Gemini 3.8 Live adds a real-time Live Avatar

Google’s Gemini 3.8 Live with Live Avatar couples the live dialogue capabilities of Gemini 3.8 with low-latency streaming video, giving enterprise agents a real-time visual presence rather than a text or voice-only interface. It processes visual and audio input simultaneously and is available in Gemini Enterprise as of the announcement.

What’s actually new

• Asynchronous tool execution with continuous presence. Live Avatar can trigger tool calls and fetch data in the background while dialogue continues, so an agent can look up an order or check in a guest without the conversation stalling.
• Native multilingual speech-to-speech across 97 languages. Lip-sync and expressions adapt as the model switches languages mid-conversation, without video fidelity loss or visual drift.
• Custom avatars from one reference image. Alongside a library of preset characters, developers can generate an animated, responsive avatar from a single high-quality reference image while preserving likeness, brand styling or character identity. Custom creation is currently limited to enterprise allowlisting.
• SynthID on the output. Google says the imperceptible watermark is woven directly into the audio and video, with a model card covering the safety approach.

Why it matters

This is the point where a realtime model stops being a voice on a phone and becomes something a customer can see. The interesting engineering is less the avatar itself than the concurrency: keeping a talking face expressive and on-script while background tool calls resolve is the hard version of realtime, and Google is shipping it as an enterprise feature rather than a research demo. Expect the same asynchronous tool pattern to show up in every competitor’s agent stack within a few months.

(Original source: Google)