Google adds a talking avatar to Gemini 3.8 Live
The enterprise feature pairs real-time video with speech, lip-syncing across 97 languages and running background tool calls.
The facts
- Gemini 3.8 Live with Live Avatar launches today in Gemini Enterprise, a week after the base Gemini 3.8 Live release
- It pairs near real-time video generation with speech, offering lip-syncing, expressions and turn-taking
- Supports native multilingual speech-to-speech sync across 97 languages, with Google saying lip-sync and expressions adapt without visual drift
- Asynchronous tool calling lets the avatar fetch data or run tasks in the background without interrupting dialogue
- Organisations can pick preset avatars or build custom ones from a reference image; custom creation is limited to enterprise allowlisting
- All output is watermarked with SynthID, embedded in both audio and video, per Google
Google DeepMind has added a real-time visual avatar to Gemini 3.8 Live, its native live-dialogue model. It gives enterprise chatbots a face that lip-syncs, expresses emotion and takes conversational turns. The feature, called Live Avatar, launched today in Gemini Enterprise. That is one week after the underlying Gemini 3.8 Live model shipped.
The pitch is customer service and interactive walkthroughs, not general chat. Google says the avatar processes visual and audio input simultaneously. It can also run tool calls asynchronously in the background — checking a guest into a hotel, for instance — while the spoken conversation continues without a pause. That distinguishes it from earlier talking-avatar demos, which had to freeze while fetching data.
Google also claims the avatar's lip-sync and expressions hold up across 97 languages. It can switch between them mid-conversation without degrading video fidelity, the company says. Enterprises get a library of preset characters. Through an allowlist, they can also generate a custom avatar from a single reference image while preserving likeness and brand styling. All generated audio and video carries Google's SynthID watermark, which the company says helps flag AI-generated content and limit misattribution.
Pricing was not announced. Google gave no numbers on latency or uptime. Nor did it back its multilingual claims with anything beyond the qualitative description in its blog post.
Why it matters
Real-time video-and-voice agents that can run background tool calls close a gap that made talking avatars feel scripted or laggy in production. Enterprises evaluating customer-service bots now have a native option that competes directly with third-party avatar vendors like Synthesia and D-ID. But it lives only inside Gemini Enterprise, not the general API. Teams outside that tier still can't build with it.
Written from the primary source above by The Neural Net’s automated desk (claude-sonnet-5, theneuralnet/pipeline@1) without human review. How this works. Corrections: check the source first — if we got it wrong, the source link is the record.