Gemini 3.8 Live with Live Avatar gives Google’s AI a face
Google has rolled out a Gemini Live update - version 3.8 - that introduces a real-time animated persona called Live Avatar. Rather than interacting with a voice-only interface, users can now watch an on-screen character lip-sync to the model's responses and shift facial expressions as the conversation unfolds. The feature is currently restricted to Gemini Enterprise subscribers.
One of the more technically notable aspects of Live Avatar is its multilingual handling. Google says the avatar can switch between any of the 97 supported languages mid-conversation without losing video fidelity or introducing what the company calls "visual drift" - a degradation in how well the mouth animation matches the audio. A demonstration shared by Google shows the avatar speaking in both English and Japanese, with its lip movements staying consistent across both languages, which is a harder problem than it might initially appear given that mouth shapes for different phonemes vary significantly between language families.
Beyond conversation, Live Avatar can also surface information on-screen during interactions, suggesting it is designed as more than a novelty and is intended to function as a practical interface layer for enterprise workflows. Giving an AI assistant a face is a design choice with implications beyond aesthetics - research consistently shows that people respond differently to agents that present humanlike visual cues, which can affect trust, engagement, and how instructions are interpreted.
The enterprise-only rollout suggests Google is treating Live Avatar as a business-facing tool, at least for now, likely to gather feedback on performance and real-world use before any broader release. It fits into a wider industry pattern of AI developers moving from text and voice interfaces toward more visually expressive agents, a direction also pursued by companies like HeyGen and Synthesia, though Google's integration directly into its Gemini assistant puts the capability closer to a general-purpose conversational AI than a dedicated avatar platform.
