Real-time lip-synced avatars. Audio in, animated face out at 25 FPS. Runs on CPU — from Raspberry Pi to H100.
Live SDK downloads, version badges, and platform matrix. For guides see docs.bithuman.ai · for release notes see docs.bithuman.ai/changelog · for source see github.com/bithuman-product.
Quickstart guide · Pricing & plans · Device matrix · API reference
bitHuman runs entirely on-device or on your own infrastructure. No video data leaves your servers. GDPR-compliant by architecture — not by policy. One CPU core renders 25 FPS lip-synced video for under $0.01/min. Cloud alternatives charge $0.10–$0.50/min for the same output.
| Device | Essence | Expression | SDK / Integration |
|---|---|---|---|
| MacBook (Apple Silicon) | ✓ | ✓ | Python, Swift, CLI, Flutter |
| MacBook (Intel) | ✓ pin 1.11.3 | — | Python, REST API |
| Windows PC | ✓ pin 1.11.3 | — | Python, REST API, Java |
| Linux (x86_64 / ARM64) | ✓ | ✓ (NVIDIA) | Python, CLI, Java, Docker |
| iPhone / iPad | ✓ | ✓ | Swift SDK, Flutter |
| Android | ✓ | — | Flutter, REST API |
| Raspberry Pi / Jetson | ✓ | — | Python, CLI |
| Cloud GPU | ✓ | ✓ | Python, Docker |
| Browser | ✓ ONNX WASM | — | browser rendering + REST API |
The avatar renders entirely in your browser via ONNX Runtime Web (WebAssembly). The server runs only the brain — no server-side video, no GPU bill. Loads in ~10 seconds on first visit; the model is cached after.
Open showcase agent →How rendering_mode=browser and rendering_mode=avatar work, when to reach for them, latency budget, browser requirements (COOP+COEP, SharedArrayBuffer), and the three-mode comparison.