Wan-Animate-2: End-to-End Character Animation in ComfyUI
ComfyUI Wiki
Wan-Animate-2 animates a character from a reference image and a driving video with a dual-branch DiT, native ComfyUI nodes and text-driven viewpoint control.
W
Wan-Animate-2
VideoCharacter AnimationOpen SourceDiTEnd-to-end character animation model from Alibaba Tongyi Lab. Feeds a driving video directly into a redesigned dual-branch Diffusion Transformer — no intermediate skeleton or expression extractors — with text-driven viewpoint control, multi-character motion driving, and a real-time Lite variant. Apache 2.0 licensed.
| Developer | Alibaba Tongyi Lab (Wan AI) |
| Release Date | 2026-08-07 |
| Architecture | Dual-branch Diffusion Transformer — Time-Align RoPE + Sparse-Ref Attention |
| License | Apache-2.0 |
| Model Size | 14B (Base and Distillation variants) |
| Capabilities | Reference-image + driving-video character animation, text-driven viewpoint control, single-to-multiple and multiple-to-multiple motion driving, real-time Lite streaming |
| Sampling | Base: 40 steps; Distillation: 10 steps, no CFG (Euler scheduler) |
| ComfyUI Support | Native — WanAnimate2ToVideo and WanAnimate2Cache nodes (experimental) |
Guides and workflows related to this model series.
No articles found.
Comments
Sign in with GitHub to join the discussion.