Google has launched Gemini 3.8 Live with Live Avatar, an enterprise feature that pairs Gemini’s live conversational model with a responsive animated on-screen persona. The company says the system can listen, see and respond through synchronized speech and video in near real time.
The release is available in Gemini Enterprise, with access through Google’s multimodal Live API. It moves the product conversation beyond text chat and voice assistants toward video-based agents designed for customer interactions, guided walkthroughs and similar workflows.
What changed
Live Avatar gives a Gemini agent a visual identity: a character with lip synchronization, facial expressions and conversational turn-taking. Google says the model processes audio and visual input simultaneously, then produces expressive audio and video responses.
The technical distinction is not only the animated face. Google says the system can make asynchronous tool calls—for example, retrieving information or taking an action in a connected system—while continuing its spoken interaction. In a hotel check-in example shared by the company, the avatar continues the dialogue while tasks run in the background.
Google also says Live Avatar supports speech-to-speech transitions across 97 languages without video drift or degraded visual fidelity. Companies can select from preset avatars, while custom avatars created from a reference image are currently limited to enterprise customers that have been allowlisted.
Why it matters for operators
For businesses already building AI agents, a visual layer could make certain self-service interactions feel less like navigating a chatbot and more like speaking to a representative. That is most relevant where conversation is central to the service: onboarding, product demonstrations, travel or hospitality flows, support triage and multilingual guidance.
The more practical capability may be uninterrupted conversation during tool execution. Voice agents often expose operational latency when they pause to search, query an internal system or complete a transaction. An agent that can acknowledge the request, collect more information and maintain the interaction while those calls execute may create a smoother handoff between conversation and action.
But a visual persona also adds a new production and governance surface. Teams will need to decide where an avatar actually improves comprehension or conversion, rather than simply adding visual novelty. They will also need brand rules for character design, scripts, escalation and disclosures—especially if custom avatars resemble real people or established brand characters.
Identity controls remain central
Google says all Live Avatar audio and video output carries its invisible SynthID watermark, intended to help identify AI-generated media. The company also says it has safeguards designed to respect identity.
Those measures matter, but they do not eliminate the deployment work for enterprises. Operators should establish approval processes for reference images, define restrictions on impersonation and ensure customers know when they are speaking with an AI system. Audit trails for tool calls and clear human escalation paths remain important, particularly in regulated or high-stakes support settings.
What to watch next
The early test is whether the feature improves business outcomes enough to justify the added design, moderation and integration complexity. Enterprises should evaluate it against simpler voice or text agents using concrete measures: task completion, resolution time, abandonment, escalation rates and customer satisfaction.
Also watch the availability of custom avatars. Google’s preset library provides a lower-risk starting point, while broader custom creation could make branded agents more useful—and raise the stakes around identity permissions and governance. The competitive question is increasingly not whether an AI agent can talk, but whether it can reliably complete work while presenting an interface customers will accept.
