Wan-Dancer-14B turns still images into dance videos

Alibaba's Wan team has published Wan-Dancer-14B, a 14-billion-parameter image-to-video model aimed squarely at a specific creative task: generating dance sequences driven by music. Rather than producing generic video clips, the model is tuned to animate a source image into choreographed movement, a niche that combines motion synthesis with the tricky problem of keeping a subject coherent across frames.

The release lands under the permissive Apache 2.0 license, which matters for anyone hoping to build on it. That licensing choice means developers and studios can experiment, fine-tune, and deploy the model commercially without the usage restrictions that accompany many video generation systems.

Why it matters

Music-driven dance generation has been one of the harder frontiers in video synthesis, requiring models to sync motion to audio timing while preserving the identity and appearance of the input subject. A dedicated open-weights model in this space gives researchers and creators a reproducible starting point instead of relying on closed APIs.

• Image-to-video generation from a single source image
• Focused on music-driven dance and choreography
• 14B parameters, released under Apache 2.0

Wan-Dancer-14B extends the Wan family's push into open video models. As always with a fresh release, the practical questions — output length, resolution, and how well motion holds up over time — will be answered as the community puts the weights, available now on Hugging Face, through its paces.