BUILT ON WAN-DANCER-14B · OPEN SOURCE · APACHE-2.0
One photo.
One song.
A full dance video.
Wan Dancer turns a single picture of anyone into a minute-long dance video that follows your music — K-pop, street, latin and more. Powered by the open-source Wan-Dancer-14B model from Alibaba's Tongyi Lab.
01:04 and the dancer is still the same person, still on beat — most video models drift before 00:20.
00:00 / THE INPUT
Three things in, one video out
01
Upload a photo
One vertical, full-body shot of the dancer — you, a friend, a character. No rigging, no extra angles.
02
Add your music
Any track. The model reads the whole song first and plans the choreography around its structure and beat.
03
Pick a style
K-pop, Chinese Classical, Street, Latin or Tap — one short prompt sets the vibe.
00:12 / FIVE STYLES
Choreography that matches the genre
Each style drives different footwork, arm lines and energy. The same photo and song rendered as K-pop and as Latin are two genuinely different routines — not one animation with a filter.
00:31 / WHY IT HOLDS UP
Most video models fall apart at 00:20. This one plans ahead.
Plans first, renders second
A global stage reads the entire track and lays the routine out as keyframes; a local stage then fills in the motion between them, frame by frame. Long-range structure comes from the plan, detail comes from the refinement.
Same dancer at 01:04
Because keyframes anchor identity and pose across the whole song, the face, outfit and body stay consistent past the one-minute mark — where autoregressive models typically drift and morph.
720p / 30fps output
Vertical-friendly resolution that's ready for TikTok, Reels and Shorts without upscaling gymnastics.
Actually open source
Weights, inference code, ComfyUI integration and LoRA fine-tuning are all public under Apache-2.0. Run it on your own GPU today.
00:47 / GET ACCESS
Online generation is almost ready
No hosted API for Wan-Dancer-14B exists yet — we're building the first browser version. Leave your email and we'll tell you the day it opens. One email, no newsletter.
01:04 / FAQ
Questions people ask
What is Wan Dancer?
Wan Dancer is an AI dance video generator built on Wan-Dancer-14B, an open-source music-to-dance model released by Alibaba's Tongyi Lab in July 2026. You provide one photo of a person and a music track; the model choreographs and renders that person dancing in sync with the music at 720p, 30fps.
Is Wan Dancer free?
The underlying Wan-Dancer-14B model is open source under Apache-2.0, so you can run it yourself for free on your own GPU (see our ComfyUI guide). Our online generator is in preparation — join the waitlist and we'll email you when it opens, including the free tier details.
How long can the dance videos be?
Over a minute while staying coherent. Wan-Dancer plans the whole routine from the full music track first (global keyframes), then refines motion between keyframes frame by frame — which is why it doesn't drift or morph the way most video models do after about 20 seconds.
Which dance styles does it support?
Five styles at release: K-pop, Chinese Classical, Street, Latin, and Tap. You pick the style with a short text prompt alongside your photo and music.
What inputs do I need?
A single reference photo (a vertical, full-body shot works best), an audio file for the music, and a one-line style prompt. No rigging, no motion capture, no video reference.
Can I use the videos commercially?
The model is Apache-2.0 licensed, which permits commercial use of the outputs you generate yourself. You are responsible for having rights to the photo and the music you use — a song's copyright is separate from the model license.