Google gives Gemini a talking face with Live Avatar

Google is giving its real-time voice AI a face. In a blog post on Thursday, the Gemini Audio team introduced Live Avatar for Gemini 3.8 Live, which turns spoken conversations with Gemini into video of an animated character that talks back.

How it works

Live Avatar combines video generation with Google's live speech model. The avatar lip-syncs as it speaks, shows natural facial expressions and handles turn-taking, so the exchange feels like a video call. Google says:

• It supports 97 languages, with lip movements and expressions adapted to each.
• It takes in audio and visual input at the same time.
• It can run tools in the background, like looking something up, without stopping the conversation.

Who can use it

Live Avatar is available in Gemini Enterprise, through its console and API. Companies can choose preset avatars or build custom ones from reference images, though custom avatars are limited to approved customers for now. Google hasn't announced pricing or plans for the consumer Gemini app.

According to Google's model card, the avatar is designed for conversations lasting a few minutes rather than hours, and it can still make mistakes, including hallucinations.

Built-in labeling

Everything Live Avatar produces carries SynthID, Google's invisible watermark for AI-generated content. That matters as realistic talking faces become easier to make and harder to tell apart from real people.

What's next

The feature builds on Gemini 3.8 Live, which launched the week before. Watch for whether Google brings avatars to consumers, and how businesses use them for support and sales. It's Google's second agent-style Gemini upgrade this week, after Gemini started making phone calls for Pixel owners. More: Google news.