ComfyUI Extension: ComfyUI-MVNT

Authored by mvnt-app

Created

Updated

2 stars

Run ComfyUI workflows without the setup

No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.

AI dance choreography from music. Generate full-body dance motion (BVH/FBX/JSON) from any audio track using MVNT's diffusion model trained with 100+ professional choreographers. Create rigged 3D characters from images, preview motion as stick-figure animations, and export photorealistic AI music videos — all inside ComfyUI. Nodes: MVNT Generate Dance, MVNT Generate Character, MVNT Export Video, MVNT Preview BVH, MVNT List Styles, MVNT Estimate Cost, MVNT Load Motion.

Looking for a different extension?

Custom Nodes (5)

README

ComfyUI-MVNT

Music in. Choreography out.

Generate 3D dance motion from any audio file using MVNT Dance Intelligence — directly inside ComfyUI.

Registry License: MIT

Demo

https://github.com/user-attachments/assets/2fcde3de-5bd5-4fe6-8931-72198dee9ace

What It Does

ComfyUI-MVNT connects ComfyUI to the MVNT Motion API, which generates music-conditioned choreography from audio input.

Drop in a track → choose a style → get an animated GLB and reference MP4 out. Works standalone or as part of larger character pipelines (Tripo, Kling, etc.).

Nodes

| Node | What it does | |---|---| | MVNT Audio Segment | Selects the audio segment to send. Duration capped at 40 seconds. | | MVNT Generate Dance | Sends audio to MVNT, polls generation, returns an animated GLB. Optionally retargets motion to a connected character GLB. | | MVNT Preview Dance 3D | Shows the animated GLB in ComfyUI's native 3D preview with companion audio controls. | | MVNT Image to T-Pose | Converts a character image into a front-facing T-pose via MVNT preprocessing — for use as a retargeting reference. | | MVNT Render Dance Video | Downloads a server-rendered MP4 from the generated dance. Two profiles: pretty (toon studio look) and kling (mannequin reference for motion transfer). |

Main Flow

Load Audio
  → MVNT Audio Segment
  → MVNT Generate Dance
  → MVNT Preview Dance 3D          ← fast 3D review
  → MVNT Render Dance Video        ← optional MP4

Optional character flow:

character image → MVNT Image to T-Pose
character GLB   → MVNT Generate Dance (retargeted output)

Example Workflows

Drag any .json from workflows/ into ComfyUI to load it.

| Workflow | Description | |---|---| | MVNT4.2 - music_to_3d_dance.json | Core flow: audio → animated GLB → 3D preview + optional MP4. | | mvnt_audio_to_dance.json | Minimal test: audio → dance output. | | mvnt_image_to_tpose.json | Character image → T-pose image. |

Installation

ComfyUI Manager (recommended)

Search MVNT in ComfyUI Manager → Install.

Comfy Registry

comfy node registry-install comfyui-mvnt

Manual

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

Restart ComfyUI after installation.

API Key

Get your API key at mvnt.studio.

Set it as an environment variable before launching ComfyUI:

export MVNT_API_KEY=mvnt_live_xxxxxxxxxxxxx

Or paste it directly into the api_key field on any node that exposes it.

Never commit API keys in workflow JSON. If a key is accidentally committed, revoke it immediately from your MVNT dashboard.

Quick Start

  1. Add an audio file to ComfyUI/input/
  2. Load workflows/MVNT4.2 - music_to_3d_dance.json
  3. In MVNT Audio Segment, set start time and duration (max 40s)
  4. In MVNT Generate Dance, choose a dance style
  5. Queue — outputs land in ComfyUI/output/:
mvnt_<id>.motion.glb
mvnt_<id>.dance.fetch.mp4   # only when MVNT Render Dance Video is connected

Character Retargeting

Connect a rigged GLB to character_glb on MVNT Generate Dance. MVNT generates the base motion, then retargets it to your character automatically. The dance_3d output becomes the retargeted animated GLB.

Works with Tripo-style GLB assets.

Advanced: Kling Motion Transfer

For AI video workflows, use video_profile: kling in MVNT Render Dance Video to get a deterministic mannequin-style MP4 suitable as a motion reference for Kling.

audio + character image
  → MVNT T-pose + Tripo model
  → MVNT Generate Dance
  → MVNT Render Dance Video (kling profile)
  → Kling motion control video

Troubleshooting

Old MVNT nodes showing up (MVNT Dance Generate, MVNT Dance (All-in-One)) Remove or rename the old custom node folder and restart ComfyUI.

Missing Tripo or Kling nodes The core audio-to-dance flow has no dependency on Tripo or Kling. Full character workflows may require those custom nodes separately.

3D preview looks different across ComfyUI versions MVNT Preview Dance 3D uses ComfyUI's native 3D preview. The exact viewer UI depends on your ComfyUI build version.

Links

License

MIT

Run ComfyUI workflows without the setup

No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.

Learn more