Gemini 3.8 Live introduces Live Avatar for enhanced enterprise interactions

Google just gave its enterprise AI customers a face. Gemini 3.8 Live with Live Avatar, announced on September 24, 2026, introduces real-time video avatars that can lip-sync, express natural facial movements, and hold fluid conversations across 97 languages, all without the uncanny-valley lag that has historically made AI avatars feel like talking to a mannequin with a WiFi connection.

The release builds on the Gemini 3.8 Live foundation launched just nine days earlier on September 15, which focused on low-latency voice processing and asynchronous workflows. Live Avatar takes those voice capabilities and wraps them in a visual layer designed specifically for enterprise use cases like interactive customer service and virtual product walkthroughs.

What Live Avatar actually does

At its core, Live Avatar is a real-time video generation system tightly coupled to Gemini’s speech-to-speech engine. When a user interacts with a Gemini-powered agent, they see a photorealistic avatar responding in real time, complete with synchronized lip movements, appropriate facial expressions, and natural turn-taking.

The multilingual piece is particularly notable. Live Avatar supports 97 languages with speech-to-speech synchronization, meaning the avatar can switch between languages mid-conversation without what Google describes as “visual drift.” In practical terms, that means the avatar’s lip movements and expressions stay calibrated to whatever language it’s speaking, rather than defaulting to a single language’s phoneme patterns.

For enterprises worried about deepfake risks and brand safety, Google has baked in several guardrails. Custom avatars must go through a verification and allowlisting process, meaning organizations can’t just generate any face they want. All generated video content carries SynthID watermarking, Google’s provenance tool that embeds an invisible identifier into AI-generated media. And the platform includes what Google calls strict data governance controls, designed to keep customer interaction data within enterprise compliance frameworks.

The enterprise AI arms race heats up

This launch positions Google more aggressively in a market where the lines between conversational AI, video generation, and enterprise software are rapidly dissolving. Microsoft has been integrating its Copilot suite deeper into enterprise workflows. Meta has pushed its open-source Llama models toward business adoption. And a growing roster of startups, from Synthesia to HeyGen, have built entire businesses around AI-generated video avatars for corporate communications.

Google’s play is integration. By tying Live Avatar directly to the Gemini model’s conversational intelligence, it can offer something the standalone avatar companies cannot: a single platform that understands context, speaks naturally, and looks human, all at once. The standalone avatar tools typically require pre-scripted content or separate language model integrations, adding latency and complexity.

Compliance features signal where the market is heading

The emphasis on SynthID watermarking and verified custom avatars reflects a broader industry shift. Regulators in the EU, US, and elsewhere are increasingly focused on AI-generated media transparency. The EU AI Act already classifies certain deepfake-adjacent technologies as high-risk, requiring disclosure and provenance tracking.

By building these compliance tools directly into the product rather than offering them as optional add-ons, Google is essentially making a bet that regulation will tighten, not loosen. The data governance controls are similarly pointed. Enterprise customers, particularly in financial services, healthcare, and government, have been cautious about deploying conversational AI precisely because of data handling concerns. Google’s framing of Live Avatar as enterprise-grade, with strict data governance baked in, is a direct response to those procurement objections.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.