Elastic-DiT support #13191
Description
Model/Pipeline/Scheduler description
Elastic-DiT was released a few hours ago: https://github.com/wangjiangshan0725/Elastic-DiT
It's supposed to greatly accelerate (~2x speed) the diffusion process of 2D image generators like qwen image and Flux with little impact to the quality of the output. This project was leaded by the creator of Hunyuan3D who used it also for the image->3d model (but this is likely out of scope for the diffusers project as it doesn't support 3D generation).
The code is available in the Github repo, the model checkpoint is hosted on HuggingFace and the paper is available on Arxiv.
Open source status
- The model implementation is available.
- The model weights are available (Only relevant if addition is not a scheduler).
Provide useful links for the implementation
https://github.com/wangjiangshan0725/Elastic-DiT
https://github.com/wangjiangshan0725/Elastic-DiT/blob/65535c3671199b0e26e96d7493cdeb71b2bcf3d8/diffsynth/pipelines/qwen_image_elastic_dit.py
https://huggingface.co/wjs0725/Elastic-DiT