SSyncDev

Wan-Dancer-14B

SyncDev model directory · Profile updated Nov 18, 2026

Compare with another model

Wan-Dancer-14B is Alibaba Wan's open music-to-dance video model, planning a whole track before refining it into minute-scale coherent dance.

Key specifications

Specification score
19.3
License
Apache-2.0

The specification score summarizes recorded features and limits. It is not a benchmark result.

Pricing conditions: No API pricing stated; open weights for self-hosting

Inputs and outputs

Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.

Accepts

  • Audio input
  • Image input
  • Text input

Produces

  • Video output

Capabilities

VisionAudioOpen weights

What is Wan-Dancer-14B?

Short dance clips are easy. A full song is hard, because most video diffusion models only see a few seconds at a time, so choreography loses the chorus, energy drifts and the dancer forgets the beat. Wan-Dancer-14B was built to generate long, rhythmic dance videos from music that hold together for minutes.

Alibaba's Wan team announced it on 13 July 2026 with weights, inference code and demos.

Plan first, then render

The trick is a two-level process.

  1. Global stage. The model listens to the full track and generates a keyframe video that plans the dance across the whole song: structure, sections, the big moves.
  2. Local stage. A second pass refines between keyframes, adding temporal detail and producing the final high-resolution video.

Because the global stage sees all of the music, repeated sections can get matching choreography and the climax can land where it should.

Inputs and controls

Input Role
Reference image The dancer's appearance
Prompt Scene and style description
Music track Drives rhythm and structure
Dance-genre settings Steer the style of movement
Seed Reproducibility

The second stage reuses the first stage's seed and image path, so a run stays consistent end to end.

Practical details

  • Hub repo: Wan-AI/Wan-Dancer-14B, also on ModelScope.
  • Size: 14B.
  • Integrations: ComfyUI, plus ModelScope Studio and Hugging Face Space demos.
  • Licence: Apache 2.0.

The Hub repository was created on 10 July, three days before the public launch.

Who it suits

Music labels and creators making visualisers, game studios prototyping dance animation, and anyone testing whether a generated performer can carry a full song. It is a specialist; for general prompt-to-video work, use the text-to-video directory.

How to test it

Pick three tracks with very different tempos and one with a clear repeated chorus. Generate full-length videos with one reference image. Check beat alignment at the start, middle and end, and whether the chorus choreography repeats sensibly. Watch the feet: ground contact is where long dance videos usually fail. Compare against other music- and audio-driven models in the audio-driven video directory and the Wan team's other releases on Wan-AI's Hugging Face listings.

How to evaluate Wan-Dancer-14B

Compare the model on the work you need to complete.

Tasks and inputs

Wan-Dancer-14B is listed for Image-to-video, Audio-Driven Video. It accepts audio, image, text and produces video. Use the publisher’s input format and preprocessing for your chosen deployment.

Capacity and cost

Choose duration, resolution and frame rate together. Compare the cost of an accepted clip, including repeat generations.

Choosing a deployment

The weights are downloadable under Apache-2.0. Check runtime support and size the complete model and working context for your expected concurrency.

Frequently asked questions

It first plans global keyframes using the full music track, then refines locally between them to produce the final high-resolution video.

These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.

kyutai

AURA-1.3B

AURA-1.3B is Kyutai's real-time streaming talking-head model: a portrait plus live speech becomes 576p, 16 fps video, chunk by chunk, on one GPU.

— context

kyutai

AURA-5B

AURA-5B is Kyutai's larger streaming talking-head model: portrait plus speech to 720p, 25 fps video in four steps per chunk, at about 9.7 fps on one H100.

— context

skywork

SkyReels-V3-A2V-19B

SkyReels-V3-A2V-19B is Skywork's open 720P talking-avatar model that animates a reference image from up to 200 seconds of driving audio.

— context

Meituan LongCat

LongCat Video Avatar 1.5

LongCat Video Avatar 1.5 generates audio-driven character video from speech, scene instructions and optional image references.

— context