> ## Documentation Index
> Fetch the complete documentation index at: https://docs.comfy.org/llms.txt
> Use this file to discover all available pages before exploring further.
# Wan Dancer: Music to Dance Video Generation
> Generate minute-scale coherent dance videos from music using Wan Dancer, a hierarchical audio-driven framework built on Wan 2.2. Input a reference image and audio to produce synchronized dance videos.
Wan Dancer is an audio-driven dance video generation model built on the Wan 2.2 architecture. It uses a hierarchical framework with global and local expert models to generate coherent, expressive dance videos synchronized to input music.
Given a reference image of a character and an audio track, Wan Dancer produces a dance video where the character performs movements that match the rhythm and style of the music. The global model captures overall dance structure and body coordination, while the local model refines fine-grained motion details for natural, high-quality output.
## Model Highlights
* **Audio-driven dance generation**: Generates dance movements directly from music without manual keyframing or motion capture
* **Hierarchical framework**: Two-stage architecture (global + local) for coherent body motion and detailed movement refinement
* **Music-synchronized output**: Dance motions are naturally aligned to the input audio's rhythm and tempo
* **Style control**: Configurable dance style and motion amplitude for creative direction
* **Minute-scale generation**: Produces long-form, coherent dance videos with temporal consistency across the full duration
## Workflow Overview
The Wan Dancer workflow takes two inputs: a reference image of the character and an audio file. It generates two video outputs: a global output preview showing the overall dance structure, and a final refined output with enhanced motion details.
Make sure your ComfyUI is updated.
* [Download ComfyUI](https://www.comfy.org/download)
* [Update Guide](/installation/update_comfyui)
Workflows in this guide can be found in the [Workflow Templates](/interface/features/template).
If you can't find them in the template, your ComfyUI may be outdated. (Desktop version's update will delay sometime)
If nodes are missing when loading a workflow, possible reasons:
1. You are not using the latest ComfyUI version (Nightly version)
2. Some nodes failed to import at startup
* The Desktop is base on ComfyUI stable release, it will auto-update when there is a new Desktop stable release available.
* [Cloud](https://cloud.comfy.org) will update after ComfyUI stable release.
So, if you find any core node missing in this document, it might be because the new core nodes have not yet been released in the latest stable version. Please wait for the next stable release.
## Wan Dancer Workflow
### 1. Download Workflow File
Update your ComfyUI to the latest version, then download and drag the workflow file into ComfyUI, or find "Wan Dancer" in the Template Library under `Workflow` → `Browse Templates` → `Video`.
Open in Comfy Cloud
Download JSON or search "Wan Dancer" in Template Library
### 2. Download Input Materials
Input character image for dance generation. Download and use this image, or replace with your own.
Reference audio track for dance synchronization. Download and use this audio, or replace with your own music.
### 3. Manually Download Models
**Diffusion Models**
* [wan2.2\_dancer\_14b\_global\_fp8\_scaled.safetensors](https://huggingface.co/Comfy-Org/Wan-Dancer/resolve/main/diffusion_models/wan2.2_dancer_14b_global_fp8_scaled.safetensors)
* [wan2.2\_dancer\_14b\_local\_fp8\_scaled.safetensors](https://huggingface.co/Comfy-Org/Wan-Dancer/resolve/main/diffusion_models/wan2.2_dancer_14b_local_fp8_scaled.safetensors)
**LoRA**
* [lightx2v\_I2V\_14B\_480p\_cfg\_step\_distill\_rank64\_bf16.safetensors](https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/Lightx2v/lightx2v_I2V_14B_480p_cfg_step_distill_rank64_bf16.safetensors)
**Text Encoder**
* [umt5\_xxl\_fp16.safetensors](https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/text_encoders/umt5_xxl_fp16.safetensors)
**CLIP Vision**
* [clip\_vision\_h.safetensors](https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/clip_vision/clip_vision_h.safetensors)
**VAE**
* [Wan2\_1\_VAE\_bf16.safetensors](https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/Wan2_1_VAE_bf16.safetensors)
```
ComfyUI/
├───📂 models/
│ ├───📂 diffusion_models/
│ │ ├─── wan2.2_dancer_14b_global_fp8_scaled.safetensors
│ │ └─── wan2.2_dancer_14b_local_fp8_scaled.safetensors
│ ├───📂 loras/
│ │ └─── lightx2v_I2V_14B_480p_cfg_step_distill_rank64_bf16.safetensors
│ ├───📂 text_encoders/
│ │ └─── umt5_xxl_fp16.safetensors
│ ├───📂 clip_vision/
│ │ └─── clip_vision_h.safetensors
│ └───📂 vae/
│ └── Wan2_1_VAE_bf16.safetensors
```
### 4. Workflow Instructions
Wan Dancer samples 149 frames at once and requires significant VRAM. The default workflow outputs a 5-second video. Make sure you have a GPU with sufficient VRAM before running this workflow.
1. **Load the reference image**: Use the `Load Image` node to upload your character image. The default input uses `blue_dancer.png`.
2. **Load the audio file**: Use the `Load Audio` node to upload your music or audio track. The default input uses `wan_dancer_reference_audio.mp3`.
3. **Configure dance parameters**:
* `dance_style`: Select the dance style for the output video
* `motion_amplitude`: Control the intensity of dance movements
* `audio_duration`: Set to match the duration of your input audio
* `final_duration`: Choose the final output video duration
4. **Configure output dimensions**:
* `width` and `height`: Set the resolution of the output video. Choose values appropriate for your GPU's available VRAM.
5. **Verify model paths**: Make sure the model selector widgets point to the correct model files:
* `global_model`: Should load `wan2.2_dancer_14b_global_fp8_scaled.safetensors`
* `local_model`: Should load `wan2.2_dancer_14b_local_fp8_scaled.safetensors`
* `lightning_lora`: Should load `lightx2v_I2V_14B_480p_cfg_step_distill_rank64_bf16.safetensors`
* `clip_vision`: Should load `clip_vision_h.safetensors`
6. **Run generation**: Click the `Run` button or use the shortcut `Ctrl(Cmd) + Enter` to execute video generation.
### 5. Output
The workflow produces two video outputs:
* **Global output preview**: Shows the overall dance structure generated by the global model
* **Final output**: The refined dance video after local model enhancement
Both outputs are saved automatically via the `SaveVideo` nodes.
## Model Information
Wan Dancer is an open source model released under the Apache 2.0 license. The model weights are available on Hugging Face:
* [Comfy-Org/Wan-Dancer](https://huggingface.co/Comfy-Org/Wan-Dancer): Global and local diffusion models
## Report Issues
If you encounter runtime errors or unexpected behavior:
* Ensure ComfyUI is updated to the latest version
* Check that all required models are downloaded and placed in the correct directories
* For Desktop/Cloud users: updates follow stable releases, so nightly-supported models may not be available immediately
Report issues to: [ComfyUI/issues](https://github.com/Comfy-Org/ComfyUI/issues)