(features, device=device, dtype=dtype, operations=operations)
+        self.postnorm = RMSNorm(features, device=device, dtype=dtype, operations=operations)
+        self.attn = Attention(features, heads, kvheads=kvheads, bias=bias, device=device, dtype=dtype, operations=operations)
+        self.mlp = SwiGLU(features, multiplier, bias, device=device, dtype=dtype, operations=operations)
+
+    def forward(self, x, vec, freqs, mask=None, transformer_options={}):
+        prescale, preshift, pregate, postscale, postshift, postgate = self.mod(vec)
+        x = x + pregate * self.attn((1 + prescale) * self.prenorm(x) + preshift, freqs, mask, transformer_options=transformer_options)
+        x = x + postgate * self.mlp((1 + postscale) * self.postnorm(x) + postshift)
+        return x
+
+
+class LastLayer(nn.Module):
+    def __init__(self, features, patch, channels, device=None, dtype=None, operations=None):
+        super().__init__()
+        self.norm = RMSNorm(features, device=device, dtype=dtype, operations=operations)
+        self.linear = operations.Linear(features, patch * patch * channels, bias=True, device=device, dtype=dtype)
+        self.modulation = SimpleModulation(features, device=device, dtype=dtype, operations=operations)
+
+    def forward(self, x, tvec):
+        scale, shift = self.modulation(tvec)
+        x = (1 + scale) * self.norm(x) + shift
+        return self.linear(x)
+
+
+class SingleStreamDiT(nn.Module):
+    def __init__(self, features=6144, tdim=256, txtdim=2560, heads=48, kvheads=12, multiplier=4,
+                 layers=28, patch=2, channels=16, bias=False, theta=1e3, txtlayers=12,
+                 txtheads=20, txtkvheads=20, image_model=None,
+                 device=None, dtype=None, operations=None, **kwargs):
+        super().__init__()
+        self.dtype = dtype
+        self.patch = patch
+        self.channels = channels
+        self.tdim = tdim
+        self.heads = heads
+        self.txtdim = txtdim
+        self.txtlayers = txtlayers
+
+        headdim = features // heads
+        axes = [headdim - 12 * (headdim // 16), 6 * (headdim // 16), 6 * (headdim // 16)]
+        assert sum(axes) == headdim, f"axes {axes} sum != headdim {headdim}"
+        self.pe_embedder = EmbedND(dim=headdim, theta=int(theta), axes_dim=axes)
+
+        self.first = operations.Linear(channels * patch ** 2, features, bias=True, device=device, dtype=dtype)
+        self.blocks = nn.ModuleList([
+            SingleStreamBlock(features, heads, multiplier, bias, kvheads, device=device, dtype=dtype, operations=operations)
+            for _ in range(layers)
+        ])
+        self.tmlp = nn.Sequential(
+            operations.Linear(tdim, features, device=device, dtype=dtype),
+            nn.GELU(approximate="tanh"),
+            operations.Linear(features, features, device=device, dtype=dtype),
+        )
+        self.txtfusion = TextFusionTransformer(txtlayers, txtdim, txtheads, multiplier, bias, txtkvheads,
+                                               device=device, dtype=dtype, operations=operations)
+        self.txtmlp = nn.Sequential(
+            RMSNorm(txtdim, device=device, dtype=dtype, operations=operations),
+            operations.Linear(txtdim, features, device=device, dtype=dtype),
+            nn.GELU(approximate="tanh"),
+            operations.Linear(features, features, device=device, dtype=dtype),
+        )
+        self.last = LastLayer(features, patch, channels, device=device, dtype=dtype, operations=operations)
+        self.tproj = nn.Sequential(
+            nn.GELU(approximate="tanh"),
+            operations.Linear(features, features * 6, device=device, dtype=dtype),
+        )
+
+    def forward(self, x, timesteps, context, attention_mask=None, transformer_options={}, **kwargs):
+        return comfy.patcher_extension.WrapperExecutor.new_class_executor(
+            self._forward,
+            self,
+            comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.DIFFUSION_MODEL, transformer_options),
+        ).execute(x, timesteps, context, attention_mask, transformer_options, **kwargs)
+
+    def _forward(self, x, timesteps, context, attention_mask=None, transformer_options={}, **kwargs):
+        temporal = x.ndim == 5
+        if temporal:
+            b5, c5, t5, h5, w5 = x.shape
+            x = x.reshape(b5 * t5, c5, h5, w5)
+        bs, c, H_orig, W_orig = x.shape
+        patch = self.patch
+        # Pad the latent up to a multiple of patch (as Flux/Lumina/QwenImage do); crop back at the end.
+        x = comfy.ldm.common_dit.pad_to_patch_size(x, (patch, patch))
+        H, W = x.shape[-2], x.shape[-1]
+        h_, w_ = H // patch, W // patch
+
+        # context arrives as (B, seq, txtlayers*txtdim); reshape to (B, txtlayers, seq, txtdim).
+        context = self._unpack_context(context)
+
+        img = rearrange(x, "b c (h ph) (w pw) -> b (h w) (c ph pw)", ph=patch, pw=patch)
+        img = self.first(img)
+
+        t = self.tmlp(timestep_embedding(timesteps, self.tdim).unsqueeze(1).to(img.dtype))
+        tvec = self.tproj(t)
+
+        context = self.txtfusion(context, mask=None, transformer_options=transformer_options)
+        context = self.txtmlp(context)
+
+        txtlen, imglen = context.shape[1], img.shape[1]
+        combined = torch.cat((context, img), dim=1)
+
+        # Position ids: text at 0, image at (0, h_idx, w_idx).
+        device = combined.device
+        txtpos = torch.zeros(bs, txtlen, 3, device=device, dtype=torch.float32)
+        imgids = torch.zeros(h_, w_, 3, device=device, dtype=torch.float32)
+        imgids[..., 1] = torch.arange(h_, device=device, dtype=torch.float32)[:, None]
+        imgids[..., 2] = torch.arange(w_, device=device, dtype=torch.float32)[None, :]
+        imgpos = imgids.reshape(1, h_ * w_, 3).repeat(bs, 1, 1)
+        pos = torch.cat((txtpos, imgpos), dim=1)
+
+        freqs = self.pe_embedder(pos)
+
+        for block in self.blocks:
+            combined = block(combined, tvec, freqs, None, transformer_options=transformer_options)
+
+        final = self.last(combined, t)
+        out = final[:, txtlen:txtlen + imglen, :]
+        out = rearrange(out, "b (h w) (c ph pw) -> b c (h ph) (w pw)",
+                        h=h_, w=w_, ph=patch, pw=patch, c=self.channels)
+        out = out[:, :, :H_orig, :W_orig]  # crop padding back off
+        if temporal:
+            out = out.reshape(b5, t5, self.channels, H_orig, W_orig).movedim(1, 2)
+        return out
+
+    def _unpack_context(self, context):
+        # context: (B, seq, txtlayers*txtdim) -> (B, seq, txtlayers, txtdim).
+        b, seq, fused = context.shape
+        if fused != self.txtlayers * self.txtdim:
+            raise ValueError(
+                f"Krea2 expects conditioning with {self.txtlayers}x{self.txtdim}={self.txtlayers * self.txtdim} "
+                f"features (a {self.txtlayers}-layer Qwen3-VL stack) but got {fused}. "
+                f"Load the text encoder with CLIPLoader type 'krea2'."
+            )
+        return context.reshape(b, seq, self.txtlayers, self.txtdim)
@@ -1085,7 +1085,7 @@ class LTXVModel(LTXBaseModel):
        )

        grid_mask = None
-        if keyframe_idxs is not None:
+        if keyframe_idxs is not None and keyframe_idxs.shape[2] > 0:
            additional_args.update({ "orig_patchified_shape": list(x.shape)})
            denoise_mask = self.patchifier.patchify(denoise_mask)[0]
            grid_mask = ~torch.any(denoise_mask < 0, dim=-1)[0]
@@ -1330,7 +1330,7 @@ class LTXVModel(LTXBaseModel):
        x = x * (1 + scale) + shift
        x = self.proj_out(x)

-        if keyframe_idxs is not None:
+        if keyframe_idxs is not None and keyframe_idxs.shape[2] > 0:
            grid_mask = kwargs["grid_mask"]
            orig_patchified_shape = kwargs["orig_patchified_shape"]
            full_x = torch.zeros(orig_patchified_shape, dtype=x.dtype, device=x.device)
@@ -1,5 +1,6 @@
 import math
 import sys
+import inspect

 import torch
 import torch.nn.functional as F
@@ -14,16 +15,16 @@ from .sub_quadratic_attention import efficient_dot_product_attention

 from comfy import model_management

-TORCH_HAS_GQA = model_management.torch_version_numeric >= (2, 5)
-
 if model_management.xformers_enabled():
    import xformers
    import xformers.ops

 SAGE_ATTENTION_IS_AVAILABLE = False
+SAGE_ATTENTION_SUPPORTS_MASK = False
 try:
    from sageattention import sageattn
    SAGE_ATTENTION_IS_AVAILABLE = True
+    SAGE_ATTENTION_SUPPORTS_MASK = "attn_mask" in inspect.signature(sageattn).parameters
 except ImportError as e:
    if model_management.sage_attention_enabled():
        if e.name == "sageattention":
@@ -89,6 +90,44 @@ def default(val, d):
        return val
    return d

+def _gqa_repeat_factor(query_heads, key_heads, value_heads):
+    if key_heads != value_heads:
+        raise ValueError(f"Key/value head count mismatch for GQA: {key_heads} != {value_heads}")
+    if query_heads == key_heads:
+        return 1
+    if query_heads % key_heads != 0:
+        raise ValueError(f"Query heads must be divisible by key/value heads for GQA: {query_heads} vs {key_heads}")
+    return query_heads // key_heads
+
+def _repeat_kv_for_gqa(k, v, query_heads, head_dim):
+    n_rep = _gqa_repeat_factor(query_heads, k.shape[head_dim], v.shape[head_dim])
+    if n_rep > 1:
+        k = k.repeat_interleave(n_rep, dim=head_dim)
+        v = v.repeat_interleave(n_rep, dim=head_dim)
+    return k, v
+
+def _heads_from_dim(tensor, dim_head, name):
+    inner_dim = tensor.shape[-1]
+    if inner_dim % dim_head != 0:
+        raise ValueError(f"{name} inner dimension {inner_dim} is not divisible by head dimension {dim_head}")
+    return inner_dim // dim_head
+
+def _reshape_qkv_to_heads(q, k, v, b, heads, dim_head, enable_gqa=False, expand_kv=True):
+    q = q.unsqueeze(3).reshape(b, -1, heads, dim_head)
+    if enable_gqa:
+        key_heads = _heads_from_dim(k, dim_head, "Key")
+        value_heads = _heads_from_dim(v, dim_head, "Value")
+    else:
+        key_heads = heads
+        value_heads = heads
+    k = k.unsqueeze(3).reshape(b, -1, key_heads, dim_head)
+    v = v.unsqueeze(3).reshape(b, -1, value_heads, dim_head)
+    if enable_gqa:
+        _gqa_repeat_factor(heads, key_heads, value_heads)
+        if expand_kv:
+            k, v = _repeat_kv_for_gqa(k, v, heads, -2)
+    return q, k, v
+

 # feedforward
 class GEGLU(nn.Module):
@@ -152,28 +191,19 @@ def attention_basic(q, k, v, heads, mask=None, attn_precision=None, skip_reshape
        b, _, dim_head = q.shape
        dim_head //= heads

-    if kwargs.get("enable_gqa", False) and q.shape[-3] != k.shape[-3]:
-        n_rep = q.shape[-3] // k.shape[-3]
-        k = k.repeat_interleave(n_rep, dim=-3)
-        v = v.repeat_interleave(n_rep, dim=-3)
-
    scale = kwargs.get("scale", dim_head ** -0.5)

    h = heads
    if skip_reshape:
-         q, k, v = map(
+        if kwargs.get("enable_gqa", False):
+            k, v = _repeat_kv_for_gqa(k, v, q.shape[-3], -3)
+        q, k, v = map(
            lambda t: t.reshape(b * heads, -1, dim_head),
            (q, k, v),
        )
    else:
-        q, k, v = map(
-            lambda t: t.unsqueeze(3)
-            .reshape(b, -1, heads, dim_head)
-            .permute(0, 2, 1, 3)
-            .reshape(b * heads, -1, dim_head)
-            .contiguous(),
-            (q, k, v),
-        )
+        q, k, v = _reshape_qkv_to_heads(q, k, v, b, heads, dim_head, kwargs.get("enable_gqa", False))
+        q, k, v = map(lambda t: t.permute(0, 2, 1, 3).reshape(b * heads, -1, dim_head).contiguous(), (q, k, v))

    # force cast to fp32 to avoid overflowing
    if attn_precision == torch.float32:
@@ -231,13 +261,16 @@ def attention_sub_quad(query, key, value, heads, mask=None, attn_precision=None,
        query = query * (kwargs["scale"] * dim_head ** 0.5)

    if skip_reshape:
+        if kwargs.get("enable_gqa", False):
+            key, value = _repeat_kv_for_gqa(key, value, query.shape[-3], -3)
        query = query.reshape(b * heads, -1, dim_head)
        value = value.reshape(b * heads, -1, dim_head)
        key = key.reshape(b * heads, -1, dim_head).movedim(1, 2)
    else:
-        query = query.unsqueeze(3).reshape(b, -1, heads, dim_head).permute(0, 2, 1, 3).reshape(b * heads, -1, dim_head)
-        value = value.unsqueeze(3).reshape(b, -1, heads, dim_head).permute(0, 2, 1, 3).reshape(b * heads, -1, dim_head)
-        key = key.unsqueeze(3).reshape(b, -1, heads, dim_head).permute(0, 2, 3, 1).reshape(b * heads, dim_head, -1)
+        query, key, value = _reshape_qkv_to_heads(query, key, value, b, heads, dim_head, kwargs.get("enable_gqa", False))
+        query = query.permute(0, 2, 1, 3).reshape(b * heads, -1, dim_head)
+        value = value.permute(0, 2, 1, 3).reshape(b * heads, -1, dim_head)
+        key = key.permute(0, 2, 3, 1).reshape(b * heads, dim_head, -1)


    dtype = query.dtype
@@ -304,19 +337,15 @@ def attention_split(q, k, v, heads, mask=None, attn_precision=None, skip_reshape
    scale = kwargs.get("scale", dim_head ** -0.5)

    if skip_reshape:
-         q, k, v = map(
+        if kwargs.get("enable_gqa", False):
+            k, v = _repeat_kv_for_gqa(k, v, q.shape[-3], -3)
+        q, k, v = map(
            lambda t: t.reshape(b * heads, -1, dim_head),
            (q, k, v),
        )
    else:
-        q, k, v = map(
-            lambda t: t.unsqueeze(3)
-            .reshape(b, -1, heads, dim_head)
-            .permute(0, 2, 1, 3)
-            .reshape(b * heads, -1, dim_head)
-            .contiguous(),
-            (q, k, v),
-        )
+        q, k, v = _reshape_qkv_to_heads(q, k, v, b, heads, dim_head, kwargs.get("enable_gqa", False))
+        q, k, v = map(lambda t: t.permute(0, 2, 1, 3).reshape(b * heads, -1, dim_head).contiguous(), (q, k, v))

    r1 = torch.zeros(q.shape[0], q.shape[1], v.shape[2], device=q.device, dtype=q.dtype)

@@ -438,7 +467,7 @@ def attention_xformers(q, k, v, heads, mask=None, attn_precision=None, skip_resh
            disabled_xformers = True

    if disabled_xformers:
-        return attention_pytorch(q, k, v, heads, mask, skip_reshape=skip_reshape, **kwargs)
+        return attention_pytorch(q, k, v, heads, mask, skip_reshape=skip_reshape, skip_output_reshape=skip_output_reshape, **kwargs)

    if skip_reshape:
        # b h k d -> b k h d
@@ -446,13 +475,12 @@ def attention_xformers(q, k, v, heads, mask=None, attn_precision=None, skip_resh
            lambda t: t.permute(0, 2, 1, 3),
            (q, k, v),
        )
+        if kwargs.get("enable_gqa", False):
+            k, v = _repeat_kv_for_gqa(k, v, q.shape[-2], -2)
    # actually do the reshaping
    else:
        dim_head //= heads
-        q, k, v = map(
-            lambda t: t.reshape(b, -1, heads, dim_head),
-            (q, k, v),
-        )
+        q, k, v = _reshape_qkv_to_heads(q, k, v, b, heads, dim_head, kwargs.get("enable_gqa", False))

    if mask is not None:
        # add a singleton batch dimension
@@ -474,7 +502,7 @@ def attention_xformers(q, k, v, heads, mask=None, attn_precision=None, skip_resh
        mask = mask_out[..., :mask.shape[-1]]
        mask = mask.expand(b, heads, -1, -1)

-    out = xformers.ops.memory_efficient_attention(q, k, v, attn_bias=mask)
+    out = xformers.ops.memory_efficient_attention(q, k, v, attn_bias=mask, scale=kwargs.get("scale", None))

    if skip_output_reshape:
        out = out.permute(0, 2, 1, 3)
@@ -498,10 +526,8 @@ def attention_pytorch(q, k, v, heads, mask=None, attn_precision=None, skip_resha
    else:
        b, _, dim_head = q.shape
        dim_head //= heads
-        q, k, v = map(
-            lambda t: t.view(b, -1, heads, dim_head).transpose(1, 2),
-            (q, k, v),
-        )
+        q, k, v = _reshape_qkv_to_heads(q, k, v, b, heads, dim_head, kwargs.get("enable_gqa", False), expand_kv=False)
+        q, k, v = map(lambda t: t.transpose(1, 2), (q, k, v))

    if mask is not None:
        # add a batch dimension if there isn't already one
@@ -511,9 +537,7 @@ def attention_pytorch(q, k, v, heads, mask=None, attn_precision=None, skip_resha
        if mask.ndim == 3:
            mask = mask.unsqueeze(1)

-    # Pass through extra SDPA kwargs (scale, enable_gqa) if provided
-    # enable_gqa requires PyTorch 2.5+; older versions use manual KV expansion above
-    sdpa_keys = ("scale", "enable_gqa") if TORCH_HAS_GQA else ("scale",)
+    sdpa_keys = ("scale", "enable_gqa")
    sdpa_extra = {k: v for k, v in kwargs.items() if k in sdpa_keys}

    if SDP_BATCH_LIMIT >= b:
@@ -541,20 +565,19 @@ def attention_pytorch(q, k, v, heads, mask=None, attn_precision=None, skip_resha

@wrap_attn
 def attention_sage(q, k, v, heads, mask=None, attn_precision=None, skip_reshape=False, skip_output_reshape=False, **kwargs):
-    if kwargs.get("low_precision_attention", True) is False:
+    if kwargs.get("low_precision_attention", True) is False or (mask is not None and not SAGE_ATTENTION_SUPPORTS_MASK):
        return attention_pytorch(q, k, v, heads, mask=mask, skip_reshape=skip_reshape, skip_output_reshape=skip_output_reshape, **kwargs)

    exception_fallback = False
    if skip_reshape:
        b, _, _, dim_head = q.shape
        tensor_layout = "HND"
+        if kwargs.get("enable_gqa", False):
+            k, v = _repeat_kv_for_gqa(k, v, q.shape[-3], -3)
    else:
        b, _, dim_head = q.shape
        dim_head //= heads
-        q, k, v = map(
-            lambda t: t.view(b, -1, heads, dim_head),
-            (q, k, v),
-        )
+        q, k, v = _reshape_qkv_to_heads(q, k, v, b, heads, dim_head, kwargs.get("enable_gqa", False))
        tensor_layout = "NHD"

    if mask is not None:
@@ -565,8 +588,12 @@ def attention_sage(q, k, v, heads, mask=None, attn_precision=None, skip_reshape=
        if mask.ndim == 3:
            mask = mask.unsqueeze(1)

+    sage_kwargs = {"is_causal": False, "tensor_layout": tensor_layout, "sm_scale": kwargs.get("scale", None), "smooth_k": False}
+    if mask is not None:
+        sage_kwargs["attn_mask"] = mask
+
    try:
-        out = sageattn(q, k, v, attn_mask=mask, is_causal=False, tensor_layout=tensor_layout)
+        out = sageattn(q, k, v, **sage_kwargs)
    except Exception as e:
        logging.error("Error running sage attention: {}, using pytorch attention instead.".format(e))
        exception_fallback = True
@@ -616,7 +643,6 @@ def attention3_sage(q, k, v, heads, mask=None, attn_precision=None, skip_reshape
                skip_output_reshape=skip_output_reshape,
                **kwargs
            )
-        q_s, k_s, v_s = q, k, v
        N = q.shape[2]
        dim_head = D
    else:
@@ -642,11 +668,15 @@ def attention3_sage(q, k, v, heads, mask=None, attn_precision=None, skip_reshape
                **kwargs
            )

-    if not skip_reshape:
-        q_s, k_s, v_s = map(
-            lambda t: t.view(B, -1, heads, dim_head).permute(0, 2, 1, 3).contiguous(),
-            (q, k, v),
-        )
+    if skip_reshape:
+        q_s = q
+        if kwargs.get("enable_gqa", False):
+            k_s, v_s = _repeat_kv_for_gqa(k, v, H, -3)
+        else:
+            k_s, v_s = k, v
+    else:
+        q_s, k_s, v_s = _reshape_qkv_to_heads(q, k, v, B, heads, dim_head, kwargs.get("enable_gqa", False))
+        q_s, k_s, v_s = map(lambda t: t.permute(0, 2, 1, 3).contiguous(), (q_s, k_s, v_s))
        B, H, L, D = q_s.shape

    try:
@@ -662,7 +692,7 @@ def attention3_sage(q, k, v, heads, mask=None, attn_precision=None, skip_reshape
                q, k, v, heads,
                mask=mask,
                attn_precision=attn_precision,
-                skip_reshape=False,
+                skip_reshape=skip_reshape,
                skip_output_reshape=skip_output_reshape,
                **kwargs
            )
@@ -681,19 +711,20 @@ def attention3_sage(q, k, v, heads, mask=None, attn_precision=None, skip_reshape
 try:
    @torch.library.custom_op("flash_attention::flash_attn", mutates_args=())
    def flash_attn_wrapper(q: torch.Tensor, k: torch.Tensor, v: torch.Tensor,
-                    dropout_p: float = 0.0, causal: bool = False) -> torch.Tensor:
-        return flash_attn_func(q, k, v, dropout_p=dropout_p, causal=causal)
+                    dropout_p: float = 0.0, causal: bool = False, softmax_scale: float = -1.0) -> torch.Tensor:
+        softmax_scale_arg = None if softmax_scale == -1.0 else softmax_scale
+        return flash_attn_func(q, k, v, dropout_p=dropout_p, causal=causal, softmax_scale=softmax_scale_arg)


    @flash_attn_wrapper.register_fake
-    def flash_attn_fake(q, k, v, dropout_p=0.0, causal=False):
+    def flash_attn_fake(q, k, v, dropout_p=0.0, causal=False, softmax_scale=-1.0):
        # Output shape is the same as q
        return q.new_empty(q.shape)
 except AttributeError as error:
    FLASH_ATTN_ERROR = error

    def flash_attn_wrapper(q: torch.Tensor, k: torch.Tensor, v: torch.Tensor,
-                    dropout_p: float = 0.0, causal: bool = False) -> torch.Tensor:
+                    dropout_p: float = 0.0, causal: bool = False, softmax_scale: float = -1.0) -> torch.Tensor:
        assert False, f"Could not define flash_attn_wrapper: {FLASH_ATTN_ERROR}"

@wrap_attn
@@ -703,10 +734,8 @@ def attention_flash(q, k, v, heads, mask=None, attn_precision=None, skip_reshape
    else:
        b, _, dim_head = q.shape
        dim_head //= heads
-        q, k, v = map(
-            lambda t: t.view(b, -1, heads, dim_head).transpose(1, 2),
-            (q, k, v),
-        )
+        q, k, v = _reshape_qkv_to_heads(q, k, v, b, heads, dim_head, kwargs.get("enable_gqa", False), expand_kv=False)
+        q, k, v = map(lambda t: t.transpose(1, 2), (q, k, v))

    if mask is not None:
        # add a batch dimension if there isn't already one
@@ -725,10 +754,16 @@ def attention_flash(q, k, v, heads, mask=None, attn_precision=None, skip_reshape
            v.transpose(1, 2),
            dropout_p=0.0,
            causal=False,
+            softmax_scale=kwargs.get("scale", -1.0),
        ).transpose(1, 2)
    except Exception as e:
        logging.warning(f"Flash Attention failed, using default SDPA: {e}")
-        out = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False)
+        sdpa_extra = {}
+        if kwargs.get("enable_gqa", False):
+            sdpa_extra["enable_gqa"] = True
+        if "scale" in kwargs:
+            sdpa_extra["scale"] = kwargs["scale"]
+        out = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False, **sdpa_extra)
    if not skip_output_reshape:
        out = (
            out.transpose(1, 2).reshape(b, -1, heads * dim_head)
@@ -1209,5 +1244,3 @@ class SpatialVideoTransformer(SpatialTransformer):
            x = self.proj_out(x)
        out = x + x_in
        return out
-
-
@@ -211,7 +211,7 @@ class TimestepEmbedder(nn.Module):
    Embeds scalar timesteps into vector representations.
    """

-    def __init__(self, hidden_size, frequency_embedding_size=256, output_size=None, dtype=None, device=None, operations=None):
+    def __init__(self, hidden_size, frequency_embedding_size=256, output_size=None, dtype=None, device=None, operations=None, max_period=10000):
        super().__init__()
        if output_size is None:
            output_size = hidden_size
@@ -221,9 +221,10 @@ class TimestepEmbedder(nn.Module):
            operations.Linear(hidden_size, output_size, bias=True, dtype=dtype, device=device),
        )
        self.frequency_embedding_size = frequency_embedding_size
+        self.max_period = max_period

    def forward(self, t, dtype, **kwargs):
-        t_freq = timestep_embedding(t, self.frequency_embedding_size).to(dtype)
+        t_freq = timestep_embedding(t, self.frequency_embedding_size, max_period=self.max_period).to(dtype)
        t_emb = self.mlp(t_freq)
        return t_emb

@@ -8,6 +8,7 @@ import torch.nn.functional as F
 from einops import rearrange, repeat
 from comfy.ldm.lightricks.model import Timesteps
 from comfy.ldm.flux.layers import EmbedND
+from comfy.ldm.flux.math import apply_rope1
 from comfy.ldm.modules.attention import optimized_attention_masked
 import comfy.model_management
 import comfy.ldm.common_dit
@@ -17,13 +18,11 @@ def apply_rotary_emb(x, freqs_cis):
    if x.shape[1] == 0:
        return x

-    t_ = x.reshape(*x.shape[:-1], -1, 1, 2)
-    t_out = freqs_cis[..., 0] * t_[..., 0] + freqs_cis[..., 1] * t_[..., 1]
-    return t_out.reshape(*x.shape).to(dtype=x.dtype)
+    return apply_rope1(x, freqs_cis)


 def swiglu(x: torch.Tensor, y: torch.Tensor) -> torch.Tensor:
-    return F.silu(x) * y
+    return F.silu(x, inplace=True).mul_(y)


 class TimestepEmbedding(nn.Module):
@@ -142,11 +141,8 @@ class Attention(nn.Module):
        key = key.transpose(1, 2)
        value = value.transpose(1, 2)

-        if self.kv_heads < self.heads:
-            key = key.repeat_interleave(self.heads // self.kv_heads, dim=1)
-            value = value.repeat_interleave(self.heads // self.kv_heads, dim=1)
-
-        hidden_states = optimized_attention_masked(query, key, value, self.heads, attention_mask, skip_reshape=True, transformer_options=transformer_options)
+        gqa_kwargs = {"enable_gqa": True} if self.kv_heads < self.heads else {}
+        hidden_states = optimized_attention_masked(query, key, value, self.heads, attention_mask, skip_reshape=True, transformer_options=transformer_options, **gqa_kwargs)
        hidden_states = self.to_out[0](hidden_states)
        return hidden_states

@@ -0,0 +1,239 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+import comfy.ldm.common_dit
+import comfy.patcher_extension
+from comfy.ldm.flux.math import apply_rope, rope
+from comfy.ldm.hidream.model import FeedForwardSwiGLU
+from comfy.ldm.modules.attention import optimized_attention
+from comfy.ldm.modules.diffusionmodules.mmdit import TimestepEmbedder
+
+from .modules import (
+    FinalLayer,
+    PatchTokenEmbedder,
+    PiTBlock,
+    PixelTokenEmbedder,
+    apply_adaln_,
+    precompute_freqs_cis_2d,
+)
+
+
+class MMDiTJointAttention(nn.Module):
+    """Joint MMDiT attention with separate Q/K/V/proj for image and text streams.
+
+    RoPE is applied to each stream before concatenation so each stream uses its own
+    2D/1D positional encoding. Concat order is [text, image] (text first).
+    """
+    def __init__(self, dim, num_heads=8, qkv_bias=False, dtype=None, device=None, operations=None):
+        super().__init__()
+        assert dim % num_heads == 0
+        self.num_heads = num_heads
+        self.head_dim = dim // num_heads
+
+        self.qkv_x = operations.Linear(dim, dim * 3, bias=qkv_bias, dtype=dtype, device=device)
+        self.qkv_y = operations.Linear(dim, dim * 3, bias=qkv_bias, dtype=dtype, device=device)
+
+        self.q_norm_x = operations.RMSNorm(self.head_dim, eps=1e-6, dtype=dtype, device=device)
+        self.k_norm_x = operations.RMSNorm(self.head_dim, eps=1e-6, dtype=dtype, device=device)
+        self.q_norm_y = operations.RMSNorm(self.head_dim, eps=1e-6, dtype=dtype, device=device)
+        self.k_norm_y = operations.RMSNorm(self.head_dim, eps=1e-6, dtype=dtype, device=device)
+
+        self.proj_x = operations.Linear(dim, dim, dtype=dtype, device=device)
+        self.proj_y = operations.Linear(dim, dim, dtype=dtype, device=device)
+
+    def forward(self, x, y, pos_img, pos_txt=None, attn_mask=None, transformer_options={}):
+        B, Nx, _ = x.shape
+        _, Ny, _ = y.shape
+        H = self.num_heads
+        D = self.head_dim
+
+        qkv_x = self.qkv_x(x).reshape(B, Nx, 3, H, D).permute(2, 0, 3, 1, 4)
+        qx, kx, vx = qkv_x.unbind(0)
+        qx = self.q_norm_x(qx)
+        kx = self.k_norm_x(kx)
+
+        qkv_y = self.qkv_y(y).reshape(B, Ny, 3, H, D).permute(2, 0, 3, 1, 4)
+        qy, ky, vy = qkv_y.unbind(0)
+        qy = self.q_norm_y(qy)
+        ky = self.k_norm_y(ky)
+
+        qx, kx = apply_rope(qx, kx, pos_img[None, None])
+        if pos_txt is not None:
+            qy, ky = apply_rope(qy, ky, pos_txt[None, None])
+
+        q_joint = torch.cat([qy, qx], dim=2)
+        k_joint = torch.cat([ky, kx], dim=2)
+        v_joint = torch.cat([vy, vx], dim=2)
+
+        out_joint = optimized_attention(
+            q_joint, k_joint, v_joint, H,
+            mask=attn_mask, skip_reshape=True, skip_output_reshape=True,
+            transformer_options=transformer_options,
+        )
+
+        out_y = out_joint[:, :, :Ny, :].transpose(1, 2).reshape(B, Ny, H * D)
+        out_x = out_joint[:, :, Ny:, :].transpose(1, 2).reshape(B, Nx, H * D)
+
+        return self.proj_x(out_x), self.proj_y(out_y)
+
+
+class MMDiTBlockT2I(nn.Module):
+    def __init__(self, hidden_size, groups, mlp_ratio=4.0, dtype=None, device=None, operations=None):
+        super().__init__()
+        self.norm_x1 = operations.RMSNorm(hidden_size, eps=1e-6, dtype=dtype, device=device)
+        self.norm_y1 = operations.RMSNorm(hidden_size, eps=1e-6, dtype=dtype, device=device)
+        self.attn = MMDiTJointAttention(hidden_size, num_heads=groups, qkv_bias=False, dtype=dtype, device=device, operations=operations)
+        self.norm_x2 = operations.RMSNorm(hidden_size, eps=1e-6, dtype=dtype, device=device)
+        self.norm_y2 = operations.RMSNorm(hidden_size, eps=1e-6, dtype=dtype, device=device)
+        mlp_hidden_dim = int(hidden_size * mlp_ratio)
+        self.mlp_x = FeedForwardSwiGLU(hidden_size, mlp_hidden_dim, multiple_of=1, dtype=dtype, device=device, operations=operations)
+        self.mlp_y = FeedForwardSwiGLU(hidden_size, mlp_hidden_dim, multiple_of=1, dtype=dtype, device=device, operations=operations)
+        self.adaLN_modulation_img = nn.Sequential(operations.Linear(hidden_size, 6 * hidden_size, bias=True, dtype=dtype, device=device))
+        self.adaLN_modulation_txt = nn.Sequential(operations.Linear(hidden_size, 6 * hidden_size, bias=True, dtype=dtype, device=device))
+
+    def forward(self, x, y, c, pos_img, pos_txt=None, attn_mask=None, transformer_options={}):
+        shift_msa_x, scale_msa_x, gate_msa_x, shift_mlp_x, scale_mlp_x, gate_mlp_x = self.adaLN_modulation_img(c).chunk(6, dim=-1)
+        shift_msa_y, scale_msa_y, gate_msa_y, shift_mlp_y, scale_mlp_y, gate_mlp_y = self.adaLN_modulation_txt(c).chunk(6, dim=-1)
+
+        x_norm = apply_adaln_(self.norm_x1(x), shift_msa_x, scale_msa_x)
+        y_norm = apply_adaln_(self.norm_y1(y), shift_msa_y, scale_msa_y)
+        attn_x, attn_y = self.attn(x_norm, y_norm, pos_img, pos_txt, attn_mask, transformer_options=transformer_options)
+        x = torch.addcmul(x, gate_msa_x, attn_x)
+        y = torch.addcmul(y, gate_msa_y, attn_y)
+
+        x = torch.addcmul(x, gate_mlp_x, self.mlp_x(apply_adaln_(self.norm_x2(x), shift_mlp_x, scale_mlp_x)))
+        y = torch.addcmul(y, gate_mlp_y, self.mlp_y(apply_adaln_(self.norm_y2(y), shift_mlp_y, scale_mlp_y)))
+        return x, y
+
+
+class PixDiT_T2I(nn.Module):
+    """PixelDiT T2I model. Hardcoded for the released 1024px Stage-3 checkpoint
+    (also runs at 512px when fed the appropriate latent size and flow_shift).
+
+    Forward:
+      x:        [B, 3, H, W] pixel-space input (no VAE)
+      timesteps:[B] in [0, 1000] (ComfyUI flow sampling convention)
+      context:  [B, Ltxt, 2304] Gemma-2-2b-it hidden states (chi_prompt prepended)
+    Returns flow-matching velocity [B, 3, H, W].
+    """
+    def __init__(
+        self,
+        in_channels=3,
+        num_groups=24,
+        hidden_size=1536,
+        pixel_hidden_size=16,
+        pixel_attn_hidden_size=1152,
+        pixel_num_groups=16,
+        patch_depth=14,
+        pixel_depth=2,
+        patch_size=16,
+        txt_embed_dim=2304,
+        txt_max_length=300,
+        use_text_rope=True,
+        text_rope_theta=10000.0,
+        image_model=None,
+        dtype=None,
+        device=None,
+        operations=None,
+        pixel_mlp_chunks=2,
+    ):
+        super().__init__()
+        self.dtype = dtype
+        self.in_channels = in_channels
+        self.out_channels = in_channels
+        self.hidden_size = hidden_size
+        self.num_groups = num_groups
+        self.patch_depth = patch_depth
+        self.pixel_depth = pixel_depth
+        self.patch_size = patch_size
+        self.pixel_hidden_size = pixel_hidden_size
+        self.pixel_attn_hidden_size = pixel_attn_hidden_size
+        self.pixel_num_groups = pixel_num_groups
+        self.txt_embed_dim = txt_embed_dim
+        self.txt_max_length = txt_max_length
+        self.use_text_rope = use_text_rope
+        self.text_rope_theta = text_rope_theta
+
+        self.pixel_embedder = PixelTokenEmbedder(self.in_channels, self.pixel_hidden_size, dtype=dtype, device=device, operations=operations)
+        self.s_embedder = PatchTokenEmbedder(self.in_channels * self.patch_size ** 2, self.hidden_size, bias=True, dtype=dtype, device=device, operations=operations)
+        self.t_embedder = TimestepEmbedder(self.hidden_size, dtype=dtype, device=device, operations=operations, max_period=10)
+        self.y_embedder = PatchTokenEmbedder(self.txt_embed_dim, self.hidden_size, bias=True, use_norm=True, dtype=dtype, device=device, operations=operations)
+        self.y_pos_embedding = nn.Parameter(torch.empty(1, self.txt_max_length, self.hidden_size, dtype=dtype, device=device))
+
+        self.patch_blocks = nn.ModuleList([
+            MMDiTBlockT2I(self.hidden_size, self.num_groups,
+                          dtype=dtype, device=device, operations=operations)
+            for _ in range(self.patch_depth)
+        ])
+        self.pixel_blocks = nn.ModuleList([
+            PiTBlock(
+                self.pixel_hidden_size,
+                self.hidden_size,
+                patch_size=self.patch_size,
+                num_heads=self.num_groups,
+                attn_hidden_size=self.pixel_attn_hidden_size,
+                attn_num_heads=self.pixel_num_groups,
+                dtype=dtype, device=device, operations=operations,
+                mlp_chunks=pixel_mlp_chunks,
+            )
+            for _ in range(self.pixel_depth)
+        ])
+
+        self.final_layer = FinalLayer(self.pixel_hidden_size, self.out_channels, dtype=dtype, device=device, operations=operations)
+
+    def _fetch_patch_pos(self, height, width, device, dtype, **rope_opts):
+        return precompute_freqs_cis_2d(self.hidden_size // self.num_groups, height, width, device=device, dtype=dtype, **rope_opts)
+
+    def _fetch_text_pos(self, length, device, dtype):
+        return rope(torch.arange(length, dtype=torch.float32, device=device).reshape(1, -1), self.hidden_size // self.num_groups, self.text_rope_theta).squeeze(0).to(dtype=dtype)
+
+    def forward(self, x, timesteps, context=None, attention_mask=None, transformer_options={}, **kwargs):
+        return comfy.patcher_extension.WrapperExecutor.new_class_executor(
+            self._forward, self, comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.DIFFUSION_MODEL, transformer_options),
+        ).execute(x, timesteps, context, attention_mask, transformer_options, **kwargs)
+
+    def _pre_patch_block(self, s, i, **kwargs):
+        """Hook for subclasses to inject per-block state into the patch stream (e.g. PiD's LQ gate)."""
+        return s
+
+    def _forward(self, x, timesteps, context=None, attention_mask=None, transformer_options={}, **kwargs):
+        H_orig, W_orig = x.shape[2], x.shape[3]
+        x = comfy.ldm.common_dit.pad_to_patch_size(x, (self.patch_size, self.patch_size))
+        B, _, H, W = x.shape
+        Hs = H // self.patch_size
+        Ws = W // self.patch_size
+        L = Hs * Ws
+
+        pos_img = self._fetch_patch_pos(Hs, Ws, x.device, x.dtype, **(transformer_options.get("rope_options") or {}))
+        x_patches = F.unfold(x, kernel_size=self.patch_size, stride=self.patch_size).transpose(1, 2)
+
+        t_emb = self.t_embedder(timesteps.view(-1), x.dtype).view(B, -1, self.hidden_size)
+
+        if context is None or context.dim() != 3:
+            raise ValueError("PixDiT_T2I requires context (text embeddings) of shape [B, L, D]")
+        Ltxt = min(context.shape[1], self.txt_max_length)
+        y = context[:, :Ltxt, :]
+        y_emb = self.y_embedder(y).view(B, Ltxt, self.hidden_size)
+        y_emb = y_emb + self.y_pos_embedding[:, :Ltxt, :].to(y_emb) # y_pos_embedding is a raw nn.Parameter
+
+        condition = F.silu(t_emb)
+        pos_txt = self._fetch_text_pos(Ltxt, x.device, x.dtype) if self.use_text_rope else None
+
+        s = self.s_embedder(x_patches)
+        for i, blk in enumerate(self.patch_blocks):
+            s = self._pre_patch_block(s, i, **kwargs)
+            s, y_emb = blk(s, y_emb, condition, pos_img, pos_txt, None, transformer_options=transformer_options)
+        s = F.silu(t_emb + s)
+
+        s_cond = s.view(B * L, self.hidden_size)
+        x_pixels = self.pixel_embedder(x, patch_size=self.patch_size)
+        for blk in self.pixel_blocks:
+            x_pixels = blk(x_pixels, s_cond, H, W, self.patch_size, mask=None, transformer_options=transformer_options)
+
+        x_pixels = self.final_layer(x_pixels)
+        C_out = self.out_channels
+        P2 = self.patch_size * self.patch_size
+        x_pixels = x_pixels.view(B, L, P2, C_out).permute(0, 3, 2, 1).reshape(B, C_out * P2, L)
+        out = F.fold(x_pixels, (H, W), kernel_size=self.patch_size, stride=self.patch_size)
+        return out[:, :, :H_orig, :W_orig]
@@ -0,0 +1,187 @@
+import torch
+import torch.nn as nn
+
+from comfy.ldm.flux.math import apply_rope, rope
+from comfy.ldm.modules.attention import optimized_attention
+from comfy.ldm.modules.diffusionmodules.mmdit import Mlp, get_1d_sincos_pos_embed_from_grid_torch
+
+
+def apply_adaln_(x, shift, scale):
+    return x.addcmul_(x, scale).add_(shift)
+
+
+def precompute_freqs_cis_2d(dim, height, width, theta=10000.0, scale=16.0,
+                            ref_grid_h=None, ref_grid_w=None,
+                            scale_x=1.0, scale_y=1.0, shift_x=0.0, shift_y=0.0,
+                            device=None, dtype=torch.float32, **kwargs):
+    """2D RoPE with x/y axis frequencies interleaved at stride 2 across head dim.
+
+    rope_options:
+      scale_x / scale_y multiply the position range (RoPE extrapolation).
+      shift_x / shift_y offset the position origin (tiled / regional inference).
+    With ref_grid_h/w set, also applies NTK-aware per-axis theta scaling
+    (rope_mode='ntk_aware'): theta_axis = theta * (current/ref)^(dim_axis/(dim_axis-2)).
+    Returns Flux-format rotation matrices of shape [H*W, dim/2, 2, 2].
+    Layout of head-dim pairs: [x_0, y_0, x_1, y_1, ..., x_{dim/4-1}, y_{dim/4-1}].
+    """
+    dim_axis = dim // 2
+    if ref_grid_h is not None and dim_axis > 2:
+        h_ntk = (height / ref_grid_h) ** (dim_axis / (dim_axis - 2))
+        w_ntk = (width / ref_grid_w) ** (dim_axis / (dim_axis - 2))
+    else:
+        h_ntk = w_ntk = 1.0
+
+    x_lin = torch.linspace(shift_x, scale * scale_x + shift_x, width, device=device)
+    y_lin = torch.linspace(shift_y, scale * scale_y + shift_y, height, device=device)
+    y_grid, x_grid = torch.meshgrid(y_lin, x_lin, indexing="ij")
+    x_rope = rope(x_grid.reshape(1, -1), dim_axis, theta * w_ntk).squeeze(0)
+    y_rope = rope(y_grid.reshape(1, -1), dim_axis, theta * h_ntk).squeeze(0)
+    out = torch.stack([x_rope, y_rope], dim=2).reshape(height * width, dim // 2, 2, 2)
+    return out.to(dtype=dtype)
+
+
+def get_2d_sincos_pos_embed(embed_dim, height, width, device=None, dtype=torch.float32):
+    """Standard 2D sin/cos absolute positional embedding (ViT-style).
+
+    first half encodes W-coordinates, second half H.
+    """
+    assert embed_dim % 4 == 0
+    grid_h = torch.arange(height, dtype=torch.float32, device=device)
+    grid_w = torch.arange(width, dtype=torch.float32, device=device)
+    grid_y, grid_x = torch.meshgrid(grid_h, grid_w, indexing="ij")
+    emb_w = get_1d_sincos_pos_embed_from_grid_torch(embed_dim // 2, grid_x.reshape(-1), device=device)
+    emb_h = get_1d_sincos_pos_embed_from_grid_torch(embed_dim // 2, grid_y.reshape(-1), device=device)
+    return torch.cat([emb_w, emb_h], dim=1).to(dtype=dtype)
+
+
+class RotaryAttention(nn.Module):
+    """Single-stream self-attention with rotary positional encoding (used inside PiTBlock)."""
+    def __init__(self, dim, num_heads=8, qkv_bias=False, dtype=None, device=None, operations=None):
+        super().__init__()
+        assert dim % num_heads == 0
+        self.num_heads = num_heads
+        self.head_dim = dim // num_heads
+        self.qkv = operations.Linear(dim, dim * 3, bias=qkv_bias, dtype=dtype, device=device)
+        self.q_norm = operations.RMSNorm(self.head_dim, eps=1e-6, dtype=dtype, device=device)
+        self.k_norm = operations.RMSNorm(self.head_dim, eps=1e-6, dtype=dtype, device=device)
+        self.proj = operations.Linear(dim, dim, dtype=dtype, device=device)
+
+    def forward(self, x, pos, mask=None, transformer_options={}):
+        B, N, C = x.shape
+        H = self.num_heads
+        D = self.head_dim
+        qkv = self.qkv(x).reshape(B, N, 3, H, D).permute(2, 0, 3, 1, 4)
+        q, k, v = qkv.unbind(0)
+        q, k = apply_rope(self.q_norm(q), self.k_norm(k), pos[None, None])
+        x = optimized_attention(q, k, v, H, mask=mask, skip_reshape=True, transformer_options=transformer_options)
+        return self.proj(x)
+
+
+class FinalLayer(nn.Module):
+    def __init__(self, hidden_size, out_channels, dtype=None, device=None, operations=None):
+        super().__init__()
+        self.norm = operations.RMSNorm(hidden_size, eps=1e-6, dtype=dtype, device=device)
+        self.linear = operations.Linear(hidden_size, out_channels, bias=True, dtype=dtype, device=device)
+
+    def forward(self, x):
+        return self.linear(self.norm(x))
+
+
+class PatchTokenEmbedder(nn.Module):
+    """Linear projection used both for patchified-image tokens and text-feature tokens."""
+    def __init__(self, in_chans, embed_dim, use_norm=False, bias=True, dtype=None, device=None, operations=None):
+        super().__init__()
+        self.proj = operations.Linear(in_chans, embed_dim, bias=bias, dtype=dtype, device=device)
+        self.norm = operations.RMSNorm(embed_dim, eps=1e-6, dtype=dtype, device=device) if use_norm else nn.Identity()
+
+    def forward(self, x):
+        return self.norm(self.proj(x))
+
+
+class PixelTokenEmbedder(nn.Module):
+    """Pixel-level embedder: lifts each RGB pixel to hidden_size and packs into per-patch sequences."""
+    def __init__(self, in_channels, hidden_size_output, dtype=None, device=None, operations=None):
+        super().__init__()
+        self.in_channels = in_channels
+        self.hidden_size_output = hidden_size_output
+        self.proj = operations.Linear(self.in_channels, self.hidden_size_output, bias=True, dtype=dtype, device=device)
+
+    def forward(self, inputs, patch_size):
+        B, _, H, W = inputs.shape
+        Hs, Ws = H // patch_size, W // patch_size
+        P2 = patch_size * patch_size
+        x = inputs.permute(0, 2, 3, 1).contiguous()
+        x = self.proj(x)
+        pos_full = get_2d_sincos_pos_embed(self.hidden_size_output, H, W, device=x.device, dtype=x.dtype).view(H, W, self.hidden_size_output)
+        x = x + pos_full.unsqueeze(0)
+        x = x.view(B, Hs, patch_size, Ws, patch_size, self.hidden_size_output)
+        return x.permute(0, 1, 3, 2, 4, 5).reshape(B * Hs * Ws, P2, self.hidden_size_output)
+
+
+class PiTBlock(nn.Module):
+    """Pixel-level transformer block.
+
+    Compresses each patch's P^2 pixel tokens → 1 attention token via a linear,
+    runs global self-attention across patches with 2D RoPE, then expands back to P^2 tokens.
+    Conditioning is per-pixel adaLN from the patch-level features.
+    """
+    def __init__(self, pixel_hidden_size, patch_hidden_size, patch_size, num_heads, mlp_ratio=4.0,
+                 attn_hidden_size=None, attn_num_heads=None, dtype=None, device=None, operations=None, mlp_chunks=1):
+        super().__init__()
+        self.pixel_dim = pixel_hidden_size
+        self.context_dim = patch_hidden_size
+        self.attn_dim = attn_hidden_size if attn_hidden_size is not None else patch_hidden_size
+        self.num_heads = attn_num_heads if attn_num_heads is not None else num_heads
+        assert self.attn_dim % self.num_heads == 0
+
+        p2 = patch_size * patch_size
+        self.compress_to_attn = operations.Linear(p2 * self.pixel_dim, self.attn_dim, bias=True, dtype=dtype, device=device)
+        self.expand_from_attn = operations.Linear(self.attn_dim, p2 * self.pixel_dim, bias=True, dtype=dtype, device=device)
+
+        self.norm1 = operations.RMSNorm(self.pixel_dim, eps=1e-6, dtype=dtype, device=device)
+        self.attn = RotaryAttention(self.attn_dim, num_heads=self.num_heads, qkv_bias=False, dtype=dtype, device=device, operations=operations)
+        self.norm2 = operations.RMSNorm(self.pixel_dim, eps=1e-6, dtype=dtype, device=device)
+        self.mlp = Mlp(self.pixel_dim, hidden_features=int(self.pixel_dim * mlp_ratio), dtype=dtype, device=device, operations=operations)
+
+        self.adaLN_modulation_msa = operations.Linear(self.context_dim, 3 * self.pixel_dim * p2, bias=True, dtype=dtype, device=device)
+        self.adaLN_modulation_mlp = operations.Linear(self.context_dim, 3 * self.pixel_dim * p2, bias=True, dtype=dtype, device=device)
+
+        self._rope_fn = precompute_freqs_cis_2d
+        self.mlp_chunks = max(1, int(mlp_chunks))
+
+    def _fetch_pos(self, height, width, device, dtype, **rope_opts):
+        return self._rope_fn(self.attn_dim // self.num_heads, height, width, device=device, dtype=dtype, **rope_opts)
+
+    def forward(self, x, s_cond, image_height, image_width, patch_size, mask=None, transformer_options={}):
+        BL, P2, _ = x.shape
+        Hs, Ws = image_height // patch_size, image_width // patch_size
+        L = Hs * Ws
+        B = BL // L
+
+        # Attention path uses only msa params; compute, use, free before mlp params allocate.
+        msa_params = self.adaLN_modulation_msa(s_cond).view(BL, P2, 3 * self.pixel_dim)
+        shift_msa, scale_msa, gate_msa = msa_params.chunk(3, dim=-1)
+
+        x_norm = apply_adaln_(self.norm1(x), shift_msa, scale_msa)
+        x_flat = x_norm.view(BL, P2 * self.pixel_dim)
+
+        x_comp = self.compress_to_attn(x_flat).view(B, L, self.attn_dim)
+        pos_comp = self._fetch_pos(Hs, Ws, x.device, x.dtype, **(transformer_options.get("rope_options") or {}))
+        attn_out = self.attn(x_comp, pos_comp, mask=mask, transformer_options=transformer_options)
+        attn_flat = self.expand_from_attn(attn_out.view(B * L, self.attn_dim))
+        attn_exp = attn_flat.view(BL, P2, self.pixel_dim)
+        x = torch.addcmul(x, gate_msa, attn_exp)
+        del msa_params, shift_msa, scale_msa, gate_msa
+
+        mlp_params = self.adaLN_modulation_mlp(s_cond).view(BL, P2, 3 * self.pixel_dim)
+        shift_mlp, scale_mlp, gate_mlp = mlp_params.chunk(3, dim=-1)
+        gate_mlp = gate_mlp.contiguous()  # detach from mlp_params so the del below frees shift+scale storage before the MLP
+        mlp_input = apply_adaln_(self.norm2(x), shift_mlp, scale_mlp)
+        del mlp_params, shift_mlp, scale_mlp
+
+        # MLP in chunks since the peak memory usage is huge here
+        chunk_size = (BL + self.mlp_chunks - 1) // self.mlp_chunks
+        for s in range(0, BL, chunk_size):
+            e = min(s + chunk_size, BL)
+            x[s:e].addcmul_(gate_mlp[s:e], self.mlp(mlp_input[s:e]))
+        return x
@@ -0,0 +1,227 @@
+"""PiD — Pixel Diffusion Decoder. Decodes a Flux/SD3/Flux2/Z-Image latent
+directly to a 4x-upscaled image in 4 distilled flow-matching steps. PixDiT_T2I
+body + LQ projection branch injected before each MMDiT patch block.
+"""
+
+from typing import List
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from .model import PixDiT_T2I
+from .modules import precompute_freqs_cis_2d
+
+
+class SigmaAwareGatePerTokenPerDim(nn.Module):
+    """gate = sigmoid(content_proj(cat[x, lq]) - exp(log_alpha) * sigma); out = x + gate * lq.
+
+    Trained init gives ~0.88 gate at sigma=0, ~0.05 at sigma=1.
+    """
+
+    def __init__(self, dim: int, dtype=None, device=None, operations=None):
+        super().__init__()
+        self.content_proj = operations.Linear(dim * 2, dim, dtype=dtype, device=device)
+        self.log_alpha = nn.Parameter(torch.empty((), dtype=dtype, device=device))
+
+    def forward(self, x: torch.Tensor, lq: torch.Tensor, sigma: torch.Tensor) -> torch.Tensor:
+        content_logit = self.content_proj(torch.cat([x, lq], dim=-1))
+        # log_alpha is a raw nn.Parameter -> doesn't auto-cast under dynamic VRAM.
+        log_alpha = self.log_alpha.to(device=x.device, dtype=torch.float32)
+        sigma_offset = -log_alpha.exp() * sigma.float().view(-1, 1, 1)
+        gate = torch.sigmoid(content_logit + sigma_offset)
+        return x + (gate * lq).to(x.dtype)
+
+
+class ResBlock(nn.Module):
+    """Pre-activation ResNet block: GN -> SiLU -> Conv -> GN -> SiLU -> Conv + skip."""
+
+    def __init__(self, channels: int, num_groups: int = 4, dtype=None, device=None, operations=None):
+        super().__init__()
+        self.block = nn.Sequential(
+            operations.GroupNorm(num_groups, channels, dtype=dtype, device=device),
+            nn.SiLU(),
+            operations.Conv2d(channels, channels, kernel_size=3, padding=1, dtype=dtype, device=device),
+            operations.GroupNorm(num_groups, channels, dtype=dtype, device=device),
+            nn.SiLU(),
+            operations.Conv2d(channels, channels, kernel_size=3, padding=1, dtype=dtype, device=device),
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        return x + self.block(x)
+
+
+class LQProjection2D(nn.Module):
+    """LQ latent -> per-block patch-aligned features for controlnet-style injection."""
+
+    def __init__(
+        self,
+        latent_channels: int,
+        hidden_dim: int = 512,
+        out_dim: int = 1536,
+        patch_size: int = 16,
+        sr_scale: int = 4,
+        latent_spatial_down_factor: int = 8,
+        num_res_blocks: int = 4,
+        num_outputs: int = 7,
+        interval: int = 2,
+        dtype=None, device=None, operations=None,
+    ):
+        super().__init__()
+        self.latent_channels = latent_channels
+        self.hidden_dim = hidden_dim
+        self.out_dim = out_dim
+        self.patch_size = patch_size
+        self.sr_scale = sr_scale
+        self.latent_spatial_down_factor = latent_spatial_down_factor
+        self.num_outputs = num_outputs
+        self.interval = interval
+
+        z_to_patch_ratio = (sr_scale * latent_spatial_down_factor) / patch_size
+        self.z_to_patch_ratio = z_to_patch_ratio
+        if z_to_patch_ratio >= 1:
+            self.latent_fold_factor = 0
+            latent_proj_in_ch = latent_channels
+        else:
+            fold_factor = int(1 / z_to_patch_ratio)
+            assert fold_factor * z_to_patch_ratio == 1.0
+            self.latent_fold_factor = fold_factor
+            latent_proj_in_ch = latent_channels * fold_factor * fold_factor
+
+        layers = [
+            operations.Conv2d(latent_proj_in_ch, hidden_dim, kernel_size=3, padding=1, dtype=dtype, device=device),
+            nn.SiLU(),
+            operations.Conv2d(hidden_dim, hidden_dim, kernel_size=3, padding=1, dtype=dtype, device=device),
+        ]
+        for _ in range(num_res_blocks):
+            layers.append(ResBlock(hidden_dim, dtype=dtype, device=device, operations=operations))
+        self.latent_proj = nn.Sequential(*layers)
+
+        self.output_heads = nn.ModuleList(
+            [operations.Linear(hidden_dim, out_dim, dtype=dtype, device=device) for _ in range(num_outputs)]
+        )
+        self.gate_modules = nn.ModuleList(
+            [SigmaAwareGatePerTokenPerDim(out_dim, dtype=dtype, device=device, operations=operations)
+             for _ in range(num_outputs)]
+        )
+
+    def is_gate_active(self, block_idx: int) -> bool:
+        return block_idx % self.interval == 0
+
+    def output_index(self, block_idx: int) -> int:
+        return block_idx // self.interval
+
+    def gate(self, x: torch.Tensor, lq_feature: torch.Tensor, sigma: torch.Tensor, out_idx: int) -> torch.Tensor:
+        return self.gate_modules[out_idx](x, lq_feature, sigma)
+
+    def _align_latent_to_patch_grid(self, lq_latent: torch.Tensor, pH: int, pW: int) -> torch.Tensor:
+        B, z_dim = lq_latent.shape[:2]
+        if self.z_to_patch_ratio >= 1:
+            if lq_latent.shape[2] != pH or lq_latent.shape[3] != pW:
+                z_aligned = F.interpolate(lq_latent, size=(pH, pW), mode="nearest")
+            else:
+                z_aligned = lq_latent
+        else:
+            f = self.latent_fold_factor
+            zH_expected, zW_expected = pH * f, pW * f
+            if lq_latent.shape[2] != zH_expected or lq_latent.shape[3] != zW_expected:
+                lq_latent = F.interpolate(lq_latent, size=(zH_expected, zW_expected), mode="nearest")
+            z_aligned = lq_latent.reshape(B, z_dim, pH, f, pW, f).permute(0, 1, 3, 5, 2, 4)
+            z_aligned = z_aligned.reshape(B, z_dim * f * f, pH, pW)
+        return self.latent_proj(z_aligned)
+
+    def forward(self, lq_latent: torch.Tensor, target_pH: int, target_pW: int) -> List[torch.Tensor]:
+        feat = self._align_latent_to_patch_grid(lq_latent, target_pH, target_pW)
+        B, C, H, W = feat.shape
+        tokens = feat.permute(0, 2, 3, 1).contiguous().view(B, H * W, C)
+        return [head(tokens) for head in self.output_heads]
+
+
+class PidNet(PixDiT_T2I):
+    """PixDiT_T2I + LQ injection (one sigma-gated feature inserted before each patch block)."""
+
+    def __init__(
+        self,
+        lq_latent_channels: int = 16,
+        lq_hidden_dim: int = 512,
+        lq_num_res_blocks: int = 4,
+        lq_interval: int = 2,
+        sr_scale: int = 4,
+        latent_spatial_down_factor: int = 8,
+        rope_ref_h: int = 1024, # NTK ref resolution in PIXEL units: 1024px / patch=16 -> grid_ref=64.
+        rope_ref_w: int = 1024,
+        image_model=None,
+        dtype=None, device=None, operations=None,
+        **pixdit_kwargs,
+    ):
+        super().__init__(dtype=dtype, device=device, operations=operations, **pixdit_kwargs)
+
+        self.rope_ref_grid_h = rope_ref_h // self.patch_size
+        self.rope_ref_grid_w = rope_ref_w // self.patch_size
+
+        # Parent's PiTBlocks were built with plain RoPE — swap in NTK-aware.
+        def _pit_rope_fn(head_dim, h, w, device=None, dtype=torch.float32, **rope_opts):
+            return precompute_freqs_cis_2d(head_dim, h, w, ref_grid_h=self.rope_ref_grid_h, ref_grid_w=self.rope_ref_grid_w, device=device, dtype=dtype, **rope_opts)
+        for blk in self.pixel_blocks:
+            blk._rope_fn = _pit_rope_fn
+
+        num_lq_outputs = (self.patch_depth + lq_interval - 1) // lq_interval
+        self.lq_proj = LQProjection2D(
+            latent_channels=lq_latent_channels,
+            hidden_dim=lq_hidden_dim,
+            out_dim=self.hidden_size,
+            patch_size=self.patch_size,
+            sr_scale=sr_scale,
+            latent_spatial_down_factor=latent_spatial_down_factor,
+            num_res_blocks=lq_num_res_blocks,
+            num_outputs=num_lq_outputs,
+            interval=lq_interval,
+            dtype=dtype,
+            device=device,
+            operations=operations,
+        )
+
+    def _fetch_patch_pos(self, height, width, device, dtype, **rope_opts):
+        return precompute_freqs_cis_2d(
+            self.hidden_size // self.num_groups,
+            height, width,
+            ref_grid_h=self.rope_ref_grid_h, ref_grid_w=self.rope_ref_grid_w,
+            device=device, dtype=dtype, **rope_opts,
+        )
+
+    def _pre_patch_block(self, s, i, pid_lq_features, pid_degrade_sigma, **kwargs):
+        if not self.lq_proj.is_gate_active(i):
+            return s
+        out_idx = self.lq_proj.output_index(i)
+        if out_idx >= len(pid_lq_features):
+            return s
+        return self.lq_proj.gate(s, pid_lq_features[out_idx], pid_degrade_sigma, out_idx)
+
+    def _forward(self, x, timesteps, context=None, attention_mask=None, transformer_options={}, lq_latent=None, degrade_sigma=None, **kwargs):
+        if lq_latent is None:
+            raise ValueError("PidNet requires lq_latent — attach via PiDConditioning")
+        expected_c = self.lq_proj.latent_channels
+        if lq_latent.shape[1] != expected_c:
+            raise ValueError(
+                f"Input latent has {lq_latent.shape[1]} channels, this model variant expects {expected_c}. "
+                f"Flux1/SD3 = 16 channels, Flux2 = 128 channels."
+            )
+        B = x.shape[0]
+        # Match the backbone's pad_to_patch_size (round up) so the LQ grid lines up with the patch stream.
+        Hs = -(-x.shape[2] // self.patch_size)
+        Ws = -(-x.shape[3] // self.patch_size)
+
+        degrade_sigma = degrade_sigma.to(device=x.device, dtype=torch.float32).reshape(-1)
+        if degrade_sigma.numel() == 1 and B > 1:
+            degrade_sigma = degrade_sigma.expand(B).contiguous()
+
+        lq_features = self.lq_proj(lq_latent=lq_latent.to(x), target_pH=Hs, target_pW=Ws)
+
+        return super()._forward(
+            x, timesteps,
+            context=context, attention_mask=attention_mask,
+            transformer_options=transformer_options,
+            pid_lq_features=lq_features,
+            pid_degrade_sigma=degrade_sigma,
+            **kwargs,
+        )
@@ -51,6 +51,9 @@ class FeedForward(nn.Module):
        return hidden_states


+# Addin this back because Nunchaku custom nodes rely on it, see comment here:
+# https://github.com/Comfy-Org/ComfyUI/pull/14178#issuecomment-4640475161
+# TODO: Eventually remove this once we natively support SVDQuants
 def apply_rotary_emb(x, freqs_cis):
    if x.shape[1] == 0:
        return x
@@ -0,0 +1,199 @@
+# TripoSplat 3D gaussian container. Operates on already-decoded
+# tensors and exposes them as render-ready tensors (render_tensors) for the generic SPLAT type.
+import torch
+import torch.nn.functional as F
+
+import comfy.model_management
+
+
+class GaussianModel:
+    def __init__(self, aabb: list, sh_degree: int = 0, mininum_kernel_size: float = 0.0,
+                 scaling_bias: float = 0.01, opacity_bias: float = 0.1,
+                 scaling_activation: str = "exp", device=None):
+        self.sh_degree = sh_degree
+        self.mininum_kernel_size = mininum_kernel_size
+        self.scaling_bias = scaling_bias
+        self.opacity_bias = opacity_bias
+        self.device = device
+        self.aabb = torch.tensor(aabb, dtype=torch.float32, device=device)
+
+        if scaling_activation == "exp":
+            self._scaling_activation = torch.exp
+            self._inverse_scaling_activation = torch.log
+        elif scaling_activation == "softplus":
+            self._scaling_activation = F.softplus
+            self._inverse_scaling_activation = lambda x: x + torch.log(-torch.expm1(-x))
+
+        self._opacity_activation = torch.sigmoid
+        self._inverse_opacity_activation = lambda x: torch.log(x / (1 - x))
+
+        self.scale_bias = self._inverse_scaling_activation(torch.tensor(self.scaling_bias)).to(self.device)
+        self.rots_bias = torch.zeros(4, device=self.device)
+        self.rots_bias[0] = 1
+        self.opacity_bias_val = self._inverse_opacity_activation(torch.tensor(self.opacity_bias)).to(self.device)
+
+        self._storage = {}
+
+    def _get_store(self, name):
+        return self._storage.get(name)
+
+    def _set_store(self, name, value):
+        self._storage[name] = value
+
+    @property
+    def _xyz(self):
+        return self._get_store("_xyz")
+    @_xyz.setter
+    def _xyz(self, value):
+        if value is None:
+            self._set_store("_xyz", None)
+            self._set_store("xyz", None)
+            return
+        self._set_store("_xyz", value)
+        self._set_store("xyz", value * self.aabb[None, 3:] + self.aabb[None, :3])
+
+    @property
+    def get_xyz(self):
+        return self._get_store("xyz")
+
+    @property
+    def _features_dc(self):
+        return self._get_store("_features_dc")
+    @_features_dc.setter
+    def _features_dc(self, value):
+        self._set_store("_features_dc", value)
+
+    @property
+    def _opacity(self):
+        return self._get_store("_opacity")
+    @_opacity.setter
+    def _opacity(self, value):
+        if value is None:
+            self._set_store("_opacity", None)
+            self._set_store("opacity", None)
+            return
+        self._set_store("_opacity", value)
+        self._set_store("opacity", self._opacity_activation(value + self.opacity_bias_val))
+
+    @property
+    def get_opacity(self):
+        return self._get_store("opacity")
+
+    @property
+    def _scaling(self):
+        return self._get_store("_scaling")
+    @_scaling.setter
+    def _scaling(self, value):
+        if value is None:
+            self._set_store("_scaling", None)
+            self._set_store("scaling", None)
+            return
+        self._set_store("_scaling", value)
+        s = self._scaling_activation(value + self.scale_bias)
+        s = torch.square(s) + self.mininum_kernel_size ** 2
+        self._set_store("scaling", torch.sqrt(s))
+
+    @property
+    def get_scaling(self):
+        return self._get_store("scaling")
+
+    @property
+    def _rotation(self):
+        return self._get_store("_rotation")
+    @_rotation.setter
+    def _rotation(self, value):
+        self._set_store("_rotation", value)
+
+    _DEFAULT_TRANSFORM = [[1, 0, 0], [0, 0, -1], [0, 1, 0]]
+
+    def render_tensors(self):
+        # Render-ready (activated, world-space) tensors for the generic SPLAT type. The axis transform
+        # (a 3x3 rotation, object frame -> viewer Y-up) is baked into positions and rotations.
+        # Returns float tensors on the intermediate device: positions (N,3), scales (N,3) linear,
+        # rotations (N,4) wxyz, opacities (N,1) in [0,1], sh (N,K,3) coefficients.
+        xyz = self.get_xyz.float()
+        scaling = self.get_scaling.float()
+        opacity = self.get_opacity.float()
+        rotation = (self._rotation + self.rots_bias[None, :]).float()
+        sh = self._features_dc.float()  # (N, K, 3)
+        T = torch.as_tensor(self._DEFAULT_TRANSFORM, dtype=torch.float32, device=xyz.device)
+        xyz = xyz @ T.T
+        rotation = _matrix_to_quat(torch.matmul(T, _quat_to_matrix(rotation)))
+        rotation = rotation / torch.linalg.norm(rotation, dim=-1, keepdim=True)
+        out_device = comfy.model_management.intermediate_device()
+        return (
+            xyz.to(out_device).contiguous(), scaling.to(out_device).contiguous(),
+            rotation.to(out_device).contiguous(), opacity.to(out_device).contiguous(),
+            sh.to(out_device).contiguous(),
+        )
+
+
+def _quat_to_matrix(q):
+    q = q / torch.linalg.norm(q, dim=-1, keepdim=True)
+    w, x, y, z = q[:, 0], q[:, 1], q[:, 2], q[:, 3]
+    R = torch.stack([
+        1 - 2*(y*y + z*z), 2*(x*y - w*z),     2*(x*z + w*y),
+        2*(x*y + w*z),     1 - 2*(x*x + z*z), 2*(y*z - w*x),
+        2*(x*z - w*y),     2*(y*z + w*x),     1 - 2*(x*x + y*y),
+    ], dim=-1).reshape(-1, 3, 3)
+    return R
+
+
+def _matrix_to_quat(R):
+    trace = R[:, 0, 0] + R[:, 1, 1] + R[:, 2, 2]
+    q = torch.zeros((R.shape[0], 4), dtype=R.dtype, device=R.device)
+    s = torch.sqrt(torch.clamp(trace + 1, min=0)) * 2
+    q[:, 0] = 0.25 * s
+    denom = torch.where(s != 0, s, torch.ones_like(s))
+    q[:, 1] = (R[:, 2, 1] - R[:, 1, 2]) / denom
+    q[:, 2] = (R[:, 0, 2] - R[:, 2, 0]) / denom
+    q[:, 3] = (R[:, 1, 0] - R[:, 0, 1]) / denom
+    m01 = (R[:, 0, 0] >= R[:, 1, 1]) & (R[:, 0, 0] >= R[:, 2, 2]) & (s == 0)
+    s1 = torch.sqrt(torch.clamp(1 + R[:, 0, 0] - R[:, 1, 1] - R[:, 2, 2], min=0)) * 2
+    q[m01, 0] = (R[m01, 2, 1] - R[m01, 1, 2]) / s1[m01]
+    q[m01, 1] = 0.25 * s1[m01]
+    q[m01, 2] = (R[m01, 0, 1] + R[m01, 1, 0]) / s1[m01]
+    q[m01, 3] = (R[m01, 0, 2] + R[m01, 2, 0]) / s1[m01]
+    m11 = (R[:, 1, 1] > R[:, 0, 0]) & (R[:, 1, 1] >= R[:, 2, 2]) & (s == 0)
+    s2 = torch.sqrt(torch.clamp(1 + R[:, 1, 1] - R[:, 0, 0] - R[:, 2, 2], min=0)) * 2
+    q[m11, 0] = (R[m11, 0, 2] - R[m11, 2, 0]) / s2[m11]
+    q[m11, 1] = (R[m11, 0, 1] + R[m11, 1, 0]) / s2[m11]
+    q[m11, 2] = 0.25 * s2[m11]
+    q[m11, 3] = (R[m11, 1, 2] + R[m11, 2, 1]) / s2[m11]
+    m21 = (R[:, 2, 2] > R[:, 0, 0]) & (R[:, 2, 2] > R[:, 1, 1]) & (s == 0)
+    s3 = torch.sqrt(torch.clamp(1 + R[:, 2, 2] - R[:, 0, 0] - R[:, 1, 1], min=0)) * 2
+    q[m21, 0] = (R[m21, 1, 0] - R[m21, 0, 1]) / s3[m21]
+    q[m21, 1] = (R[m21, 0, 2] + R[m21, 2, 0]) / s3[m21]
+    q[m21, 2] = (R[m21, 1, 2] + R[m21, 2, 1]) / s3[m21]
+    q[m21, 3] = 0.25 * s3[m21]
+    return q / torch.linalg.norm(q, dim=-1, keepdim=True)
+
+
+def build_gaussian_models(decoder, points_pred: dict, pred: dict):
+    # Assemble GaussianModels from the elastic decoder layout. decoder is the ElasticGaussianFixedlenDecoder
+    # (carries layout / rep_config / _get_offset)
+    x = points_pred
+    offset = decoder._get_offset(pred['features'])
+    h = pred["features"]
+    ret = []
+    for i in range(h.shape[0]):
+        g = GaussianModel(
+            sh_degree=0,
+            aabb=[-0.5, -0.5, -0.5, 1.0, 1.0, 1.0],
+            mininum_kernel_size=decoder.rep_config['filter_kernel_size_3d'],
+            scaling_bias=decoder.rep_config['scaling_bias'],
+            opacity_bias=decoder.rep_config['opacity_bias'],
+            scaling_activation=decoder.rep_config['scaling_activation'],
+            device=h.device,
+        )
+        _x = x["points"][i, :, None, :]
+        for k, v in decoder.layout.items():
+            if k == '_xyz':
+                setattr(g, k, (offset[i] + _x).flatten(0, 1))
+            elif k in ('_xyz_center', '_offset_scale'):
+                continue
+            else:
+                feats = h[i][:, v['range'][0]:v['range'][1]].reshape(-1, *v['shape']).flatten(0, 1)
+                setattr(g, k, feats * decoder.rep_config['lr'][k])
+        ret.append(g)
+    return ret
@@ -0,0 +1,326 @@
+# TripoSplat flow-matching denoiser (LatentSeqMMFlowModel). Registered as a ModelType.FLOW arch and
+# driven by the standard KSampler; jointly denoises the (B, 8192, 16) latent and a (B, 1, 5) camera token
+# carried as a 2-element nested latent.
+import numpy as np
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+import comfy.model_management
+import comfy.patcher_extension
+import comfy.rmsnorm
+from comfy.ldm.modules.attention import optimized_attention
+from comfy.ldm.flux.math import apply_rope
+
+
+class MultiHeadRMSNorm(nn.Module):
+    def __init__(self, dim, heads, dtype=None, device=None):
+        super().__init__()
+        self.gamma = nn.Parameter(torch.empty(heads, dim, dtype=dtype, device=device))
+
+    def forward(self, x):
+        x = comfy.rmsnorm.rms_norm(x)
+        return x * comfy.model_management.cast_to(self.gamma, x.dtype, x.device)
+
+
+# Positional embeddings
+
+class RePo3DRotaryEmbedding(nn.Module):
+    def __init__(self, model_channels, num_heads, head_dim, repo_hidden_ratio=0.125, max_freq=16.0,
+                 dtype=None, device=None, operations=None):
+        super().__init__()
+        self.num_heads = num_heads
+        self.head_dim = head_dim
+        repo_hidden_size = int(model_channels * repo_hidden_ratio)
+        self.norm = operations.LayerNorm(model_channels, dtype=dtype, device=device)
+        self.gate_map = operations.Linear(model_channels, repo_hidden_size, bias=False, dtype=dtype, device=device)
+        self.content_map = operations.Linear(model_channels, repo_hidden_size, bias=False, dtype=dtype, device=device)
+        self.act = nn.SiLU()
+        self.final_map = operations.Linear(repo_hidden_size, 3 * num_heads, bias=False, dtype=dtype, device=device)
+        self.dim_0 = 2 * (head_dim // 6)
+        self.dim_1 = 2 * (head_dim // 6)
+        self.dim_2 = head_dim - self.dim_0 - self.dim_1
+        dims = [self.dim_0, self.dim_1, self.dim_2]
+        freqs_list = []
+        for d in dims:
+            freq_dim = d // 2
+            freqs_list.append(torch.linspace(1.0, float(max_freq), steps=freq_dim, dtype=torch.float32))
+        self.freqs_0 = nn.Parameter(freqs_list[0])
+        self.freqs_1 = nn.Parameter(freqs_list[1])
+        self.freqs_2 = nn.Parameter(freqs_list[2])
+
+    def forward(self, hidden_states):
+        h = self.norm(hidden_states)
+        feat = self.act(self.gate_map(h)) * self.content_map(h)
+        out = self.final_map(feat)
+        B, L, _ = out.shape
+        delta_pos = out.reshape(B, L, self.num_heads, 3)
+        f0 = comfy.model_management.cast_to(self.freqs_0, torch.float32, out.device)
+        f1 = comfy.model_management.cast_to(self.freqs_1, torch.float32, out.device)
+        f2 = comfy.model_management.cast_to(self.freqs_2, torch.float32, out.device)
+        ang_0 = delta_pos[..., 0].unsqueeze(-1) * f0 * torch.pi
+        ang_1 = delta_pos[..., 1].unsqueeze(-1) * f1 * torch.pi
+        ang_2 = delta_pos[..., 2].unsqueeze(-1) * f2 * torch.pi
+        ang = torch.cat([ang_0, ang_1, ang_2], dim=-1).float()  # (B, L, heads, head_dim/2)
+        cos, sin = ang.cos(), ang.sin()
+        return torch.stack([cos, -sin, sin, cos], dim=-1).reshape(*ang.shape, 2, 2)
+
+
+class PcdAbsolutePositionEmbedder(nn.Module):
+    # Sinusoidal absolute position embedding. Two fixed schedules are used in TripoSplat:
+    # "pow2" (flow-model latent anchors) and "log2" (octree / gaussian decoders).
+    def __init__(self, channels: int, in_channels: int = 3, max_res: int = 16, schedule: str = "pow2"):
+        super().__init__()
+        self.channels = channels
+        self.in_channels = in_channels
+        self.max_res = max_res
+        self.schedule = schedule
+        self.freq_dim = channels // in_channels // 2
+
+    def _freqs(self, device):
+        if self.schedule == "pow2":
+            freqs_2exp = torch.arange(self.max_res, dtype=torch.float32, device=device)
+            res_dim = max(0, self.freq_dim - self.max_res)
+            freqs_res = (torch.arange(res_dim, dtype=torch.float32, device=device) / max(res_dim, 1) * self.max_res
+                         if res_dim > 0 else torch.empty(0, device=device))
+            freqs = torch.cat([freqs_2exp, freqs_res], dim=0)[:self.freq_dim]
+            return torch.pow(2.0, freqs) * 2.0  # *2 folds this schedule's 2*pi into the shared *pi below
+        logs = torch.linspace(0.0, float(self.max_res), steps=self.freq_dim, dtype=torch.float32, device=device)
+        return torch.pow(2.0, logs)
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        orig_dtype = x.dtype
+        x = x.float()
+        *dims, D = x.shape
+        out = torch.outer(x.reshape(-1), self._freqs(x.device)) * torch.pi
+        out = torch.cat([out.sin(), out.cos()], dim=-1).reshape(*dims, -1)
+        if out.shape[-1] < self.channels:
+            out = torch.cat([out, torch.zeros(*dims, self.channels - out.shape[-1],
+                                              device=out.device, dtype=out.dtype)], dim=-1)
+        return out.to(orig_dtype)
+
+
+def attention(q, k, v, transformer_options=None):
+    # q, k, v: (B, L, heads, dim) -> (B, L, heads, dim). Shared optimized_attention call convention.
+    out = optimized_attention(q.transpose(1, 2), k.transpose(1, 2), v.transpose(1, 2), heads=q.shape[2],
+                              skip_reshape=True, skip_output_reshape=True, low_precision_attention=False,
+                              transformer_options=transformer_options)
+    return out.transpose(1, 2)
+
+
+# Transformer building blocks
+
+class MLP(nn.Module):
+    def __init__(self, in_channels, hidden_channels, out_channels, dtype=None, device=None, operations=None):
+        super().__init__()
+        self.mlp = nn.Sequential(
+            operations.Linear(in_channels, hidden_channels, dtype=dtype, device=device),
+            nn.GELU(approximate="tanh"),
+            operations.Linear(hidden_channels, out_channels, dtype=dtype, device=device),
+        )
+
+    def forward(self, x):
+        return self.mlp(x)
+
+
+class RopeMultiHeadAttention(nn.Module):
+    def __init__(self, channels, num_heads, qkv_bias=True, qk_rms_norm=False, use_rope=False,
+                 dtype=None, device=None, operations=None):
+        super().__init__()
+        self.channels = channels
+        self.num_heads = num_heads
+        self.head_dim = channels // num_heads
+        self.qk_rms_norm = qk_rms_norm
+        self.use_rope = use_rope
+        self.qkv = operations.Linear(channels, channels * 3, bias=qkv_bias, dtype=dtype, device=device)
+        if self.qk_rms_norm:
+            self.q_norm = MultiHeadRMSNorm(self.head_dim, num_heads, dtype=dtype, device=device)
+            self.k_norm = MultiHeadRMSNorm(self.head_dim, num_heads, dtype=dtype, device=device)
+        self.out = operations.Linear(channels, channels, dtype=dtype, device=device)
+
+    def forward(self, x, rope_emb=None, transformer_options=None):
+        B, L, C = x.shape
+        qkv = self.qkv(x).reshape(B, L, 3, self.num_heads, self.head_dim)
+        q, k, v = qkv.unbind(2)
+        if self.use_rope:
+            q, k = apply_rope(q, k, rope_emb)
+        if self.qk_rms_norm:
+            q = self.q_norm(q)
+            k = self.k_norm(k)
+        h = attention(q, k, v, transformer_options)  # (B, L, heads, dim)
+        return self.out(h.reshape(B, L, C))
+
+
+class UnifiedTransformerBlock(nn.Module):
+    def __init__(self, channels, num_heads, mlp_ratio=4.0,
+                 use_rope=False, qk_rms_norm=False, qkv_bias=True,
+                 modulation=True, share_mod=False,
+                 dtype=None, device=None, operations=None):
+        super().__init__()
+        self.modulation = modulation
+        self.share_mod = share_mod
+        self.norm1 = operations.LayerNorm(channels, elementwise_affine=not modulation, eps=1e-6, dtype=dtype, device=device)
+        self.norm2 = operations.LayerNorm(channels, elementwise_affine=not modulation, eps=1e-6, dtype=dtype, device=device)
+        self.attn = RopeMultiHeadAttention(channels, num_heads=num_heads,
+                                           qkv_bias=qkv_bias, use_rope=use_rope, qk_rms_norm=qk_rms_norm,
+                                           dtype=dtype, device=device, operations=operations)
+        self.mlp = MLP(channels, int(channels * mlp_ratio), channels, dtype=dtype, device=device, operations=operations)
+        if modulation:
+            if not share_mod:
+                self.adaLN_modulation = nn.Sequential(
+                    nn.SiLU(), operations.Linear(channels, 6 * channels, bias=True, dtype=dtype, device=device))
+            self.shift_table = nn.Parameter(torch.empty(1, 6 * channels, dtype=dtype, device=device))
+
+    def forward(self, x, mod=None, rotary_emb=None, transformer_options=None):
+        if self.modulation:
+            if not self.share_mod:
+                mod = self.adaLN_modulation(mod)
+            mod = mod + comfy.model_management.cast_to(self.shift_table, mod.dtype, mod.device)
+            shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = mod.chunk(6, dim=1)
+            h = torch.addcmul(shift_msa.unsqueeze(1), self.norm1(x), 1 + scale_msa.unsqueeze(1))
+            x = torch.addcmul(x, self.attn(h, rope_emb=rotary_emb, transformer_options=transformer_options), gate_msa.unsqueeze(1))
+            h = torch.addcmul(shift_mlp.unsqueeze(1), self.norm2(x), 1 + scale_mlp.unsqueeze(1))
+            x = torch.addcmul(x, self.mlp(h), gate_mlp.unsqueeze(1))
+        else:
+            x = x + self.attn(self.norm1(x), rope_emb=rotary_emb, transformer_options=transformer_options)
+            x = x + self.mlp(self.norm2(x))
+        return x
+
+
+class TimestepEmbedder(nn.Module):
+    def __init__(self, hidden_size, frequency_embedding_size=256, dtype=None, device=None, operations=None):
+        super().__init__()
+        self.mlp = nn.Sequential(
+            operations.Linear(frequency_embedding_size, hidden_size, bias=True, dtype=dtype, device=device),
+            nn.SiLU(),
+            operations.Linear(hidden_size, hidden_size, bias=True, dtype=dtype, device=device),
+        )
+        self.frequency_embedding_size = frequency_embedding_size
+
+    @staticmethod
+    def timestep_embedding(t, dim, max_period=10000):
+        half = dim // 2
+        freqs = torch.exp(-np.log(max_period) * torch.arange(start=0, end=half, dtype=torch.float32) / half).to(device=t.device)
+        args = t[:, None].float() * freqs[None]
+        embedding = torch.cat([torch.cos(args), torch.sin(args)], dim=-1)
+        if dim % 2:
+            embedding = torch.cat([embedding, torch.zeros_like(embedding[:, :1])], dim=-1)
+        return embedding
+
+    def forward(self, t):
+        emb = self.timestep_embedding(t, self.frequency_embedding_size)
+        return self.mlp(emb.to(self.mlp[0].weight.dtype))
+
+
+class LatentSeqMMFlowModel(nn.Module):
+    def __init__(self, image_model=None, q_token_length=8192, in_channels=16, model_channels=1024,
+                 cond_channels=1280, out_channels=16, num_blocks=24, num_refiner_blocks=2,
+                 num_heads=None, num_head_channels=64, cam_channels=5, cond2_channels=128,
+                 mlp_ratio=4, share_mod=True, qk_rms_norm=True,
+                 dtype=None, device=None, operations=None, **kwargs):
+        super().__init__()
+        self.dtype = dtype
+        self.q_token_length = q_token_length
+        self.in_channels = in_channels
+        self.cam_channels = cam_channels
+        self.model_channels = model_channels
+        self.cond_channels = cond_channels
+        self.cond2_channels = cond2_channels
+        self.out_channels = out_channels
+        self.num_blocks = num_blocks
+        self.num_refiner_blocks = num_refiner_blocks
+        self.num_heads = num_heads or model_channels // num_head_channels
+        self.mlp_ratio = mlp_ratio
+        self.share_mod = share_mod
+        self.qk_rms_norm = qk_rms_norm
+
+        factory_kwargs = dict(dtype=dtype, device=device)
+        op_kwargs = dict(operations=operations, **factory_kwargs)
+
+        self.t_embedder = TimestepEmbedder(model_channels, **op_kwargs)
+        if share_mod:
+            self.adaLN_modulation = nn.Sequential(nn.SiLU(), operations.Linear(model_channels, 6 * model_channels, bias=True, **factory_kwargs))
+
+        self.input_layer = operations.Linear(in_channels, model_channels, **factory_kwargs)
+        self.cond_embedder = operations.Linear(cond_channels, model_channels, **factory_kwargs)
+        self.cond_embedder2 = operations.Linear(cond2_channels, model_channels, **factory_kwargs) if cond2_channels is not None else None
+
+        # Fixed Sobol (low-discrepancy) 3D anchor positions for the latent tokens, used as positional encoding.
+        # The embedder is parameter-free and the anchors are fixed, precompute once.
+        sobol_seq = torch.quasirandom.SobolEngine(dimension=3, scramble=True, seed=123).draw(q_token_length)
+        pos_emb = PcdAbsolutePositionEmbedder(model_channels)(sobol_seq.unsqueeze(0))
+        self.register_buffer("pos_emb", pos_emb, persistent=False)
+
+        # RePo3DRotaryEmbedding layers for the refiner and main blocks
+        repo_kwargs = dict(num_heads=self.num_heads, head_dim=num_head_channels, **op_kwargs)
+        self.noise_repo_layers = nn.ModuleList(
+            [RePo3DRotaryEmbedding(model_channels, **repo_kwargs) for _ in range(num_refiner_blocks)])
+        self.context_repo_layers = nn.ModuleList(
+            [RePo3DRotaryEmbedding(model_channels, **repo_kwargs) for _ in range(num_refiner_blocks)])
+        self.repo_layers = nn.ModuleList(
+            [RePo3DRotaryEmbedding(model_channels, **repo_kwargs) for _ in range(num_blocks)])
+
+        # Refiner blocks
+        block_kwargs = dict(num_heads=self.num_heads, mlp_ratio=self.mlp_ratio, use_rope=True, qk_rms_norm=self.qk_rms_norm, **op_kwargs)
+        self.noise_refiner = nn.ModuleList(
+            [UnifiedTransformerBlock(model_channels, modulation=True, share_mod=self.share_mod, **block_kwargs) for _ in range(num_refiner_blocks)])
+        self.context_refiner = nn.ModuleList(
+            [UnifiedTransformerBlock(model_channels, modulation=False, **block_kwargs) for _ in range(num_refiner_blocks)])
+
+        self.cam_refiner = MLP(self.cam_channels, model_channels, model_channels, **op_kwargs)
+
+        self.blocks = nn.ModuleList(
+            [UnifiedTransformerBlock(model_channels, modulation=True, share_mod=self.share_mod, **block_kwargs) for _ in range(num_blocks)])
+
+        self.shift_table = nn.Parameter(torch.empty(1, 2, model_channels, **factory_kwargs))
+        self.out_layer = operations.Linear(model_channels, out_channels, **factory_kwargs)
+        self.cam_out_layer = operations.Linear(model_channels, cam_channels, **factory_kwargs)
+
+    def forward(self, x, t, context=None, ref_latents=None, transformer_options={}, **kwargs):
+        return comfy.patcher_extension.WrapperExecutor.new_class_executor(
+            self._forward,
+            self,
+            comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.DIFFUSION_MODEL, transformer_options)
+        ).execute(x, t, context, ref_latents, transformer_options, **kwargs)
+
+    def _forward(self, x, t, context=None, ref_latents=None, transformer_options={}, **kwargs):
+        # x is the unpacked nested latent: [latent (B,8192,in_channels), camera (B,1,cam_channels)].
+        # context == feature1.
+        z, camera = x[0], x[1]
+        feat1 = context
+
+        h_x = self.input_layer(z)
+        h_cond = self.cond_embedder(feat1)
+        if ref_latents is not None and self.cond_embedder2 is not None:
+            # Flatten the Flux2 VAE latent (B,128,h,w) to a token sequence and front-pad to feat1's length
+            # (the pad count = feat1's prefix tokens: DINOv3 cls + registers), then add to the context.
+            feat2 = ref_latents[0].flatten(2).transpose(1, 2)
+            feat2 = F.pad(feat2, (0, 0, feat1.shape[1] - feat2.shape[1], 0))
+            h_cond = h_cond + self.cond_embedder2(feat2.to(h_cond.dtype))
+        t_emb = self.t_embedder(t)
+        t_mod = self.adaLN_modulation(t_emb) if self.share_mod else t_emb
+
+        h_x = h_x + self.pos_emb.to(z)
+
+        for i, block in enumerate(self.noise_refiner):
+            h_x = block(h_x, mod=t_mod, rotary_emb=self.noise_repo_layers[i](h_x), transformer_options=transformer_options)
+
+        for i, block in enumerate(self.context_refiner):
+            h_cond = block(h_cond, mod=None, rotary_emb=self.context_repo_layers[i](h_cond), transformer_options=transformer_options)
+
+        cam = camera.to(z)
+        h_cam = self.cam_refiner(cam)
+        h = torch.cat([h_x, h_cond, h_cam], dim=1)
+
+        for i, block in enumerate(self.blocks):
+            h = block(h, mod=t_mod, rotary_emb=self.repo_layers[i](h), transformer_options=transformer_options)
+
+        h_x = F.layer_norm(h[:, :z.shape[1]].float(), h.shape[-1:]).to(z)
+        h_cam = F.layer_norm(h[:, -cam.shape[1]:].float(), h.shape[-1:]).to(z)
+
+        shift, scale = (comfy.model_management.cast_to(self.shift_table, t_emb.dtype, t_emb.device) + t_emb.unsqueeze(1)).chunk(2, dim=1)
+        scale = 1 + scale
+        h_x = torch.addcmul(shift, h_x, scale)
+        h_cam = torch.addcmul(shift, h_cam, scale)
+
+        return self.out_layer(h_x), self.cam_out_layer(h_cam)
@@ -0,0 +1,91 @@
+# Live preview for TripoSplat: decode an x0 estimate into a coarse gaussian splat and render it with a perspective orbit camera.
+import numpy as np
+from PIL import Image
+
+_C0 = 0.28209479177387814
+_LATENT_TOKENS = 8192  # q_token_length
+_LATENT_CH = 16        # in_channels
+_OBJECT_TO_VIEWER = np.array([[1, 0, 0], [0, 0, -1], [0, 1, 0]], np.float32)  # object frame -> viewer Y-up frame
+
+
+def _view_matrix(yaw_deg, pitch_deg):
+    y, p = np.radians(yaw_deg), np.radians(pitch_deg)
+    Ry = np.array([[np.cos(y), 0, np.sin(y)], [0, 1, 0], [-np.sin(y), 0, np.cos(y)]], np.float32)
+    Rx = np.array([[1, 0, 0], [0, np.cos(p), -np.sin(p)], [0, np.sin(p), np.cos(p)]], np.float32)
+    return Rx @ Ry
+
+
+def render_splat(xyz, rgb, scale, opacity=None, yaw=35.0, pitch=30.0, size=320, min_px=2, gain=1.0,
+                 max_px=9, min_opacity=0.0, fov=35.0, dist=2.2):
+    # Project gaussian centers with a perspective camera and paint each as a filled disk whose screen
+    # radius follows the gaussian's world-space scale, composited with a nearest-wins z-buffer.
+    # gain scales the footprint (≈ std spanned), `min_px`/`max_px` clamp the on-screen radius.
+
+    pts = xyz.astype(np.float32) @ _OBJECT_TO_VIEWER.T
+    v = pts @ _view_matrix(yaw, pitch).T
+    zc = v[:, 2] + dist
+    keep = zc > 1e-2
+    if opacity is not None and min_opacity > 0.0:  # culls gaussians with very low opacity
+        keep = keep & (opacity > min_opacity)
+    v, zc, scale = v[keep], zc[keep], scale[keep]
+    col = (np.clip(rgb, 0, 1)[:, :3] * 255).astype(np.uint8)[keep]
+    if v.shape[0] == 0:
+        return Image.fromarray(np.zeros((size, size, 3), np.uint8))
+    f = (size / 2) / np.tan(np.radians(fov) / 2)
+    cx = size / 2 + f * v[:, 0] / zc
+    cy = size / 2 + f * v[:, 1] / zc
+    radius = np.clip(np.round(f * scale / zc * gain), min_px, max_px).astype(np.int32)
+
+    # Expand each splat to its disk pixels, bucketed by integer radius so it stays vectorized.
+    px, py, pz, pc = [], [], [], []
+    for r in range(int(radius.min()), int(radius.max()) + 1):
+        m = radius == r
+        if not m.any():
+            continue
+        dy, dx = np.mgrid[-r:r + 1, -r:r + 1]
+        disk = (dx * dx + dy * dy) <= r * r
+        ox, oy = dx[disk], dy[disk]
+        px.append((cx[m, None] + ox).ravel())
+        py.append((cy[m, None] + oy).ravel())
+        pz.append(np.repeat(zc[m], ox.size))
+        pc.append(np.repeat(col[m], ox.size, axis=0))
+    px, py = np.concatenate(px), np.concatenate(py)
+    pz, pc = np.concatenate(pz), np.concatenate(pc)
+    xi = np.clip(px, 0, size - 1).astype(np.int64)
+    yi = np.clip(py, 0, size - 1).astype(np.int64)
+
+    # Nearest-wins z-buffer: pack (quantized depth, source index), per-pixel min picks the closest
+    # splat, then decode the winning index back to its color.
+    pid = yi * size + xi
+    q = np.clip((pz * 1024.0).astype(np.int64), 0, (1 << 20) - 1)  # near = small
+    key = (q << 32) | np.arange(pid.size, dtype=np.int64)
+    buf = np.full(size * size, 1 << 62, np.int64)
+    np.minimum.at(buf, pid, key)
+    img = np.zeros((size * size, 3), np.uint8)
+    hit = buf < (1 << 62)
+    img[hit] = pc[buf[hit] & 0xFFFFFFFF]
+    return Image.fromarray(img.reshape(size, size, 3))
+
+
+def _extract_latent(x0):
+    # x0 from the sampler callback is the nested latent packed to (B, 1, TOKENS*CH + 1*5);
+    # the plain single-latent case is (B, TOKENS, CH). Return the (B, TOKENS, CH) latent stream.
+    if x0.ndim == 3 and x0.shape[1] == _LATENT_TOKENS and x0.shape[2] == _LATENT_CH:
+        return x0
+    flat = x0.reshape(x0.shape[0], -1)
+    return flat[:, :_LATENT_TOKENS * _LATENT_CH].reshape(x0.shape[0], _LATENT_TOKENS, _LATENT_CH)
+
+
+def decode_x0_to_image(decoder, x0, cfg):
+    # Decode x0 at a coarse octree level / few gaussians and render a preview image.
+    latent = _extract_latent(x0)
+    fsm = decoder.first_stage_model
+    gaussian = fsm.decode(latent.to(decoder.device, decoder.vae_dtype),
+                          num_gaussians=cfg.get("gaussians", 16384), level=cfg.get("level", 5))[0]
+    xyz = gaussian.get_xyz.float().cpu().numpy()
+    rgb = gaussian._features_dc.float().cpu().numpy()[:, 0, :] * _C0 + 0.5
+    scale = gaussian.get_scaling.float().cpu().numpy().max(axis=1)  # per-splat world radius (largest axis)
+    opacity = gaussian.get_opacity.float().cpu().numpy()[:, 0]
+    return render_splat(xyz, rgb, scale, opacity=opacity, yaw=cfg.get("yaw", 35.0), pitch=cfg.get("pitch", 30.0),
+                        size=cfg.get("size", 320), min_px=1, gain=1.0, max_px=cfg.get("point_size", 3),
+                        min_opacity=0.01)
@@ -0,0 +1,382 @@
+# TripoSplat gaussian decoder ("VAE"): an octree probability decoder picks point coords, then an
+# elastic-gaussian decoder predicts per-point gaussian params. OctreeGaussianDecoder.decode() returns
+# a Gaussian. The octree sampler uses the global torch RNG (no generator) like upstream, so seed it for repeatable decodes.
+import numpy as np
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+import comfy.model_management
+import comfy.ops
+from .gaussian import build_gaussian_models
+from .model import MultiHeadRMSNorm, MLP, PcdAbsolutePositionEmbedder, attention
+
+
+# Quasi-random sampling utilities (pure functions, dtype/device-agnostic)
+
+PRIMES = [2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47, 53]
+
+
+def radical_inverse(base, n):
+    val = 0
+    inv_base = 1.0 / base
+    inv_base_n = inv_base
+    while n > 0:
+        digit = n % base
+        val += digit * inv_base_n
+        n //= base
+        inv_base_n *= inv_base
+    return val
+
+
+def halton_sequence(dim, n):
+    return [radical_inverse(PRIMES[i], n) for i in range(dim)]
+
+
+def hammersley_sequence(dim, n, num_samples):
+    return [n / num_samples] + halton_sequence(dim - 1, n)
+
+
+def sample_probs(probs, counts, generator=None):
+    # Systematic resampling: distribute counts[r] draws across the P bins of row r
+    batch_shape = counts.shape
+    R = counts.numel()
+    P = probs.size(-1)
+    device = probs.device
+    probs = probs.reshape(R, P).to(torch.float32).clamp_min(0)
+    counts = counts.reshape(R).to(device=device, dtype=torch.long)
+
+    row_sums = probs.sum(1, keepdim=True)
+    probs = torch.where(row_sums == 0, probs.new_tensor(1.0 / P), probs / row_sums.clamp_min(1))
+    cdf = probs.cumsum(dim=1).clamp(max=1.0 - 1e-12)
+
+    Nmax = int(counts.max())
+    if Nmax == 0:
+        return counts.new_zeros(*batch_shape, P)
+    cnt = counts.clamp_min(1).float().unsqueeze(1)                              # (R, 1)
+    grid = torch.arange(Nmax, device=device, dtype=torch.float32).unsqueeze(0)  # (1, Nmax)
+    u = (torch.rand(R, 1, generator=generator).to(device) + grid) / cnt         # (R, Nmax) systematic samples (CPU-seeded)
+    idx = torch.searchsorted(cdf, u.clamp(max=1.0 - 1e-12)).clamp_max(P - 1)
+    weight = (grid < counts.unsqueeze(1)).to(cdf.dtype)                         # mask out j >= counts[r]
+    out = torch.zeros(R, P, dtype=torch.float32, device=device)
+    out.scatter_add_(1, idx, weight)
+    return out.to(torch.long).view(*batch_shape, P)
+
+
+class MultiHeadAttention(nn.Module):
+    def __init__(self, channels, num_heads, ctx_channels=None, type="self", qkv_bias=True, qk_rms_norm=False,
+                 dtype=None, device=None, operations=None):
+        super().__init__()
+        assert channels % num_heads == 0
+        self.channels = channels
+        self.head_dim = channels // num_heads
+        self.ctx_channels = ctx_channels if ctx_channels is not None else channels
+        self.num_heads = num_heads
+        self._type = type
+        self.qk_rms_norm = qk_rms_norm
+        if self._type == "self":
+            self.to_qkv = operations.Linear(channels, channels * 3, bias=qkv_bias, dtype=dtype, device=device)
+        else:
+            self.to_q = operations.Linear(channels, channels, bias=qkv_bias, dtype=dtype, device=device)
+            self.to_kv = operations.Linear(self.ctx_channels, channels * 2, bias=qkv_bias, dtype=dtype, device=device)
+        if self.qk_rms_norm:
+            self.q_rms_norm = MultiHeadRMSNorm(self.head_dim, num_heads, dtype=dtype, device=device)
+            self.k_rms_norm = MultiHeadRMSNorm(self.head_dim, num_heads, dtype=dtype, device=device)
+        self.to_out = operations.Linear(channels, channels, dtype=dtype, device=device)
+
+    def forward(self, x, context=None):
+        B, L, C = x.shape
+        if self._type == "self":
+            q, k, v = self.to_qkv(x).reshape(B, L, 3, self.num_heads, -1).unbind(dim=2)
+        else:
+            Lkv = context.shape[1]
+            q = self.to_q(x).reshape(B, L, self.num_heads, -1)
+            k, v = self.to_kv(context).reshape(B, Lkv, 2, self.num_heads, -1).unbind(dim=2)
+        if self.qk_rms_norm:
+            q = self.q_rms_norm(q)
+            k = self.k_rms_norm(k)
+        h = attention(q, k, v)
+        return self.to_out(h.reshape(B, L, -1))
+
+
+# Octree probability decoder
+
+class LevelEmbedder(nn.Module):
+    def __init__(self, hidden_size, frequency_embedding_size=256, max_period=1024,
+                 dtype=None, device=None, operations=None):
+        super().__init__()
+        self.mlp = nn.Sequential(
+            operations.Linear(frequency_embedding_size, hidden_size, bias=True, dtype=dtype, device=device),
+            nn.SiLU(),
+            operations.Linear(hidden_size, hidden_size, bias=True, dtype=dtype, device=device),
+        )
+        self.frequency_embedding_size = frequency_embedding_size
+        self.max_period = max_period
+
+    @staticmethod
+    def level_embedding(t, dim, max_period=1024):
+        half = dim // 2
+        freqs = torch.exp(-np.log(max_period) * torch.arange(start=0, end=half, dtype=torch.float32) / half).to(device=t.device)
+        args = t[:, None].float() * freqs[None] * 2 * torch.pi
+        embedding = torch.cat([torch.cos(args), torch.sin(args)], dim=-1)
+        if dim % 2:
+            embedding = torch.cat([embedding, torch.zeros_like(embedding[:, :1])], dim=-1)
+        return embedding
+
+    def forward(self, t):
+        emb = self.level_embedding(t, self.frequency_embedding_size, self.max_period)
+        return self.mlp(emb.to(self.mlp[0].weight.dtype))
+
+
+class ModulatedTransformerCrossOnlyBlock(nn.Module):
+    def __init__(self, channels, ctx_channels, num_heads, mlp_ratio=4.0, share_mod=False,
+                 qk_rms_norm_cross=True, qkv_bias=True, dtype=None, device=None, operations=None):
+        super().__init__()
+        self.share_mod = share_mod
+        self.norm1 = operations.LayerNorm(channels, elementwise_affine=False, eps=1e-6, dtype=dtype, device=device)
+        self.norm2 = operations.LayerNorm(channels, elementwise_affine=False, eps=1e-6, dtype=dtype, device=device)
+        self.cross_attn = MultiHeadAttention(channels, ctx_channels=ctx_channels, num_heads=num_heads,
+                                             type="cross", qkv_bias=qkv_bias,
+                                             qk_rms_norm=qk_rms_norm_cross, dtype=dtype, device=device, operations=operations)
+        self.mlp = MLP(channels, int(channels * mlp_ratio), channels, dtype=dtype, device=device, operations=operations)
+        if not share_mod:
+            self.adaLN_modulation = nn.Sequential(
+                nn.SiLU(), operations.Linear(channels, 6 * channels, bias=True, dtype=dtype, device=device))
+
+    def forward(self, x, mod, context):
+        if self.share_mod:
+            shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = mod.chunk(6, dim=1)
+        else:
+            shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = self.adaLN_modulation(mod).chunk(6, dim=1)
+        h = torch.addcmul(shift_msa.unsqueeze(1), self.norm1(x), 1 + scale_msa.unsqueeze(1))
+        x = torch.addcmul(x, self.cross_attn(h, context), gate_msa.unsqueeze(1))
+        h = torch.addcmul(shift_mlp.unsqueeze(1), self.norm2(x), 1 + scale_mlp.unsqueeze(1))
+        x = torch.addcmul(x, self.mlp(h), gate_mlp.unsqueeze(1))
+        return x
+
+
+class OctreeProbabilityFixedlenDecoder(nn.Module):
+    # Cross-attention transformer over octree coords -> per-node 8-way child occupancy logits.
+    def __init__(self, model_channels=1024, cond_channels=16, num_blocks=4, num_heads=16,
+                 num_head_channels=64, mlp_ratio=4.0, share_mod=True,
+                 qk_rms_norm_cross=True, dtype=None, device=None, operations=None):
+        super().__init__()
+        self.model_channels = model_channels
+        self.cond_channels = cond_channels
+        self.num_blocks = num_blocks
+        self.num_heads = num_heads or model_channels // num_head_channels
+        self.mlp_ratio = mlp_ratio
+        self.share_mod = share_mod
+        self.qk_rms_norm_cross = qk_rms_norm_cross
+        self.input_layer = operations.Linear(model_channels, model_channels, dtype=dtype, device=device)
+        self.l_embedder = LevelEmbedder(model_channels, dtype=dtype, device=device, operations=operations)
+        if share_mod:
+            self.adaLN_modulation = nn.Sequential(
+                nn.SiLU(), operations.Linear(model_channels, 6 * model_channels, bias=True, dtype=dtype, device=device))
+        if cond_channels is not None:
+            self.blocks = nn.ModuleList([
+                ModulatedTransformerCrossOnlyBlock(
+                    model_channels, ctx_channels=cond_channels, num_heads=self.num_heads,
+                    mlp_ratio=self.mlp_ratio, qk_rms_norm_cross=self.qk_rms_norm_cross,
+                    share_mod=self.share_mod, dtype=dtype, device=device, operations=operations)
+                for _ in range(num_blocks)
+            ])
+        self.out_proj = operations.Linear(model_channels, 8, dtype=dtype, device=device)
+        self.in_proj = operations.Linear(3, model_channels, dtype=dtype, device=device)
+        self.pos_embedder = PcdAbsolutePositionEmbedder(channels=model_channels, in_channels=3, max_res=10, schedule="log2")
+
+    def forward(self, x, l, cond):
+        d = next(self.parameters()).dtype
+        B, L, _ = x.shape
+        h = self.in_proj(x.to(d)) + self.pos_embedder(x.reshape(-1, 3)).reshape(B, L, -1).to(d)
+        h = self.input_layer(h)
+        l_emb = self.l_embedder(l)
+        if self.share_mod:
+            l_emb = self.adaLN_modulation(l_emb)
+        cond = cond.to(d)
+        for block in self.blocks:
+            h = block(h, l_emb, cond)
+        h = F.layer_norm(h.float(), h.shape[-1:]).to(d)
+        logits = self.out_proj(h)
+        return {"logits": logits, "probs": torch.softmax(logits, dim=-1)}
+
+    @staticmethod
+    def sample(model, cond, num_points, level, temperature=1.0, generator=None):
+        B = cond.shape[0]
+        device = cond.device
+        child_offset = torch.tensor([[i, j, k] for k in [0, 1] for j in [0, 1] for i in [0, 1]],
+                                    dtype=torch.long, device=device)
+        prev_coords_int = torch.zeros(B, 1, 3, dtype=torch.long, device=device)
+        prev_counts = torch.full((B, 1), num_points, dtype=torch.long, device=device)
+        prev_log_probs = torch.zeros(B, 1, dtype=torch.float32, device=device)
+        batch_indices_range = torch.arange(B, device=device).unsqueeze(1)
+
+        for lv in range(1, level + 1):
+            res_p = 1 << (lv - 1)
+            res = 1 << lv
+            parent_coords_norm = (prev_coords_int.to(torch.float32) + 0.5) / res_p
+            res_tensor = torch.full((B,), res, dtype=torch.long, device=device)
+            pred_logits = model(parent_coords_norm, res_tensor, cond)["logits"] / temperature
+            pred_probs = torch.softmax(pred_logits, dim=-1)
+            pred_log_probs = torch.log_softmax(pred_logits, dim=-1)
+            sampled = sample_probs(pred_probs, prev_counts, generator=generator).flatten(1, 2)
+            pred_log_probs = pred_log_probs.flatten(1, 2)
+            prev_log_probs_expanded = prev_log_probs.repeat_interleave(8, dim=1)
+            child_coords_int = (prev_coords_int[:, :, None, :] * 2 + child_offset[None, None, :, :]).flatten(1, 2)
+            mask = sampled > 0
+            max_valid = mask.sum(dim=1).max().item()
+            scatter_indices = mask.cumsum(dim=1) - 1
+            valid_scatter_indices = scatter_indices[mask]
+            valid_batch_indices = batch_indices_range.expand_as(mask)[mask]
+            next_prev_coords_int = torch.zeros(B, max_valid, 3, dtype=child_coords_int.dtype, device=device)
+            next_prev_coords_int[valid_batch_indices, valid_scatter_indices] = child_coords_int[mask]
+            next_prev_counts = torch.zeros(B, max_valid, dtype=sampled.dtype, device=device)
+            next_prev_counts[valid_batch_indices, valid_scatter_indices] = sampled[mask]
+            next_prev_log_probs = torch.zeros(B, max_valid, dtype=prev_log_probs.dtype, device=device)
+            next_prev_log_probs[valid_batch_indices, valid_scatter_indices] = (prev_log_probs_expanded + pred_log_probs)[mask]
+            prev_coords_int = next_prev_coords_int
+            prev_counts = next_prev_counts
+            prev_log_probs = next_prev_log_probs
+
+        res = 1 << level
+        prev_log_probs = torch.repeat_interleave(prev_log_probs.flatten(0, 1), prev_counts.flatten(0, 1), dim=0).reshape(B, num_points)
+        coords_int = torch.repeat_interleave(prev_coords_int.flatten(0, 1), prev_counts.flatten(0, 1), dim=0).reshape(B, num_points, -1)
+        rand = torch.rand(coords_int.shape, dtype=torch.float32, generator=generator).to(device)
+        coords_norm = (coords_int.to(torch.float32) + rand) / res
+        return {"points": coords_norm, "log_probs": prev_log_probs}
+
+
+# Elastic gaussian decoder
+
+class TransformerCrossBlock(nn.Module):
+    def __init__(self, channels, ctx_channels, num_heads, mlp_ratio=4.0,
+                 qk_rms_norm=True, qk_rms_norm_cross=True, qkv_bias=True,
+                 dtype=None, device=None, operations=None):
+        super().__init__()
+        self.norm1 = operations.LayerNorm(channels, elementwise_affine=False, eps=1e-6, dtype=dtype, device=device)
+        self.norm2 = operations.LayerNorm(channels, elementwise_affine=True, eps=1e-6, dtype=dtype, device=device)
+        self.norm3 = operations.LayerNorm(channels, elementwise_affine=False, eps=1e-6, dtype=dtype, device=device)
+        self.self_attn = MultiHeadAttention(channels, num_heads=num_heads, type="self", qkv_bias=qkv_bias,
+                                            qk_rms_norm=qk_rms_norm, dtype=dtype, device=device, operations=operations)
+        self.cross_attn = MultiHeadAttention(channels, ctx_channels=ctx_channels, num_heads=num_heads, type="cross",
+                                             qkv_bias=qkv_bias, qk_rms_norm=qk_rms_norm_cross, dtype=dtype, device=device, operations=operations)
+        self.mlp = MLP(channels, int(channels * mlp_ratio), channels, dtype=dtype, device=device, operations=operations)
+
+    def forward(self, x, context):
+        x = x + self.self_attn(self.norm1(x))
+        x = x + self.cross_attn(self.norm2(x), context)
+        x = x + self.mlp(self.norm3(x))
+        return x
+
+
+class ElasticGaussianFixedlenDecoder(nn.Module):
+    # Cross-attention transformer over sampled octree points -> per-point gaussian params.
+    def __init__(self, in_channels=3, model_channels=1024, cond_channels=16, num_blocks=16, num_heads=16,
+                 num_head_channels=64, mlp_ratio=4.0, *, representation_config=None,
+                 qk_rms_norm=True, qk_rms_norm_cross=True, dtype=None, device=None, operations=None):
+        super().__init__()
+        self.rep_config = representation_config or dict(
+            lr=dict(_xyz=1.0, _features_dc=1.0, _opacity=1.0, _scaling=1.0, _rotation=0.1),
+            perturb_offset=True, perturbe_size=1.5, offset_scale=0.05, num_gaussians=32,
+            filter_kernel_size_3d=0.0009, scaling_bias=0.004, opacity_bias=0.1,
+            scaling_activation="softplus",
+        )
+        self.out_channels = self._calc_layout()
+        self.model_channels = model_channels
+        self.cond_channels = cond_channels
+        self.num_blocks = num_blocks
+        self.num_heads = num_heads or model_channels // num_head_channels
+        self.mlp_ratio = mlp_ratio
+        self.input_layer = operations.Linear(model_channels, model_channels, dtype=dtype, device=device)
+        if cond_channels is not None:
+            self.blocks = nn.ModuleList([
+                TransformerCrossBlock(model_channels, ctx_channels=cond_channels,
+                                      num_heads=self.num_heads, mlp_ratio=self.mlp_ratio,
+                                      qk_rms_norm=qk_rms_norm, qk_rms_norm_cross=qk_rms_norm_cross,
+                                      dtype=dtype, device=device, operations=operations)
+                for _ in range(num_blocks)
+            ])
+        self.in_proj = operations.Linear(in_channels, model_channels, dtype=dtype, device=device)
+        self.pos_embedder = PcdAbsolutePositionEmbedder(channels=model_channels, in_channels=3, max_res=10, schedule="log2")
+        self.out_proj = operations.Linear(model_channels, self.out_channels, dtype=dtype, device=device)
+        self._build_perturbation()
+
+    def _calc_layout(self):
+        ng = self.rep_config['num_gaussians']
+        self.layout = {
+            '_xyz':         {'shape': (ng, 3),    'size': ng * 3},
+            '_features_dc': {'shape': (ng, 1, 3), 'size': ng * 3},
+            '_scaling':     {'shape': (ng, 3),    'size': ng * 3},
+            '_rotation':    {'shape': (ng, 4),    'size': ng * 4},
+            '_opacity':     {'shape': (ng, 1),    'size': ng},
+        }
+        self.layout['_offset_scale'] = {'shape': (ng, 1), 'size': ng}
+        start = 0
+        for k, v in self.layout.items():
+            v['range'] = (start, start + v['size'])
+            start += v['size']
+        return start
+
+    def _build_perturbation(self):
+        ng = self.rep_config['num_gaussians']
+        perturbation = torch.tensor([hammersley_sequence(3, i, ng) for i in range(ng)]).float()
+        perturbation = torch.atanh((perturbation * 2 - 1) / self.rep_config['perturbe_size'])
+        self.register_buffer('points_offset_perturbation', perturbation)
+        base = torch.tensor(self.rep_config['offset_scale'])
+        self.register_buffer('base_offset_scale', torch.log(torch.exp(base) - 1.0))
+
+    def _get_offset(self, h):
+        B = h.shape[0]
+        r = self.layout['_offset_scale']['range']
+        _offset_scale = F.softplus(
+            h[:, :, r[0]:r[1]].reshape(B, -1, *self.layout['_offset_scale']['shape'])
+            + comfy.model_management.cast_to(self.base_offset_scale, h.dtype, h.device))
+
+        r = self.layout['_xyz']['range']
+        offset = h[:, :, r[0]:r[1]].reshape(B, -1, *self.layout['_xyz']['shape'])
+        offset = offset * self.rep_config['lr']['_xyz']
+        if self.rep_config['perturb_offset']:
+            offset = offset + comfy.model_management.cast_to(self.points_offset_perturbation, offset.dtype, offset.device)
+        offset = torch.tanh(offset) * 0.5 * self.rep_config['perturbe_size']
+        offset = offset * _offset_scale
+        return offset
+
+    def forward(self, x=None, cond=None):
+        pcd = x["points"]
+        d = next(self.parameters()).dtype
+        B, L, _ = pcd.shape
+        h = self.in_proj(pcd.to(d)) + self.pos_embedder(pcd.reshape(-1, 3)).reshape(B, L, -1).to(d)
+        h = self.input_layer(h)
+        cond = cond.to(d)
+        for block in self.blocks:
+            h = block(h, cond)
+        h = F.layer_norm(h.float(), h.shape[-1:]).to(h.dtype)
+        return {"features": self.out_proj(h)}
+
+
+# Combined octree gaussian decoder (comfy first-stage model)
+
+class OctreeGaussianDecoder(nn.Module):
+    _MAX_VOXEL_LEVEL = 8
+
+    def __init__(self, dtype=None, device=None, operations=None):
+        super().__init__()
+        if operations is None:
+            operations = comfy.ops.disable_weight_init
+        self.octree = OctreeProbabilityFixedlenDecoder(dtype=dtype, device=device, operations=operations)
+        self.gs = ElasticGaussianFixedlenDecoder(dtype=dtype, device=device, operations=operations)
+
+    @property
+    def gaussians_per_point(self) -> int:
+        return self.gs.rep_config['num_gaussians']
+
+    def decode(self, latent: torch.Tensor, num_gaussians: int, level: int = None, generator=None):
+        # level defaults to the full octree depth, a lower level is cheaper (coarser) for live previews.
+        # generator (a CPU torch.Generator) makes the octree sampling reproducible without touching global RNG.
+        level = self._MAX_VOXEL_LEVEL if level is None else level
+        num_decoder_tokens = max(1, num_gaussians // self.gaussians_per_point)
+        points_pred = OctreeProbabilityFixedlenDecoder.sample(
+            self.octree, latent, num_points=num_decoder_tokens, level=level, temperature=1.0, generator=generator,
+        )
+        pred = self.gs(x=points_pred, cond=latent)
+        return build_gaussian_models(self.gs, points_pred, pred)  # one GaussianModel per batch item
@@ -8,7 +8,7 @@ from einops import rearrange

 from comfy.ldm.modules.attention import optimized_attention
 from comfy.ldm.flux.layers import EmbedND
-from comfy.ldm.flux.math import apply_rope1
+from comfy.ldm.flux.math import apply_rope1, rope
 import comfy.ldm.common_dit
 import comfy.model_management
 import comfy.patcher_extension
@@ -570,6 +570,14 @@ class WanModel(torch.nn.Module):
                full_ref = self.ref_conv(full_ref).flatten(2).transpose(1, 2)
                x = torch.concat((full_ref, x), dim=1)

+        # In-context reference (Bernini)
+        context_latents = kwargs.get("context_latents", None)
+        main_len = x.shape[1]
+        if context_latents is not None:
+            for lat in context_latents:
+                cl = self.patch_embedding(lat.float().to(x.device)).to(x.dtype).flatten(2).transpose(1, 2)
+                x = torch.cat([x, cl], dim=1)
+
        # context
        context = self.text_embedding(context)

@@ -599,6 +607,9 @@ class WanModel(torch.nn.Module):
        # head
        x = self.head(x, e)

+        if context_latents is not None:
+            x = x[:, :main_len]
+
        if full_ref is not None:
            x = x[:, full_ref.shape[1]:]

@@ -606,7 +617,7 @@ class WanModel(torch.nn.Module):
        x = self.unpatchify(x, grid_sizes)
        return x

-    def rope_encode(self, t, h, w, t_start=0, steps_t=None, steps_h=None, steps_w=None, device=None, dtype=None, transformer_options={}):
+    def rope_encode(self, t, h, w, t_start=0, steps_t=None, steps_h=None, steps_w=None, device=None, dtype=None, transformer_options={}, source_id=0):
        patch_size = self.patch_size
        t_len = ((t + (patch_size[0] // 2)) // patch_size[0])
        h_len = ((h + (patch_size[1] // 2)) // patch_size[1])
@@ -638,6 +649,13 @@ class WanModel(torch.nn.Module):
        img_ids = img_ids.reshape(1, -1, img_ids.shape[-1])

        freqs = self.rope_embedder(img_ids).movedim(1, 2)
+
+        # In-context reference: a non-zero source_id composes an extra rotation into the spatial rope
+        if source_id:
+            d = self.dim // self.num_heads
+            pos = torch.tensor([[float(source_id)]], device=freqs.device, dtype=torch.float32)
+            id_rot = rope(pos, d, self.rope_embedder.theta).reshape(1, 1, 1, d // 2, 2, 2).to(freqs.dtype)
+            freqs = torch.einsum('...ij,...jk->...ik', freqs, id_rot)
        return freqs

    def forward(self, x, timestep, context, clip_fea=None, time_dim_concat=None, transformer_options={}, **kwargs):
@@ -661,6 +679,15 @@ class WanModel(torch.nn.Module):
            t_len += 1

        freqs = self.rope_encode(t_len, h, w, device=x.device, dtype=x.dtype, transformer_options=transformer_options)
+
+        # In-context reference: one rope block per stream, each with it's own source_id (1, 2, ...) to distinguish from the target (id 0).
+        context_latents = kwargs.get("context_latents", None)
+        if context_latents is not None:
+            context_latents = [comfy.ldm.common_dit.pad_to_patch_size(lat, self.patch_size) for lat in context_latents]
+            for i, lat in enumerate(context_latents):
+                freqs = torch.cat([freqs, self.rope_encode(lat.shape[-3], lat.shape[-2], lat.shape[-1], device=x.device, dtype=x.dtype, transformer_options=transformer_options, source_id=i + 1)], dim=1)
+            kwargs = {**kwargs, "context_latents": context_latents}
+
        return self.forward_orig(x, timestep, context, clip_fea=clip_fea, freqs=freqs, transformer_options=transformer_options, **kwargs)[:, :, :t, :h, :w]

    def unpatchify(self, x, grid_sizes):
@@ -1631,13 +1658,15 @@ class SCAILWanModel(WanModel):

        self.patch_embedding_pose = operations.Conv3d(in_dim, dim, kernel_size=patch_size, stride=patch_size, device=device, dtype=torch.float32)

-    def forward_orig(self, x, t, context, clip_fea=None, freqs=None, transformer_options={}, pose_latents=None, reference_latent=None, **kwargs):
+    def forward_orig(self, x, t, context, clip_fea=None, freqs=None, transformer_options={}, pose_latents=None, reference_latent=None, ref_mask_latents=None, sam_latents=None, **kwargs):

        if reference_latent is not None:
            x = torch.cat((reference_latent, x), dim=2)

        # embeddings
        x = self.patch_embedding(x.float()).to(x.dtype)
+        if ref_mask_latents is not None:  # SCAIL-2 additive mask stream (one identity mask frame per reference, then video)
+            x = x + self.patch_embedding_mask(ref_mask_latents.float()).to(x.dtype)
        grid_sizes = x.shape[2:]
        transformer_options["grid_sizes"] = grid_sizes
        x = x.flatten(2).transpose(1, 2)
@@ -1645,6 +1674,8 @@ class SCAILWanModel(WanModel):
        scail_pose_seq_len = 0
        if pose_latents is not None:
            scail_x = self.patch_embedding_pose(pose_latents.float()).to(x.dtype)
+            if sam_latents is not None:  # SCAIL-2 additive mask stream
+                scail_x = scail_x + self.patch_embedding_mask(sam_latents.float()).to(x.dtype)
            scail_x = scail_x.flatten(2).transpose(1, 2)
            scail_pose_seq_len = scail_x.shape[1]
            x = torch.cat([x, scail_x], dim=1)
@@ -1695,16 +1726,44 @@ class SCAILWanModel(WanModel):

        return x

-    def rope_encode(self, t, h, w, t_start=0, steps_t=None, steps_h=None, steps_w=None, device=None, dtype=None, pose_latents=None, reference_latent=None, transformer_options={}):
+    # ref_mask_flag is a scalar bool (CONDConstant, SCAIL-2 only). False => replacement mode,
+    # which places ref/pose via H/W rope shifts instead of the animation-mode temporal offset.
+    # reference_latent may stack several frames: the last is the primary reference adjacent to the video, the earlier frames are additional references.
+    def rope_encode(self, t, h, w, t_start=0, steps_t=None, steps_h=None, steps_w=None, device=None, dtype=None, pose_latents=None, reference_latent=None, ref_mask_flag=None, transformer_options={}):
+        ref_t_patches = 0
+        if reference_latent is not None:
+            ref_t_patches = (reference_latent.shape[2] + (self.patch_size[0] // 2)) // self.patch_size[0]
+
+        if ref_mask_flag is not None and not bool(ref_mask_flag):
+            REF_ROPE_H = 120.0
+            POSE_ROPE_W = 120.0
+
+            main_t_patches = t - ref_t_patches
+            video_t_start = max(ref_t_patches - 1, 0)
+
+            parts = []
+            if ref_t_patches > 0:
+                ref_tf = {"rope_options": {"shift_y": REF_ROPE_H, "shift_x": 0.0, "scale_y": 1.0, "scale_x": 1.0}}
+                parts.append(super().rope_encode(ref_t_patches, h, w, t_start=0, device=device, dtype=dtype, transformer_options=ref_tf))
+            if main_t_patches > 0:
+                parts.append(super().rope_encode(main_t_patches, h, w, t_start=video_t_start, device=device, dtype=dtype, transformer_options=transformer_options))
+
+            if pose_latents is not None:
+                F_pose, H_pose, W_pose = pose_latents.shape[-3], pose_latents.shape[-2], pose_latents.shape[-1]
+                h_scale = h / H_pose
+                w_scale = w / W_pose
+                h_shift = (h_scale - 1) / 2
+                w_shift = (w_scale - 1) / 2
+                pose_tf = {"rope_options": {"shift_y": h_shift, "shift_x": POSE_ROPE_W + w_shift, "scale_y": h_scale, "scale_x": w_scale}}
+                parts.append(super().rope_encode(F_pose, H_pose, W_pose, t_start=video_t_start, device=device, dtype=dtype, transformer_options=pose_tf))
+
+            return torch.cat(parts, dim=1)
+
        main_freqs = super().rope_encode(t, h, w, t_start=t_start, steps_t=steps_t, steps_h=steps_h, steps_w=steps_w, device=device, dtype=dtype, transformer_options=transformer_options)

        if pose_latents is None:
            return main_freqs

-        ref_t_patches = 0
-        if reference_latent is not None:
-            ref_t_patches = (reference_latent.shape[2] + (self.patch_size[0] // 2)) // self.patch_size[0]
-
        F_pose, H_pose, W_pose = pose_latents.shape[-3], pose_latents.shape[-2], pose_latents.shape[-1]

        # if pose is at half resolution, scale_y/scale_x=2 stretches the position range to cover the same RoPE extent as the main frames
@@ -1719,12 +1778,16 @@ class SCAILWanModel(WanModel):

        return torch.cat([main_freqs, pose_freqs], dim=1)

-    def _forward(self, x, timestep, context, clip_fea=None, time_dim_concat=None, transformer_options={}, pose_latents=None, **kwargs):
+    def _forward(self, x, timestep, context, clip_fea=None, time_dim_concat=None, transformer_options={}, pose_latents=None, ref_mask_latents=None, sam_latents=None, **kwargs):
        bs, c, t, h, w = x.shape
        x = comfy.ldm.common_dit.pad_to_patch_size(x, self.patch_size)

        if pose_latents is not None:
            pose_latents = comfy.ldm.common_dit.pad_to_patch_size(pose_latents, self.patch_size)
+        if ref_mask_latents is not None:  # SCAIL-2
+            ref_mask_latents = comfy.ldm.common_dit.pad_to_patch_size(ref_mask_latents, self.patch_size)
+        if sam_latents is not None:  # SCAIL-2
+            sam_latents = comfy.ldm.common_dit.pad_to_patch_size(sam_latents, self.patch_size)

        t_len = t
        if time_dim_concat is not None:
@@ -1737,5 +1800,15 @@ class SCAILWanModel(WanModel):
            reference_latent = comfy.ldm.common_dit.pad_to_patch_size(kwargs.pop("reference_latent"), self.patch_size)
            t_len += reference_latent.shape[2]

-        freqs = self.rope_encode(t_len, h, w, device=x.device, dtype=x.dtype, transformer_options=transformer_options, pose_latents=pose_latents, reference_latent=reference_latent)
-        return self.forward_orig(x, timestep, context, clip_fea=clip_fea, freqs=freqs, transformer_options=transformer_options, pose_latents=pose_latents, reference_latent=reference_latent, **kwargs)[:, :, :t, :h, :w]
+        ref_mask_flag = kwargs.pop("ref_mask_flag", None)  # SCAIL-2
+
+        freqs = self.rope_encode(t_len, h, w, device=x.device, dtype=x.dtype, transformer_options=transformer_options, pose_latents=pose_latents, reference_latent=reference_latent, ref_mask_flag=ref_mask_flag)
+        return self.forward_orig(x, timestep, context, clip_fea=clip_fea, freqs=freqs, transformer_options=transformer_options, pose_latents=pose_latents, reference_latent=reference_latent, ref_mask_latents=ref_mask_latents, sam_latents=sam_latents, **kwargs)[:, :, :t, :h, :w]
+
+
+class SCAIL2WanModel(SCAILWanModel):
+    """SCAIL-2: SCAIL-Preview + an additive binary multi-identity mask stream."""
+
+    def __init__(self, model_type="scail2", patch_size=(1, 2, 2), in_dim=20, mask_in_dim=28, dim=5120, operations=None, device=None, dtype=None, **kwargs):
+        super().__init__(model_type=model_type, patch_size=patch_size, in_dim=in_dim, dim=dim, operations=operations, device=device, dtype=dtype, **kwargs)
+        self.patch_embedding_mask = operations.Conv3d(mask_in_dim, dim, kernel_size=patch_size, stride=patch_size, device=device, dtype=torch.float32)
@@ -326,6 +326,17 @@ def model_lora_keys_unet(model, key_map={}):
                key_map["transformer.{}".format(key_lora)] = k
                key_map["lycoris_{}".format(key_lora.replace(".", "_"))] = k #SimpleTuner lycoris format

+    if isinstance(model, comfy.model_base.Krea2):
+        diffusers_keys = comfy.utils.krea2_to_diffusers(model.model_config.unet_config, output_prefix="diffusion_model.")
+        for k in diffusers_keys:
+            if k.endswith(".weight"):
+                to = diffusers_keys[k]
+                key_lora = k[:-len(".weight")]
+                key_map["diffusion_model.{}".format(key_lora)] = to
+                key_map["transformer.{}".format(key_lora)] = to
+                key_map["lycoris_{}".format(key_lora.replace(".", "_"))] = to
+                key_map[key_lora] = to
+
    if isinstance(model, comfy.model_base.Lumina2):
        diffusers_keys = comfy.utils.z_image_to_diffusers(model.model_config.unet_config, output_prefix="diffusion_model.")
        for k in diffusers_keys:
@@ -357,6 +368,12 @@ def model_lora_keys_unet(model, key_map={}):
                key_lora = k[len("diffusion_model."):-len(".weight")]
                key_map["transformer.{}".format(key_lora)] = k

+    if isinstance(model, (comfy.model_base.LTXV, comfy.model_base.LTXAV)):
+        for k in sdk:
+            if k.startswith("diffusion_model.") and k.endswith(".weight"):
+                key_lora = k[len("diffusion_model."):-len(".weight")]
+                key_map["{}".format(key_lora)] = k
+
    return key_map


@@ -4,6 +4,7 @@ import dataclasses
 import torch
 from typing import NamedTuple

+import comfy_aimdo.host_buffer
 from comfy.quant_ops import QuantizedTensor


@@ -17,21 +18,18 @@ class TensorFileSlice(NamedTuple):
 def read_tensor_file_slice_into(tensor, destination, stream=None, destination2=None):

    if isinstance(tensor, QuantizedTensor):
-        if not isinstance(destination, QuantizedTensor):
-            return False
-        if tensor._layout_cls != destination._layout_cls:
-            return False
-
-        if not read_tensor_file_slice_into(tensor._qdata, destination._qdata, stream=stream,
+        if not read_tensor_file_slice_into(tensor._qdata,
+                                           destination._qdata if destination is not None else None, stream=stream,
                                           destination2=(destination2._qdata if destination2 is not None else None)):
            return False

-        dst_orig_dtype = destination._params.orig_dtype
-        destination._params.copy_from(tensor._params, non_blocking=False)
-        destination._params = dataclasses.replace(destination._params, orig_dtype=dst_orig_dtype)
+        if destination is not None:
+            dst_orig_dtype = destination._params.orig_dtype
+            destination._params.copy_from(tensor._params, non_blocking=False)
+            destination._params = dataclasses.replace(destination._params, orig_dtype=dst_orig_dtype)
        if destination2 is not None:
            dst_orig_dtype = destination2._params.orig_dtype
-            destination2._params.copy_from(destination._params, non_blocking=True)
+            destination2._params.copy_from(destination._params if destination is not None else tensor._params, non_blocking=True)
            destination2._params = dataclasses.replace(destination2._params, orig_dtype=dst_orig_dtype)
        return True

@@ -39,10 +37,15 @@ def read_tensor_file_slice_into(tensor, destination, stream=None, destination2=N
    if info is None:
        return False

+    if destination is not None and destination.device.type != "cpu" and destination2 is None:
+        destination2 = destination
+        destination = None
+
    file_obj = info.file_ref
-    if (destination.device.type != "cpu"
-            or file_obj is None
-            or destination.numel() * destination.element_size() < info.size
+    if (file_obj is None
+            or (destination is None and destination2 is None)
+            or (destination is not None and (destination.device.type != "cpu" or destination.numel() * destination.element_size() < info.size))
+            or (destination2 is not None and (destination2.device.type == "cpu" or destination2.numel() * destination2.element_size() < info.size))
            or tensor.numel() * tensor.element_size() != info.size
            or tensor.storage_offset() != 0
            or not tensor.is_contiguous()):
@@ -51,6 +54,14 @@ def read_tensor_file_slice_into(tensor, destination, stream=None, destination2=N
    if info.size == 0:
        return True

+    if destination is None:
+        stream_ptr = getattr(stream, "cuda_stream", 0) if stream is not None else 0
+        comfy_aimdo.host_buffer.read_file_to_device(file_obj, info.offset, info.size,
+                                                    stream_ptr, destination2.data_ptr(),
+                                                    destination2.device.index,
+                                                    mark_cold=False)
+        return True
+
    hostbuf = getattr(destination.untyped_storage(), "_comfy_hostbuf", None)
    if hostbuf is not None:
        stream_ptr = getattr(stream, "cuda_stream", 0) if stream is not None else 0
@@ -63,6 +74,9 @@ def read_tensor_file_slice_into(tensor, destination, stream=None, destination2=N
                                    device=None if destination2 is None else destination2.device.index)
        return True

+    if not hasattr(file_obj, "seek") or not hasattr(file_obj, "readinto"):
+        return False
+
    buf_type = ctypes.c_ubyte * info.size
    view = memoryview(buf_type.from_address(destination.data_ptr()))

@@ -21,6 +21,7 @@ import comfy.ldm.hunyuan3dv2_1.hunyuandit
 import torch
 import logging
 import comfy.ldm.lightricks.av_model
+import comfy.ldm.lightricks.symmetric_patchifier
 import comfy.context_windows
 from comfy.ldm.modules.diffusionmodules.openaimodel import UNetModel, Timestep
 from comfy.ldm.cascade.stage_c import StageC
@@ -46,12 +47,18 @@ import comfy.ldm.wan.model_animate
 import comfy.ldm.wan.ar_model
 import comfy.ldm.wan.model_wandancer
 import comfy.ldm.hunyuan3d.model
+import comfy.ldm.triposplat.model
 import comfy.ldm.hidream.model
 import comfy.ldm.chroma.model
 import comfy.ldm.chroma_radiance.model
+import comfy.ldm.pixeldit.model
+import comfy.ldm.pixeldit.pid
 import comfy.ldm.ace.model
 import comfy.ldm.omnigen.omnigen2
+import comfy.ldm.boogu.model
 import comfy.ldm.qwen_image.model
+import comfy.ldm.ideogram4.model
+import comfy.ldm.krea2.model
 import comfy.ldm.kandinsky5.model
 import comfy.ldm.anima.model
 import comfy.ldm.ace.ace_step15
@@ -61,6 +68,7 @@ import comfy.ldm.ernie.model
 import comfy.ldm.sam3.detector
 import comfy.ldm.hidream_o1.model
 from comfy.ldm.hidream_o1.conditioning import build_extra_conds
+import comfy.ldm.depth_anything_3.model

 import comfy.model_management
 import comfy.patcher_extension
@@ -1198,6 +1206,127 @@ class LTXAV(BaseModel):
    def scale_latent_inpaint(self, sigma, noise, latent_image, **kwargs):
        return latent_image

+    def map_context_window_to_modalities(self, primary_indices, latent_shapes, dim):
+        result = [primary_indices]
+        if len(latent_shapes) < 2:
+            return result
+
+        video_total = latent_shapes[0][dim]
+
+        for i in range(1, len(latent_shapes)):
+            mod_total = latent_shapes[i][dim]
+            # Map each primary index to its proportional range of modality indices and
+            # concatenate in order. Preserves wrapped/strided geometry so the modality
+            # attends to the same temporal regions as the primary window.
+            mod_indices = []
+            seen = set()
+            for v_idx in primary_indices:
+                a_start = min(int(round(v_idx * mod_total / video_total)), mod_total - 1)
+                a_end = min(int(round((v_idx + 1) * mod_total / video_total)), mod_total)
+                if a_end <= a_start:
+                    a_end = a_start + 1
+                for a in range(a_start, a_end):
+                    if a not in seen:
+                        seen.add(a)
+                        mod_indices.append(a)
+            result.append(mod_indices)
+
+        return result
+
+    @staticmethod
+    def _get_guide_entries(conds):
+        for cond_list in conds:
+            if cond_list is None:
+                continue
+            for cond_dict in cond_list:
+                model_conds = cond_dict.get('model_conds', {})
+                entries = model_conds.get('guide_attention_entries')
+                if entries is not None and hasattr(entries, 'cond') and entries.cond:
+                    return entries.cond
+        return None
+
+    def resize_cond_for_context_window(self, cond_key, cond_value, window, x_in, device, retain_index_list=[]):
+        # Audio denoise mask — slice using audio modality window
+        if cond_key == "audio_denoise_mask" and hasattr(window, 'modality_windows') and window.modality_windows:
+            audio_window = window.modality_windows.get(1)
+            if audio_window is not None and hasattr(cond_value, "cond") and isinstance(cond_value.cond, torch.Tensor):
+                sliced = audio_window.get_tensor(cond_value.cond, device, dim=2)
+                return cond_value._copy_with(sliced)
+
+        # Video denoise mask — split into video + guide portions, slice each
+        if cond_key == "denoise_mask" and hasattr(cond_value, "cond") and isinstance(cond_value.cond, torch.Tensor):
+            cond_tensor = cond_value.cond
+            guide_count = cond_tensor.size(window.dim) - x_in.size(window.dim)
+            if guide_count > 0:
+                T_video = x_in.size(window.dim)
+                video_mask = cond_tensor.narrow(window.dim, 0, T_video)
+                guide_mask = cond_tensor.narrow(window.dim, T_video, guide_count)
+                sliced_video = window.get_tensor(video_mask, device, retain_index_list=retain_index_list)
+                suffix_indices = window.guide_frames_indices
+                if suffix_indices:
+                    idx = tuple([slice(None)] * window.dim + [suffix_indices])
+                    sliced_guide = guide_mask[idx].to(device)
+                    return cond_value._copy_with(torch.cat([sliced_video, sliced_guide], dim=window.dim))
+                else:
+                    return cond_value._copy_with(sliced_video)
+
+        # Keyframe indices — regenerate pixel coords for window, select guide positions
+        if cond_key == "keyframe_idxs":
+            kf_local_pos = window.guide_kf_local_positions
+            if not kf_local_pos:
+                return cond_value._copy_with(cond_value.cond[:, :, :0, :])  # empty
+            H, W = x_in.shape[3], x_in.shape[4]
+            window_len = len(window.index_list)
+            # account for causal_window_fix anchor in coord space size
+            anchor_idx = getattr(window, 'causal_anchor_index', None)
+            if anchor_idx is not None and anchor_idx >= 0:
+                window_len += 1
+            patchifier = self.diffusion_model.patchifier
+            latent_coords = patchifier.get_latent_coords(window_len, H, W, 1, cond_value.cond.device)
+            scale_factors = self.diffusion_model.vae_scale_factors
+            pixel_coords = comfy.ldm.lightricks.symmetric_patchifier.latent_to_pixel_coords(
+                latent_coords,
+                scale_factors,
+                causal_fix=self.diffusion_model.causal_temporal_positioning)
+            tokens = []
+            for pos in kf_local_pos:
+                tokens.extend(range(pos * H * W, (pos + 1) * H * W))
+            pixel_coords = pixel_coords[:, :, tokens, :]
+
+            # Adjust spatial end positions for dilated (downscaled) guides.
+            # Each guide entry may have a different downscale factor; expand the
+            # per-entry factor to cover all tokens belonging to that entry.
+            downscale_factors = window.guide_downscale_factors
+            overlap_info = window.guide_overlap_info
+            if downscale_factors:
+                per_token_factor = []
+                for (entry_idx, overlap_count), dsf in zip(overlap_info, downscale_factors):
+                    per_token_factor.extend([dsf] * (overlap_count * H * W))
+                factor_tensor = torch.tensor(per_token_factor, device=pixel_coords.device, dtype=pixel_coords.dtype)
+                spatial_end_offset = (factor_tensor.unsqueeze(0).unsqueeze(0).unsqueeze(-1) - 1) * torch.tensor(
+                    scale_factors[1:], device=pixel_coords.device, dtype=pixel_coords.dtype,
+                ).view(1, -1, 1, 1)
+                pixel_coords[:, 1:, :, 1:] += spatial_end_offset
+
+            B = cond_value.cond.shape[0]
+            if B > 1:
+                pixel_coords = pixel_coords.expand(B, -1, -1, -1)
+            return cond_value._copy_with(pixel_coords)
+
+        # Guide attention entries — adjust per-guide counts based on window overlap
+        if cond_key == "guide_attention_entries":
+            overlap_info = window.guide_overlap_info
+            H, W = x_in.shape[3], x_in.shape[4]
+            new_entries = []
+            for entry_idx, overlap_count in overlap_info:
+                e = cond_value.cond[entry_idx]
+                new_entries.append({**e,
+                    "pre_filter_count": overlap_count * H * W,
+                    "latent_shape": [overlap_count, H, W]})
+            return cond_value._copy_with(new_entries)
+
+        return None
+
 class HunyuanVideo(BaseModel):
    def __init__(self, model_config, model_type=ModelType.FLOW, device=None):
        super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.hunyuan_video.model.HunyuanVideo)
@@ -1397,6 +1526,53 @@ class ZImagePixelSpace(Lumina2):
        BaseModel.__init__(self, model_config, model_type, device=device, unet_model=comfy.ldm.lumina.model.NextDiTPixelSpace)
        self.memory_usage_factor_conds = ("ref_latents",)

+
+class PixelDiTT2I(BaseModel):
+    def __init__(self, model_config, model_type=ModelType.FLOW, device=None):
+        super().__init__(model_config, model_type, device=device,
+                         unet_model=comfy.ldm.pixeldit.model.PixDiT_T2I)
+
+    def extra_conds(self, **kwargs):
+        out = super().extra_conds(**kwargs)
+        attention_mask = kwargs.get("attention_mask", None)
+        if attention_mask is not None:
+            out["attention_mask"] = comfy.conds.CONDRegular(attention_mask)
+        return out
+
+
+class PiD(PixelDiTT2I):
+    def __init__(self, model_config, model_type=ModelType.FLOW, device=None):
+        BaseModel.__init__(self, model_config, model_type, device=device,
+                           unet_model=comfy.ldm.pixeldit.pid.PidNet)
+
+    def extra_conds(self, **kwargs):
+        out = super().extra_conds(**kwargs)
+        lq_latent = kwargs.get("lq_latent", None)
+        if lq_latent is not None:
+            out["lq_latent"] = comfy.conds.CONDRegular(lq_latent)
+        degrade_sigma = kwargs.get("degrade_sigma", None)
+        if degrade_sigma is not None:
+            out["degrade_sigma"] = comfy.conds.CONDRegular(degrade_sigma)
+        return out
+
+    def resize_cond_for_context_window(self, cond_key, cond_value, window, x_in, device, retain_index_list=[]):
+        if cond_key == "lq_latent" and hasattr(cond_value, "cond") and isinstance(cond_value.cond, torch.Tensor):
+            lq = cond_value.cond
+            dim = window.dim
+            if dim >= lq.ndim:
+                return None
+            lq_proj = self.diffusion_model.lq_proj
+            ratio = lq_proj.sr_scale * lq_proj.latent_spatial_down_factor
+            # Map x window indices -> lq indices (deduplicated, sorted, in-bounds).
+            lq_size = lq.size(dim)
+            lq_indices = sorted({i // ratio for i in window.index_list if 0 <= i // ratio < lq_size})
+            if not lq_indices:
+                return None
+            idx = tuple([slice(None)] * dim + [lq_indices])
+            return cond_value._copy_with(lq[idx].to(device))
+        return super().resize_cond_for_context_window(cond_key, cond_value, window, x_in, device, retain_index_list=retain_index_list)
+
+
 class WAN21(BaseModel):
    def __init__(self, model_config, model_type=ModelType.FLOW, image_to_video=False, device=None):
        super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.wan.model.WanModel)
@@ -1467,8 +1643,26 @@ class WAN21(BaseModel):
        if reference_latents is not None:
            out['reference_latent'] = comfy.conds.CONDRegular(self.process_latent_in(reference_latents[-1])[:, :, 0])

+        # In-context reference conditioning (Bernini)
+        context_latents = kwargs.get("context_latents", None)
+        if context_latents is not None:
+            out['context_latents'] = comfy.conds.CONDList([self.process_latent_in(l) for l in context_latents])
+
        return out

+    def resize_cond_for_context_window(self, cond_key, cond_value, window, x_in, device, retain_index_list=[]):
+        # In-context cond slicing (Bernini)
+        if cond_key == "context_latents" and isinstance(getattr(cond_value, "cond", None), list):
+            dim = window.dim
+            out = []
+            for lat in cond_value.cond:
+                if lat.ndim > dim and lat.shape[dim] > 1 and lat.shape[dim] == x_in.shape[dim]:
+                    out.append(window.get_tensor(lat, device, dim=dim, retain_index_list=retain_index_list))
+                else:
+                    out.append(lat.to(device))
+            return cond_value._copy_with(out)
+        return super().resize_cond_for_context_window(cond_key, cond_value, window, x_in, device, retain_index_list=retain_index_list)
+

 class WAN21_CausalAR(WAN21):
    def __init__(self, model_config, model_type=ModelType.FLOW, device=None):
@@ -1677,10 +1871,14 @@ class WAN21_SCAIL(WAN21):

        reference_latents = kwargs.get("reference_latents", None)
        if reference_latents is not None:
-            ref_latent = self.process_latent_in(reference_latents[-1])
-            ref_mask = torch.ones_like(ref_latent[:, :4])
-            ref_latent = torch.cat([ref_latent, ref_mask], dim=1)
-            out['reference_latent'] = comfy.conds.CONDRegular(ref_latent)
+            # SCAIL-2 multi-reference: reference_latents[0] is the primary ref, [1:] are additional
+            # references. Stack as [additional..., primary] so the primary stays adjacent to the video.
+            ordered = list(reference_latents[1:]) + list(reference_latents[:1])
+            stacked = []
+            for lat in ordered:
+                lat = self.process_latent_in(lat)
+                stacked.append(torch.cat([lat, torch.ones_like(lat[:, :4])], dim=1))
+            out['reference_latent'] = comfy.conds.CONDRegular(torch.cat(stacked, dim=2))

        pose_latents = kwargs.get("pose_video_latent", None)
        if pose_latents is not None:
@@ -1703,6 +1901,99 @@ class WAN21_SCAIL(WAN21):

        return out

+class WAN21_SCAIL2(WAN21_SCAIL):
+    """SCAIL-2: SCAIL-Preview + an additive binary multi-identity mask stream."""
+
+    def __init__(self, model_config, model_type=ModelType.FLOW, image_to_video=False, device=None):
+        super(WAN21, self).__init__(model_config, model_type, device=device, unet_model=comfy.ldm.wan.model.SCAIL2WanModel)
+        self.memory_usage_factor_conds = ("reference_latent", "pose_latents", "ref_mask_latents", "sam_latents")
+        self.memory_usage_shape_process = {
+            "pose_latents": lambda shape: [shape[0], shape[1], 1.5, shape[-2], shape[-1]],
+            "sam_latents":  lambda shape: [shape[0], shape[1], 1.5, shape[-2], shape[-1]],
+        }
+        self.image_to_video = image_to_video
+
+    def extra_conds(self, **kwargs):
+        out = super().extra_conds(**kwargs)
+
+        driving_mask_28ch = kwargs.get("driving_mask_28ch", None)
+        if driving_mask_28ch is not None:
+            out['sam_latents'] = comfy.conds.CONDRegular(driving_mask_28ch.movedim(1, 2).contiguous())
+
+        # ref_mask_28ch holds one identity mask per stacked reference frame (additional refs first, then the primary ref), followed by zeros over the video frames.
+        ref_mask_28ch = kwargs.get("ref_mask_28ch", None)
+        if ref_mask_28ch is not None:
+            out['ref_mask_latents'] = comfy.conds.CONDRegular(ref_mask_28ch.movedim(1, 2).contiguous())
+
+        ref_mask_flag = kwargs.get("ref_mask_flag", None)
+        if ref_mask_flag is not None:
+            out['ref_mask_flag'] = comfy.conds.CONDConstant(ref_mask_flag)
+
+        return out
+
+    def extra_conds_shapes(self, **kwargs):
+        out = super().extra_conds_shapes(**kwargs)
+        driving_mask_28ch = kwargs.get("driving_mask_28ch", None)
+        if driving_mask_28ch is not None:
+            s = driving_mask_28ch.shape
+            out['sam_latents'] = [s[0], 28, s[1], s[3], s[4]]
+        ref_mask_28ch = kwargs.get("ref_mask_28ch", None)
+        if ref_mask_28ch is not None:
+            s = ref_mask_28ch.shape
+            out['ref_mask_latents'] = [s[0], 28, s[1], s[3], s[4]]
+        return out
+
+    def resize_cond_for_context_window(self, cond_key, cond_value, window, x_in, device, retain_index_list=[]):
+        if cond_key in ("sam_latents", "pose_latents"):
+            # Return sliced view omitting retain_index_list
+            return comfy.context_windows.slice_cond(cond_value, window, x_in, device, temporal_dim=2, temporal_offset=0)
+        if cond_key == "ref_mask_latents" and hasattr(cond_value, "cond") and isinstance(cond_value.cond, torch.Tensor):
+            # The ref mask is N leading ref frames padded with frames of zeros, so just grab the first frames for all windows
+            full_ref_mask = cond_value.cond
+            video_frame_count = x_in.shape[2]
+            ref_frame_count = full_ref_mask.shape[2] - video_frame_count
+            if ref_frame_count < 1:
+                return None
+            window_length = len(window.index_list)
+
+            # Account for the causal anchor frame if it exists
+            anchor_index = getattr(window, "causal_anchor_index", None)
+            if anchor_index is not None and anchor_index >= 0:
+                window_length += 1
+
+            window_ref_mask = full_ref_mask[:, :, :window_length + ref_frame_count].to(device)
+            return cond_value._copy_with(window_ref_mask)
+
+        return super().resize_cond_for_context_window(cond_key, cond_value, window, x_in, device, retain_index_list=retain_index_list)
+
+    def concat_cond(self, **kwargs):
+        # The 4 extra channels are the history_mask (1 at clean-anchor frames).
+        noise = kwargs.get("noise", None)
+        extra_channels = self.diffusion_model.patch_embedding.weight.shape[1] - noise.shape[1]
+        if extra_channels != 4:
+            return super().concat_cond(**kwargs)
+
+        mask = kwargs.get("concat_mask", kwargs.get("denoise_mask", None))
+        if mask is None:
+            return torch.zeros_like(noise)[:, :4]
+
+        device = kwargs["device"]
+        if mask.shape[1] != 4:
+            mask = torch.mean(mask, dim=1, keepdim=True)
+        mask = 1.0 - mask
+        mask = utils.common_upscale(mask.to(device), noise.shape[-1], noise.shape[-2], "bilinear", "center")
+        if mask.shape[-3] < noise.shape[-3]:
+            mask = torch.nn.functional.pad(mask, (0, 0, 0, 0, 0, noise.shape[-3] - mask.shape[-3]), mode='constant', value=0)
+        if mask.shape[1] == 1:
+            mask = mask.repeat(1, 4, 1, 1, 1)
+        mask = utils.resize_to_batch_size(mask, noise.shape[0])
+        return mask
+
+    def scale_latent_inpaint(self, sigma, noise, latent_image, **kwargs):
+        # Hold anchor constant across all sigmas instead of base sigma*noise + (1-sigma)*latent_image.
+        return latent_image
+
+
 class WAN22_WanDancer(WAN21):
    def __init__(self, model_config, model_type=ModelType.FLOW, image_to_video=True, device=None):
        super(WAN21, self).__init__(model_config, model_type, device=device, unet_model=comfy.ldm.wan.model_wandancer.WanDancerModel)
@@ -1757,6 +2048,24 @@ class Hunyuan3Dv2_1(BaseModel):
            out['guidance'] = comfy.conds.CONDRegular(torch.FloatTensor([guidance]))
        return out

+class TripoSplat(BaseModel):
+    def __init__(self, model_config, model_type=ModelType.FLOW, device=None):
+        super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.triposplat.model.LatentSeqMMFlowModel)
+
+    def extra_conds(self, **kwargs):
+        out = super().extra_conds(**kwargs)
+        cross_attn = kwargs.get("cross_attn", None) # DINOv3 token sequence -> cross-attention context.
+        if cross_attn is not None:
+            out['c_crossattn'] = comfy.conds.CONDRegular(cross_attn)
+        ref_latents = kwargs.get("reference_latents", None) # Flux2 VAE image latent -> additive second conditioning.
+        if ref_latents is not None:
+            out['ref_latents'] = comfy.conds.CONDList(list(ref_latents))
+        latent_shapes = kwargs.get("latent_shapes", None) # {latent, camera} nested latent
+        if latent_shapes is not None:
+            out['latent_shapes'] = comfy.conds.CONDConstant(latent_shapes)
+        return out
+
+
 class HiDream(BaseModel):
    def __init__(self, model_config, model_type=ModelType.FLOW, device=None):
        super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.hidream.model.HiDreamImageTransformer2DModel)
@@ -1918,6 +2227,11 @@ class Omnigen2(BaseModel):
            out['ref_latents'] = list([1, 16, sum(map(lambda a: math.prod(a.size()), ref_latents)) // 16])
        return out

+class Boogu(Omnigen2):
+    def __init__(self, model_config, model_type=ModelType.FLOW, device=None):
+        super(Omnigen2, self).__init__(model_config, model_type, device=device, unet_model=comfy.ldm.boogu.model.BooguTransformer2DModel)
+        self.memory_usage_factor_conds = ("ref_latents",)
+
 class QwenImage(BaseModel):
    def __init__(self, model_config, model_type=ModelType.FLUX, device=None):
        super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.qwen_image.model.QwenImageTransformer2DModel)
@@ -1950,6 +2264,32 @@ class QwenImage(BaseModel):
            out['ref_latents'] = list([1, 16, sum(map(lambda a: math.prod(a.size()), ref_latents)) // 16])
        return out

+class Ideogram4(BaseModel):
+    def __init__(self, model_config, model_type=ModelType.FLOW, device=None):
+        super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.ideogram4.model.Ideogram4Transformer2DModel)
+
+    def extra_conds(self, **kwargs):
+        out = super().extra_conds(**kwargs)
+        attention_mask = kwargs.get("attention_mask", None)
+        if attention_mask is not None:
+            if torch.numel(attention_mask) != attention_mask.sum():
+                out['attention_mask'] = comfy.conds.CONDRegular(attention_mask)
+        cross_attn = kwargs.get("cross_attn", None)
+        if cross_attn is not None:
+            out['c_crossattn'] = comfy.conds.CONDRegular(cross_attn)
+        return out
+
+class Krea2(BaseModel):
+    def __init__(self, model_config, model_type=ModelType.FLUX, device=None):
+        super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.krea2.model.SingleStreamDiT)
+
+    def extra_conds(self, **kwargs):
+        out = super().extra_conds(**kwargs)
+        cross_attn = kwargs.get("cross_attn", None)
+        if cross_attn is not None:
+            out['c_crossattn'] = comfy.conds.CONDRegular(cross_attn)
+        return out
+
 class HunyuanImage21(BaseModel):
    def __init__(self, model_config, model_type=ModelType.FLOW, device=None):
        super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.hunyuan_video.model.HunyuanVideo)
@@ -2143,6 +2483,12 @@ class RT_DETR_v4(BaseModel):
    def __init__(self, model_config, model_type=ModelType.FLOW, device=None):
        super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.rt_detr.rtdetr_v4.RTv4)

+
+class DepthAnything3(BaseModel):
+    def __init__(self, model_config, model_type=ModelType.FLOW, device=None):
+        super().__init__(model_config, model_type, device=device,
+                         unet_model=comfy.ldm.depth_anything_3.model.DepthAnything3Net)
+
 class ErnieImage(BaseModel):
    def __init__(self, model_config, model_type=ModelType.FLOW, device=None):
        super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.ernie.model.ErnieImageModel)
@@ -313,6 +313,10 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
                    dit_config["use_x0"] = True
                else:
                    dit_config["use_x0"] = False
+                if "{}__sequential__".format(key_prefix) in state_dict_keys: # sequential txt_ids
+                    dit_config["use_sequential_txt_ids"] = True
+                else:
+                    dit_config["use_sequential_txt_ids"] = False
        else:
            dit_config["guidance_embed"] = "{}guidance_in.in_layer.weight".format(key_prefix) in state_dict_keys
            dit_config["yak_mlp"] = '{}double_blocks.0.img_mlp.gate_proj.weight'.format(key_prefix) in state_dict_keys
@@ -463,6 +467,23 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
            dit_config["extra_per_block_abs_pos_emb_type"] = "learnable"
        return dit_config

+    # PiD (Pixel Diffusion Decoder). Must check BEFORE plain PixelDiT_T2I.
+    _lq_w_key = '{}lq_proj.latent_proj.0.weight'.format(key_prefix)
+    if _lq_w_key in state_dict_keys:
+        in_ch = int(state_dict[_lq_w_key].shape[1])
+        _gate_prefix = '{}lq_proj.gate_modules.'.format(key_prefix)
+        num_gates = len({k[len(_gate_prefix):].split('.')[0]
+                         for k in state_dict_keys if k.startswith(_gate_prefix)})
+        dit_config = {"image_model": "pid",
+                      "lq_latent_channels": in_ch,
+                      "latent_spatial_down_factor": 16 if in_ch >= 64 else 8}
+        if num_gates > 0:
+            dit_config["lq_interval"] = (14 + num_gates - 1) // num_gates
+        return dit_config
+
+    if '{}core.pixel_embedder.proj.weight'.format(key_prefix) in state_dict_keys:  # PixelDiT T2I
+        return {"image_model": "pixeldit_t2i"}
+
    if '{}cap_embedder.1.weight'.format(key_prefix) in state_dict_keys and '{}noise_refiner.0.attention.k_norm.weight'.format(key_prefix) in state_dict_keys:  # Lumina 2
        dit_config = {}
        dit_config["image_model"] = "lumina2"
@@ -609,6 +630,8 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
            dit_config["model_type"] = "humo"
        elif '{}face_adapter.fuser_blocks.0.k_norm.weight'.format(key_prefix) in state_dict_keys:
            dit_config["model_type"] = "animate"
+        elif '{}patch_embedding_mask.weight'.format(key_prefix) in state_dict_keys:
+            dit_config["model_type"] = "scail2"
        elif '{}patch_embedding_pose.weight'.format(key_prefix) in state_dict_keys:
            dit_config["model_type"] = "scail"
        elif '{}patch_embedding_global.weight'.format(key_prefix) in state_dict_keys:
@@ -659,6 +682,9 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
        dit_config["guidance_cond_proj_dim"] = None#f"{key_prefix}t_embedder.cond_proj.weight" in state_dict_keys
        return dit_config

+    if '{}cam_out_layer.weight'.format(key_prefix) in state_dict_keys and '{}repo_layers.0.final_map.weight'.format(key_prefix) in state_dict_keys:  # TripoSplat
+        return {"image_model": "triposplat"}
+
    if '{}t_embedder1.mlp.0.weight'.format(key_prefix) in state_dict_keys and '{}x_embedder.proj1.weight'.format(key_prefix) in state_dict_keys:  # HiDream-O1
        return {"image_model": "hidream_o1"}

@@ -735,6 +761,16 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):

        return dit_config

+    if '{}double_stream_layers.0.img_instruct_attn.processor.img_to_q.weight'.format(key_prefix) in state_dict_keys:  # Boogu-Image (OmniGen2 derivative + dual-stream stage)
+        dit_config = {}
+        dit_config["image_model"] = "boogu"
+        dit_config["hidden_size"] = state_dict['{}x_embedder.weight'.format(key_prefix)].shape[0]
+        dit_config["num_layers"] = count_blocks(state_dict_keys, '{}single_stream_layers.'.format(key_prefix) + '{}.')
+        dit_config["num_double_stream_layers"] = count_blocks(state_dict_keys, '{}double_stream_layers.'.format(key_prefix) + '{}.')
+        dit_config["num_refiner_layers"] = count_blocks(state_dict_keys, '{}noise_refiner.'.format(key_prefix) + '{}.')
+        dit_config["instruction_feat_dim"] = state_dict['{}time_caption_embed.caption_embedder.0.weight'.format(key_prefix)].shape[0]
+        return dit_config
+
    if '{}time_caption_embed.timestep_embedder.linear_1.bias'.format(key_prefix) in state_dict_keys:  # Omnigen2
        dit_config = {}
        dit_config["image_model"] = "omnigen2"
@@ -791,6 +827,28 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
            dit_config["default_ref_method"] = "negative_index"
        return dit_config

+    if '{}embed_image_indicator.weight'.format(key_prefix) in state_dict_keys:  # Ideogram 4
+        dit_config = {}
+        dit_config["image_model"] = "ideogram4"
+        dit_config["in_channels"] = state_dict['{}input_proj.weight'.format(key_prefix)].shape[1]
+        dit_config["num_layers"] = count_blocks(state_dict_keys, '{}layers.'.format(key_prefix) + '{}.')
+        return dit_config
+
+    if '{}txtfusion.projector.weight'.format(key_prefix) in state_dict_keys:  # Krea 2 (K2)
+        dit_config = {}
+        dit_config["image_model"] = "krea2"
+        head_dim = 128
+        first_w = state_dict['{}first.weight'.format(key_prefix)]  # (features, channels*patch^2)
+        dit_config["features"] = first_w.shape[0]
+        dit_config["channels"] = first_w.shape[1] // (2 * 2)  # patch=2
+        dit_config["patch"] = 2
+        dit_config["layers"] = count_blocks(state_dict_keys, '{}blocks.'.format(key_prefix) + '{}.')
+        dit_config["heads"] = state_dict['{}blocks.0.attn.wq.weight'.format(key_prefix)].shape[0] // head_dim
+        dit_config["kvheads"] = state_dict['{}blocks.0.attn.wk.weight'.format(key_prefix)].shape[0] // head_dim
+        dit_config["txtlayers"] = state_dict['{}txtfusion.projector.weight'.format(key_prefix)].shape[1]
+        dit_config["txtdim"] = state_dict['{}txtfusion.layerwise_blocks.0.prenorm.scale'.format(key_prefix)].shape[0]
+        return dit_config
+
    if '{}visual_transformer_blocks.0.cross_attention.key_norm.weight'.format(key_prefix) in state_dict_keys: # Kandinsky 5
        dit_config = {}
        model_dim = state_dict['{}visual_embeddings.in_layer.bias'.format(key_prefix)].shape[0]
@@ -829,6 +887,95 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
        dit_config["enc_h"] = state_dict['{}encoder.pan_blocks.1.cv4.conv.weight'.format(key_prefix)].shape[0]
        return dit_config

+    # Depth Anything 3 (repackaged to ComfyUI's native Dinov2Model layout via scripts/convert_da3.py)
+    if '{}backbone.embeddings.patch_embeddings.projection.weight'.format(key_prefix) in state_dict_keys:
+        dit_config = {}
+        dit_config["image_model"] = "DepthAnything3"
+
+        patch_w = state_dict['{}backbone.embeddings.patch_embeddings.projection.weight'.format(key_prefix)]
+        embed_dim = patch_w.shape[0]
+        depth = count_blocks(state_dict_keys, '{}backbone.encoder.layer.'.format(key_prefix) + '{}.')
+
+        # Backbone preset is determined by embed_dim (matches vits/vitb/vitl/vitg).
+        backbone_name = {384: "vits", 768: "vitb", 1024: "vitl", 1536: "vitg"}.get(embed_dim)
+        if backbone_name is None:
+            return None
+        dit_config["backbone_name"] = backbone_name
+
+        # Detect DA3 extensions on top of vanilla DINOv2.
+        has_camera_token = '{}backbone.embeddings.camera_token'.format(key_prefix) in state_dict_keys
+        # qk-norm shows up as `attention.q_norm.weight` on enabled blocks.
+        qknorm_indices = [
+            i for i in range(depth)
+            if '{}backbone.encoder.layer.{}.attention.q_norm.weight'.format(key_prefix, i) in state_dict_keys
+        ]
+        qknorm_start = qknorm_indices[0] if qknorm_indices else -1
+
+        # The DA3 main-series configs always set alt_start == qknorm_start == rope_start.
+        # cat_token=True is implied by the presence of camera_token.
+        if has_camera_token:
+            dit_config["alt_start"] = qknorm_start
+            dit_config["rope_start"] = qknorm_start
+            dit_config["qknorm_start"] = qknorm_start
+            dit_config["cat_token"] = True
+        else:
+            dit_config["alt_start"] = -1
+            dit_config["rope_start"] = -1
+            dit_config["qknorm_start"] = -1
+            dit_config["cat_token"] = False
+
+        # Detect head type and config.
+        has_aux = '{}head.scratch.refinenet1_aux.out_conv.weight'.format(key_prefix) in state_dict_keys
+        dit_config["head_dim_in"] = state_dict['{}head.projects.0.weight'.format(key_prefix)].shape[1]
+        dit_config["head_features"] = state_dict['{}head.scratch.refinenet1.out_conv.weight'.format(key_prefix)].shape[0]
+        dit_config["head_out_channels"] = [
+            state_dict['{}head.projects.{}.weight'.format(key_prefix, i)].shape[0]
+            for i in range(4)
+        ]
+        if has_aux:
+            # DualDPT: dim_in = 2 * embed_dim (because cat_token doubles token width).
+            dit_config["head_type"] = "dualdpt"
+            dit_config["head_output_dim"] = 2
+            dit_config["head_use_sky_head"] = False
+        else:
+            dit_config["head_type"] = "dpt"
+            dit_config["head_output_dim"] = state_dict[
+                '{}head.scratch.output_conv2.2.weight'.format(key_prefix)
+            ].shape[0]
+            dit_config["head_use_sky_head"] = (
+                '{}head.scratch.sky_output_conv2.0.weight'.format(key_prefix) in state_dict_keys
+            )
+
+        # out_layers: hard-coded per upstream YAML config (depth-aware default).
+        if depth >= 24:
+            # vitl: depths used vary between DA3-Large (DualDPT) and Mono/Metric (DPT).
+            if has_aux:
+                dit_config["out_layers"] = [11, 15, 19, 23]
+            else:
+                dit_config["out_layers"] = [4, 11, 17, 23]
+        else:
+            # vits/vitb: 12 blocks
+            dit_config["out_layers"] = [5, 7, 9, 11]
+
+        # Camera encoder/decoder presence (multi-view + pose path).
+        has_cam_enc = '{}cam_enc.token_norm.weight'.format(key_prefix) in state_dict_keys
+        has_cam_dec = '{}cam_dec.fc_t.weight'.format(key_prefix) in state_dict_keys
+        dit_config["has_cam_enc"] = has_cam_enc
+        dit_config["has_cam_dec"] = has_cam_dec
+        if has_cam_enc:
+            cam_enc_w = state_dict.get(
+                '{}cam_enc.pose_branch.fc2.weight'.format(key_prefix)
+            )
+            if cam_enc_w is not None:
+                dit_config["cam_dim_out"] = cam_enc_w.shape[0]
+        if has_cam_dec:
+            cam_dec_w = state_dict.get(
+                '{}cam_dec.fc_t.weight'.format(key_prefix)
+            )
+            if cam_dec_w is not None:
+                dit_config["cam_dec_dim_in"] = cam_dec_w.shape[1]
+        return dit_config
+
    if '{}layers.0.mlp.linear_fc2.weight'.format(key_prefix) in state_dict_keys: # Ernie Image
        dit_config = {}
        dit_config["image_model"] = "ernie"
@@ -534,8 +534,10 @@ try:
 except:
    pass

-if torch.cuda.is_available() and torch.backends.cudnn.is_available() and PerformanceFeature.AutoTune in args.fast:
-    torch.backends.cudnn.benchmark = True
+
+def set_cudnn_benchmark():
+    if torch.cuda.is_available() and torch.backends.cudnn.is_available():
+        torch.backends.cudnn.benchmark = PerformanceFeature.AutoTune in args.fast

 try:
    if torch_version_numeric >= (2, 5):
@@ -641,15 +643,19 @@ def free_pins(size, evict_active=False):
    return freed_total

 def ensure_pin_budget(size, evict_active=False):
-    shortfall = size + comfy.memory_management.RAM_CACHE_HEADROOM / 2 - psutil.virtual_memory().available
+    if args.high_ram:
+        return True
+    if args.fast_disk:
+        shortfall = TOTAL_PINNED_MEMORY + size - MAX_PINNED_MEMORY
+    else:
+        shortfall = size + max(comfy.memory_management.RAM_CACHE_HEADROOM / 2, 2048 * 1024 ** 2) - psutil.virtual_memory().available
    if shortfall <= 0:
        return True

    to_free = shortfall + PIN_PRESSURE_HYSTERESIS
    return free_pins(to_free, evict_active=evict_active) >= shortfall

-def ensure_pin_registerable(size, evict_active=False):
-    shortfall = TOTAL_PINNED_MEMORY + size - MAX_PINNED_MEMORY
+def free_registrations(shortfall, evict_active=True):
    if MAX_PINNED_MEMORY <= 0:
        return False
    if shortfall <= 0:
@@ -658,12 +664,22 @@ def ensure_pin_registerable(size, evict_active=False):
    shortfall += REGISTERABLE_PIN_HYSTERESIS
    for loaded_model in reversed(current_loaded_models):
        model = loaded_model.model
-        if model is not None and model.is_dynamic() and (evict_active or not model.model.dynamic_pins[model.load_device]["active"]):
+        if model is not None and model.is_dynamic() and not model.model.dynamic_pins[model.load_device]["active"]:
            shortfall -= model.unregister_inactive_pins(shortfall)
            if shortfall <= 0:
                return True
+    if evict_active:
+        for loaded_model in current_loaded_models:
+            model = loaded_model.model
+            if model is not None and model.is_dynamic() and model.model.dynamic_pins[model.load_device]["active"]:
+                shortfall -= model.unregister_inactive_pins(shortfall)
+                if shortfall <= 0:
+                    return True
    return shortfall <= REGISTERABLE_PIN_HYSTERESIS

+def ensure_pin_registerable(size, evict_active=True):
+    return free_registrations(TOTAL_PINNED_MEMORY + size - MAX_PINNED_MEMORY, evict_active=evict_active)
+
 class LoadedModel:
    def __init__(self, model: ModelPatcher):
        self._set_model(model)
@@ -803,9 +819,9 @@ def free_memory(memory_required, device, keep_loaded=[], for_dynamic=False, pins
    for x in can_unload_sorted:
        i = x[-1]
        memory_to_free = 1e32
-        if current_loaded_models[i].model.is_dynamic() and (not DISABLE_SMART_MEMORY or device is None):
+        if not DISABLE_SMART_MEMORY or device is None:
            memory_to_free = 0 if device is None else memory_required - get_free_memory(device)
-            if for_dynamic:
+            if current_loaded_models[i].model.is_dynamic() and for_dynamic:
                #don't actually unload dynamic models for the sake of other dynamic models
                #as that works on-demand.
                memory_required -= current_loaded_models[i].model.loaded_size()
@@ -817,6 +833,10 @@ def free_memory(memory_required, device, keep_loaded=[], for_dynamic=False, pins
    for i in sorted(unloaded_model, reverse=True):
        unloaded_models.append(current_loaded_models.pop(i))

+    if not for_dynamic and pins_required > 0:
+        ensure_pin_budget(pins_required)
+        ensure_pin_registerable(pins_required)
+
    if len(unloaded_model) > 0:
        soft_empty_cache()
    elif device is not None:
@@ -879,15 +899,19 @@ def load_models_gpu(models, memory_required=0, force_patch_weights=False, minimu
            model_to_unload.model_finalizer.detach()

    total_memory_required = {}
+    total_pins_required = {}
    for loaded_model in models_to_load:
        device = loaded_model.device
        total_memory_required[device] = total_memory_required.get(device, 0) + loaded_model.model_memory_required(device)
+        if not loaded_model.model.is_dynamic():
+            total_pins_required[device] = total_pins_required.get(device, 0) + loaded_model.model_memory()

    for device in total_memory_required:
        if device != torch.device("cpu"):
            free_memory(total_memory_required[device] * 1.1 + extra_mem,
                        device,
-                        for_dynamic=free_for_dynamic)
+                        for_dynamic=free_for_dynamic,
+                        pins_required=total_pins_required.get(device, 0))

    for device in total_memory_required:
        if device != torch.device("cpu"):
@@ -938,8 +962,6 @@ def loaded_models(only_currently_used=False):
 def cleanup_models_gc():
    do_gc = False

-    reset_cast_buffers()
-
    for i in range(len(current_loaded_models)):
        cur = current_loaded_models[i]
        if cur.is_dead():
@@ -1283,7 +1305,6 @@ STREAM_CAST_BUFFERS = {}
 LARGEST_CASTED_WEIGHT = (None, 0)
 STREAM_AIMDO_CAST_BUFFERS = {}
 LARGEST_AIMDO_CASTED_WEIGHT = (None, 0)
-STREAM_PIN_BUFFERS = {}

 DEFAULT_AIMDO_CAST_BUFFER_RESERVATION_SIZE = 16 * 1024 ** 3

@@ -1326,42 +1347,13 @@ def get_aimdo_cast_buffer(offload_stream, device):
        STREAM_AIMDO_CAST_BUFFERS[offload_stream] = cast_buffer
    return cast_buffer

-def get_pin_buffer(offload_stream):
-    pin_buffer = STREAM_PIN_BUFFERS.get(offload_stream, None)
-    if pin_buffer is None:
-        pin_buffer = comfy_aimdo.host_buffer.HostBuffer(0, 0, pinned_hostbuf_size(8 * 1024**3), mark_cold=False)
-        STREAM_PIN_BUFFERS[offload_stream] = pin_buffer
-    elif offload_stream is not None:
-        event = getattr(pin_buffer, "_comfy_event", None)
-        if event is not None:
-            event.synchronize()
-            delattr(pin_buffer, "_comfy_event")
-    return pin_buffer
-
-def resize_pin_buffer(pin_buffer, size):
-    global TOTAL_PINNED_MEMORY
-    old_size = pin_buffer.size
-    if size <= old_size:
-        return True
-    growth = size - old_size
-    comfy.memory_management.extra_ram_release(comfy.memory_management.RAM_CACHE_HEADROOM)
-    ensure_pin_budget(growth, evict_active=True)
-    ensure_pin_registerable(growth, evict_active=True)
-    try:
-        pin_buffer.extend(size=size, reallocate=True)
-    except RuntimeError:
-        return False
-    TOTAL_PINNED_MEMORY += pin_buffer.size - old_size
-    return True
-
 def reset_cast_buffers():
-    global TOTAL_PINNED_MEMORY
    global LARGEST_CASTED_WEIGHT
    global LARGEST_AIMDO_CASTED_WEIGHT

    LARGEST_CASTED_WEIGHT = (None, 0)
    LARGEST_AIMDO_CASTED_WEIGHT = (None, 0)
-    for offload_stream in set(STREAM_CAST_BUFFERS) | set(STREAM_AIMDO_CAST_BUFFERS) | set(STREAM_PIN_BUFFERS):
+    for offload_stream in set(STREAM_CAST_BUFFERS) | set(STREAM_AIMDO_CAST_BUFFERS):
        if offload_stream is not None:
            offload_stream.synchronize()
    synchronize()
@@ -1370,20 +1362,24 @@ def reset_cast_buffers():
        mmap_obj.bounce()
    DIRTY_MMAPS.clear()

-    for pin_buffer in STREAM_PIN_BUFFERS.values():
-        TOTAL_PINNED_MEMORY -= pin_buffer.size
-    TOTAL_PINNED_MEMORY = max(0, TOTAL_PINNED_MEMORY)
-
    for loaded_model in current_loaded_models:
        model = loaded_model.model
        if model is not None and model.is_dynamic():
-            model.model.dynamic_pins[model.load_device]["active"] = False
+            pin_state = model.model.dynamic_pins[model.load_device]
+
+            if pin_state["active"]:
+                *_, buckets = pin_state["weights"]
+                for size, bucket in list(buckets.items()):
+                    bucket[:] = [ entry for entry in bucket if entry[-1] is not None ]
+                    if not bucket:
+                        del buckets[size]
+
+            pin_state["active"] = False
            model.partially_unload_ram(1e30, subsets=[ "patches" ])
-            model.model.dynamic_pins[model.load_device]["patches"] = (comfy_aimdo.host_buffer.HostBuffer(0, 8 * 1024 * 1024, pinned_hostbuf_size(model.model_size())), [], [-1], [0])
+            model.model.dynamic_pins[model.load_device]["patches"] = (comfy_aimdo.host_buffer.HostBuffer(0, 8 * 1024 * 1024, pinned_hostbuf_size(model.model_size())), [], [-1], [0], [0], {})

    STREAM_CAST_BUFFERS.clear()
    STREAM_AIMDO_CAST_BUFFERS.clear()
-    STREAM_PIN_BUFFERS.clear()
    soft_empty_cache()

 def get_offload_stream(device):
@@ -1436,7 +1432,7 @@ def cast_to_gathered(tensors, r, non_blocking=False, stream=None, r2=None):
       if hasattr(wf_context, "as_context"):
           wf_context = wf_context.as_context(stream)

-    dest_views = comfy.memory_management.interpret_gathered_like(tensors, r)
+    dest_views = comfy.memory_management.interpret_gathered_like(tensors, r) if r is not None else [None] * len(tensors)
    dest2_views = comfy.memory_management.interpret_gathered_like(tensors, r2) if r2 is not None else None
    with wf_context:
        for tensor in tensors:
@@ -1448,9 +1444,10 @@ def cast_to_gathered(tensors, r, non_blocking=False, stream=None, r2=None):
                continue
            storage = tensor._qdata.untyped_storage() if isinstance(tensor, comfy.quant_ops.QuantizedTensor) else tensor.untyped_storage()
            mark_mmap_dirty(storage)
-            dest_view.copy_(tensor, non_blocking=non_blocking)
+            if dest_view is not None:
+                dest_view.copy_(tensor, non_blocking=non_blocking)
            if dest2_view is not None:
-                dest2_view.copy_(dest_view, non_blocking=non_blocking)
+                dest2_view.copy_(tensor if dest_view is None else dest_view, non_blocking=non_blocking)


 def cast_to(weight, dtype=None, device=None, non_blocking=False, copy=False, stream=None, r=None):
@@ -1501,6 +1498,8 @@ if not args.disable_pinned_memory:
 PINNING_ALLOWED_TYPES = set(["Tensor", "Parameter", "QuantizedTensor"])

 def pinned_hostbuf_size(size):
+    if args.high_ram:
+        return max(0, int(size * 2))
    return max(0, int(min(size, MAX_PINNED_MEMORY) * 2))

 def discard_cuda_async_error():
@@ -1723,6 +1722,13 @@ def is_device_xpu(device):
 def is_device_cuda(device):
    return is_device_type(device, 'cuda')

+def set_torch_device(device):
+    """Set the current device for the given torch device. Supports CUDA and XPU."""
+    if is_device_cuda(device):
+        torch.cuda.set_device(device)
+    elif is_device_xpu(device):
+        torch.xpu.set_device(device)
+
 def is_directml_enabled():
    global directml_enabled
    if directml_enabled:
@@ -379,10 +379,11 @@ class ModelPatcher:
    def get_clone_model_override(self):
        return self.model, (self.backup, self.backup_buffers, self.object_patches_backup, self.pinned)

-    def clone(self, disable_dynamic=False, model_override=None):
+    def clone(self, disable_dynamic=False, model_override=None, force_deepcopy=False):
        class_ = self.__class__
-        if self.is_dynamic() and disable_dynamic:
-            class_ = ModelPatcher
+        if self.is_dynamic() and disable_dynamic or force_deepcopy:
+            if self.is_dynamic() and disable_dynamic:
+                class_ = ModelPatcher
            if model_override is None:
                if self.cached_patcher_init is None:
                    raise RuntimeError("Cannot create non-dynamic delegate: cached_patcher_init is not initialized.")
@@ -1721,8 +1722,8 @@ class ModelPatcherDynamic(ModelPatcher):
        """
        if device not in self.model.dynamic_pins:
            self.model.dynamic_pins[device] = {
-                "weights": (comfy_aimdo.host_buffer.HostBuffer(0, 0, 0), [], [-1], [0]),
-                "patches": (comfy_aimdo.host_buffer.HostBuffer(0, 0, 0), [], [-1], [0]),
+                "weights": (comfy_aimdo.host_buffer.HostBuffer(0, 0, 0), [], [-1], [0], [0], {}),
+                "patches": (comfy_aimdo.host_buffer.HostBuffer(0, 0, 0), [], [-1], [0], [0], {}),
                "hostbufs_initialized": False,
                "failed": False,
                "active": False,
@@ -1799,8 +1800,8 @@ class ModelPatcherDynamic(ModelPatcher):
            pin_state = self.model.dynamic_pins[self.load_device]
            if not pin_state["hostbufs_initialized"]:
                hostbuf_size = comfy.model_management.pinned_hostbuf_size(self.model_size())
-                pin_state["weights"] = (comfy_aimdo.host_buffer.HostBuffer(0, 64 * 1024 * 1024, hostbuf_size), [], [-1], [0])
-                pin_state["patches"] = (comfy_aimdo.host_buffer.HostBuffer(0, 8 * 1024 * 1024, hostbuf_size), [], [-1], [0])
+                pin_state["weights"] = (comfy_aimdo.host_buffer.HostBuffer(0, 64 * 1024 * 1024, hostbuf_size), [], [-1], [0], [0], {})
+                pin_state["patches"] = (comfy_aimdo.host_buffer.HostBuffer(0, 8 * 1024 * 1024, hostbuf_size), [], [-1], [0], [0], {})
                pin_state["hostbufs_initialized"] = True
            pin_state["failed"] = False
            pin_state["active"] = True
@@ -1942,18 +1943,16 @@ class ModelPatcherDynamic(ModelPatcher):
        return freed

    def loaded_ram_size(self):
-        return (self.model.dynamic_pins[self.load_device]["weights"][0].size +
-                self.model.dynamic_pins[self.load_device]["patches"][0].size)
+        return (self.model.dynamic_pins[self.load_device]["weights"][0].size)

    def pinned_memory_size(self):
-        return (self.model.dynamic_pins[self.load_device]["weights"][3][0] +
-                self.model.dynamic_pins[self.load_device]["patches"][3][0])
+        return (self.model.dynamic_pins[self.load_device]["weights"][3][0])

    def unregister_inactive_pins(self, ram_to_unload, subsets=[ "weights", "patches" ]):
        freed = 0
        pin_state = self.model.dynamic_pins[self.load_device]
        for subset in subsets:
-            hostbuf, stack, stack_split, pinned_size = pin_state[subset]
+            hostbuf, stack, stack_split, pinned_size, *_ = pin_state[subset]
            split = stack_split[0]
            while split >= 0:
                module, offset = stack[split]
@@ -1978,10 +1977,12 @@ class ModelPatcherDynamic(ModelPatcher):
        freed = 0
        pin_state = self.model.dynamic_pins[self.load_device]
        for subset in subsets:
-            hostbuf, stack, stack_split, pinned_size = pin_state[subset]
+            hostbuf, stack, stack_split, pinned_size, *_ = pin_state[subset]
            while len(stack) > 0:
                module, offset = stack.pop()
                size = module._pin.numel() * module._pin.element_size()
+                module._pin_balancer_entry[-1] = None
+                del module._pin_balancer_entry
                del module._pin
                hostbuf.truncate(offset, do_unregister=module._pin_registered)
                stack_split[0] = min(stack_split[0], len(stack) - 1)
@@ -1,4 +1,5 @@
 import comfy_aimdo.model_vbar
+import comfy.memory_management
 import comfy.model_management
 import comfy.ops

@@ -50,7 +51,17 @@ def prefetch_queue_pop(queue, device, module):
            if hasattr(s, "_v"):
                comfy_modules.append(s)

+        registerable_size = 0
+        for s in comfy_modules:
+            registerable_size += comfy.memory_management.vram_aligned_size([s.weight, s.bias])
+            for param_key in ("weight", "bias"):
+                lowvram_fn = getattr(s, param_key + "_lowvram_function", None)
+                if lowvram_fn is not None:
+                    registerable_size += lowvram_fn.memory_required()
+
        offload_stream = comfy.ops.cast_modules_with_vbar(comfy_modules, None, device, None, True)
+        if not comfy.model_management.args.fast_disk:
+            comfy.model_management.ensure_pin_registerable(registerable_size)
        comfy.model_management.sync_stream(device, offload_stream)
        queue[0] = (offload_stream, (prefetch, comfy_modules))

@@ -17,7 +17,7 @@ class MultiGPUThreadPool:
    """Persistent thread pool for multi-GPU work distribution.

    Maintains one worker thread per extra GPU device. Each thread calls
-    torch.cuda.set_device() once at startup so that compiled kernel caches
+    set_torch_device() once at startup so that compiled kernel caches
    (inductor/triton) stay warm across diffusion steps.
    """

@@ -37,7 +37,7 @@ class MultiGPUThreadPool:

    def _worker_loop(self, device: torch.device, work_q: queue.Queue, result_q: queue.Queue):
        try:
-            torch.cuda.set_device(device)
+            comfy.model_management.set_torch_device(device)
        except Exception as e:
            logging.error(f"MultiGPUThreadPool: failed to set device {device}: {e}")
            while True:
@@ -54,6 +54,8 @@ class MultiGPUThreadPool:
            try:
                result = fn(*args, **kwargs)
                result_q.put((result, None))
+            except comfy.model_management.InterruptProcessingException as e:
+                result_q.put((None, e))
            except Exception as e:
                result_q.put((None, e))

@@ -76,8 +76,6 @@ except:

 cast_to = comfy.model_management.cast_to #TODO: remove once no more references

-STREAM_PIN_BUFFER_HEADROOM = 8 * 1024 * 1024
-
 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)

@@ -94,9 +92,6 @@ def cast_modules_with_vbar(comfy_modules, dtype, device, bias_dtype, non_blockin
    offload_stream = None
    cast_buffer = None
    cast_buffer_offset = 0
-    stream_pin_hostbuf = None
-    stream_pin_offset = 0
-    stream_pin_queue = []

    def ensure_offload_stream(module, required_size, check_largest):
        nonlocal offload_stream
@@ -130,22 +125,6 @@ def cast_modules_with_vbar(comfy_modules, dtype, device, bias_dtype, non_blockin
        cast_buffer_offset += buffer_size
        return buffer

-    def get_stream_pin_buffer_offset(buffer_size):
-        nonlocal stream_pin_hostbuf
-        nonlocal stream_pin_offset
-
-        if buffer_size == 0 or offload_stream is None:
-            return None
-
-        if stream_pin_hostbuf is None:
-            stream_pin_hostbuf = comfy.model_management.get_pin_buffer(offload_stream)
-            if stream_pin_hostbuf is None:
-                return None
-
-        offset = stream_pin_offset
-        stream_pin_offset += buffer_size
-        return offset
-
    for s in comfy_modules:
        signature = comfy_aimdo.model_vbar.vbar_fault(s._v)
        resident = comfy_aimdo.model_vbar.vbar_signature_compare(signature, s._v_signature)
@@ -184,33 +163,29 @@ def cast_modules_with_vbar(comfy_modules, dtype, device, bias_dtype, non_blockin
        if xfer_dest is None:
            xfer_dest = get_cast_buffer(dest_size)

-        def cast_maybe_lowvram_patch(xfer_source, xfer_dest, stream):
+        def cast_maybe_lowvram_patch(xfer_source, xfer_dest, stream, xfer_dest2=None):
            if xfer_source is not None:
                if getattr(xfer_source, "is_lowvram_patch", False):
-                    xfer_source.prepare(xfer_dest, stream, copy=True, commit=False)
-                else:
-                    comfy.model_management.cast_to_gathered(xfer_source, xfer_dest, non_blocking=non_blocking, stream=stream)
+                    if xfer_dest is not None:
+                        xfer_source.prepare(xfer_dest, stream, copy=True, commit=False)
+                        xfer_source = [ xfer_dest ]
+                        xfer_dest = xfer_dest2
+                        xfer_dest2 = None
+                    elif xfer_dest2 is not None:
+                        xfer_source.prepare(xfer_dest2, stream, copy=True, commit=False)
+                        return
+                    else:
+                        return
+                comfy.model_management.cast_to_gathered(xfer_source, xfer_dest, non_blocking=non_blocking, stream=stream, r2=xfer_dest2)

        def handle_pin(m, pin, source, dest, subset="weights", size=None):
            if pin is not None:
                cast_maybe_lowvram_patch([pin], dest, offload_stream)
                return
-            if signature is None:
+            if signature is None or args.high_ram:
                comfy.pinned_memory.pin_memory(m, subset=subset, size=size)
                pin = comfy.pinned_memory.get_pin(m, subset=subset)
-                if pin is not None:
-                    if isinstance(source, list):
-                        comfy.model_management.cast_to_gathered(source, pin, non_blocking=non_blocking, stream=offload_stream, r2=dest)
-                    else:
-                        cast_maybe_lowvram_patch(source, pin, None)
-                        cast_maybe_lowvram_patch([ pin ], dest, offload_stream)
-                    return
-            if pin is None:
-                pin_offset = get_stream_pin_buffer_offset(size)
-                if pin_offset is not None:
-                    stream_pin_queue.append((source, pin_offset, size, dest))
-                    return
-            cast_maybe_lowvram_patch(source, dest, offload_stream)
+            cast_maybe_lowvram_patch(source, pin, offload_stream, xfer_dest2=dest)

        handle_pin(s, pin, xfer_source, xfer_dest, size=dest_size)

@@ -232,23 +207,6 @@ def cast_modules_with_vbar(comfy_modules, dtype, device, bias_dtype, non_blockin
        prefetch["needs_cast"] = needs_cast
        s._prefetch = prefetch

-    if stream_pin_offset > 0:
-        if stream_pin_hostbuf.size < stream_pin_offset:
-            if not comfy.model_management.resize_pin_buffer(stream_pin_hostbuf, stream_pin_offset + STREAM_PIN_BUFFER_HEADROOM):
-                for xfer_source, _, _, xfer_dest in stream_pin_queue:
-                    cast_maybe_lowvram_patch(xfer_source, xfer_dest, offload_stream)
-                return offload_stream
-        stream_pin_tensor = comfy_aimdo.torch.hostbuf_to_tensor(stream_pin_hostbuf)
-        stream_pin_tensor.untyped_storage()._comfy_hostbuf = stream_pin_hostbuf
-        for xfer_source, pin_offset, pin_size, xfer_dest in stream_pin_queue:
-            pin = stream_pin_tensor[pin_offset:pin_offset + pin_size]
-            if isinstance(xfer_source, list):
-                comfy.model_management.cast_to_gathered(xfer_source, pin, non_blocking=non_blocking, stream=offload_stream, r2=xfer_dest)
-            else:
-                cast_maybe_lowvram_patch(xfer_source, pin, None)
-                comfy.model_management.cast_to_gathered([ pin ], xfer_dest, non_blocking=non_blocking, stream=offload_stream)
-        stream_pin_hostbuf._comfy_event = offload_stream.record_event()
-
    return offload_stream


@@ -300,7 +258,7 @@ def resolve_cast_module_with_vbar(s, dtype, device, bias_dtype, compute_dtype, w
            if (want_requant and len(fns) == 0 or update_weight):
                seed = comfy.utils.string_to_seed(s.seed_key)
                if isinstance(orig, QuantizedTensor):
-                    y = QuantizedTensor.from_float(x, s.layout_type, scale="recalculate", stochastic_rounding=seed)
+                    y = orig.requantize_from_float(x, scale="recalculate", stochastic_rounding=seed)
                else:
                    y = comfy.float.stochastic_rounding(x, orig.dtype, seed=seed)
            if want_requant and len(fns) == 0:
@@ -343,21 +301,21 @@ 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"):
+    if hasattr(s, "_v") and comfy.model_management.is_device_cpu(device):

        #vbar doesn't support CPU weights, but some custom nodes have weird paths
        #that might switch the layer to the CPU and expect it to work. We have to take
        #a clone conservatively as we are mmapped and some SFT files are packed misaligned
        #If you are a custom node author reading this, please move your layer to the GPU
        #or declare your ModelPatcher as CPU in the first place.
-        if comfy.model_management.is_device_cpu(device):
-            materialize_meta_param(s, ["weight", "bias"])
-            weight = s.weight.to(dtype=dtype, copy=True)
-            if isinstance(weight, QuantizedTensor):
-                weight = weight.dequantize()
-            bias = s.bias.to(dtype=bias_dtype, copy=True) if s.bias is not None else None
-            return format_return((weight, bias, (None, None, None)), offloadable)
+        materialize_meta_param(s, ["weight", "bias"])
+        weight = s.weight.to(dtype=dtype, copy=True)
+        if isinstance(weight, QuantizedTensor):
+            weight = weight.dequantize()
+        bias = s.bias.to(dtype=bias_dtype, copy=True) if s.bias is not None else None
+        return format_return((weight, bias, (None, None, None)), offloadable)

+    elif hasattr(s, "_v") and s.weight.device != device:
        prefetched = hasattr(s, "_prefetch")
        offload_stream = None
        offload_device = None
@@ -1133,6 +1091,19 @@ def _load_quantized_module(module, super_load, state_dict, prefix, local_metadat
            if ts is None or bs is None:
                raise ValueError(f"Missing NVFP4 scales for layer {layer_name}")
            scales = {"scale": ts, "block_scale": bs}
+        elif module.quant_format == "int8_tensorwise":
+            scale = pop_scale("weight_scale")
+            if scale is None:
+                raise ValueError(f"Missing INT8 weight scale for layer {layer_name}")
+            scales = {"scale": scale}
+            params_conf = layer_conf.get("params", {})
+            if not isinstance(params_conf, dict):
+                params_conf = {}
+            if layer_conf.get("convrot", params_conf.get("convrot", False)):
+                scales["convrot"] = True
+                scales["convrot_groupsize"] = int(
+                    layer_conf.get("convrot_groupsize", params_conf.get("convrot_groupsize", 256))
+                )
        else:
            raise ValueError(f"Unsupported quantization format: {module.quant_format}")

@@ -1175,6 +1146,10 @@ def _quantized_weight_state_dict(module, sd, prefix, extra_quant_conf=None, extr
        quant_conf = {"format": module.quant_format}
        if getattr(module, '_full_precision_mm_config', False):
            quant_conf["full_precision_matrix_mult"] = True
+        params = getattr(module.weight, "_params", None)
+        if module.quant_format == "int8_tensorwise" and getattr(params, "convrot", False):
+            quant_conf["convrot"] = True
+            quant_conf["convrot_groupsize"] = getattr(params, "convrot_groupsize", 256)
        if extra_quant_conf:
            quant_conf.update(extra_quant_conf)
        sd[f"{prefix}comfy_quant"] = torch.tensor(list(json.dumps(quant_conf).encode("utf-8")), dtype=torch.uint8)
@@ -1227,8 +1202,33 @@ 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, want_requant=False):
-                weight, bias, offload_stream = cast_bias_weight(self, input, offloadable=True, compute_dtype=compute_dtype, want_requant=want_requant)
+            def forward_comfy_cast_weights(
+                self,
+                input,
+                compute_dtype=None,
+                want_requant=False,
+                weight_only_quant=False,
+            ):
+                if weight_only_quant:
+                    weight, bias, offload_stream = cast_bias_weight(
+                        self,
+                        input=None,
+                        dtype=self.weight.dtype,
+                        device=input.device,
+                        bias_dtype=input.dtype,
+                        offloadable=True,
+                        compute_dtype=compute_dtype,
+                        want_requant=True,
+                    )
+                    weight = weight.to(dtype=input.dtype)
+                else:
+                    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
@@ -1247,9 +1247,10 @@ def mixed_precision_ops(quant_config={}, compute_dtype=torch.bfloat16, full_prec
                    not getattr(self, 'comfy_force_cast_weights', False) and
                    len(self.weight_function) == 0 and len(self.bias_function) == 0
                )
+                quantize_input = QUANT_ALGOS.get(getattr(self, 'quant_format', None), {}).get("quantize_input", True)

                # Training path: quantized forward with compute_dtype backward via autograd function
-                if (input.requires_grad and _use_quantized):
+                if (input.requires_grad and _use_quantized and quantize_input):

                    weight, bias, offload_stream = cast_bias_weight(
                        self,
@@ -1271,7 +1272,7 @@ def mixed_precision_ops(quant_config={}, compute_dtype=torch.bfloat16, full_prec
                    return output

                # Inference path (unchanged)
-                if _use_quantized:
+                if _use_quantized and quantize_input:

                    # Reshape 3D tensors to 2D for quantization (needed for NVFP4 and others)
                    input_reshaped = input.reshape(-1, input_shape[2]) if input.ndim == 3 else input
@@ -1285,7 +1286,13 @@ 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, want_requant=isinstance(input, QuantizedTensor))
+                weight_only_quant = _use_quantized and not quantize_input and isinstance(self.weight, QuantizedTensor)
+                output = self.forward_comfy_cast_weights(
+                    input,
+                    compute_dtype,
+                    want_requant=isinstance(input, QuantizedTensor),
+                    weight_only_quant=weight_only_quant,
+                )

                # Reshape output back to 3D if input was 3D
                if reshaped_3d:
@@ -1301,8 +1308,7 @@ def mixed_precision_ops(quant_config={}, compute_dtype=torch.bfloat16, full_prec

            def set_weight(self, weight, inplace_update=False, seed=None, return_weight=False, **kwargs):
                if getattr(self, 'layout_type', None) is not None:
-                    # dtype is now implicit in the layout class
-                    weight = QuantizedTensor.from_float(weight, self.layout_type, scale="recalculate", stochastic_rounding=seed, inplace_ops=True).to(self.weight.dtype)
+                    weight = self.weight.requantize_from_float(weight, scale="recalculate", stochastic_rounding=seed, inplace_ops=True).to(self.weight.dtype)
                else:
                    weight = weight.to(self.weight.dtype)
                if return_weight:
@@ -1,17 +1,55 @@
+import bisect
+
 import comfy.model_management
 import comfy.memory_management
+import comfy.utils
 import comfy_aimdo.host_buffer
 import comfy_aimdo.torch
 import torch

 from comfy.cli_args import args

+def _add_to_bucket(module, buckets, size, priority):
+    bucket = buckets.setdefault(size, [])
+    entry = [-priority, 0, module]
+    entry[1] = id(entry)
+    bisect.insort(bucket, entry)
+    module._pin_balancer_entry = entry
+
+def _steal_pin(module, stack, buckets, size, priority):
+    bucket = buckets.get(size)
+    if bucket is None:
+        return False
+
+    while bucket and bucket[-1][-1] is None:
+        bucket.pop()
+    if not bucket:
+        del buckets[size]
+        return False
+
+    if priority <= -bucket[-1][0]:
+        return False
+
+    *_, victim = bucket.pop()
+    module._pin = victim._pin
+    module._pin_registered = victim._pin_registered
+    module._pin_stack_index = victim._pin_stack_index
+    stack[module._pin_stack_index] = (module, stack[module._pin_stack_index][1])
+
+    victim._pin_registered = False
+    del victim._pin
+    del victim._pin_stack_index
+    del victim._pin_balancer_entry
+
+    _add_to_bucket(module, buckets, size, priority)
+    return True
+
 def get_pin(module, subset="weights"):
    pin = getattr(module, "_pin", None)
    if pin is None or module._pin_registered or args.disable_pinned_memory:
        return pin

-    _, _, stack_split, pinned_size = module._pin_state[subset]
+    _, _, stack_split, pinned_size, *_ = module._pin_state[subset]
    size = pin.nbytes
    comfy.model_management.ensure_pin_registerable(size)

@@ -31,33 +69,51 @@ def pin_memory(module, subset="weights", size=None):
        return

    pin = get_pin(module, subset)
-    if pin is not None or pin_state["failed"]:
+    if pin is not None:
        return

-    hostbuf, stack, stack_split, pinned_size = pin_state[subset]
+    hostbuf, stack, stack_split, pinned_size, counter, buckets = pin_state[subset]
    if size is None:
        size = comfy.memory_management.vram_aligned_size([ module.weight, module.bias ])
    offset = hostbuf.size
-    registerable_size = size + max(0, hostbuf.size - pinned_size[0])
+    registerable_size = size
+    priority = getattr(module, "_pin_balancer_priority", None)
+
+    if priority is None:
+        priority = comfy.utils.bit_reverse_range(counter[0], 16)
+        counter[0] += 1
+        module._pin_balancer_priority = priority

    comfy.memory_management.extra_ram_release(comfy.memory_management.RAM_CACHE_HEADROOM)
    if (not comfy.model_management.ensure_pin_budget(size) or
        not comfy.model_management.ensure_pin_registerable(registerable_size)):
-        pin_state["failed"] = True
-        return False
+        return _steal_pin(module, stack, buckets, size, priority)

+    extended = False
    try:
-        hostbuf.extend(size=size)
+        hostbuf.extend(size=size, register=False)
+        extended = True
+        pin = comfy_aimdo.torch.hostbuf_to_tensor(hostbuf)[offset:offset + size]
+        pin.untyped_storage()._comfy_hostbuf = hostbuf
+        if torch.cuda.cudart().cudaHostRegister(pin.data_ptr(), size, 1) != 0:
+            comfy.model_management.discard_cuda_async_error()
+            comfy.model_management.free_registrations(size)
+            if torch.cuda.cudart().cudaHostRegister(pin.data_ptr(), size, 1) != 0:
+                comfy.model_management.discard_cuda_async_error()
+                del pin
+                hostbuf.truncate(offset, do_unregister=False)
+                return _steal_pin(module, stack, buckets, size, priority)
    except RuntimeError:
-        pin_state["failed"] = True
-        return False
+        if extended:
+            hostbuf.truncate(offset, do_unregister=False)
+        return _steal_pin(module, stack, buckets, size, priority)

-    module._pin = comfy_aimdo.torch.hostbuf_to_tensor(hostbuf)[offset:offset + size]
-    module._pin.untyped_storage()._comfy_hostbuf = hostbuf
+    module._pin = pin
    stack.append((module, offset))
    module._pin_registered = True
    module._pin_stack_index = len(stack) - 1
    stack_split[0] = max(stack_split[0], module._pin_stack_index)
    comfy.model_management.TOTAL_PINNED_MEMORY += size
    pinned_size[0] += size
+    _add_to_bucket(module, buckets, size, priority)
    return True
@@ -10,6 +10,7 @@ try:
        QuantizedLayout,
        TensorCoreFP8Layout as _CKFp8Layout,
        TensorCoreNVFP4Layout as _CKNvfp4Layout,
+        TensorWiseINT8Layout as _CKTensorWiseINT8Layout,
        register_layout_op,
        register_layout_class,
        get_layout_class,
@@ -47,6 +48,9 @@ except ImportError as e:
    class _CKNvfp4Layout:
        pass

+    class _CKTensorWiseINT8Layout:
+        pass
+
    def register_layout_class(name, cls):
        pass

@@ -174,6 +178,7 @@ class TensorCoreFP8E5M2Layout(_TensorCoreFP8LayoutBase):

 # Backward compatibility alias - default to E4M3
 TensorCoreFP8Layout = TensorCoreFP8E4M3Layout
+TensorWiseINT8Layout = _CKTensorWiseINT8Layout


 # ==============================================================================
@@ -184,6 +189,7 @@ register_layout_class("TensorCoreFP8Layout", TensorCoreFP8Layout)
 register_layout_class("TensorCoreFP8E4M3Layout", TensorCoreFP8E4M3Layout)
 register_layout_class("TensorCoreFP8E5M2Layout", TensorCoreFP8E5M2Layout)
 register_layout_class("TensorCoreNVFP4Layout", TensorCoreNVFP4Layout)
+register_layout_class("TensorWiseINT8Layout", _CKTensorWiseINT8Layout)
 if _CK_MXFP8_AVAILABLE:
    register_layout_class("TensorCoreMXFP8Layout", TensorCoreMXFP8Layout)

@@ -214,6 +220,13 @@ if _CK_MXFP8_AVAILABLE:
        "group_size": 32,
    }

+QUANT_ALGOS["int8_tensorwise"] = {
+    "storage_t": torch.int8,
+    "parameters": {"weight_scale"},
+    "comfy_tensor_layout": "TensorWiseINT8Layout",
+    "quantize_input": False,
+}
+

 # ==============================================================================
 # Re-exports for backward compatibility
@@ -226,6 +239,7 @@ __all__ = [
    "TensorCoreFP8E4M3Layout",
    "TensorCoreFP8E5M2Layout",
    "TensorCoreNVFP4Layout",
+    "TensorWiseINT8Layout",
    "QUANT_ALGOS",
    "register_layout_op",
 ]
@@ -464,10 +464,7 @@ def _calc_cond_batch_multigpu(model: BaseModel, conds: list[list[dict]], x_in: t

    def _handle_batch(device: torch.device, batch_tuple: tuple[comfy.hooks.HookGroup, tuple], results: list[thread_result]):
        try:
-            # TODO: non-NVIDIA support -- guard with `if device.type == "cuda":` once
-            # we extend multigpu QA beyond CUDA. Unconditional call crashes on
-            # XPU/NPU/MPS/CPU/DirectML backends.
-            torch.cuda.set_device(device)
+            comfy.model_management.set_torch_device(device)
            model_current: BaseModel = model_options["multigpu_clones"][device].model
            # run every hooked_to_run separately
            with torch.no_grad():
@@ -16,6 +16,7 @@ import comfy.ldm.cosmos.vae
 import comfy.ldm.wan.vae
 import comfy.ldm.wan.vae2_2
 import comfy.ldm.hunyuan3d.vae
+import comfy.ldm.triposplat.vae
 import comfy.ldm.ace.vae.music_dcae_pipeline
 import comfy.ldm.cogvideo.vae
 import comfy.ldm.hunyuan_video.vae
@@ -49,6 +50,7 @@ import comfy.text_encoders.lt
 import comfy.text_encoders.hunyuan_video
 import comfy.text_encoders.cosmos
 import comfy.text_encoders.lumina2
+import comfy.text_encoders.pixeldit
 import comfy.text_encoders.wan
 import comfy.text_encoders.hidream
 import comfy.text_encoders.ace
@@ -56,6 +58,8 @@ import comfy.text_encoders.omnigen2
 import comfy.text_encoders.qwen_image
 import comfy.text_encoders.hunyuan_image
 import comfy.text_encoders.z_image
+import comfy.text_encoders.krea2
+import comfy.text_encoders.ideogram4
 import comfy.text_encoders.ovis
 import comfy.text_encoders.kandinsky5
 import comfy.text_encoders.jina_clip_2
@@ -64,6 +68,8 @@ import comfy.text_encoders.anima
 import comfy.text_encoders.ace15
 import comfy.text_encoders.longcat_image
 import comfy.text_encoders.qwen35
+import comfy.text_encoders.qwen3vl
+import comfy.text_encoders.boogu
 import comfy.text_encoders.ernie
 import comfy.text_encoders.gemma4
 import comfy.text_encoders.cogvideo
@@ -462,6 +468,9 @@ class CLIP:
    def decode(self, token_ids, skip_special_tokens=True):
        return self.tokenizer.decode(token_ids, skip_special_tokens=skip_special_tokens)

+    def is_dynamic(self):
+        return self.patcher.is_dynamic()
+
 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
@@ -893,6 +902,16 @@ class VAE:
                #Force cast it for --disable-dynamic-vram users until there is a true core fix.
                if not comfy.memory_management.aimdo_enabled:
                    self.disable_offload = True
+            elif "gs.base_offset_scale" in sd and "octree.out_proj.weight" in sd:  # TripoSplat octree gaussian decoder
+                self.first_stage_model = comfy.ldm.triposplat.vae.OctreeGaussianDecoder()
+                self.latent_channels = 16
+                self.latent_dim = 1
+                self.working_dtypes = [torch.float16, torch.bfloat16, torch.float32]
+                # The generic VAE.encode/decode path isn't used: VAEDecodeTripoSplat calls the gaussian
+                # decoder directly (structured GaussianSplat objects, not a tensor and reserves VRAM itself from num_gaussians.
+                def _no_generic_io(*args, **kwargs):
+                    raise RuntimeError("TripoSplat gaussian decoder: use the 'TripoSplat Decode' (VAEDecodeTripoSplat)")
+                self.memory_used_encode = self.memory_used_decode = _no_generic_io
            else:
                logging.warning("WARNING: No VAE weights detected, VAE not initalized.")
                self.first_stage_model = None
@@ -1235,6 +1254,11 @@ class VAE:
        except:
            return None

+    def is_dynamic(self):
+        # A VAE built from a state dict with no detectable VAE weights returns early
+        # from __init__ ("No VAE weights detected") before self.patcher is assigned.
+        patcher = getattr(self, "patcher", None)
+        return patcher is not None and patcher.is_dynamic()

 class StyleModel:
    def __init__(self, model, device="cpu"):
@@ -1285,6 +1309,10 @@ class CLIPType(Enum):
    LONGCAT_IMAGE = 26
    COGVIDEOX = 27
    LENS = 28
+    PIXELDIT = 29
+    IDEOGRAM4 = 30
+    BOOGU = 31
+    KREA2 = 32



@@ -1338,6 +1366,8 @@ class TEModel(Enum):
    GEMMA_4_31B = 31
    T5_GEMMA = 32
    GPT_OSS_20B = 33
+    QWEN3VL_4B = 34
+    QWEN3VL_8B = 35


 def detect_te_model(sd):
@@ -1399,6 +1429,8 @@ def detect_te_model(sd):
        if weight.shape[0] == 5120:
            return TEModel.QWEN35_27B
        return TEModel.QWEN35_2B
+    if "model.visual.deepstack_merger_list.0.norm.weight" in sd:  # DeepStack is unique to Qwen3-VL
+        return TEModel.QWEN3VL_4B if sd["model.visual.merger.linear_fc2.weight"].shape[0] == 2560 else TEModel.QWEN3VL_8B
    if "model.layers.0.post_attention_layernorm.weight" in sd:
        weight = sd['model.layers.0.post_attention_layernorm.weight']
        if 'model.layers.0.self_attn.q_norm.weight' in sd:
@@ -1528,8 +1560,12 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip
            clip_target.tokenizer = variant.tokenizer
            tokenizer_data["tokenizer_json"] = clip_data[0].get("tokenizer_json", None)
        elif te_model == TEModel.GEMMA_2_2B:
-            clip_target.clip = comfy.text_encoders.lumina2.te(**llama_detect(clip_data))
-            clip_target.tokenizer = comfy.text_encoders.lumina2.LuminaTokenizer
+            if clip_type == CLIPType.PIXELDIT:
+                clip_target.clip = comfy.text_encoders.pixeldit.pixeldit_te(**llama_detect(clip_data))
+                clip_target.tokenizer = comfy.text_encoders.pixeldit.PixelDiTGemma2Tokenizer
+            else:
+                clip_target.clip = comfy.text_encoders.lumina2.te(**llama_detect(clip_data))
+                clip_target.tokenizer = comfy.text_encoders.lumina2.LuminaTokenizer
            tokenizer_data["spiece_model"] = clip_data[0].get("spiece_model", None)
        elif te_model == TEModel.GEMMA_3_4B:
            clip_target.clip = comfy.text_encoders.lumina2.te(**llama_detect(clip_data), model_type="gemma3_4b")
@@ -1579,8 +1615,12 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip
            clip_target.clip = comfy.text_encoders.ovis.te(**llama_detect(clip_data))
            clip_target.tokenizer = comfy.text_encoders.ovis.OvisTokenizer
        elif te_model == TEModel.QWEN3_8B:
-            clip_target.clip = comfy.text_encoders.flux.klein_te(**llama_detect(clip_data), model_type="qwen3_8b")
-            clip_target.tokenizer = comfy.text_encoders.flux.KleinTokenizer8B
+            if clip_type == CLIPType.IDEOGRAM4:
+                clip_target.clip = comfy.text_encoders.ideogram4.te(**llama_detect(clip_data))
+                clip_target.tokenizer = comfy.text_encoders.ideogram4.Ideogram4Tokenizer
+            else:
+                clip_target.clip = comfy.text_encoders.flux.klein_te(**llama_detect(clip_data), model_type="qwen3_8b")
+                clip_target.tokenizer = comfy.text_encoders.flux.KleinTokenizer8B
        elif te_model == TEModel.JINA_CLIP_2:
            clip_target.clip = comfy.text_encoders.jina_clip_2.JinaClip2TextModelWrapper
            clip_target.tokenizer = comfy.text_encoders.jina_clip_2.JinaClip2TokenizerWrapper
@@ -1589,6 +1629,28 @@ def load_text_encoder_state_dicts(state_dicts=[], embedding_directory=None, clip
            qwen35_type = {TEModel.QWEN35_08B: "qwen35_08b", TEModel.QWEN35_2B: "qwen35_2b", TEModel.QWEN35_4B: "qwen35_4b", TEModel.QWEN35_9B: "qwen35_9b", TEModel.QWEN35_27B: "qwen35_27b"}[te_model]
            clip_target.clip = comfy.text_encoders.qwen35.te(**llama_detect(clip_data), model_type=qwen35_type)
            clip_target.tokenizer = comfy.text_encoders.qwen35.tokenizer(model_type=qwen35_type)
+        elif te_model in (TEModel.QWEN3VL_4B, TEModel.QWEN3VL_8B):
+            if clip_type == CLIPType.IDEOGRAM4 and te_model == TEModel.QWEN3VL_8B:  # Ideogram4 reuses the full Qwen3-VL-8B (13-layer tap for conditioning + multimodal generate).
+                clip_data[0] = comfy.utils.state_dict_prefix_replace(clip_data[0], {"model.language_model.": "model.", "model.visual.": "visual.", "lm_head.": "model.lm_head."})
+                clip_target.clip = comfy.text_encoders.ideogram4.te_qwen3vl(**llama_detect(clip_data))
+                clip_target.tokenizer = comfy.text_encoders.ideogram4.Ideogram4Qwen3VLTokenizer
+            elif clip_type == CLIPType.BOOGU and te_model == TEModel.QWEN3VL_8B:  # Boogu-Image: full Qwen3-VL-8B, last hidden state, no-think template.
+                clip_data[0] = comfy.utils.state_dict_prefix_replace(clip_data[0], {"model.language_model.": "model.", "model.visual.": "visual.", "lm_head.": "model.lm_head."})
+                clip_target.clip = comfy.text_encoders.boogu.te(**llama_detect(clip_data))
+                clip_target.tokenizer = comfy.text_encoders.boogu.BooguTokenizer
+            elif clip_type == CLIPType.KREA2 and te_model == TEModel.QWEN3VL_4B:  # Krea2: full Qwen3-VL-4B (12-layer tap for conditioning + multimodal generate).
+                clip_data[0] = comfy.utils.state_dict_prefix_replace(clip_data[0], {"model.language_model.": "model.", "model.visual.": "visual.", "lm_head.": "model.lm_head."})
+                clip_target.clip = comfy.text_encoders.krea2.te(**llama_detect(clip_data))
+                clip_target.tokenizer = comfy.text_encoders.krea2.Krea2Tokenizer
+            elif clip_type in (CLIPType.FLUX, CLIPType.FLUX2):  # Flux2 Klein reuses the Qwen3-VL LM (3-layer tap -> 12288); visual unused.
+                klein_model_type = "qwen3_8b" if te_model == TEModel.QWEN3VL_8B else "qwen3_4b"
+                clip_target.clip = comfy.text_encoders.flux.klein_te(**llama_detect(clip_data), model_type=klein_model_type)
+                clip_target.tokenizer = comfy.text_encoders.flux.KleinTokenizer8B if te_model == TEModel.QWEN3VL_8B else comfy.text_encoders.flux.KleinTokenizer
+            else:
+                clip_data[0] = comfy.utils.state_dict_prefix_replace(clip_data[0], {"model.language_model.": "model.", "model.visual.": "visual.", "lm_head.": "model.lm_head."})
+                qwen3vl_type = {TEModel.QWEN3VL_4B: "qwen3vl_4b", TEModel.QWEN3VL_8B: "qwen3vl_8b"}[te_model]
+                clip_target.clip = comfy.text_encoders.qwen3vl.te(**llama_detect(clip_data), model_type=qwen3vl_type)
+                clip_target.tokenizer = comfy.text_encoders.qwen3vl.tokenizer(model_type=qwen3vl_type)
        elif te_model == TEModel.QWEN3_06B:
            clip_target.clip = comfy.text_encoders.anima.te(**llama_detect(clip_data))
            clip_target.tokenizer = comfy.text_encoders.anima.AnimaTokenizer
@@ -543,18 +543,24 @@ class SDTokenizer:
    def _try_get_embedding(self, embedding_name:str):
        '''
        Takes a potential embedding name and tries to retrieve it.
-        Returns a Tuple consisting of the embedding and any leftover string, embedding can be None.
+        Returns a Tuple consisting of the embedding, the cleaned embedding name, and any leftover string, embedding can be None.
        '''
        split_embed = embedding_name.split()
        embedding_name = split_embed[0]
        leftover = ' '.join(split_embed[1:])
+
+        match = re.search(r'[<\[]', embedding_name)
+        if match is not None:
+            leftover = embedding_name[match.start():] + (" " + leftover if leftover else "")
+            embedding_name = embedding_name[:match.start()]
+
        embed = load_embed(embedding_name, self.embedding_directory, self.embedding_size, self.embedding_key)
        if embed is None:
            stripped = embedding_name.strip(',')
            if len(stripped) < len(embedding_name):
                embed = load_embed(stripped, self.embedding_directory, self.embedding_size, self.embedding_key)
-                return (embed, "{} {}".format(embedding_name[len(stripped):], leftover))
-        return (embed, leftover)
+                return (embed, embedding_name, "{} {}".format(embedding_name[len(stripped):], leftover))
+        return (embed, embedding_name, leftover)

    def pad_tokens(self, tokens, amount):
        if self.pad_left:
@@ -585,7 +591,7 @@ class SDTokenizer:
        tokens = []
        for weighted_segment, weight in parsed_weights:
            to_tokenize = unescape_important(weighted_segment)
-            split = re.split(' {0}|\n{0}'.format(self.embedding_identifier), to_tokenize)
+            split = re.split(r'(?<=\s){}'.format(re.escape(self.embedding_identifier)), to_tokenize)
            to_tokenize = [split[0]]
            for i in range(1, len(split)):
                to_tokenize.append("{}{}".format(self.embedding_identifier, split[i]))
@@ -595,7 +601,7 @@ class SDTokenizer:
                # if we find an embedding, deal with the embedding
                if word.startswith(self.embedding_identifier) and self.embedding_directory is not None:
                    embedding_name = word[len(self.embedding_identifier):].strip('\n')
-                    embed, leftover = self._try_get_embedding(embedding_name)
+                    embed, embedding_name, leftover = self._try_get_embedding(embedding_name)
                    if embed is None:
                        logging.warning(f"warning, embedding:{embedding_name} does not exist, ignoring")
                    else:
@@ -24,12 +24,16 @@ import comfy.text_encoders.qwen_image
 import comfy.text_encoders.hunyuan_image
 import comfy.text_encoders.kandinsky5
 import comfy.text_encoders.z_image
+import comfy.text_encoders.ideogram4
+import comfy.text_encoders.boogu
+import comfy.text_encoders.krea2
 import comfy.text_encoders.anima
 import comfy.text_encoders.ace15
 import comfy.text_encoders.longcat_image
 import comfy.text_encoders.ernie
 import comfy.text_encoders.cogvideo
 import comfy.text_encoders.hidream_o1
+import comfy.text_encoders.pixeldit

 from . import supported_models_base
 from . import latent_formats
@@ -844,6 +848,8 @@ class Lens(supported_models_base.BASE):
    unet_extra_config = {}
    latent_format = latent_formats.Flux2

+    memory_usage_factor = 4.0
+
    supported_inference_dtypes = [torch.bfloat16, torch.float32] # fp16 causes NaNs

    vae_key_prefix = ["vae."]
@@ -1201,6 +1207,72 @@ class ZImagePixelSpace(ZImage):
    def get_model(self, state_dict, prefix="", device=None):
        return model_base.ZImagePixelSpace(self, device=device)

+class PixelDiTT2I(supported_models_base.BASE):
+    unet_config = {
+        "image_model": "pixeldit_t2i",
+    }
+
+    unet_extra_config = {}
+
+    sampling_settings = {
+        "shift": 4.0,  # 1024px stage 3 default; 2.0 for 512px
+    }
+
+    latent_format = latent_formats.PixelDiTPixel
+    memory_usage_factor = 0.04
+    supported_inference_dtypes = [torch.bfloat16, torch.float32]
+
+    vae_key_prefix = ["vae."]
+    text_encoder_key_prefix = ["text_encoders."]
+
+    def get_model(self, state_dict, prefix="", device=None):
+        return model_base.PixelDiTT2I(self, device=device)
+
+    def process_unet_state_dict(self, state_dict):
+        # pixel_dim from pixel_embedder.proj.weight = (pixel_dim, in_channels); p2 derived per-weight from total // (6 * pixel_dim).
+        pixel_dim = next(v for k, v in state_dict.items() if k.endswith("pixel_embedder.proj.weight")).shape[0]
+
+        out = {}
+        marker = ".adaLN_modulation.0."
+        for k, v in state_dict.items():
+            if k.startswith("_repa_projector") or k.startswith("net_ema."):
+                continue
+            if k.startswith("core."):
+                k = k[len("core."):]
+            elif k.startswith("net."):
+                k = k[len("net."):]
+            if "pixel_blocks." in k and marker in k:
+                # Split into msa (chunks 0-2) and mlp (chunks 3-5) for the two-Linear PiTBlock to reduce peak VRAM
+                p2 = v.shape[0] // (6 * pixel_dim)
+                trail = v.shape[1:]  # () for bias, (in_dim,) for weight
+                vv = v.view(p2, 6, pixel_dim, *trail)
+                base, suffix = k.split(marker)
+                out[f"{base}.adaLN_modulation_msa.{suffix}"] = vv[:, 0:3].reshape(3 * p2 * pixel_dim, *trail).contiguous()
+                out[f"{base}.adaLN_modulation_mlp.{suffix}"] = vv[:, 3:6].reshape(3 * p2 * pixel_dim, *trail).contiguous()
+            else:
+                out[k] = v
+        return out
+
+    def clip_target(self, state_dict={}):
+        return supported_models_base.ClipTarget(
+            comfy.text_encoders.pixeldit.PixelDiTGemma2Tokenizer,
+            comfy.text_encoders.pixeldit.PixelDiTGemma2TE,
+        )
+
+class PiD(PixelDiTT2I):
+    unet_config = {
+        "image_model": "pid",
+    }
+
+    sampling_settings = {
+        "shift": 1.5, # close approximation of the original distill 4 steps [0.999, 0.866, 0.634, 0.342, 0]
+    }
+
+    memory_usage_factor = 0.04
+
+    def get_model(self, state_dict, prefix="", device=None):
+        return model_base.PiD(self, device=device)
+
 class WAN21_T2V(supported_models_base.BASE):
    unet_config = {
        "image_model": "wan2.1",
@@ -1380,6 +1452,17 @@ class WAN21_SCAIL(WAN21_T2V):
        out = model_base.WAN21_SCAIL(self, image_to_video=False, device=device)
        return out

+
+class WAN21_SCAIL2(WAN21_T2V):
+    unet_config = {
+        "image_model": "wan2.1",
+        "model_type": "scail2",
+    }
+
+    def get_model(self, state_dict, prefix="", device=None):
+        out = model_base.WAN21_SCAIL2(self, image_to_video=False, device=device)
+        return out
+
 class WAN22_WanDancer(WAN21_T2V):
    unet_config = {
        "image_model": "wan2.1",
@@ -1469,6 +1552,30 @@ class Hunyuan3Dv2mini(Hunyuan3Dv2):

    latent_format = latent_formats.Hunyuan3Dv2mini

+class TripoSplat(supported_models_base.BASE):
+    # Image -> 3D gaussian splat flow denoiser
+    unet_config = {
+        "image_model": "triposplat",
+    }
+
+    unet_extra_config = {}
+
+    sampling_settings = {
+        "shift": 3.0,
+    }
+
+    memory_usage_factor = 0.6
+
+    latent_format = latent_formats.TripoSplat
+
+    supported_inference_dtypes = [torch.float16, torch.bfloat16, torch.float32]
+
+    def get_model(self, state_dict, prefix="", device=None):
+        return model_base.TripoSplat(self, device=device)
+
+    def clip_target(self, state_dict={}):
+        return None
+
 class HiDream(supported_models_base.BASE):
    unet_config = {
        "image_model": "hidream",
@@ -1653,6 +1760,94 @@ class Omnigen2(supported_models_base.BASE):
        hunyuan_detect = comfy.text_encoders.hunyuan_video.llama_detect(state_dict, "{}qwen25_3b.transformer.".format(pref))
        return supported_models_base.ClipTarget(comfy.text_encoders.omnigen2.Omnigen2Tokenizer, comfy.text_encoders.omnigen2.te(**hunyuan_detect))

+class Boogu(Omnigen2):
+    unet_config = {
+        "image_model": "boogu",
+    }
+
+    sampling_settings = {
+        "multiplier": 1.0,
+        "shift": 3.16,
+    }
+
+    memory_usage_factor = 2.15
+
+    def get_model(self, state_dict, prefix="", device=None):
+        out = model_base.Boogu(self, device=device)
+        return out
+
+    def clip_target(self, state_dict={}):
+        pref = self.text_encoder_key_prefix[0]
+        hunyuan_detect = comfy.text_encoders.hunyuan_video.llama_detect(state_dict, "{}qwen3vl_8b.transformer.".format(pref))
+        return supported_models_base.ClipTarget(comfy.text_encoders.boogu.BooguTokenizer, comfy.text_encoders.boogu.te(**hunyuan_detect))
+
+class Ideogram4(supported_models_base.BASE):
+    unet_config = {
+        "image_model": "ideogram4",
+    }
+
+    sampling_settings = {
+        "multiplier": 1.0,
+        "shift": 1.0,
+    }
+
+    memory_usage_factor = 11.6
+
+    unet_extra_config = {
+        "num_attention_heads": 18,
+        "attention_head_dim": 256,
+        "intermediate_size": 12288,
+        "adaln_dim": 512,
+        "llm_features_dim": 53248,
+        "rope_theta": 5000000,
+        "mrope_section": [24, 20, 20],
+        "norm_eps": 1e-5,
+    }
+    latent_format = latent_formats.Flux2
+
+    supported_inference_dtypes = [torch.bfloat16, torch.float32]
+
+    vae_key_prefix = ["vae."]
+    text_encoder_key_prefix = ["text_encoders."]
+
+    def get_model(self, state_dict, prefix="", device=None):
+        out = model_base.Ideogram4(self, device=device)
+        return out
+
+    def clip_target(self, state_dict={}):
+        pref = self.text_encoder_key_prefix[0]
+        hunyuan_detect = comfy.text_encoders.hunyuan_video.llama_detect(state_dict, "{}qwen3vl_8b.transformer.".format(pref))
+        return supported_models_base.ClipTarget(comfy.text_encoders.ideogram4.Ideogram4Tokenizer, comfy.text_encoders.ideogram4.te(**hunyuan_detect))
+
+
+class Krea2(supported_models_base.BASE):
+    unet_config = {
+        "image_model": "krea2",
+    }
+
+    sampling_settings = {
+        "multiplier": 1.0,
+        "shift": 1.15,
+    }
+
+    memory_usage_factor = 2.2
+
+    latent_format = latent_formats.Wan21
+
+    supported_inference_dtypes = [torch.bfloat16, torch.float16, torch.float32]
+
+    vae_key_prefix = ["vae."]
+    text_encoder_key_prefix = ["text_encoders."]
+
+    def get_model(self, state_dict, prefix="", device=None):
+        out = model_base.Krea2(self, device=device)
+        return out
+
+    def clip_target(self, state_dict={}):
+        pref = self.text_encoder_key_prefix[0]
+        hunyuan_detect = comfy.text_encoders.hunyuan_video.llama_detect(state_dict, "{}qwen3vl_4b.transformer.".format(pref))
+        return supported_models_base.ClipTarget(comfy.text_encoders.krea2.Krea2Tokenizer, comfy.text_encoders.krea2.te(**hunyuan_detect))
+
 class QwenImage(supported_models_base.BASE):
    unet_config = {
        "image_model": "qwen_image",
@@ -1913,6 +2108,23 @@ class RT_DETR_v4(supported_models_base.BASE):
        return None


+class DepthAnything3(supported_models_base.BASE):
+    unet_config = {
+        "image_model": "DepthAnything3",
+    }
+
+    # Mono path: no num_heads / num_head_channels needed.
+    unet_extra_config = {}
+
+    supported_inference_dtypes = [torch.float16, torch.bfloat16, torch.float32]
+
+    def get_model(self, state_dict, prefix="", device=None):
+        return model_base.DepthAnything3(self, device=device)
+
+    def clip_target(self, state_dict={}):
+        return None
+
+
 class ErnieImage(supported_models_base.BASE):
    unet_config = {
        "image_model": "ernie",
@@ -2111,6 +2323,8 @@ models = [
    CosmosI2VPredict2,
    ZImagePixelSpace,
    ZImage,
+    PiD,
+    PixelDiTT2I,
    Lumina2,
    WAN22_T2V,
    WAN21_CausalAR_T2V,
@@ -2125,10 +2339,12 @@ models = [
    WAN22_Animate,
    WAN21_FlowRVS,
    WAN21_SCAIL,
+    WAN21_SCAIL2,
    WAN22_WanDancer,
    Hunyuan3Dv2mini,
    Hunyuan3Dv2,
    Hunyuan3Dv2_1,
+    TripoSplat,
    HiDream,
    HiDreamO1,
    Chroma,
@@ -2136,7 +2352,10 @@ models = [
    ACEStep,
    ACEStep15,
    Omnigen2,
+    Boogu,
    QwenImage,
+    Ideogram4,
+    Krea2,
    Flux2,
    Lens,
    Kandinsky5Image,
@@ -2150,4 +2369,5 @@ models = [
    CogVideoX_I2V,
    CogVideoX_T2V,
    SVD_img2vid,
+    DepthAnything3,
 ]
@@ -0,0 +1,58 @@
+"""Boogu-Image text encoder: full Qwen3-VL-8B, last hidden state (4096-dim).
+
+Boogu uses the final hidden state of Qwen3-VL as the per-token instruction feature
+(num_instruction_feature_layers=1, reduce_type=mean -> just the last layer).
+The model itself is the standard Qwen3-VL TE, only the chat template differs
+(a fixed system prompt and no <think> block).
+"""
+
+import comfy.text_encoders.qwen3vl
+from comfy import sd1_clip
+
+
+# System prompts from the reference pipeline (pipeline_boogu.py).
+# T2I (non-empty instruction, no image) uses the helpful-assistant prompt
+# everything else (the CFG negative / "drop" condition, and any image case) uses the TI2I "describe" prompt.
+BOOGU_T2I_SYSTEM = "You are a helpful assistant that generates high-quality images based on user instructions. The instructions are as follows."
+BOOGU_DROP_SYSTEM = "Describe the key features of the input image (color, shape, size, texture, objects, background), then explain how the user's text instruction should alter or modify the image. Generate a new image that meets the user's requirements while maintaining consistency with the original input where appropriate."
+
+
+class BooguTokenizer(comfy.text_encoders.qwen3vl.Qwen3VLTokenizer):
+    def __init__(self, embedding_directory=None, tokenizer_data={}):
+        super().__init__(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data, model_type="qwen3vl_8b")
+        # apply_chat_template without add_generation_prompt
+        self.llama_template = "<|im_start|>system\n" + BOOGU_T2I_SYSTEM + "<|im_end|>\n<|im_start|>user\n{}<|im_end|>\n"
+        self.llama_template_images = "<|im_start|>system\n" + BOOGU_DROP_SYSTEM + "<|im_end|>\n<|im_start|>user\n<|vision_start|><|image_pad|><|vision_end|>{}<|im_end|>\n"
+        # Reference SYSTEM_PROMPT_DROP: used for the empty negative/uncond instruction.
+        self.llama_template_drop = "<|im_start|>system\n" + BOOGU_DROP_SYSTEM + "<|im_end|>\n<|im_start|>user\n{}<|im_end|>\n"
+
+    def tokenize_with_weights(self, text, return_word_ids=False, llama_template=None, images=[], prevent_empty_text=False, thinking=True, **kwargs):
+        if llama_template is None and len(images) == 0 and text.strip() == "":
+            llama_template = self.llama_template_drop
+        # Boogu conditions on the no-think template; thinking=True drops the empty <think> block qwen3vl adds by default.
+        return super().tokenize_with_weights(text, return_word_ids=return_word_ids, llama_template=llama_template, images=images, prevent_empty_text=prevent_empty_text, thinking=thinking, **kwargs)
+
+
+class BooguQwen3VLClipModel(comfy.text_encoders.qwen3vl.Qwen3VLClipModel):
+    def __init__(self, device="cpu", dtype=None, attention_mask=True, model_options={}, model_type="qwen3vl_8b"):
+        super().__init__(device=device, dtype=dtype, attention_mask=attention_mask, model_options=model_options, model_type=model_type)
+        # apply the final RMSNorm to the tapped last layer
+        self.layer_norm_hidden_state = True
+
+
+class BooguTEModel(sd1_clip.SD1ClipModel):
+    def __init__(self, device="cpu", dtype=None, model_options={}):
+        clip_model = lambda **kw: BooguQwen3VLClipModel(**kw, model_type="qwen3vl_8b")
+        super().__init__(device=device, dtype=dtype, name="qwen3vl_8b", clip_model=clip_model, model_options=model_options)
+
+
+def te(dtype_llama=None, llama_quantization_metadata=None):
+    class BooguTEModel_(BooguTEModel):
+        def __init__(self, device="cpu", dtype=None, model_options={}):
+            if dtype_llama is not None:
+                dtype = dtype_llama
+            if llama_quantization_metadata is not None:
+                model_options = model_options.copy()
+                model_options["quantization_metadata"] = llama_quantization_metadata
+            super().__init__(device=device, dtype=dtype, model_options=model_options)
+    return BooguTEModel_
@@ -12,7 +12,7 @@ import torch.nn.functional as F

 import comfy.ops
 from comfy import sd1_clip
-from comfy.ldm.modules.attention import TORCH_HAS_GQA, optimized_attention_for_device
+from comfy.ldm.modules.attention import optimized_attention_for_device
 from comfy.text_encoders.llama import RMSNorm, apply_rope


@@ -110,10 +110,6 @@ def _attention_with_sinks(q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, sin
    putting the sink logit in the mask at that column.
    """

-    if num_kv_groups > 1 and not TORCH_HAS_GQA:
-        k = k.repeat_interleave(num_kv_groups, dim=1)
-        v = v.repeat_interleave(num_kv_groups, dim=1)
-
    B, _, S_q, D = q.shape
    H_kv = k.shape[1]
    S_kv = k.shape[-2]
@@ -0,0 +1,120 @@
+"""Ideogram 4 text encoder: Qwen3-VL-8B language model, 13-layer tap.
+
+Ideogram 4 conditions on the concatenation of hidden states from 13 layers of
+Qwen3-VL (layers 0,3,...,33,35), giving a 4096*13 = 53248-dim feature per token.
+"""
+
+import os
+
+from transformers import Qwen2Tokenizer
+
+import comfy.text_encoders.llama
+import comfy.text_encoders.qwen3vl
+from comfy import sd1_clip
+
+# Reference taps outputs of layers (0,3,...,35); comfy captures layer inputs, offset by +1.
+IDEOGRAM4_TAP_LAYERS = [1, 4, 7, 10, 13, 16, 19, 22, 25, 28, 31, 34, 36]
+
+
+class Qwen3VLTokenizer(sd1_clip.SDTokenizer):
+    def __init__(self, embedding_directory=None, tokenizer_data={}):
+        tokenizer_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), "qwen25_tokenizer")
+        super().__init__(tokenizer_path, pad_with_end=False, embedding_directory=embedding_directory,
+                         embedding_size=4096, embedding_key='qwen3vl_8b', tokenizer_class=Qwen2Tokenizer,
+                         has_start_token=False, has_end_token=False, pad_to_max_length=False,
+                         max_length=99999999, min_length=1, pad_token=151643, tokenizer_data=tokenizer_data)
+
+
+class Ideogram4Tokenizer(sd1_clip.SD1Tokenizer):
+    def __init__(self, embedding_directory=None, tokenizer_data={}):
+        super().__init__(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data,
+                         name="qwen3vl_8b", tokenizer=Qwen3VLTokenizer)
+
+        self.llama_template = "<|im_start|>user\n{}<|im_end|>\n<|im_start|>assistant\n"
+
+    def tokenize_with_weights(self, text, return_word_ids=False, llama_template=None, **kwargs):
+        if text.startswith('<|im_start|>'):
+            llama_text = text
+        elif llama_template is None:
+            llama_text = self.llama_template.format(text)
+        else:
+            llama_text = llama_template.format(text)
+        return super().tokenize_with_weights(llama_text, return_word_ids=return_word_ids, disable_weights=True, **kwargs)
+
+
+# Qwen3-VL-8B = 5e6 (vs plain Qwen3-8B's 1e6)
+# final_norm/lm_head off -> Ideogram only reads raw tapped hidden states
+QWEN3VL_8B_CONFIG = {"rope_theta": 5000000.0, "final_norm": False, "lm_head": False}
+
+
+class Qwen3VL8BModel(sd1_clip.SDClipModel):
+    def __init__(self, device="cpu", layer="hidden", layer_idx=None, dtype=None, attention_mask=True, model_options={}):
+        super().__init__(device=device, layer=IDEOGRAM4_TAP_LAYERS, layer_idx=None,
+                         textmodel_json_config=dict(QWEN3VL_8B_CONFIG),
+                         dtype=dtype, special_tokens={"pad": 151643}, layer_norm_hidden_state=False,
+                         model_class=comfy.text_encoders.llama.Qwen3_8B,
+                         enable_attention_masks=attention_mask, return_attention_masks=attention_mask,
+                         model_options=model_options)
+
+
+class Ideogram4TEModel(sd1_clip.SD1ClipModel):
+    def __init__(self, device="cpu", dtype=None, model_options={}):
+        super().__init__(device=device, dtype=dtype, name="qwen3vl_8b", clip_model=Qwen3VL8BModel, model_options=model_options)
+
+    def encode_token_weights(self, token_weight_pairs):
+        out, pooled, extra = super().encode_token_weights(token_weight_pairs)
+        b, n, seq, h = out.shape # (B, n_taps=13, seq, 4096) stacked in ascending layer order.
+        out = out.permute(0, 2, 3, 1).reshape(b, seq, h * n) # (B, seq, 4096*13). permute -> (B, seq, H, taps).
+        return out, pooled, extra
+
+
+def te(dtype_llama=None, llama_quantization_metadata=None):
+    class Ideogram4TEModel_(Ideogram4TEModel):
+        def __init__(self, device="cpu", dtype=None, model_options={}):
+            if dtype_llama is not None:
+                dtype = dtype_llama
+            if llama_quantization_metadata is not None:
+                model_options = model_options.copy()
+                model_options["quantization_metadata"] = llama_quantization_metadata
+            super().__init__(device=device, dtype=dtype, model_options=model_options)
+    return Ideogram4TEModel_
+
+
+# Full Qwen3-VL-8B variant with vision
+
+class Ideogram4Qwen3VLClipModel(comfy.text_encoders.qwen3vl.Qwen3VLClipModel):
+    def __init__(self, device="cpu", dtype=None, attention_mask=True, model_options={}):
+        super().__init__(device=device, layer=IDEOGRAM4_TAP_LAYERS, layer_idx=None, dtype=dtype,
+                         attention_mask=attention_mask, model_options=model_options, model_type="qwen3vl_8b")
+
+
+class Ideogram4Qwen3VLTEModel(sd1_clip.SD1ClipModel):
+    def __init__(self, device="cpu", dtype=None, model_options={}):
+        super().__init__(device=device, dtype=dtype, name="qwen3vl_8b", clip_model=Ideogram4Qwen3VLClipModel, model_options=model_options)
+
+    def encode_token_weights(self, token_weight_pairs):
+        out, pooled, extra = super().encode_token_weights(token_weight_pairs)
+        b, n, seq, h = out.shape  # (B, n_taps=13, seq, 4096), ascending layer order.
+        out = out.permute(0, 2, 3, 1).reshape(b, seq, h * n)  # (B, seq, 4096*13 = 53248).
+        return out, pooled, extra
+
+
+class Ideogram4Qwen3VLTokenizer(comfy.text_encoders.qwen3vl.Qwen3VLTokenizer):
+    def __init__(self, embedding_directory=None, tokenizer_data={}):
+        super().__init__(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data, model_type="qwen3vl_8b")
+
+    def tokenize_with_weights(self, text, return_word_ids=False, llama_template=None, images=[], prevent_empty_text=False, thinking=True, **kwargs):
+        # Ideogram 4 conditions on the no-think template; default thinking=True drops the empty think block qwen3vl adds.
+        return super().tokenize_with_weights(text, return_word_ids=return_word_ids, llama_template=llama_template, images=images, prevent_empty_text=prevent_empty_text, thinking=thinking, **kwargs)
+
+
+def te_qwen3vl(dtype_llama=None, llama_quantization_metadata=None):
+    class Ideogram4Qwen3VLTEModel_(Ideogram4Qwen3VLTEModel):
+        def __init__(self, device="cpu", dtype=None, model_options={}):
+            if dtype_llama is not None:
+                dtype = dtype_llama
+            if llama_quantization_metadata is not None:
+                model_options = model_options.copy()
+                model_options["quantization_metadata"] = llama_quantization_metadata
+            super().__init__(device=device, dtype=dtype, model_options=model_options)
+    return Ideogram4Qwen3VLTEModel_
@@ -0,0 +1,84 @@
+"""Krea 2 (K2) text encoder: Qwen3-VL-4B, 12-layer tap.
+
+K2 conditions on a stack of hidden states from 12 layers of Qwen3-VL-4B
+(reference taps ``hidden_states[2,5,8,...,35]``), kept as a ``(B, 12, seq, 2560)`` tensor and
+consumed by the DiT's internal ``txtfusion`` adapter. Comfy carries conditioning as a 3D tensor,
+so the 12-layer stack is flattened to ``(B, seq, 12*2560)`` here and unpacked inside the model.
+"""
+
+import numbers
+
+import torch
+
+import comfy.text_encoders.qwen3vl
+from comfy import sd1_clip
+
+# tap k == hidden_states[k] (no offset).
+KREA2_TAP_LAYERS = [2, 5, 8, 11, 14, 17, 20, 23, 26, 29, 32, 35]
+
+# Identical system template to Qwen-Image; Krea2 strips the system+user-opening prefix.
+KREA2_TEMPLATE = "<|im_start|>system\nDescribe the image by detailing the color, shape, size, texture, quantity, text, spatial relationships of the objects and background:<|im_end|>\n<|im_start|>user\n{}<|im_end|>\n<|im_start|>assistant\n"
+
+
+class Krea2Tokenizer(comfy.text_encoders.qwen3vl.Qwen3VLTokenizer):
+    def __init__(self, embedding_directory=None, tokenizer_data={}):
+        super().__init__(embedding_directory=embedding_directory, tokenizer_data=tokenizer_data, model_type="qwen3vl_4b")
+        self.llama_template = KREA2_TEMPLATE  # conditioning template; image text-gen uses qwen3vl's default image template.
+
+    def tokenize_with_weights(self, text, return_word_ids=False, llama_template=None, images=[], prevent_empty_text=False, thinking=True, **kwargs):
+        # Krea2 conditions on the no-think template; thinking=True drops the empty <think> block qwen3vl adds.
+        return super().tokenize_with_weights(text, return_word_ids=return_word_ids, llama_template=llama_template, images=images, prevent_empty_text=prevent_empty_text, thinking=thinking, **kwargs)
+
+
+class Krea2Qwen3VLClipModel(comfy.text_encoders.qwen3vl.Qwen3VLClipModel):
+    def __init__(self, device="cpu", dtype=None, attention_mask=True, model_options={}):
+        super().__init__(device=device, layer=KREA2_TAP_LAYERS, layer_idx=None, dtype=dtype,
+                         attention_mask=attention_mask, model_options=model_options, model_type="qwen3vl_4b")
+
+
+class Krea2TEModel(sd1_clip.SD1ClipModel):
+    def __init__(self, device="cpu", dtype=None, model_options={}):
+        super().__init__(device=device, dtype=dtype, name="qwen3vl_4b", clip_model=Krea2Qwen3VLClipModel, model_options=model_options)
+
+    def encode_token_weights(self, token_weight_pairs, template_end=-1):
+        out, pooled, extra = super().encode_token_weights(token_weight_pairs)  # out: (B, 12, seq, 2560)
+        tok_pairs = token_weight_pairs["qwen3vl_4b"][0]
+
+        # Strip the system + user-opening prefix
+        count_im_start = 0
+        if template_end == -1:
+            for i, v in enumerate(tok_pairs):
+                elem = v[0]
+                if not torch.is_tensor(elem) and isinstance(elem, numbers.Integral):
+                    if elem == 151644 and count_im_start < 2:
+                        template_end = i
+                        count_im_start += 1
+            if out.shape[2] > (template_end + 3):
+                if tok_pairs[template_end + 1][0] == 872:      # "user"
+                    if tok_pairs[template_end + 2][0] == 198:   # "\n"
+                        template_end += 3
+
+        out = out[:, :, template_end:]
+
+        b, n, seq, h = out.shape
+        # Flatten the 12-layer axis into the feature dim: (B, seq, 12*2560). Unpacked in the model.
+        out = out.permute(0, 2, 1, 3).reshape(b, seq, n * h)
+
+        if "attention_mask" in extra:
+            extra["attention_mask"] = extra["attention_mask"][:, template_end:]
+            if extra["attention_mask"].sum() == torch.numel(extra["attention_mask"]):
+                extra.pop("attention_mask")
+
+        return out, pooled, extra
+
+
+def te(dtype_llama=None, llama_quantization_metadata=None):
+    class Krea2TEModel_(Krea2TEModel):
+        def __init__(self, device="cpu", dtype=None, model_options={}):
+            if llama_quantization_metadata is not None:
+                model_options = model_options.copy()
+                model_options["quantization_metadata"] = llama_quantization_metadata
+            if dtype_llama is not None:
+                dtype = dtype_llama
+            super().__init__(device=device, dtype=dtype, model_options=model_options)
+    return Krea2TEModel_
--- a/Show More
+++ b/Show More