Google Adds Live Avatar to Gemini 3.8 Live for Enterprise

Google has extended its live conversational AI model with a visual layer. Gemini 3.8 Live with Live Avatar, announced on September 24, 2026, builds on the Gemini 3.8 Live launch from the previous week by adding near real-time video generation alongside the existing speech capability. The result, according to the company, is an assistant that listens, sees and speaks through an animated persona rather than through audio alone.

The feature is aimed squarely at enterprise deployments rather than the consumer Gemini app. Google positions it for customer service and interactive walkthroughs, citing lip-syncing, expression and turn-taking as the qualities meant to make the exchange feel closer to a conversation with a person than a scripted bot response. The company also says the underlying model can call tools in the background while a conversation continues, so a task such as checking in a hotel guest can trigger a data lookup without visibly pausing the dialogue.

Multilingual support is built in: Google states the avatar can switch between languages mid-conversation across 97 languages while keeping lip-sync and expression consistent. Businesses can draw on a library of preset avatars or, through enterprise allowlisting, submit a reference image to generate a custom animated avatar that keeps brand styling or character likeness. Google says every output carries a SynthID watermark woven into the audio and video, intended to keep AI-generated content identifiable.

The numbers

• Languages: 97, the count Google says the avatar can switch between mid-conversation while keeping lip-sync intact, from the newsroom post.

Live Avatar is available now inside Gemini Enterprise, with API documentation published alongside the announcement. Google frames it as a natural extension of the Live model line rather than a separate product, following the Gemini 3.8 Live and Live Extended Thinking release the week before and the Gemini 3.8 Flash and Flash Cyber launches earlier in the cycle. Custom avatar creation remains gated behind enterprise allowlisting rather than opening to every customer immediately, which keeps the rollout narrower than the base Live Avatar feature itself.

For a company already paying for Gemini Enterprise, the practical question is not whether the avatar looks convincing in a demo video but whether video generation holds up under the volume a real support queue produces, and whether the SynthID watermark satisfies the disclosure rules a given industry or region already requires of synthetic media in customer-facing settings.