Gemini 3.8 Live Now Talks Back With a Face, Not Just a Voice

Google made Gemini 3.8 Live with Live Avatar generally available inside Gemini Enterprise on 24 September, with endpoints live in the US and EU. The feature pairs Gemini's existing live voice conversation with a streaming video avatar, an animated person or character whose mouth and expressions stay locked to what it is saying in real time, at 24 frames per second.

The avatar can keep talking naturally while background tools run, look at whatever a user shares on camera or screen, and lip sync across 97 languages. Enterprise customers pick from a library of pre built avatars by default. Building a custom one is possible but sits behind a strict allowlist and identity verification process, which is Google trying to head off the obvious deepfake risk before launch rather than after a headline forces the issue. Every generated frame of audio and video also carries an invisible SynthID watermark.

The interesting design decision is putting a face on an assistant at all. Text and voice interfaces have got comfortable existing without one, and a talking avatar is the kind of feature that can tip into uncanny very fast if the lip sync or timing is even slightly off. Google is betting that a well made avatar reduces the friction of a voice only interaction, particularly for training, customer service and other scenarios where seeing a response land, not just hearing it, changes how much someone trusts what they are being told.

For anyone building on top of Gemini Enterprise, this is a genuinely new interface option rather than a visual skin on the existing chat. It changes what a script for an AI interaction needs to account for. Pacing, gesture and expression are suddenly design decisions in a way they never were for a text box, and that is closer to directing a piece of motion work than writing a prompt.