Gemini Live Avatar adds a visual face to enterprise voice agents

Google introduced Gemini 3.8 Live with Live Avatar on 24 September through Gemini Enterprise. The feature couples live dialogue with streaming video, allowing an avatar’s expressions and lip movements to follow generated speech. Google says it can transition across 97 languages.

The system supports asynchronous tool calling, so an agent can retrieve information without completely stopping the spoken interaction. Organisations can choose prepared avatars, while custom creation from a reference image begins through enterprise allowlisting. Audio and video output include SynthID. A visual presence is a presentation layer over the agent’s capabilities; it does not establish a human representative or change the need to check the actions an underlying tool can perform. The initial access is enterprise-focused.