💃 Wan-Dancer: Dance videos longer than a minute
The Tongyi Lab team (Alibaba Group) has introduced Wan-Dancer—a hierarchical framework for generating coordinated dance videos lasting more than one minute based on music and photos. The system utilizes a two-stage planning process to eliminate character identity loss and temporal drift.
🌍 A breakthrough in long-form video generation. Unlike models limited to 20 seconds, Wan-Dancer ensures rhythmic accuracy and temporal coherence over long durations.
👤 It is now possible to create full one-minute clips from a single photograph without the fear that the character will lose their form or rhythm halfway through the video.
Source 1: https://humanaigc.github.io/wan-dancer-project/ Source 2: https://github.com/Wan-Video/Wan-Dancer