Alibaba's Tongyi open-source music-driven portrait dance video generation model, Wan-Dancer

Alibaba's Tongyi Wanxiang open-source music-driven portrait dance video generation model, Wan-Dancer, allows users to upload photos of themselves and music to generate dance videos with precise beats and smooth movements. The model overcomes minute-level timing bottlenecks, supporting continuous 720p HD output from 15 seconds to 3 minutes, covering various dance styles. Wan-Dancer employs a hierarchical architecture of global keyframe planning and local timing refinement, effectively solving the problem of motion drift in long sequences.