LiveAvatar is HeyGen's platform for low-latency talking avatars over WebRTC/LiveKit. FULL mode manages STT, LLM, TTS, and turn-taking end-to-end; LITE mode exposes a media-server websocket so an external stack (for example OpenAI GPT-Live-1) can feed audio while the browser watches lip-synced video. In September 2026 HeyGen open-sourced GPT-Live + HyperFrames demos that use LITE mode as the face layer.