Jetha Chan / ComfyUI

02 / 3D & SKELETAL ANIMATION

A character.
A motion in mind.

UniMate brings prompt-driven skeletal animation to ComfyUI. Start with a rigged GLB, generate a motion clip, and export an animated GLB on the original asset.

UniMate on GitHub Get started
FROM A PROMPT TO A PERFORMANCE.Pipeline
RIG / MOTIONGLB → GLB
INPUT / RIGGED GLB
Motion prompt
OUTPUT / ANIMATED GLB

One skeleton. A new motion clip.

Skeletal animation workflow60 frames · 30 fps

THE PIPELINE

Five nodes, from rig to result.

  1. 01

    Load Rigged GLB

    Read a self-contained, skinned asset from ComfyUI input.

  2. 02

    Prepare UniMate Rig

    Build topology conditioning from the bind pose.

  3. 03

    Load UniMate Model

    Load your installed .unimate model bundle.

  4. 04

    Generate Motion

    Sample 60 frames at 30 fps from a motion prompt.

  5. 05

    Export UniMate GLB

    Save the generated clip with provenance metadata.

Connect the model loader to Generate Motion; the rig passes from Load to Prepare to Generate to Export. Example API workflow

THE ASSET CONTRACT

Keep the character.
Generate the performance.

Your original asset, animated

Mesh data, skin weights, inverse binds, materials, and textures stay in the source asset. Existing source clips are replaced by one generated clip.

The input must already be rigged. Automatic rigging and retargeting are outside this pack; clips are not automatically looped.

A defined input format

One skin and one connected skeleton with 5–70 joints. Triangle geometry, up to four skin influences, embedded PNG/JPEG textures, and positive uniform scales.

No FBX, morph targets, or sparse/compressed geometry. The GLB asset limit is 256 MiB.

Full supported-asset contract

GET STARTED

Install the nodes.
Bring the model.

Python 3.11+, ComfyUI’s current extension API, and Blender are required. The documented tested Blender version is 5.1.1.

1. Install in ComfyUI’s environment

cd ComfyUI/custom_nodes
git clone https://github.com/jethac/ComfyUI-UniMate.git comfy-unimate
cd comfy-unimate
python -m pip install -r requirements.txt

2. Configure Blender and the model bundle

Set UNIMATE_BLENDER to your Blender executable before starting ComfyUI. Download the pinned UniMate and FLAN-T5 files, build the bundle, and put it in ComfyUI/models/unimate/.

The pack does not install Blender or download models during execution. Once prepared, the bundle supports offline inference.

Model downloads and bundle instructions

3. Connect the workflow

Put a supported rigged GLB in ComfyUI/input/. Connect the rig and model nodes, choose the source facing direction, and enter a motion prompt.

Independent ComfyUI integration of UniMate . Node pack: MIT. UniMate weights: MIT. FLAN-T5: Apache-2.0. Imported assets retain their source licenses.