UniMate for Agents

UniMate for Agents

Animates a rigged skeleton from a text sentence and returns an animated GLB, motion data and a preview, through a REST API, MCP tools and a Gradio page. It suits agent builders, game developers and animators.

Made by townbox on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: This Space is a text-to-animation service built for AI agents. You give it a rigged skeleton and a sentence, and it returns a 60-frame motion of 2 seconds at 30 fps. The result comes as an animated GLB, raw motion features, per-frame joint positions and a skeleton preview. It uses the UniMate model, runs on zero-a10g hardware at Hugging Face and is tagged MCP.

Who it suits: It suits developers who want an agent to animate a character or robot from a prompt, and animators who want quick motion drafts for different skeletons.

Industry | Who uses it | How they use it | Example
Game development | Game developers | Draft motions for a rigged creature | Ask for a quadruped to trot forward
Robotics and simulation | Robotics teams | Preview movement ideas for a robot skeleton | Animate the bundled robot skeleton from a sentence
Animation | Animators | Get starting motion, then edit it | Keep some joints and regenerate the rest under a new prompt
AI agents | Agent builders | Give an agent an animation tool over MCP | Let an agent call animate and fetch the GLB

You can skip it if your rig has more than 71 joints, since the source limits rigs to that size.

Where to use it: You can use it in these places:

• A browser: open the Space's page on Hugging Face for the landing page and Gradio interface.
• MCP clients such as Claude Code: connect over streamable HTTP to use the animation tools.
• Any REST client: the service has OpenAPI docs and an agent guide.
• Windows, macOS or Linux: run your own copy with Python 3.10.

Use case | Who | What they get back
Text to animation | Agents and developers | An animated GLB plus motion features and joint positions
In-betweening | Animators | A clip that keeps chosen frames and regenerates the rest
Motion expansion | Animators | One continuous motion chained from two to six prompts

How to set it up: You need these first:

• A Hugging Face account token, sent as a bearer header, because GPU time is billed to the caller
• Python 3.10 and pip, for a local copy
• Blender, if you want to prepare your own rig with the repository's script

To run a copy of the Space, follow the steps below.

1. Install the Hugging Face tools.
2. Download the Space's files into a folder.
3. Move into that folder.
4. Install Gradio at the version the Space uses, then the other packages.
5. Start the app.

pip install -U huggingface_hub
hf download townbox/unimate-for-agents --repo-type space --local-dir unimate-for-agents
cd unimate-for-agents
pip install gradio==6.30.0
pip install -r requirements.txt
python app.py

To add the MCP tools to Claude Code, the README gives a claude mcp add command that points at the Space's MCP address and passes your token in a header.

How to use it: Send a skeleton and a sentence, then fetch the files from the finished job. Set BASE to the address of the Space or your copy first. The calls below list the skeletons, ask for a trot and download the result.

curl -s -H "Authorization: Bearer $HF_TOKEN" $BASE/api/assets
curl -s -X POST $BASE/api/animate -H "Authorization: Bearer $HF_TOKEN" -H 'content-type: application/json' \
  -d '{"asset": "unitree_go2", "prompt": "An object trots forward.", "num_samples": 2, "seed": 7}'
curl -s -H "Authorization: Bearer $HF_TOKEN" -o trot.glb "$BASE/api/jobs/<job_id>/files/sample0_animation.glb"

Tool | What it does
animate | Animates a skeleton from a text prompt
inbetween | Keeps some frames of an earlier result and regenerates the rest
edit_joints | Keeps some joints over the whole clip and regenerates the others
expand | Chains two to six prompts into one continuous motion
list_assets | Lists the skeletons you can animate
get_result | Fetches the result of a job
view_animation | Lets an agent look at an animation
service_info | Returns information about the service
guide | Returns the guide for agents

Example: You send the bundled robot skeleton unitree_go2 with the prompt "An object trots forward." and ask for two samples. You get back animated GLB files, motion features, joint positions and a skeleton preview for each sample.

For newcomers and experts: If you are new to this, try one of the three bundled skeletons, the robot, the eagle or the shark, with a short prompt. Prompts work best in the training style, starting with the words An object and describing one motion.

If you know the field, the model is a flow-matching transformer that reads the structure of each skeleton, so one model animates different rigs from text without separate training. On the GPU, only the sampling loop for the flow equation runs, and the caller's token pays for it. The editing modes hold the kept frames or joints fixed at each sampling step and denoise the rest under the new prompt.

Send a skeleton and a sentence
The model samples motion with flow matching
GPU time is billed to your token
You get GLB, motion data and preview

Good to know: One call takes about 10 seconds for one sample on a warm worker, with about 7 seconds of ZeroGPU time. The bundled skeletons are a quadruped robot with 18 joints, an eagle with 51 and a shark with 16.

• License: cc-by-nc-4
• Cost to run: the caller's own Hugging Face ZeroGPU quota, because GPU time is billed to the caller and not to the Space
• Data it reaches: your token in the request header, your prompts, and any zipped rig you upload
• Not for: rigs with more than 71 joints

On Hugging Face, 13 November 2026

• Likes: 3
• Built with: gradio
• License: cc-by-nc-4.0

Links

• Space: https://huggingface.co/spaces/townbox/unimate-for-agents
• Files: https://huggingface.co/spaces/townbox/unimate-for-agents/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.