andrisgauracs/litert.js-mocap — explained in plain English
Analysis updated 2026-05-18
Capture your own body motion through a webcam and drive a 3D character with it live.
Record a motion session and export it as BVH to animate a character in Blender.
Swap in a custom character model from Mixamo or Ready Player Me by mapping its bones.
Prototype browser based motion capture without any server or cloud dependency.
| andrisgauracs/litert.js-mocap | angelsswordstudios/angelssword-adventurers-creator | avacocloud/avaco-railway | |
|---|---|---|---|
| Stars | 37 | 37 | 37 |
| Language | JavaScript | JavaScript | JavaScript |
| Setup difficulty | easy | easy | moderate |
| Complexity | 3/5 | 2/5 | 3/5 |
| Audience | developer | general | ops devops |
Figures from each repo's GitHub metadata at analysis time.
Works best in a WebGPU capable browser like recent Chrome, falls back to a slower CPU mode otherwise.
LiteRT Mocap turns your webcam into a motion capture rig that runs entirely inside your web browser. It watches your body through the camera, figures out the position of 33 points on your body using Google's BlazePose AI model, and uses that to drive a rigged 3D character on screen in real time. Everything happens locally after you install it, there is no server involved and no data sent to the cloud, and it even works offline. To try it, you install the project's dependencies and run one command, which starts a local website you open in your browser. The app works best in Chrome or another browser that supports WebGPU for faster processing, and it automatically falls back to running on the CPU with a visible notice if WebGPU is not available. A heads up display shows the frame rate and how long each prediction takes. The pose detection model ships with the project already, so nothing extra needs to be downloaded to get started. Two ready made 3D characters are included as well, one playful robot and one standard humanoid figure, both taken from a well known open source three.js example collection. You can also bring your own character model, such as one exported from Mixamo or Ready Player Me, by dropping in its file and adding a small entry that maps its bone names to the tracked body parts. At the core of the project is a piece of math called retargeting, which converts the raw body point positions the AI model detects into rotations for each bone of the 3D character, working from the body's center outward so the character's pose stays consistent. Extra touches include a mirror mode so the character moves like your reflection, smart handling for when only your upper body is visible in frame, and the ability to record a session and export it either as a JSON file or as a BVH file that can be imported into Blender for further animation work. The pose model comes from Google's MediaPipe project under the Apache 2.0 license, the runtime is Google's LiteRT.js also under Apache 2.0, and the included 3D characters come from the three.js examples project under the MIT license.
A browser based motion capture app that uses your webcam and an AI pose model to animate a 3D character in real time, fully offline.
Mainly JavaScript. The stack also includes JavaScript, LiteRT.js, Three.js.
The pose model and runtime are Apache 2.0 licensed and the included character models are MIT licensed, both permissive and allowing commercial use.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.