Google is giving its conversational AI something it has largely been missing: an actual face. Gemini 3.8 Live with Live Avatar combines real-time conversation with generated video, creating digital characters that can see, listen and respond while appearing to speak naturally on screen.
Launching through Gemini Enterprise, Live Avatar builds on the Gemini 3.8 Live model introduced last week. Rather than presenting users with the familiar voice assistant experience, Google is synchronising generated video with speech to create an animated visual persona capable of maintaining a conversation with relatively low latency.
The interesting part isn’t simply having an AI-generated person talking back to you. Google says the system processes audio and visual information together, allowing an avatar to react to what it sees and hears while maintaining facial expressions, lip-sync and conversational turn-taking. The obvious applications are customer support, virtual assistants, interactive demonstrations and guided experiences where staring at a text box or listening to a disembodied voice isn’t particularly engaging.
Gemini 3.8 Live Avatar can also continue a conversation while carrying out other tasks. Its asynchronous tool execution allows an agent to call external tools or retrieve information in the background without effectively freezing the interaction. Google demonstrated this with a hotel check-in scenario, where the avatar continues speaking while the necessary systems are accessed behind the scenes.
Language support is another major part of the launch. Google says Live Avatar can move between 97 languages during a conversation, adapting its speech synchronisation and facial movements without disrupting the generated video. If it works consistently outside controlled demonstrations, that could make the technology particularly useful for multinational customer-service operations.
Businesses aren’t restricted to Google’s preset characters either. Live Avatar supports customised appearances generated from a reference image, potentially allowing companies to create characters that fit an existing brand or visual identity. Custom avatar creation, however, is currently limited to enterprises that have been specifically allowlisted.
Predictably, realistic AI-generated people also introduce questions around impersonation and transparency. Google says Live Avatar’s generated audio and video are embedded with SynthID, its invisible watermarking technology designed to identify AI-generated media.
For now, Gemini 3.8 Live with Live Avatar is firmly aimed at enterprise customers rather than becoming your new animated Gemini companion on a phone. But it points toward where conversational AI is heading: assistants that don’t just talk like someone is there, but increasingly look like someone is there too.
