Alibaba's Tongyi Wanxiang team released the Wan video generation model series.

Wan2.2 is an open-weight diffusion transformer model family that generates text-to-video and image-to-video clips, including a mixture-of-experts variant for efficiency.

Hosted on GitHub at Wan-Video/Wan2.2 and on Hugging Face and ModelScope under Wan-AI.

Released in July 2025, following the initial Wan2.1 release in February 2025.

It provides one of the strongest openly licensed video generation models, competitive with closed commercial systems.

Built on PyTorch, runnable via Diffusers or the official Wan inference repo, and integrated into ComfyUI.

Released under the Apache 2.0 license by Alibaba; fully open weights with no gating.