New to this? Read it in simple words
- Google’s Gemini AI can now appear as a talking video character.
- The character moves its lips in time with its speech and works in 97 languages.
- Only business customers can use it. Custom faces from a photo need Google’s approval.
- Every video has a hidden watermark that shows it was made by AI.
- Avatar
- A digital character that represents a person or a company on a screen.
- Watermark
- A hidden mark inside a picture, video, or sound that shows where it came from.
- Voice agent
- An AI that talks with people by voice and can do tasks for them.
Google announced Gemini 3.8 Live with Live Avatar on September 24, about a week after it launched the Gemini 3.8 Live voice models. The new feature gives an AI voice agent a face that moves and speaks in near real time.
A face for voice agents
Live Avatar pairs Google’s live voice model with generated video. The avatar lip-syncs to its own speech, shows facial expressions, and takes turns in a conversation without long pauses.
It can watch a live camera feed or a shared screen while it listens. It can also call tools and fetch data in the background, so the conversation does not stop while it looks something up.
Google says it understands and speaks 97 languages and detects the language automatically. Its examples include hotel check-in, insurance claims, and guided walkthroughs.
Custom faces need Google’s approval. Every stream carries a SynthID watermark.
Only for businesses, for now
Live Avatar is available only in Gemini Enterprise, Google’s service for companies, with US and EU endpoints. Developers build with it through the Gemini Live API.
Companies can choose from a library of ready-made avatars, both realistic and cartoon-like. Google’s cloud blog shows a custom avatar built from a single reference photo and a voice sample.
Custom avatars are available only to companies that Google adds to an allowlist after a verification process. Cox Automotive says it used the technology for a car-shopping assistant on Autotrader.
Watermarks, and a question of trust
Google says every audio and video stream carries SynthID, its invisible watermark for AI-made content. It says the allowlist for custom avatars is there to protect identity and prevent misuse.
Neither of Google’s announcements explains what consent a real person must give before their face becomes an avatar.
Engadget called the avatars “creepy”. It wrote that, given people’s dislike of AI slop, they may not be universally popular.
Sources
Every fact in this story comes from the sources below. Open them to check our work.
- 1
- 2Primary source · September 25, 2026Power your agents: Gemini 3.8 Live with Live Avatar is now generally available Google Cloud
- 3
- 4Research · September 25, 2026Google has given Gemini 3.8 Live a face that can lip-sync, react, and use tools in the background Digital Trends
- 5
Features, language support and customer examples are Google’s own claims. We found no independent tests of the avatars yet.
Google’s Gemini can now talk to customers as a video avatar. Silicon AI News, September 26, 2026. https://siliconainews.com/stories/gemini-live-avatar-enterprise
Google’s Gemini can now talk to customers as a video avatar, but one review called it “creepy” and consent rules are unclear. How we rate the mood
The mood shows how the news frames the event for the tech and AI world: as an opportunity (6 and above) or as a loss (below 5). It is not our verdict on who is right.