Google has added a real-time video avatar to Gemini 3.8 Live, the conversational model it launched only last week, letting the chatbot appear on screen with a synced, expressive face while it talks. The feature, called Live Avatar, is rolling out today inside Gemini Enterprise rather than the consumer Gemini app.
Google’s Shuo-yiin Chang, a research scientist, and CJ Zheng, a software engineer on the Gemini Audio team, described the feature in a company blog post published Thursday. They wrote that Live Avatar processes voice and video together so the model can listen, watch, and respond with lip-syncing and facial expressions in near real time, rather than answering with text or audio alone.
The system keeps working while it thinks. Google says Live Avatar can call external tools and pull data in the background mid-conversation, so an avatar walking someone through a hotel check-in, the company’s own example, does not freeze up while it fetches a reservation. That kind of asynchronous tool calling is table stakes for text-based agents already; extending it to a live video face is the harder engineering problem Google is claiming to have solved.
Multilingual support is broad on paper: Google says the avatar’s lip-sync and expressions adapt across 97 languages without a drop in video quality. None of these claims come with independent benchmarks or a third party’s usability testing. They are Google’s own description of its own launch.
Businesses can swap in a custom avatar built from a single reference photo, preserving likeness and brand styling, but only through an enterprise allowlist, so most developers cannot try that piece yet. Google says every Live Avatar output is watermarked with SynthID, its system for marking AI-generated audio and video, and points to a model card for more detail on safety testing.
The timing puts Google in a real-time avatar race that already has other entrants. Meta has been pushing its own Muse Realtime Avatar technology this month, aimed at more consumer-facing conversational use, while OpenAI and others have shipped increasingly fluid voice modes without a visual face attached. Google’s bet is that enterprises want a persona customers can see, not just hear, for customer service and interactive walkthroughs.
Restricting the launch to Gemini Enterprise, rather than shipping it to consumer Gemini users first, suggests Google wants paying business customers to stress-test the avatar’s reliability, including how it handles interruptions, before opening it more broadly. Google has not said when, or if, Live Avatar will reach the consumer app or a wider developer tier. Teams building customer-facing chat interfaces should watch that gap: whoever ships a believable, low-latency avatar to consumers first sets the bar the rest of the market gets measured against.
Reported by Google’s official blog (blog.google) on 24 September 2026.
