Google introduces Gemini 3.8 Live with an AI avatar
Google is bringing Gemini 3.8 Live to a visual avatar experience that listens, sees and responds with voice and video in near real time. The feature is arriving in Gemini Enterprise with support for 97 languages, background task execution and custom avatars for businesses.
Google has added a near-real-time visual presence to Gemini 3.8 Live: its new Live Avatar listens, sees and responds with voice, video and an animated visual persona. The feature is already available to businesses through Gemini Enterprise.
The avatar syncs its lips with the audio, changes its expressions and respects conversational turn-taking. It does not simply display a face: it processes video and sound inputs at the same time to sustain exchanges that feel more like a human conversation.
What Live Avatar can do
The offering is designed for enterprise AI agents. It could, for example, assist a customer, guide someone through a process or explain step by step how to check in at a hotel.
One of its most relevant features is asynchronous tool execution. This means the avatar can look up data or trigger actions in the background while continuing to talk to you, instead of falling silent until the task is complete.
If a guest asks whether a room is available, the agent could check the hotel’s system while keeping the conversation going. The answer arrives without breaking the flow of the dialogue.
Conversations in 97 languages
Live Avatar includes native speech-to-speech synchronization across multiple languages. It can switch between 97 languages, adapting lip movements and facial expressions so the visuals stay aligned with the audio.
This makes it possible to design services for customers in different countries without creating a separate visual experience for each language. Google says that changing languages should not reduce video fidelity or create visible shifts between the voice and the face.
Custom avatars for each company
Organizations can choose from a library of predesigned avatars or create a custom one. Using a reference image, developers can generate an animated avatar that preserves visual features, character identity or branding elements.
Custom avatar creation is still limited to companies included in a pre-authorization system known as allowlisting. It is not an open option for every Gemini Enterprise account.
Watermarks for detecting generated content
Google says Live Avatar includes measures to make it clear when an image or voice has been generated by AI. All content created by its products includes an invisible watermark called SynthID, embedded directly into the audio and video.
The watermark is designed to help detect the content’s artificial origin and reduce the risks of misinformation or incorrect attribution. It does not stop the avatar from appearing natural, but it adds a technical signal that can be used to identify it.
Gemini 3.8 Live with Live Avatar is available today in Gemini Enterprise, along with its API documentation. The important thing now will be seeing how it performs outside demonstrations: its usefulness will depend on the quality of its responses, how well it integrates with each company’s systems and how clearly users understand that they are talking to an AI.