ComfyUI-MVNT

Music in, choreography out. Generate full-body dance sequences from audio using MVNT AI — the world's first AI system purpose-built for dance — directly inside ComfyUI.

MVNT's diffusion-based model is trained alongside 100+ professional choreographers with studio-captured motion data. It outputs production-ready .BVH / .FBX / .JSON at 60 fps with music-synced, style-aware choreography and physics-aware refinement.

Demo

MVNT + Kling AI — Motion Transfer

https://github.com/user-attachments/assets/mvnt_kling_usecase.mp4

Audio → AI dance choreography → photorealistic video via Kling motion transfer

https://github.com/mvnt-app/ComfyUI-MVNT/releases/download/v1.1.0/mvnt_kling_usecase.mp4

MVNT Studio — AI Dance Generation

https://github.com/mvnt-app/ComfyUI-MVNT/releases/download/v1.1.0/mS-demo-comp_480.mov

K-pop quality dance from any music track in ~15 seconds

Nodes

Node	Description
MVNT Generate Dance	Audio → full-body dance motion (BVH / FBX / JSON)
MVNT Generate Character	Image or text prompt → rigged 3D GLB character
MVNT Export Video	Character image + motion reference → photorealistic AI video
MVNT Preview BVH	Render stick-figure animation from BVH → IMAGE frames
MVNT List Styles	Fetch available dance styles
MVNT Estimate Cost	Estimate credits and generation time before running
MVNT Load Motion	Load a downloaded motion file for downstream nodes

Example Workflows

Ready-to-use workflows are in the workflows/ folder. Drag any .json file into ComfyUI to load it.

Workflow	Description
Audio to Dance	Basic: load audio → generate dance → view BVH data
Dance with Preview	Generate dance + render stick-figure preview as IMAGE frames
Full Music Video	End-to-end: audio → dance → 3D character → AI video export
Compare Styles	Side-by-side Male vs Female dance from the same audio

Installation

Option A: ComfyUI Manager (recommended)

Search for MVNT in the ComfyUI Manager and click Install.

Option B: Comfy Registry

comfy node registry-install comfyui-mvnt

Option C: Manual

cd ComfyUI/custom_nodes
git clone https://github.com/mvnt-app/ComfyUI-MVNT.git
cd ComfyUI-MVNT
pip install -r requirements.txt

Restart ComfyUI.

API Key Setup

1. Get API access at mvnt.studio
2. Set the environment variable before launching ComfyUI:

export MVNT_API_KEY=mk_live_xxxxxxxxxxxxx

Or pass the key directly to each node via the api_key input.

Quick Start

Simple: Audio → Dance

1. Add Load Audio → MVNT Generate Dance → MVNT Load Motion
2. Pick a style (All / Male / Female)
3. Queue — the node submits audio to the MVNT API, polls for completion, and saves the motion file to output/

With Preview

Chain MVNT Load Motion → MVNT Preview BVH → PreviewImage to see a stick-figure animation of the choreography directly in ComfyUI.

Full Music Video Pipeline

1. Load Audio → MVNT Generate Dance (produces BVH motion)
2. Load Image → MVNT Generate Character (creates rigged 3D character from any photo)
3. MVNT Export Video (combines character + motion into a photorealistic AI video)

API

The nodes wrap the MVNT Motion API — a REST API with webhook support and async job processing.

Endpoint	Description
POST /v1/generations	Submit audio for dance generation
GET /v1/generations/:id	Poll generation status
GET /v1/generations/:id/output	Download BVH / FBX / JSON output
POST /v1/characters	Generate rigged 3D character from image or text
POST /v1/videos	Motion-transfer video export
GET /v1/styles	List available dance styles
POST /v1/estimate	Estimate cost and generation time

Environments:

• Production: https://api.mvnt.world/v1
• Sandbox: https://api-sandbox.mvnt.world/v1

For integration support, reach out at hello@mvnt.world.

Links

• MVNT — Product & research
• MVNT Studio — Creative suite for AI dance
• MVNT API Docs — API reference
• Unreal Dance Editor — Dance generation native in Unreal Engine

License

MIT