* Equal contribution
†Project lead
Guangzhou Quwan Network Technology
2025/10/15: 🔥🔥🔥 We released Playmate2, a novel DiT framework for generating long-duration high-quality audio-driven videos. Playmate2 also supports multi-character animation. To the best of our knowledge, Playmate2 is the first training-free approach capable of enabling audio-driven animation for three or more characters. Codes and models will release, please stay tuned!2025/05/07: 🎉🎉🎉 Super stoked to share that our paper has been accepted to ICML 2025!2025/04/28: ✨✨✨ Created a GitHub repository for the project.2025/02/11: 🚀🚀🚀 Our paper is in public on arxiv.
- Inference code for the first stage.
- Release pretrained models for the first stage.
- Inference code for the second stage.
- Release pretrained models for the second stage.
- Release the training code for the first stage.
- Release the training code for the second stage.
001.mp4 |
002.mp4 |
003.mp4 |
004.mp4 |
005.mp4 |
006.mp4 |
| Angry | Disgusted | Contempt | Fear | Happy | Sad | Surprised |
|---|---|---|---|---|---|---|
007.mp4 |
||||||
008.mp4 |
||||||
009.mp4 |
||||||
010.mp4 |
||||||
011.mp4 |
||||||
Explore more examples.
If you find our work useful for your research, please consider citing the paper:
@inproceedings{maplaymate,
title={Playmate: Flexible Control of Portrait Animation via 3D-Implicit Space Guided Diffusion},
author={Ma, Xingpei and Cai, Jiaran and Guan, Yuansheng and Huang, Shenneng and Zhang, Qiang and Zhang, Shunsi},
booktitle={Forty-second International Conference on Machine Learning}
}