Overview
Google is introducing video capabilities to its recently launched Gemini 3.8 Live models, aiming to power interactive voice agents for customer service and sales support.
Last week, Google launched Gemini 3.8 Live and Live Extended Thinking models, presenting them as advanced live dialogue models designed to give developers building blocks for reliable voice agents while enabling more natural voice commands.
Features
Google has expanded that framework with Gemini 3.8 Live with Live Avatar. The feature pairs near real-time visual presence with live dialogue models to create a dynamic visual persona capable of listening, seeing, and speaking.
Demonstration videos released by Google showcase both realistic and cartoon-style avatars featuring precise lip-syncing, natural facial expressions, and fluid turn-taking. The avatars communicate using facial expressions while utilizing Gemini's reasoning capabilities to trigger tool calls and fetch data in the background during active conversations.
Availability
Google plans to offer a library of preset avatars, alongside customization options for organizations to build proprietary characters using high-quality reference images. These customized options require enterprise allowlisting. Additionally, all Live Avatars incorporate SynthID watermarking to ensure AI-generated content remains transparent.
The Live Avatar feature is aimed at Google's Gemini Enterprise customers, meaning users can expect to encounter these interactive sales and support agents across various screens in the near future.
