DreamID-V: Bridging the Image-to-Video Gap for High-Fidelity Face Swapping via Diffusion Transformer

🌐 Project Page | 📜 Arxiv | 🤗 Models |

DreamID-V: Bridging the Image-to-Video Gap for High-Fidelity Face Swapping via Diffusion Transformer
Xu Guo ^* , Fulong Ye ^* , Xinghui Li ^*, Pengqi Tu, Pengze Zhang, Qichao Sun, Songtao Zhao ^†, Xiangwang Hou ^† Qian He
^* Equal contribution, ^† Corresponding author
Tsinghua University | Intelligent Creation Team, ByteDance

🔥 News

• [01/08/2026] 🔥 Thanks HM-RunningHub for supporting ComfyUI!
• [01/06/2026] 🔥 Our paper is released!
• [01/05/2026] 🔥 Our code is released!
• [12/17/2025] 🔥 Our project is released!
• [08/11/2025] 🎉 Our image version DreamID is accepted by SIGGRAPH Asia 2025!

💡 Usage Tips

• Reference Image Preparation: Please upload cropped face images (recommended resolution: 512x512) as reference. Avoid using full-body photos to ensure optimal identity preservation.
• Inference Steps: For simple scenes, you can reduce the sampling steps to 20 to significantly decrease inference time.

  Note: Our internal model based on Seedance1.0 achieves high quality in under 8 steps. Feel free to experience it at CapCut.

• Best Quality: For the highest fidelity results, we recommend using a resolution of 1280x720.
• Known Issue (Pose Detection): You may encounter the error no pose detected in the reference video due to limitations in the current pose extractor. We are actively working on integrating a more robust solution. Pull Requests are highly welcome!

⚡️ Quickstart

Model Preparation

Models	Download Link	Notes
DreamID-V	🤗 Huggingface	Supports 480P & 720P
Wan-2.1	🤗 Huggingface	VAE & Text encoder

Installation

Install dependencies:

# Ensure torch >= 2.4.0
pip install -r requirements.txt

DreamID-V-Wan-1.3B

• Single-GPU inference

python generate_dreamidv.py \
    --size 832*480 \
    --ckpt_dir wan2.1-1.3B path \
    --dreamidv_ckpt dreamidv.pth path  \
    --sample_steps 50 \
    --base_seed 42

• Multi-GPU inference using FSDP + xDiT USP

pip install "xfuser>=0.4.1"
torchrun --nproc_per_node=2 generate_dreamidv.py \
    --size 832*480 \
    --ckpt_dir wan2.1-1.3B path \
    --dreamidv_ckpt dreamidv.pth path  \
    --sample_steps 50 \
    --dit_fsdp \
    --t5_fsdp \
    --ulysses_size 2 \
    --ring_size 1 \
    --base_seed 42

👍 Acknowledgements

Our work builds upon and is greatly inspired by several outstanding open-source projects, including Wan2.1, Phantom, OpenHumanVid, Follow-Your-Emoji. We sincerely thank the authors and contributors of these projects for generously sharing their excellent codes and ideas.

📧 Contact

If you have any comments or questions regarding this open-source project, please open a new issue or contact Xu Guo and Fulong Ye.

⭐ Citation

If you find our work helpful, please consider citing our paper and leaving valuable stars

@misc{guo2026dreamidvbridgingimagetovideogaphighfidelity,
      title={DreamID-V:Bridging the Image-to-Video Gap for High-Fidelity Face Swapping via Diffusion Transformer}, 
      author={Xu Guo and Fulong Ye and Xinghui Li and Pengqi Tu and Pengze Zhang and Qichao Sun and Songtao Zhao and Xiangwang Hou and Qian He},
      year={2026},
      eprint={2601.01425},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2601.01425}, 
}