CONFIRMED: Gemini just got a face
Google confirmed on September 24 the launch of Gemini 3.8 Live with Live Avatar, a visual layer bolted onto the real-time dialogue model it had released just days earlier. Instead of only listening and answering by voice, the system now generates video of an animated character, cartoon-like or realistic, whose mouth and expressions track the speech at up to 24 frames per second. The core shift is not a narrow technical tweak: it is Google trying to turn voice conversation into persistent visual presence, something Meta and other players had already been testing with their own avatars.
How it works in practice
According to Google, Live Avatar processes audio and video simultaneously, lets users interrupt at any point without losing context, and keeps lip-sync accurate across 97 languages, including mid-conversation language switches. The feature most relevant to enterprises is asynchronous tool execution: the avatar can keep talking while it fetches data or runs a background task, such as confirming a hotel booking without freezing the dialogue. That brings the feature closer to a real agent who resolves things while talking, rather than a chatbot that stalls waiting on a system call.
Where it should show up first
The feature is exclusive to Gemini Enterprise and targets use cases like customer service, virtual reception, product walkthroughs, and multilingual support, with a library of ready-made avatars or custom faces built from a reference photo for approved organizations. For small and midsize businesses currently limited to text or voice support, the jump to a synced talking face could raise the perceived quality of support, but it also requires calibrated expectations: sessions are short, cost runs higher than audio alone, and for now the feature is only available through Google-approved enterprise allowlisting, not self-serve.
The limit Google itself acknowledges
Google applies a SynthID watermark to all generated video and audio, keeps custom faces under review, and restricts the feature to the Enterprise tier, a sign the company is aware of the reputational risk of poorly received AI avatars. Independent coverage has already described some avatars as looking unnatural, and the short-session design itself suggests Google is still testing compute cost limits before widening access. The more powerful Live Extended Thinking version remains in private preview without the avatar feature.
Our take
Gemini 3.8 Live with Live Avatar is both defensive and offensive: it keeps Google from falling behind in the race for visually present assistants, and it moves on the enterprise support market before other players lock in this layer. For Brazilian businesses evaluating support automation, the feature is not yet self-serve or publicly priced, so the practical next step is watching for wider access before committing budget to this front.
Sources
Google, official Gemini 3.8 Live with Live Avatar announcement: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-live-with-live-avatar/
Engadget, independent coverage of the launch: https://www.engadget.com/ai/google-adds-creepy-avatars-to-gemini-3-8-lives-agents