Wan-Animate-2 (and Lite)
Open character-animation model that takes the driving video directly, with text-controlled camera viewpoint and a Lite variant for real-time streaming.
Redesigned diffusion transformer removes the intermediate motion extractor used in Wan2.2-Animate, improving identity preservation and motion fidelity; text prompts can change the output camera angle. Base and distilled weights, Apache 2.0.
- Date
- Friday, 7 August 2026
- Lab
- Alibaba (Qwen)
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| Paper | arXiv 2608.06009 Revised v2 on 2026-08-08 | company |
README release notes date the inference scripts and weights 2026-08-07 (a first Hugging Face repo appeared 2026-07-14).
Sources
- github.com/Wan-Video/Wan-Animate-2
- arxiv.org/abs/2608.06009
- huggingface.co/Wan-AI/Wan2.2-Animate-2-14B
This record was checked against its sources on 6 October 2026. How we check