Theme
AI Visuals
Wan-Dancer-14B
Wan-Dancer-14B is a music-to-dance video model and framework from Tongyi Lab, Alibaba Group.
The released workflow takes a reference image, a music file, and a dance-style prompt through two stages: global keyframe planning followed by local refinement. The official model page and repository include weights, inference code, setup instructions, examples across five dance styles, and the linked research paper. Use this as a first read, not a recommendation. Open the original project before trusting details like terms, limits, privacy, cost, setup, or safety.
What it is
Music-to-dance video generation
Wan-Dancer uses a reference image, full music track, and style prompt to generate dance video rather than treating movement as a short, isolated clip.
Why it stands out
Long-form planning before refinement
Its first stage plans a lower-detail keyframe video across the track. A second stage uses that global result to generate the final higher-resolution video with local temporal detail.
Availability
Public weights and inference code
Readers can inspect the 14B model files, GitHub code, installation steps, generation scripts, example inputs, and paper. Hugging Face currently lists no hosted Inference Provider for the model.
Why it matters
What makes it useful
Long dance videos can lose rhythm, identity, or motion variety as they extend beyond a short clip. Wan-Dancer plans movement against the full track first, then refines shorter temporal regions without discarding that global structure.
What to know
Where it fits
Open it when comparing generative-video systems for choreography, dance visualization, music-led character motion, or longer image-conditioned human video. It is more specialized than a general prompt-to-video model or broad talking-avatar tool.
Notable points
What stands out
The project reports minute-scale 720p video at 30 frames per second and improved long-range stability across five dance genres. These results come from the project paper and examples.
Before using
What to review
The official test setup uses Ubuntu 22.04, Python 3.10.14, and eight NVIDIA A800 80GB GPUs, alongside specific CUDA, PyTorch, Diffusers, FlashAttention, xFuser, and Transformers versions.
The local stage depends on the global-stage video and its own prompt file, so plan for two generation runs and verify the paths and settings for each script.
Check whether you have permission to use the reference image and music in the way you plan to share the result.
Hosted demos, model-provider availability, and setup instructions can change; check the official pages before planning around them.
Reader fit
Who may find it relevant
Creators and researchers exploring music-driven choreography or longer dance-video generation.
Technical teams able to inspect or run a demanding two-stage 14B video workflow.
Less relevant for readers seeking a lightweight local app, a simple consumer editor, or general avatar speech generation.
Editorial note
Why LifeHubber lists it
Wan-Dancer turns a full song into part of the generation plan, giving readers a concrete way to examine how rhythm and choreography are carried beyond a short clip.
Source links
Source materials
Reader note
Before relying on this entry
LifeHubber lists entries to help readers inspect AI projects, not to endorse them or prove they are safe, suitable, accurate, maintained, or right for a specific use. We do not verify every entry in depth. Before relying on anything listed, review the original materials, terms, privacy practices, limits, and risks that matter for your situation.
What to explore next
Compare dance generation with talking-avatar video.
Wan-Dancer follows music into choreography. LongCat Video Avatar shifts the question toward lip sync, single- and multi-person scenes, and video continuation.
More in Music / Image Gen Models
Keep browsing this category
Explore more media-generation model resources.
ACE-Step 1.5
ace-step/ACE-Step-1.5
A locally runnable music model with 2B and XL 4B variants for full-song generation, reference-guided creation, repainting, accompaniment, stem separation, and lightweight style training.
Boogu Image
boogu-project/Boogu-Image
A 10B research model family for text-to-image generation and instruction-based image editing, with Base, Turbo, Edit, and Edit-Turbo checkpoints, bilingual Chinese-English text rendering, local inference code, and public demos.
Fooocus
lllyasviel/Fooocus
A local image-generation interface built around prompt-focused SDXL workflows, with Windows downloads, Colab access, inpainting, outpainting, image prompts, and presets.
For project maintainers
Listed here? You can use the badge.
If you maintain a project with a current LifeHubber listing, you may add the optional “Listed on LifeHubber AI Resources” badge to its README, docs, or website. No introduction or permission request is needed.
