Google DeepMind launches Gemini 3.8 Live with Live Avatar for enterprise agents

Google DeepMind announced the release of Gemini 3.8 Live with Live Avatar, a multimodal model that combines real-time video streaming with speech to create interactive, visually present enterprise agents. The feature is designed to make conversations feel more natural by enabling agents to listen, see, and speak with dynamic visual personas. According to the announcement, Live Avatar supports precise lip-syncing, natural expressions, and fluid turn-taking to enhance user engagement.

The model integrates near real-time video generation with speech, allowing agents to process visual and audio inputs simultaneously. This enables continuous, uninterrupted dialogue even as background tasks are executed asynchronously. For example, the announcement describes a use case where an agent checks in a hotel guest while continuing the conversation without interruption.

Live Avatar supports native multilingual speech-to-speech synchronization across 97 languages. The feature dynamically adapts lip-sync and expressions during language transitions, maintaining video fidelity and avoiding visual drift. The announcement includes a demonstration of seamless mid-conversation language switching.

Organizations can customize Live Avatar to align with brand identity. Developers can generate fully animated, responsive avatars from high-quality reference images while preserving likeness, brand styling, or character identity. Custom avatar creation is currently available only through enterprise allowlisting, according to the post.

The company emphasizes trust and transparency, noting that all output generated by the AI is watermarked with SynthID, an imperceptible watermark embedded directly into audio and video output. This is intended to help detect AI-generated content and minimize misinformation and misattribution. The announcement links to a model card for further details on safety and responsible deployment.

Gemini 3.8 Live with Live Avatar is available in Gemini Enterprise. The announcement provides a link to API documentation for developers to begin using the feature.