A browser driving game you control with your hands and voice, powered by models running fully local.
Steer by holding both hands up like a steering wheel. Speak commands to accelerate, brake, toggle headlights, and play music. No cloud calls, no server round-trips. Everything runs in your browser tab.
Two models run in parallel, entirely client-side:
- MediaPipe Hand Landmarker tracks your hand positions via webcam at ~30 fps. The angle between your two wrists drives the steering.
- LFM2.5-Audio-1.5B runs in a Web Worker with ONNX Runtime Web. It listens for speech via the Silero VAD and transcribes each utterance on-device. Matched keywords control game state.
The audio model loads from Hugging Face and is cached in IndexedDB after the first run, so subsequent starts are instant.
| Say | Effect |
|---|---|
speed / fast / go |
Accelerate to 120 km/h |
slow / stop / brake |
Decelerate to 0 km/h |
lights on |
Enable headlights |
lights off |
Disable headlights |
music / play |
Start the techno beat |
stop music / silence |
Stop the beat |
- Chrome 113+ or Edge 113+ (WebGPU required for fast audio inference; falls back to WASM)
- Webcam and microphone access
- Node.js 18+
npm install
npm run devThen open http://localhost:3001.
On first load the audio model (~900 MB at Q4 quantization) downloads from Hugging Face and is cached in your browser. Hand detection assets load from CDN and MediaPipe's model storage.
Browser tab
├── main thread
│ ├── MediaPipe HandLandmarker (webcam → hand angles → steering)
│ ├── Canvas 2D renderer (road, scenery, dashboard, HUD)
│ └── Web Audio API (procedural techno synthesizer)
└── audio-worker.js (Web Worker)
├── Silero VAD (mic → speech segments)
└── LFM2.5-Audio-1.5B ONNX (speech segment → transcript → keyword)
The game loop runs on requestAnimationFrame. Hand detection is throttled to ~30 fps so it does not block rendering. Voice processing happens off the main thread and delivers results via postMessage.
Join the Liquid AI Discord Community and ask.