DyDiT

The official implementation of two papers:

• "2025ICLR Dynamic Diffusion Transformer"
• Journal version "2025ArXiv DyDiT++: Dynamic Diffusion Transformers for Efficient Visual Generation"

DiT.vs.DyDiT.mp4

🚀 News

• 2025.04.10: The extended journal version has been released.
• 2025.03.26: We release the code of training and text-to-image generation model, DyFLUX.
• 2025.01.23: "Dynamic Diffusion Transformer" is accepted by ICLR 2025!!! We will update the code and paper soon.
• 2024.12.19: We release the code for inference.
• 2024.10.04: Our paper is released.

🔧 Usage

We provide detailed instructions to run our code. Please cd DyDiT or cd DyFLUX for more information.

🤔 Cite DyDiT

If you found our work useful, please consider citing us.

@article{zhao2024dynamic,
  title={Dynamic diffusion transformer},
  author={Zhao, Wangbo and Han, Yizeng and Tang, Jiasheng and Wang, Kai and Song, Yibing and Huang, Gao and Wang, Fan and You, Yang},
  journal={ICLR},
  year={2025}
}


@misc{zhao2025dyditdynamicdiffusiontransformers,
      title={DyDiT++: Dynamic Diffusion Transformers for Efficient Visual Generation}, 
      author={Wangbo Zhao and Yizeng Han and Jiasheng Tang and Kai Wang and Hao Luo and Yibing Song and Gao Huang and Fan Wang and Yang You},
      year={2025},
      eprint={2504.06803},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2504.06803}, 
}

☎️ Contact

If you're interested in collaborating with us, feel free to reach out via email at wangbo.zhao96@gmail.com.