Skip to content

Add SkyReels-V2 pipelines #11518

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
wants to merge 38 commits into
base: main
Choose a base branch
from
Draft

Conversation

tolgacangoz
Copy link
Contributor

@tolgacangoz tolgacangoz commented May 7, 2025

Thanks for the opportunity to fix #11374!

TODOs:
FlowMatchUniPCMultistepScheduler: Just copy-pasted (for now).
SkyReelsV2DiffusionForcingPipeline
SkyReelsV2DiffusionForcingImageToVideoPipeline
SkyReelsV2Pipeline
SkyReelsV2ImageToVideoPipeline
⬜ Did you make sure to update the documentation with your changes?
⬜ Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag members/contributors who may be interested in your PR.

tolgacangoz and others added 7 commits May 7, 2025 21:53
…usion forcing

- Introduced the drafts of `SkyReelsV2TextToVideoPipeline`, `SkyReelsV2ImageToVideoPipeline`, `SkyReelsV2DiffusionForcingPipeline`, and `FlowUniPCMultistepScheduler`.
@ukaprch
Copy link

ukaprch commented May 8, 2025

It's about time. Thanks.

tolgacangoz added 22 commits May 8, 2025 20:01
Replaces custom attention implementations with `SkyReelsV2AttnProcessor2_0` and the standard `Attention` module.
Updates `WanAttentionBlock` to use `FP32LayerNorm` and `FeedForward`.
Removes the `model_type` parameter, simplifying model architecture and attention block initialization.
Introduces new classes `SkyReelsV2ImageEmbedding` and `SkyReelsV2TimeTextImageEmbedding` for enhanced image and time-text processing. Refactors the `SkyReelsV2Transformer3DModel` to integrate these embeddings, updating the constructor parameters for better clarity and functionality. Removes unused classes and methods to streamline the codebase.
…ds and begin reorganizing the forward pass.
…hod, integrating rotary embeddings and improving attention handling. Removes the deprecated `rope_apply` function and streamlines the attention mechanism for better integration and clarity.
…ethod by updating parameter names for clarity, integrating attention masks, and improving the handling of encoder hidden states.
…ethod by enhancing the handling of time embeddings and encoder hidden states. Updates parameter names for clarity and integrates rotary embeddings, ensuring better compatibility with the model's architecture.
…ing components and streamline the text-to-video generation process. Updates class documentation and adjusts parameter handling for improved clarity and functionality.
…parameter handling and improving integration.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

[Feature request] Integrate SkyReels-V2 support in diffusers
2 participants