Google has rolled out Gemini 3.8 Live with Live Avatar, bringing interactive video agents featuring real-time lip-syncing and dynamic facial expressions to Gemini Enterprise customers. The system supports 97 languages with automatic detection and visual understanding.
Engineering Native Speech-to-Speech and Visual Understanding
Gemini 3.8 Live establishes a native speech-to-speech foundation. This architecture allows the model to handle interruption recovery naturally, letting users jump into a sentence without dropping conversation context or severing active backend transactions.
Complementing the audio engine is the newly deployed Live Avatar feature, which made its general availability debut in Gemini Enterprise endpoints across the US and EU. The model processes live camera feeds and screen shares concurrently with audio streams. It can pull up information on-screen while it talks, bridging the gap between verbal dialogue and visual data presentation.
According to Google, Live Avatar transitions across its supported 97 languages without degrading video fidelity or introducing visual drift. Demonstration footage exhibits the system speaking in English and Japanese while preserving precise mouth animation synchronization across both linguistic models.
Enterprise Customization and Background Tool Execution
To satisfy these requirements, the platform supports asynchronous background tool calling. The model executes tools and API calls in the background while keeping the conversation alive, acknowledging user requests and maintaining verbal continuity while heavy server tasks finish processing.
Organizations can choose from a curated library of diverse, preset avatars or generate custom iterations. Developers utilize a high-quality reference image to build a fully animated, responsive avatar that preserves reference likeness and brand styling. However, Google restricts custom avatar creation to enterprise allowlisting and verification processes to safeguard identity.
Security, Watermarking, and Deployment Protocols
To mitigate these risks, Google embeds its invisible SynthID watermark directly into the audio and video streams of all Live Avatar outputs. This imperceptible signature remains woven into the generated content, ensuring that synthetic media stays detectable by automated verification systems.
Under the current deployment structure, Gemini 3.8 Live with Live Avatar is available through Gemini Enterprise with provisioned throughput, enterprise compliance standards, and strict data governance policies. While Gemini 3.8 Live Extended Thinking remains confined to private preview status, the Live Avatar production architecture is fully operational for web, mobile, and interactive kiosk environments via updated API documentation.
