Google has added a real-time visual avatar to its Gemini 3.8 Live conversational AI, creating an agent that can listen, speak, and display synchronized facial expressions during dialogue. The feature, called Live Avatar, is available starting today for Gemini Enterprise customers and targets customer service, interactive walkthroughs, and other virtual agent deployments.
Live Avatar pairs near real-time video generation with speech to produce a dynamic visual persona. The system synchronizes lip movements and expressions with spoken audio across 97 languages, using native multilingual speech-to-speech processing. Google said the avatar maintains video fidelity without visual drift when switching between languages.
How the multimodal system works
The feature processes visual and audio inputs simultaneously. A user can speak and show objects or gestures to the camera, and the avatar responds with matched facial expressions and voice, creating what Google describes as a more natural turn-taking rhythm than audio-only agents.
Behind the visual layer, Gemini's reasoning engine handles asynchronous tool calls. The avatar can trigger background data fetches or API calls while continuing an active conversation. A customer service avatar, for example, could pull up an order record without pausing or interrupting the spoken exchange.
Customization and brand control
Organizations can select from a library of preset avatars or create custom ones from a reference image. The custom option preserves likeness, brand styling, or character identity from a single high-quality photo, generating a fully animated and responsive avatar. Custom avatar creation requires enterprise allowlisting at this stage.
All audio and video output carries a SynthID watermark, an imperceptible marker woven directly into the content. Google said this helps detect AI-generated material to reduce risks of misinformation and misattribution.
Why this matters for creatives, marketers, and customer support teams
For customer support leaders, Live Avatar offers a way to add visual presence to automated agents without building synchronization infrastructure in-house. The asynchronous tool execution means a single avatar can handle complex queries - checking inventory, pulling account details, and processing returns - while maintaining eye contact and natural speech flow. Marketing and product teams can test branded avatars for interactive demos or onboarding walkthroughs. The 97-language support removes a common scaling barrier for global campaigns, though custom avatars currently require Google's approval through the allowlist process.
Your membership also unlocks:
