CPU3DAI tools for 3D, video and audio

BlinkMimic: Video Generator — What It Does and What You Need to Run It

BlinkMimic is an open source video generator that the authors describe as a full-body motion tracking system for the Unitree G1 robot. The tool's job is to transfer reference animation onto a humanoid robot model, and the whole process, according to the developer, takes about three minutes on a single GPU.

What it does

The tool performs full-body motion tracking for the Unitree G1. It is built on the LAFAN1 dataset: the description mentions 40 clips from this set. The authors place the project in the field of humanoid robotics and world models, and the repository tags include humanoid-robotics, lafan1, mjlab, motion-tracking, mujoco-warp, supermimic, unitree-g1 and world-model.

The code is written in Python. The latest release is v1.0.0.

What you need to run it

The platform, as described by the author, is CUDA. The repository description states verbatim: "All runs on one RTX 4090 (driver 580, CUDA 12.8, torch 2.9.0), the GPU otherwise idle. 40 LAFAN1 clips; every". That is, it runs on a single RTX 4090 graphics card with driver 580, CUDA 12.8 and torch 2.9.0.

The code license is MIT. It permits use, modification and distribution of the code, including as part of other projects, provided the license text and attribution are retained.

The repository was created on November 2, 2026, and the last code change was on November 3, 2026.

Who it suits

The tool is aimed at those working with humanoid robotics and motion transfer tasks, in particular with the Unitree G1 robot. It will be useful to researchers and developers who need to quickly get full-body motion tracking on a single graphics card.

The project is unlikely to suit those looking for a universal video generator for arbitrary scenes or characters: its scope is limited to motion tracking for a specific robot and is tied to the LAFAN1 dataset.

BlinkMimic occupies a narrow niche at the intersection of motion generation and robotics. The open source code and MIT license make it possible to embed it in your own projects, and the requirements for running it are limited to a single RTX 4090 graphics card with the specified driver and library versions.

BlinkMimic — video generator pipeline Body motion tracking for Unitree G1 Input data LAFAN1 clips Capture data Processing World model MuJoCo-Warp Tracking Body motions Supermimic Output Unitree G1 motions Finished video Mode: CUDA Mode: Python Platform: CUDA RTX 4090 · driver 580 · CUDA 12.8 · torch 2.9.0 License: MIT · Open source · Repo: github.com/fuyichen0725/blinkmimic
How the BlinkMimic pipeline works. The diagram is drawn from the tool’s fact sheet.

Fact sheet

Code licenseMIT source
PlatformCUDA per the author’s description source
LanguagePython source
Last code change2026-11-03 source
Repository created2026-11-02 source
Changes often — as of 2026-11-20
Latest releasev1.0.0 source
Release date2026-11-03 source
GitHub stars316 source
Forks24 source

Values are collected automatically from official sources and were checked on 2026-11-20. Each one links to its source, and values taken from the developer’s pages also carry a verbatim quote — hover over the note. Pricing and versions are shown as of the check date and change most often; verify on the vendor’s site before buying.

See also