Google said on Sept. 24, 2026, that Gemini 3.8 Live with Live Avatar is now available in Gemini Enterprise, adding an animated visual persona to the company’s live conversational AI. The launch is important because it changes the surface of the product: Gemini is no longer only speaking or typing back, but doing so through a face, expressions and synchronized mouth movement in real time.
The Verge reported the same launch and said the feature is currently limited to Gemini Enterprise customers. That makes the rollout easier to understand as a business product move rather than a consumer novelty. For organizations already using Gemini in customer support or other guided interactions, Google is now offering a more embodied interface; the launch does not extend the feature to the consumer Gemini app.
What changed
Google’s announcement says Live Avatar pairs near real-time video generation with speech so the system can “listen, see and speak” as a dynamic persona. The company says the avatar can lip-sync, show natural expressions and keep turn-taking fluid. In practical terms, that means the model is not just generating words. It is generating a live visual performance that sits on top of the conversation.
That distinction matters because enterprise AI products often fail at the handoff between a model’s answer and the user’s next action. A plain chat box can be enough for information retrieval, but it can feel thin when the task is conversational or service-oriented. Google is betting that a face, expressions and lip-sync can make the interaction feel less robotic and easier to follow. That is an editorial inference, not a measured claim in the retrieved material, but it is the obvious product logic behind the launch.
Google frames the use cases around customer service and interactive walkthroughs. Those are the kinds of tasks where a live, branded agent can hold attention while still giving users instructions, policy details or procedural help. The appeal is not just visual polish. If the interface makes the exchange feel continuous, users may be more willing to stay engaged long enough to finish a task.
How it works, according to Google
The mechanism behind the avatar is more than video generation. Google says Live Avatar is built on Gemini’s live dialogue stack and can process visual and audio inputs together. It also says the feature supports asynchronous tool calling, which lets the avatar continue talking while background tools fetch data or complete a request. Google’s example is a hotel check-in flow, where the system keeps the conversation going while it works behind the scenes.
That background execution is the more consequential technical change. Many enterprise assistants become awkward when they have to stop and wait for a backend system. A live avatar that can keep speaking while tools run could reduce that pause and make orchestration feel smoother. But it also raises the bar for the surrounding system: developers still need reliable permissions, integration with backend services and a human escalation path for requests the model cannot resolve cleanly.
Google also says Live Avatar supports multilingual speech-to-speech synchronization across 97 languages and can switch between languages without degrading video fidelity or introducing visual drift. That is a strong claim about scale and visual consistency, but it remains a vendor claim in the material retrieved here. The evidence does not include independent language-by-language benchmarks, latency measurements or third-party testing, so readers should treat the multilingual capability as part of Google’s launch description rather than as independently verified performance data.
Where the limits are
Google says organizations can choose from preset avatars and, in some cases, generate a custom avatar from a high-quality reference image while preserving likeness, brand styling or character identity. The company also says custom avatar creation is currently available only through enterprise allowlisting. That is an important boundary: the feature is designed for controlled deployment, not open-ended avatar creation.
Google says Live Avatar’s generated audio and video carry SynthID watermarks and that the feature includes safeguards intended to respect identity. Those controls matter for enterprises that care about provenance, impersonation risk and brand governance. But they are still Google’s own claims. The retrieved material does not show external validation of the watermark’s detection reliability in this launch context.
For decision-makers, the practical question is simple: does your use case actually need a live visual presence, or do you mainly need a better model behind the scenes? If the answer is the latter, a live avatar may add presentation without solving the bottleneck. If the answer is the former, the new feature could matter because it combines conversation, visual signaling and background tool use in a single interface. Either way, the launch is a reminder that enterprise AI products are moving from text boxes toward managed, branded interaction layers.
If you deploy Gemini Enterprise, check whether your account is on the allowlist for custom avatars and whether your workflow can support human escalation before you rely on a branded face, so you do not add polish without a workable support path.
