Google Unveils Gemini 3.8 Live with Real-Time Live Avatar

Real-Time Visual Presence for Enterprise Agents

Google has announced Gemini 3.8 Live with Live Avatar, integrating near real-time streaming video with its native live dialogue models. Released in Gemini Enterprise, the system couples low-latency visual generation with speech to give AI agents a dynamic visual persona featuring precise lip-syncing and natural expressions.

This capability builds directly upon the recent Gemini 3.8 Live launch. By processing visual and audio streams simultaneously, the new tier is designed to handle interactions like customer service or interactive walkthroughs with fluid turn-taking and continuous visual presence.

Asynchronous Execution and Multilingual Scale

Beyond visual rendering, the model incorporates advanced reasoning capabilities that support asynchronous tool calling. This allows the system to fetch data and trigger background tasks while maintaining an active, uninterrupted dialogue flow with the user.

The platform also features native multilingual speech-to-speech synchronization. According to the announcement, the model can seamlessly transition across 97 languages mid-conversation without degrading video fidelity, visual drift, or sync quality.

Customization and Safety Protocols

Organizations can choose from a preset library of avatars or utilize enterprise allowlisting to generate custom animated personas from a high-quality reference image, maintaining specific brand identities.

To address transparency and security, all audio and video outputs generated by the feature integrate SynthID watermarking. This imperceptible watermark remains embedded to help identify synthetic media, complemented by an accompanying model card detailing Google’s safety frameworks.

Source

Official announcement or documentation