Alibaba's Wan2.2 Adds Control to Open Video

Alibaba's Platform for AI (PAI) team has released a new open-source model for video generation, Wan2.2-VACE-Fun-A14B. This 14-billion-parameter model isn't just another text-to-video generator; its primary focus is on providing creators with a high degree of control over the final output.

The key differentiator for this Wan2.2 variant is its ability to condition video generation on more than just a text prompt. Users can provide structural inputs like depth maps, Canny edge outlines, sketches, and human pose skeletons to guide the creative process. This allows for much more precise control over scene composition and character movement than is possible with text alone.

Under the Hood

The model is part of the "Video-Audio-Caption-Editing" (VACE) project and uses a parameter-efficient method called "Fun-tuning" to adapt the base model for these specialized control tasks. According to the team, this approach makes it more efficient to train and adapt the model for specific creative needs.

Released under the permissive Apache 2.0 license, Wan2.2-VACE-Fun-A14B joins a growing field of open models that are making advanced video synthesis more accessible. By focusing on controllability, Alibaba is providing a valuable tool for developers and artists who need to move beyond simple prompts and direct their creative process with greater precision.