Google Adds Real-Time Video Avatars to Gemini 3.8 Live for Enterprise Voice Agents

Google has rolled out Live Avatar, a new feature that adds real-time video personas to its Gemini 3.8 Live and Live Extended Thinking dialogue models, according to a company blog post. The addition follows last week's launch of Gemini 3.8 Live, which Google described as its most advanced live dialogue system for building voice agents.

Live Avatar pairs near real-time visual presence with Gemini's existing voice capabilities, creating what Google calls "an experience that listens, sees and speaks with a dynamic visual persona." The company's blog demonstrates both realistic and cartoon-style avatars performing lip-syncing, facial expressions, and what Google terms "fluid turn-taking" during conversation.

The feature is aimed squarely at enterprise customer service and sales applications. Because the avatars run on Gemini's underlying reasoning models, Google says they can trigger tool calls and retrieve data in the background while maintaining an active spoken dialogue with a user — for example, pulling account information mid-conversation without breaking the visual interaction.

Availability and customization

Google will provide a library of preset avatars for organizations to deploy out of the box. Enterprises can also build custom avatars from reference images to match "brand styling or character identity," though this customization tier is gated behind enterprise allowlisting — it is not available to general developers or consumer accounts.

Every avatar output is watermarked using Google's SynthID system, which the company says is intended to keep AI-generated video content identifiable and address concerns about synthetic media transparency. Google describes this as part of "strict safeguards designed to respect identity" for the feature.

Access restrictions

Live Avatar is currently limited to Gemini Enterprise customers. Google has not disclosed pricing for the feature, nor has it published specific latency, resolution, or benchmark figures for how the avatars perform compared to competing video-agent products from other vendors.

No technical specifications — frame rate, resolution, supported languages, or per-minute cost — were included in Google's announcement. The company has also not stated whether Live Avatar will expand beyond enterprise allowlisting to broader developer access on a defined timeline.

What this means

This is a feature layered on top of the existing Gemini 3.8 Live model family, not a new model release — the underlying dialogue and reasoning models are unchanged. Google is packaging its live voice AI into a more commercially deployable form factor by adding a face to it, targeting a specific enterprise use case: replacing or augmenting human customer service and sales reps with video-capable agents.

The SynthID watermarking and identity safeguards suggest Google is anticipating backlash over synthetic human-like avatars appearing in customer-facing roles, a concern reflected in early public reaction to demo videos showing photorealistic AI personas. Whether businesses adopt Live Avatar at scale will likely hinge less on the technology's technical fluency and more on whether customers tolerate — or actively reject — interacting with an AI face during support and sales interactions. Google has not published data on user acceptance, only its own product claims.