← All models

Wan-Dancer-14B

Published Publisher Alibaba

Overview

Wan-Dancer-14B is an open-weight music-to-dance video generation model from Alibaba's Tongyi Lab. It takes a reference image, music and a dance-style text prompt to generate a rhythmic dance video. (humanaigc.github.io)

Capabilities and uses

  • Uses full-track musical context in a two-stage workflow: global keyframe planning followed by local temporal refinement.
  • The authors report generating videos longer than one minute at 720p and 30 fps; these are reported experimental results, not a guarantee for every input.
  • The supplied examples cover Chinese classical, K-pop, street, tap and Latin dance. Potential uses include dance-video creation and music-driven character animation. (arxiv.org)

Access and limitations

The model weights and inference code are available for self-hosting under Apache 2.0. The documented workflow requires separate global and local generation stages, with the global video supplied to the refinement stage. Longer videos may require more inference steps and therefore additional computation. The model card does not specify a minimum GPU memory requirement or guarantee perfect identity consistency, motion quality or beat synchronization. Users are responsible for obtaining appropriate rights to reference images and music. (huggingface.co)

Release date

July 13, 2026, according to the official model announcement. The associated paper was first submitted on July 10, 2026. (huggingface.co)

Pricing

As of November 16, 2026, no official hosted API price or input/output billing rate for this exact model was verified in either Alibaba Cloud's mainland China pricing documentation or its international pricing documentation. Do not apply prices for other Wan models to this release. Self-hosting still incurs the user's infrastructure costs. (alibabacloud.com)

Official sources