Google DeepMind has launched Live Avatar, a new mode for Gemini 3.8 Live that gives its conversational AI a visual presence. Instead of just hearing a voice, users see an animated face that speaks back, with lip movements, facial expressions and turn-taking timed to match the conversation in near real time.
The feature arrives in Gemini Enterprise and follows last week's launch of Gemini 3.8 Live itself. Enterprises choose from a set of preset avatars, which can look like realistic humans or stylised animated characters. Custom avatars, built from a company's own reference images, are currently limited to allowlisted enterprise accounts rather than open to everyone.
A notable design choice is that tools and APIs can keep running in the background while the avatar continues talking, something Google describes as asynchronous function calling. In practice, that means the avatar does not have to go silent or freeze while it looks something up or completes an action, which is a common failure point in earlier voice-and-video agents. Language support spans 97 languages, including mid-conversation switching between them.
On safety, Google says every piece of audio and video the avatars generate carries an invisible SynthID watermark, a step aimed at making AI-generated avatar output identifiable even once it has been shared elsewhere.
The target use cases are customer service and interactive walkthroughs, areas where a visible, responsive face can make automated support feel less like a chatbot and more like talking to a person. It is a clear sign that Google sees avatar-based interaction, not just voice, as the next layer for enterprise AI agents.
Source: Google DeepMind
