Google adds talking video avatars to Gemini for enterprises

Google's Gemini can now show up to a conversation with a face. The company on Thursday introduced Gemini 3.8 Live with Live Avatar, which pairs its real-time voice model with near-real-time generated video so that an enterprise AI agent appears as an animated person that listens, looks and speaks.

The feature is available now in Gemini Enterprise, Google's AI platform for businesses, and developers can build with it through the API. It arrives a week after Google launched Gemini 3.8 Live, the voice model underneath.

What Live Avatar does

Google's pitch is that human conversation is multimodal, and agents should be too. Live Avatar processes visual and audio inputs at the same time, and responds with lip-synced speech, facial expressions and natural turn-taking. The company suggests customer service desks and interactive product walkthroughs as the first uses. One demo shows an avatar checking a guest into a hotel. Other demo clips show a range of characters, each with its own look and voice, reacting to what they see and hear in near real time.

It also does work in the background. Using asynchronous tool calling, the avatar can fetch data or trigger actions while it keeps talking, so the conversation does not stall while it waits for a booking system or database to respond.

Language support is broad. Google says Live Avatar can switch between 97 languages mid-conversation, adjusting lip-sync and expressions without losing video quality or introducing visual drift. Google has not published the full language list in the announcement, so Indian companies will want to check which Indian languages are covered and how well before planning deployments.

Custom faces, with guardrails

Organisations can pick from a library of preset avatars or build their own from a high-quality reference image, keeping a likeness, brand style or character identity. Custom avatar creation is limited to enterprises that Google has put on an allowlist.

That restriction is deliberate. A tool that can animate a face from a single photo is also a tool that can be abused. Google says every output from its AI products is watermarked with SynthID, an invisible marker embedded in the audio and video so that generated content can be detected later. It has also released a model card describing its safety approach.

Why Indian businesses may take note

India has one of the largest customer support industries in the world, and banks, telecom operators and e-commerce firms here already run voice bots in several languages. A video layer on top of those bots is the obvious next step Google is selling, whether on a website, an app or a kiosk.

There are open questions. Google has not published pricing for Live Avatar in the blog post, and near-real-time video generation is compute-heavy, so costs per conversation could be significantly higher than for voice-only agents. Regulated sectors will also want clarity on how avatars disclose that customers are talking to AI, especially under India's recent rules on synthetically generated content.

For now, Live Avatar is an enterprise product, not a feature in the consumer Gemini app. But it shows where Google wants customer-facing AI to go: from a voice on a phone line to a face on a screen that can hold a conversation in dozens of languages.