Google gives Gemini Live a face with new talking AI avatars

Google has announced Gemini 3.8 Live, a model update that pairs the AI's real-time voice with a selectable avatar designed to make interactions feel less like talking to a machine.

The avatars perform lip-syncing to Gemini’s responses using low-latency video generation that is superimposed onto the speech capabilities of the model.

Google is targeting this capability directly toward business customers rather than consumer users. Research scientist Shuo-yiin Chang and software engineer CJ Zheng outlined the intent behind the development as making AI more “visually present” in applications such as customer service, virtual receptionists, and employee training.

Companies can choose from a library of preset avatars at launch or build custom ones, including branded elements like a company logo on the avatar's uniform.

All videos produced by the avatar will include an unobtrusive SynthID watermark in both the audio and video streams, as Google claims this will reduce misinformation and misattribution as the line between AI-created content and actual content blurs with time.

In addition, Google announced that its new Gemini 3.8 Live can support 97 different languages automatically and switch between them during a conversation for more fluent interaction.

Gemini 3.8 Live Extended Thinking, a more sophisticated version of the previous product, has been reported to outperform its competitors GPT-Live-1 Astra and Grok Voice Think Fast 2.0 on several benchmarks while being cheaper to use.