mirror of
https://github.com/comfyanonymous/ComfyUI.git
synced 2026-04-02 23:13:42 +08:00
abfea891ef
May also resolve #9784 — the mask normalization fixes a class of dimensionality mismatches that can cause the `y, x = torch.where(mask)` crash in `get_mask_aabb`, though the root cause in that report is unconfirmed. ## Summary `resolve_areas_and_cond_masks_multidim` assumes 2D spatial masks. This breaks for 1D audio models (StableAudio1, ACEAudio15) because upstream code (`ConditioningSetMask`, `set_mask_for_conditioning`) unconditionally unsqueezes masks with `ndim < 3`, corrupting valid `[B, L]` masks into `[1, B, L]` before they reach the sampler. This PR: - Normalizes masks to `[batch, *spatial_dims]` using `dims` as the source of truth - Adds a 1D resize path via `F.interpolate(mode='linear')` - Guards `set_area_to_bounds` with `len(dims) == 2` to prevent crashes on non-2D masks (the existing `get_mask_aabb` and `H, W, Y, X` unpacking are 2D-only) The root cause is the hardcoded `if len(mask.shape) < 3` in `nodes.py:242` and `hooks.py:725`. Fixing it there would require threading latent dimensionality into the conditioning nodes — a much larger change. Normalizing in `resolve_areas_and_cond_masks_multidim` where `dims` is already available is the minimal fix. Fully backwards compatible for existing 2D image and 3D video workflows. ## Test plan - [x] 26 unit tests covering 1D/2D/3D mask normalization, resize, and `set_area_to_bounds` guard (`tests-unit/comfy_test/samplers_test.py`) - [x] 2D image regression with hook masking: [lorahookmasking.json](https://github.com/Kosinkadink/ComfyUI/blob/workflows/lorahookmasking.json) - [x] 2D image with `set_area_to_bounds` ("mask bounds" mode) — no crash, correct area computation - [x] 1D audio with conditioning mask: [acestep-1.5-prompt-lora-blending.json](https://github.com/ryanontheinside/ComfyUI_RyanOnTheInside/blob/main/examples/ace1.5/acestep-1.5-prompt-lora-blending.json) (requires custom nodes that patch this function pending upstream) |
||
|---|---|---|
| .. | ||
| audio_encoders | ||
| cldm | ||
| comfy_types | ||
| extra_samplers | ||
| image_encoders | ||
| k_diffusion | ||
| ldm | ||
| sd1_tokenizer | ||
| t2i_adapter | ||
| taesd | ||
| text_encoders | ||
| weight_adapter | ||
| cli_args.py | ||
| clip_config_bigg.json | ||
| clip_model.py | ||
| clip_vision_config_g.json | ||
| clip_vision_config_h.json | ||
| clip_vision_config_vitl_336_llava.json | ||
| clip_vision_config_vitl_336.json | ||
| clip_vision_config_vitl.json | ||
| clip_vision_siglip2_base_naflex.json | ||
| clip_vision_siglip_384.json | ||
| clip_vision_siglip_512.json | ||
| clip_vision.py | ||
| conds.py | ||
| context_windows.py | ||
| controlnet.py | ||
| diffusers_convert.py | ||
| diffusers_load.py | ||
| float.py | ||
| gligen.py | ||
| hooks.py | ||
| latent_formats.py | ||
| lora_convert.py | ||
| lora.py | ||
| memory_management.py | ||
| model_base.py | ||
| model_detection.py | ||
| model_management.py | ||
| model_patcher.py | ||
| model_sampling.py | ||
| nested_tensor.py | ||
| ops.py | ||
| options.py | ||
| patcher_extension.py | ||
| pinned_memory.py | ||
| pixel_space_convert.py | ||
| quant_ops.py | ||
| rmsnorm.py | ||
| sample.py | ||
| sampler_helpers.py | ||
| samplers.py | ||
| sd1_clip_config.json | ||
| sd1_clip.py | ||
| sd.py | ||
| sdxl_clip.py | ||
| supported_models_base.py | ||
| supported_models.py | ||
| utils.py | ||
| windows.py | ||