Google DeepMind Adds Real-Time Video Avatars to Gemini 3.8 Live

What happened

Google DeepMind today launched Live Avatar, a real-time visual avatar feature built on top of Gemini 3.8 Live, the native audio dialogue model the company released one week earlier. The feature pairs near real-time video generation with speech, giving Gemini's conversational AI a visible, animated persona that maintains lip-sync, facial expressions, and turn-taking during live conversations.

Live Avatar is available starting today exclusively within Gemini Enterprise. It is not being released as a standalone consumer product.

Technical details

According to Google DeepMind, Live Avatar processes visual and audio inputs simultaneously and generates matching video output in near real time. The system supports native multilingual speech-to-speech synchronization, with the company claiming lip-sync and expressions adapt seamlessly across 97 languages without degrading video fidelity or introducing visual drift when switching languages mid-conversation.

The feature also supports asynchronous tool execution: Live Avatar can trigger background tool calls — such as looking up data during a hotel check-in scenario cited by Google DeepMind — while continuing active dialogue, avoiding interruptions to conversational flow.

Enterprises get access to a library of preset avatars out of the box. Custom avatars, generated from a single high-quality reference image while preserving likeness and brand styling, are currently restricted to enterprise allowlisting — meaning general developer access is not yet available.

All audio and video output is watermarked with SynthID, Google's imperceptible watermarking system, intended to keep AI-generated content detectable and reduce misattribution risk.

What's missing

Google DeepMind has not disclosed pricing for Live Avatar, nor published specific latency figures, benchmark scores, or technical details about the underlying video generation architecture. The company has not specified compute requirements or usage limits for enterprise customers. No word on whether Live Avatar will expand beyond Gemini Enterprise to the standard Gemini API or consumer apps.

What this means

This is a feature launch layered on an existing model — Gemini 3.8 Live, which shipped a week prior — rather than a new foundation model release. The move signals Google's push to compete in the enterprise conversational-agent market against products like Microsoft's Copilot and various avatar-based customer service platforms, where visual presence is often seen as a differentiator for use cases like virtual receptionists, walkthroughs, and support agents.

The restriction to Gemini Enterprise and allowlisted custom avatars suggests Google is prioritizing controlled, monetizable enterprise deployment over broad developer access — likely due to both commercial strategy and the reputational risk of unrestricted avatar generation from user-supplied images. The SynthID watermarking emphasis reflects continued sensitivity around deepfake-style video generation, particularly for a feature explicitly designed to create realistic, branded human-like personas at scale.