Origin
Alibaba’s video generation models go by the name Wan. The weights are published under the Wan-AI organisation on Hugging Face and the Wan-Video organisation on GitHub. Wikipedia’s Alibaba article attributes the Wan 2.1 first-and-last-frame model to Alibaba Cloud and says it was released in April 2025 under the Apache licence. Alibaba Group itself was founded on 28 June 1999 in Hangzhou.
The Wan technical report went up on arXiv on 26 March 2025 under the title “Wan: Open and Advanced Large-Scale Video Generative Models”. It describes Wan as an open suite of video foundation models built on the diffusion transformer approach.
How it got here
The Wan2.1 repository says the inference code and weights of Wan2.1 were released on 25 February 2025. That first release had text-to-video models at 1.3B and 14B parameters, and 14B image-to-video models at 480p and 720p. The Diffusers library integrated them in early March 2025. A first-and-last-frame model came on 17 April 2025, and the VACE video creation and editing models on 14 May 2025.
Wan2.2 came on 28 July 2025, according to its model card. It included a 5B model that handles both text and image input, and two A14B models, for text-to-video and image-to-video, with a mixture-of-experts design.
The most recent release we track is Wan-Dancer, a 14B model that generates long, rhythmic dance videos from music. Its model card dates it to 13 July 2026.
Licensing pattern
Every Wan release we track uses Apache 2.0, with no change of licence across Wan2.1, Wan2.2 and Wan-Dancer. The Wan2.1 and Wan2.2 model cards both say the models in the repository are licensed under Apache 2.0, and the Wan-Dancer card says the same for that project.
The cards add a note on outputs. The team claims no rights over the content you generate, and you are accountable for your use of the models, which must not involve content that violates applicable laws, causes harm to individuals or groups, or spreads misinformation.