Support Comfy 0.3.69+ and Add Lumina2 (ZImage Turbo) Compatibility #69
Open
EnragedAntelope wants to merge 8 commits into
Open
Support Comfy 0.3.69+ and Add Lumina2 (ZImage Turbo) Compatibility #69EnragedAntelope wants to merge 8 commits into
EnragedAntelope wants to merge 8 commits into
Conversation
- Update imports to use comfy.ldm.flux.layers instead of comfy.ldm.chroma.layers for DoubleStreamBlock and SingleStreamBlock (they are now set to None in chroma.layers) - Pass latent_shapes parameter through executor.execute in NAGCFGGuider.sample() to match base CFGGuider implementation in latest ComfyUI Fixes: - TypeError: NoneType takes no arguments (import issue) - TypeError: NAGCFGGuider.inner_sample() got an unexpected keyword argument 'latent_shapes'
…ck-016HZBLZssSEyCn5ne5dT9k9 Fix ComfyUI 0.3.69+ compatibility issues
Add None check for guidance parameter before calling .detach() to prevent AttributeError: 'NoneType' object has no attribute 'detach' when queue is interrupted. When guidance is None, use zeros_like(distill_timestep) as fallback, matching the pattern used in other model implementations (flux, hunyuan_video).
1. Fix TypeError: Chroma.forward_orig() got multiple values for argument 'attn_mask' - Remove incorrect txt_ids_negative parameter from else block - The else block calls base Chroma.forward_orig which doesn't accept txt_ids_negative - Previous fix in e751545 was based on misunderstanding of the code flow 2. Add defensive checks for IndexError during queue interruption - Check img_ids is not None and has correct dimensions before concatenation - Handle edge case where origin_bsz <= 0 - Prevents "Dimension out of range" error when img_ids has unexpected shape - Follows same pattern as flux/model.py This properly handles queue interruption scenarios while maintaining compatibility with base Chroma class in non-NAG mode.
**Root cause:** When a queue with NAG active is interrupted, the restoration of self.forward_orig and block.forward methods never executes, leaving them in the NAG state. When a new queue starts without NAG active, the else block calls self.forward_orig which is still the NAG version, causing parameter misalignment and IndexError. **Fixes applied to all models:** 1. **Chroma** (chroma/model.py): - Wrap NAG setup in try/finally to ensure restoration even on interruption - In else block, call Chroma.forward_orig directly instead of self.forward_orig - Add defensive checks for img_ids dimensions 2. **Flux** (flux/model.py): - Same try/finally protection - Call Flux.forward_orig directly in else block 3. **HunyuanVideo** (hunyuan_video/model.py): - Same try/finally protection - Call HunyuanVideo.forward_orig directly in else block 4. **Wan** (wan/model.py - both NAGWanModel and NAGVaceWanModel): - Wrap forward_orig call in try/finally - Initialize forward_orig_ and cross_attns_forward before if block - Restore in finally block only if NAG was active This ensures that even if queue is interrupted mid-processing, all modified methods are restored to their original state, preventing parameter misalignment errors in subsequent queue runs.
…ck-016HZBLZssSEyCn5ne5dT9k9 Claude/fix comfyui node pack 016 hzbl zss s ey cn5ne5d t9k9
This adds NAG (Normalized Attention Guidance) support for Lumina 2 and Z-Image Turbo models which use the NextDiT architecture. - Add lumina/model.py with NAGNextDiT wrapper and NAGNextDiTSwitch - Add lumina/layers.py with utility function for attention-level NAG - Register NextDiT model type in samplers.py The implementation applies NAG at the output level by running the model with both positive and negative conditions, then extrapolating between the outputs using the NAG guidance formula.
…-QMcu0 Add NAG support for Lumina/NextDiT models (Z-Image Turbo)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.