SEEDS: update noise decomposition and refactor #9633
Merged
Conversation
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
- Update the decomposition to reflect interval dependency - Extract phi computations into functions - Use torch.lerp for interpolation
Thor-ATX
pushed a commit
to asteriafilmco/ComfyUI
that referenced
this pull request
Sep 3, 2025
…e-update * commit '4f5812b93712e0f52ae8fe80a89e8b5e7d0fa309': (77 commits) Update template to 0.1.73 (comfyanonymous#9686) ImageScaleToMaxDimension node. (comfyanonymous#9689) Accept prompt_id in interrupt handler (comfyanonymous#9607) uso -> uxo/uno as requested. (comfyanonymous#9688) USO style reference. (comfyanonymous#9677) Enable Convolution AutoTuning (comfyanonymous#9301) Implement the USO subject identity lora. (comfyanonymous#9674) Probably not necessary anymore. (comfyanonymous#9646) SEEDS: update noise decomposition and refactor (comfyanonymous#9633) convert Primitive nodes to V3 schema (comfyanonymous#9372) convert nodes_stability.py to V3 schema (comfyanonymous#9497) convert Video nodes to V3 schema (comfyanonymous#9489) convert Stable Cascade nodes to V3 schema (comfyanonymous#9373) ComfyUI version 0.3.56 Lower ram usage on windows. (comfyanonymous#9628) ComfyUI v0.3.55 Update template to 0.1.70 (comfyanonymous#9620) Trim audio to video when saving video. (comfyanonymous#9617) Support the 5B fun inpaint model. (comfyanonymous#9614) Support wan2.2 5B fun control model. (comfyanonymous#9611) ...
toxicwind
pushed a commit
to toxicwind/ComfyUI
that referenced
this pull request
Oct 12, 2025
- Update the decomposition to reflect interval dependency - Extract phi computations into functions - Use torch.lerp for interpolation
adlerfaulkner
pushed a commit
to LucaLabsInc/ComfyUI
that referenced
this pull request
Oct 16, 2025
- Update the decomposition to reflect interval dependency - Extract phi computations into functions - Use torch.lerp for interpolation
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.
Noise decomposition
The current noise decomposition seems right for the default
r, but not for other cases. For instance, settingr=1in SEEDS-2 will fully use a new noise, instead of depending on the noise from the first stage that already covers the current interval [s, t]. This seems not to match the claim that "overlapping time intervals need to be dependent."Therefore, I tried to follow the existed derivation to obtain another result.
In this decomposition, when
r=1,z1dominates the current interval. In SEEDS-2, this appears to produce relatively sharper results at low CFG and in the low NFE regime whenr > 0.5.Test with V-prediction ztsnr model and simple scheduler.
Other changes
h_phi_kvariable.