Gemini 3.8 Live adds a lip-synced video avatar for enterprise
Google DeepMind has introduced Gemini 3.8 Live with Live Avatar, a week after launching Gemini 3.8 Live itself. The new feature pairs near real-time video generation with speech, giving the native live dialogue model a dynamic visual persona that listens, sees and speaks with precise lip-syncing, natural expressions and fluid turn-taking. It is aimed at enterprises building interactive customer service or walkthrough experiences, and is available starting today in Gemini Enterprise. Because the model processes visual and audio inputs simultaneously, it can hold a conversation while also working: with asynchronous tool calling, Live Avatar can trigger tool calls and fetch data in the background while continuing active dialogue, handling complex tasks such as checking in a hotel guest without breaking the flow of conversation. The feature also targets global deployments through native multilingual speech-to-speech synchronization, dynamically adapting lip-sync and expressions across 97 languages, including mid-conversation language switches, without degrading video fidelity or introducing visual drift. Organizations get a library of preset avatars plus the option to customize their own: from a high-quality reference image, developers can generate a fully animated, responsive avatar that preserves reference likeness, brand styling or character identity, though custom avatar creation is currently available only through enterprise allowlisting. On trust and safety, Google DeepMind says all output from its AI products, including Live Avatar's audio and video, is watermarked with SynthID, an imperceptible watermark meant to keep AI-generated content detectable and help limit misinformation and misattribution.
Key facts
- Live Avatar adds near real-time, lip-synced video presence to Gemini 3.8 Live, available today in Gemini Enterprise
- Asynchronous tool calling lets the avatar fetch data and run tasks in the background while dialogue continues uninterrupted
- Speech-to-speech synchronization adapts lip-sync and expressions across 97 languages without losing video fidelity
- Custom avatars can be generated from a reference image but are currently limited to enterprise allowlisting
- All Live Avatar audio and video output is watermarked with SynthID to keep AI-generated content detectable
Why it matters
Live Avatar turns a voice-only conversational AI into one with a visible, expressive face, closing a gap between chatbots and human-staffed video interactions. It arrives just a week after the underlying Gemini 3.8 Live launch, and combines existing pieces (real-time video generation, live dialogue, tool calling, multilingual speech) into a single enterprise-facing product rather than introducing a new capability from scratch.
Who it affects
The feature targets enterprises building customer-facing agents, such as customer service desks or interactive walkthroughs, through Gemini Enterprise. Organizations that want a branded look can request custom avatars, though that path currently requires enterprise allowlisting rather than open self-service.
How to use it
Gemini 3.8 Live with Live Avatar is available starting today inside Gemini Enterprise, with API documentation provided to get started. Teams can choose from a library of preset avatars or, once allowlisted, generate a custom animated avatar from a high-quality reference image that preserves likeness and brand styling. No pricing or subscription details for the feature were given.
How solid is it
This is Google DeepMind's own product announcement, published on its blog, describing capabilities and a same-day availability date without third-party verification. It names no customers, executives or technical specifications such as latency figures or model architecture, so claims about fidelity and tool-calling performance rest on the company's own description.
Risks and caveats
The claims of seamless 97-language transitions and uninterrupted background tool calls are the company's own characterization, not independently tested. Custom avatar generation from a reference image raises identity and likeness concerns that Google DeepMind addresses by restricting the feature to allowlisted enterprise customers and by watermarking all output with SynthID, though no timeline was given for wider availability of custom avatars.
“Live Avatar can trigger tool calls and fetch data in the background while continuing active dialogue, handling complex tasks while ensuring an uninterrupted conversational flow.”
— Google DeepMind blog post