Gemini's Live Avatar switches between 97 languages, but only on Gemini Enterprise

Gemini's Live Avatar switches between 97 languages, but only on Gemini Enterprise

Google has given Gemini a talking face that moves between 97 languages in the middle of a conversation, and it runs in Gemini Enterprise and nowhere else. Gemini 3.8 Live with Live Avatar went live on September 24 with its lip sync and expressions adapting as it switches language, and Google names no other place to get it.

Gemini 3.8 Flash and 3.8 Flash Cyber title graphic

Google's title card for the Gemini 3.8 Flash and 3.8 Flash Cyber launch. Live Avatar runs on Gemini 3.8 Live, a different model in the same generation. Image: Google

That single-tier availability is the shape of the whole release. Live Avatar is sold as something a company puts in front of its own customers, not something you can switch on in the Gemini app, and Google announces no consumer version. Google Cloud's own post says the feature is generally available there with US and EU endpoints.

What Google says it does

The feature pairs what Google calls near real-time video generation with speech, producing what it describes as an experience that "listens, sees, and speaks with a dynamic visual persona". Google claims precise lip syncing, natural expressions and fluid turn-taking, and says the model processes visual and audio input at the same time. Google's wording throughout is "near real-time" rather than real-time.

On the languages, Google says Live Avatar "can seamlessly transition across 97 languages without degrading video fidelity or introducing visual drift". The claim covers speech-to-speech synchronisation, so the lip movements and expressions are meant to follow the switch rather than lagging behind it.

There is also asynchronous tool calling, which lets the avatar "trigger tool calls and fetch data in the background while continuing active dialogue". Google's worked example is checking a guest into a hotel: the avatar keeps talking while the lookup happens. Its other examples are customer service and interactive walkthroughs.

Custom avatars need an allowlist

Gemini Enterprise customers get a library of preset avatars, and developers can generate their own from what Google calls "a high-quality reference image", preserving "reference likeness, brand styling, or character identity". That route is gated: "Custom avatar creation is currently available only through enterprise allowlisting."

Google says every output from the feature carries a SynthID watermark, "woven directly into the audio and video output" so generated content stays detectable. Google describes that as part of "strict safeguards designed to respect identity", and points to its model card for the rest.

The published detail stops at the feature list

Google gives no latency figure, no frame rate, no resolution, no price and no benchmark score of any kind. Live Avatar is a layer on Gemini 3.8 Live, the live dialogue model Google launched on September 15, and the published numbers stop at 97 languages.

A convincing talking head that changes language mid-conversation is the interface the industry has been demonstrating for years. What most readers have actually used is closer to an avatar standing in for a webcam in a Teams call or a Vision Pro Persona on a video call, and neither of those answers back. This one does, in 97 languages, for anyone whose employer pays for Gemini Enterprise.

Leave a Comment

Your email address will not be published. Required fields are marked *