Summary

Wan-Dancer is a hierarchical music-to-dance video generation framework from Tongyi Lab for creating long, coherent dance videos from music and reference imagery. It is designed to push beyond the short duration limit of many video diffusion systems by producing high-definition 720p, 30 FPS dance videos that can exceed one minute.

Key Features

Generates coherent music-to-dance videos at minute-scale durations.
Produces 720p, 30 FPS dance videos across multiple dance genres.
Uses hierarchical global keyframe planning followed by local temporal refinement.
Leverages full-track musical context for long-range rhythmic synchronization.
Applies time-mapped RoPE embeddings for precise music-to-motion alignment.
Uses optical-flow-based training signals to improve motion continuity.
Supports customization with LoRA for specific choreographic motions.
Demonstrates keyframe applications such as outfit changes and manual movement control.

About Wan-Dancer


The system decouples generation into global keyframe planning and local temporal refinement, using the full musical track to preserve long-range structure. It adds time-mapped RoPE embeddings for rhythm alignment, optical-flow-based losses for continuity, and motion-speed control to keep fast choreography detailed and temporally stable.


Wan-Dancer is useful for creators and researchers building music-driven character performance, dance video synthesis, and controllable long-form human video generation. The project demonstrates multiple dance genres, minute-scale outputs, customized choreography with LoRA, diversity controls, outfit changes, movement keyframes, and comparisons with prior state-of-the-art dance generation methods.

Get more likes & reach the top of search results by adding this button on your site!

Embed button preview - Light theme
Embed button preview - Dark theme
TurboType Banner

Subscribe to the AI Search Newsletter

Get top updates in AI to your inbox every weekend. It's free!