88Z AI, 15 July — Alibaba Tongyi Wanxiang has released the open-source model Wan-Dancer, which generates dance videos from uploaded portrait photos and music. The model features precise beat alignment and smooth movements, overcoming long-video generation bottlenecks by supporting 720p HD output for durations ranging from 15 seconds to 3 minutes, while using a hierarchical architecture to address motion drift.
At a glance
- Supports 720p HD dance video generation from 15 seconds to 3 minutes
- Uses hierarchical architecture to solve motion drift in long sequences
The update
Alibaba Tongyi Wanxiang has officially open-sourced Wan-Dancer, a music-driven model for generating portrait dance videos. By simply providing a portrait photo and a music track, users can create dance videos with precise beat synchronization and natural, fluid body movements. This technology breaks through the performance bottlenecks traditional models face when generating long-sequence videos, capable of stably producing coherent 720p HD footage ranging from 15 seconds to 3 minutes in duration, covering various dance styles. Technically, Wan-Dancer employs a hierarchical architecture that combines global keyframe planning with local temporal refinement. This structure effectively addresses the common issue of motion drift in long-sequence video generation, ensuring action consistency and visual stability for characters throughout extended dance routines, thereby providing creators with a high-quality video generation tool.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- Wan-DancerSource published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.