Gemini 3.8 Live gets video avatars

Google DeepMind has added Live Avatar to Gemini 3.8 Live, its native live dialogue model line, launched last week. The feature pairs near real-time video generation with speech so an agent can listen, see and respond as an animated on-screen persona with lip-sync, facial expressions and turn-taking.

Two details stand out for people building voice agents. Asynchronous tool calling lets the avatar trigger tool calls and fetch data in the background while the dialogue continues — DeepMind's example is checking a guest into a hotel without breaking conversational flow. And native speech-to-speech synchronization covers 97 languages, with the company saying lip-sync and expressions adapt through mid-conversation language switches without video fidelity loss or visual drift.

Organizations get a library of preset avatars; custom avatars can be generated from a single high-quality reference image, preserving likeness and brand styling, but that path is available only through enterprise allowlisting. All audio and video output carries a SynthID watermark, which DeepMind frames as a safeguard against misattribution.

Live Avatar is available starting today in Gemini Enterprise, with API documentation and a model card published.

DeepMind released no measured latency, quality or throughput numbers — "near real time" is the only performance characterization offered — and the announcement is illustrated with demo videos rather than benchmarks. Pricing, availability outside Gemini Enterprise, and the criteria for custom-avatar allowlisting are not stated.