Google made Live Avatar in Gemini 3.8 Live generally available in Gemini Enterprise. The capability connects real-time voice conversation with an AI-generated video character, including lip synchronization, turn-taking and tool use during the conversation.
At this stage, the launch primarily targets organizations developing interactive service, training or communication experiences. It is not a general launch for all consumer Gemini users.
What is Live Avatar?
Live Avatar adds a video layer to a real-time Gemini conversation.
Instead of receiving only voice, users can see a character speaking with lip movements matching the speech. Google says the system is also designed for more natural conversation turn-taking, rather than just playing a prerecorded video.
The system supports 97 languages in speech-to-speech conversation.
All video output includes SynthID, Google's invisible marking intended to identify AI-generated content.
More than a talking character
One important point in the launch is the integration with tool calling.
Gemini can take actions through tools during a conversation, including asynchronous tool calling. This means the system can continue interacting while an external action runs in the background.
Google presents this capability as a basis for agents able not just to talk to users, but to perform a task in real time.
The company also notes Agent Development Kit support, allowing developers to incorporate Live Avatar into broader agentic environments.
Custom avatars
Gemini Enterprise includes a library of ready-made avatars.
Organizations can also create a custom character from a reference image, but this capability is not freely open to every customer. According to Google, custom avatar creation is available through an enterprise allowlist.
This is important for companies building a branded virtual representative or a character representing a particular person.
Google says it operates safeguards designed to address identity use, but their practical impact will become clear as usage expands.
Availability and infrastructure
Live Avatar is now in General Availability within Gemini Enterprise.
Google lists endpoints in the United States and European Union, along with provisioned throughput for customers needing more stable capacity.
By contrast, Live Extended Thinking with Avatar remains in private preview.
Not all capabilities appearing around Live Avatar are therefore GA at the same time.
Where it could be used
Live Avatar primarily suits experiences where speech and visual connection are part of the interaction.
Possible examples are customer service, training, onboarding, sales and internal support.
Tool calling during conversation also makes it relevant to agents needing to gather information, initiate a process or update a system during the interaction.
How it differs from ordinary Avatar generators
The market already includes tools generating video with talking characters.
The distinction Google seeks to emphasize is real-time interactivity: Live Avatar is intended not merely to generate a video from text, but to participate in a conversation and operate tools along the way.
This positions it against existing avatar and voice AI services while connecting it directly to Gemini's enterprise ecosystem.
What is still missing
Google did not publish a full pricing breakdown for every Live Avatar use scenario in the research examined.
Questions of latency under heavy load, quality across languages and actual adoption also remain open.
Reaching GA means the service passed a significant product stage, but does not prove that every enterprise use is suitable for it today.
Why the launch matters
Combining voice, video and agentic tools makes the boundary between a chatbot, voice representative and video character less clear.
For Google, Live Avatar is a way to turn Gemini Enterprise from a conversational interface into a tool able to represent an organization to a user in real time.
The next step will be to see whether companies choose to deploy these avatars at scale and how they handle questions of identity, trust, privacy and cost.