Use MVNT inside ComfyUI

Install the comfyui-mvnt custom node package, connect your MVNT API key, and build an audio-to-dance workflow directly inside ComfyUI.

The comfyui-mvnt package adds MVNT motion generation as native nodes inside ComfyUI. Once installed, you can wire audio files through a chain of MVNT nodes to produce a 3D preview and a dance video — no separate API client or scripting required. The source repository is at github.com/mvnt-app/ComfyUI-MVNT.

Install the package

Choose whichever method fits your ComfyUI setup.

ComfyUI Manager

1. Open ComfyUI Manager from the ComfyUI interface.
2. Search for MVNT.
3. Select comfyui-mvnt from the results and click Install.
4. Restart ComfyUI when the install completes.

Comfy Registry

Run the following command from your terminal, then restart ComfyUI:

comfy node registry-install comfyui-mvnt

Manual

Clone the repository into your ComfyUI custom nodes directory, install the Python dependencies, then restart ComfyUI:

cd ComfyUI/custom_nodes
git clone https://github.com/mvnt-app/ComfyUI-MVNT.git comfyui-mvnt
cd comfyui-mvnt
python3 -m pip install -r requirements.txt

Set up your API key

You can provide your MVNT API key in two ways.

Environment variable — set this before launching ComfyUI:

export MVNT_API_KEY=mvnt_live_xxxxxxxxxxxx

Node field — paste your key directly into the api_key field on any MVNT node in the ComfyUI canvas.

You can create and manage API keys in MVNT Studio Platform.

Never export or commit a workflow JSON file that contains an API key. ComfyUI workflow exports can embed node field values, including the api_key field. If a key appears in a workflow file, logs, or Git history, revoke it immediately in MVNT Studio Platform and generate a new one. Use the environment variable approach to keep keys out of workflow files entirely.

Build a basic workflow

This section shows how to generate 3D dance motion from music using the ComfyUI toolbox. The workflow is a node-based pipeline where each chained node prepares, submits, and previews the audio-to-motion generation result.

1. Load Audio

   

   Add a Load Audio node and point it to your input audio file. Currently, only .wav files are supported. This node is a standard ComfyUI node that passes audio data downstream.

2. MVNT Audio Segment

   

   Connect the audio output to an MVNT Audio Segment node. This node trims or segments the audio to a range suitable for dance generation.

   • start_sec: The segment start time, in seconds.
   • duration_sec: The duration of the generated audio segment, in seconds. Defaults to 10 seconds, supports up to 40 seconds, and can be adjusted in 0.1-second increments.
3. MVNT Generate Dance

   

   Wire the segmented audio into an MVNT Generate Dance node. This node submits the audio to the MVNT API and returns the generation outputs when the job completes. This node uses the MVNT model to generate 3D motion from music. By default, generation uses the MVNT original character. To generate with a custom character, see the Character control section.

   1. Choose the desired motion style in the style field.
   2. Enter your issued API key in api_key.
   3. Optional advanced input: set a seed value. The default is -1.

   You can now preview or save the generated output.

4. MVNT Preview Dance 3D

   

   Connect the generation output to an MVNT Preview Dance 3D node to view the result in the ComfyUI canvas.

5. MVNT Render Dance Video

   

   Render the generated output as a 2D video. For 2D video control using the kling video_profile, see the Video control section.

   • video_profile: Specify the rendering settings to use. The default is Video Reference.
6. Save Video

   

   Use filename_prefix, format, and codec to configure the file name and rendering information. When generation completes, you will find two output files in your ComfyUI output directory:

   mvnt_<job_id>.motion.glb
   mvnt_<job_id>.dance.mp4
   

Character Control

Use character control when you want the generated dance to run on your own character instead of the MVNT original character. MVNT Generate Dance accepts an optional character_glb input. When a compatible character GLB is connected, MVNT generates the motion first, then retargets that motion to the supplied character. The dance_3d output becomes the animated custom-character GLB.

1. Load Image

   

   Add a Load Image node and select a clear, full-body character image. A front-facing image with the full body visible works best for creating a clean character reference.

2. MVNT Image-to-T-pose

   

   Connect the image to an MVNT Image to T-Pose node. This node converts the source character image into a front-facing T-pose image that can be used by a compatible character or rigging workflow.

3. Animate a custom character with Tripo

   Use Tripo’s model generation workflow to turn the T-pose character image into a rigged GLB that MVNT can animate.

   • Add Tripo’s Image to Model and Rig Model nodes.

Video Control

Use video control when you want the MVNT model to drive motion in a 2D image-to-video workflow. Generate dance motion with MVNT Generate Dance, use the kling video profile output as the motion reference, then follow Kling’s Motion Control workflow to animate your target image.

• reference_image: Upload the image you want to animate.
• reference_video: The kling video profile output from MVNT Generate Dance is used automatically.