Talking Avatar gives AI voice agents a face. It takes the audio an agent produces and returns an avatar that speaks it in real time, with lip sync, blinks and facial expressions arriving together with the voice. Because the avatar follows the audio itself, it works with any text-to-speech voice and needs no text or timing data.
Open Source (MIT): a 3D talking avatar that runs entirely in the browser using its own 1 MB lip-sync model. It needs no server, account or API key, includes male and female avatars, supports custom .glb models and ships with a React component.
Realistic Avatars: photorealistic avatars rendered in the cloud and streamed to a web page over WebRTC or into a LiveKit room. Interruptions stop the avatar mid-sentence and report how much the user heard, and sessions run for up to 3 hours at 25 fps.
Voice Agents: a no-code builder for complete agents with a prompt, knowledge base, avatar and voice, shared by link, website widget or iframe.
Talking Avatar works with OpenAI, ElevenLabs, Azure, Fish Audio and Cartesia voices, and with LiveKit Agents, Pipecat, Daily and OpenAI Realtime. Developers can use the Node.js SDK, a Python plugin for LiveKit Agents, or the REST and WebSocket APIs.
Common uses include AI customer service, tutors, interviewers, shopping assistants, sales role-play and kiosks. The open-source edition is free; hosted avatars cost $0.02 per minute, billed per second, and new accounts get free credit.
No comments or reviews, maybe you want to be first?