Last Updated: 24 September 2026
Talking to an AI assistant usually means reading text or listening to a voice. Google is adding another part to that experience: a face. Gemini 3.8 Live with Live Avatar creates animated characters that can speak during a conversation, move their faces naturally and react to what the user is saying or showing through a camera.
The new feature arrives only days after the release of Gemini 3.8 Live. It is designed mainly for business uses such as customer support, guided services and virtual assistants. The avatar can also continue talking when it needs to access another tool, giving users fewer silent pauses during tasks that require information from another system.
Live Avatar takes the existing voice experience and adds generated video to it. Instead of hearing an AI voice without a visual presence, a user can see an animated person or character speaking during the conversation. The mouth movements are created to follow the generated speech, and the face can change its expression as the conversation moves forward.
The system also receives different forms of input together. It can listen to speech and process visual information at the same time. In practice, someone could speak to an agent and show something through a camera during the same interaction. The response can then arrive through speech and animated video rather than through a separate text box or static image.
That combination opens the door to AI services that feel closer to a normal guided conversation. A hotel could use an avatar during check in. A support service could guide a customer through a problem. A company could also create an interactive guide that explains a product as the person asks questions instead of forcing them through fixed menus.
Live conversations become awkward when an AI agent suddenly stops to search another system. The latest Gemini feature tries to reduce those pauses through background tool use. It can send a request to another connected service and keep the dialogue active during the wait, rather than treating every outside action as a reason to interrupt the exchange.
Imagine asking a hotel assistant to confirm a reservation. The system may need to search booking information before giving a final answer. During that search, the avatar can continue speaking with the guest and handle another part of the conversation. When the reservation data arrives, it becomes available to the ongoing exchange instead of starting a completely separate interaction.
For customer service teams, this could change how automated conversations are designed. Many requests involve several systems that have to communicate with each other. Keeping the dialogue active gives the user something closer to the flow of speaking with a human agent, who can continue asking questions even when another process is taking place.
One of the most practical parts of Live Avatar is its language support. The system can speak across 97 languages and change from one to another inside an active conversation. There is no need for the user to open a language menu each time. That could make a single virtual agent useful in services where customers arrive with very different language preferences.
The visual side also changes together with the voice. Lip movement is generated to follow the language being spoken, and the avatar adjusts its expressions during the switch. This is harder than simply translating text because the speech and face need to remain synchronized. A badly timed mouth movement would make the generated character much less convincing during direct conversation.
There are clear business uses for this. A travel service may serve visitors speaking several languages in one location. An international support center could receive customers from many regions. A multilingual avatar gives those services another option for handling the first part of an interaction before a human agent is needed for a more difficult case.
The feature supports language changes inside a session, though real performance will still depend on the language, the request and the audio conditions around the person speaking.
The first release is clearly built around business use. Companies can place a visual AI agent inside customer support systems or other digital services. A user could speak with the avatar rather than typing every request into a chat box. The same agent can also receive visual information, which makes it possible to guide a person through situations that are difficult to explain through text.
The feature is built for situations where people need more than a quick answer. It can guide someone through a service, respond to questions as they appear and keep the conversation moving when more information is needed. That makes it especially useful in places where users would normally depend on menus, forms or a human assistant to complete a process.
Uses include:
These are situations where users often need several steps of help rather than one short answer. The avatar gives businesses another interface for delivering that guidance without depending completely on menus or typed commands.
Companies are not limited to one standard digital face. Gemini Enterprise includes a collection of preset avatars with different appearances and voices. Organizations that need something more specific can also create a custom character from a reference image. The generated version can keep visual elements from that image, giving businesses more control over how the assistant appears to customers.
Custom avatars could be useful when an organization already has a recognizable character or a defined visual identity. The character can also have its own voice and expressive style. This means two businesses using the same underlying AI could present very different experiences to their users instead of displaying an identical virtual assistant in every service.
There is an important restriction. Custom avatar creation is not freely available to every customer. Access currently requires enterprise allowlisting. That process creates an additional check before an organization can generate an avatar from its own reference image, especially when the image represents a specific identity that needs protection against misuse.
Preset characters are therefore the easier starting point. More controlled visual identities are possible, but businesses need additional approval before using the custom creation option.
Generated faces and voices create obvious questions and concers about whether people can tell when content comes from AI. Every audio and video output produced by Live Avatar includes SynthID. The watermark is designed to remain hidden during normal use but can help identify material created by an AI system. It is applied directly to generated audio and video rather than appearing as a visible label.
The feature also has technical limits. The model card lists possible hallucinations as a known issue, meaning an avatar can still provide incorrect information. Occasional slowness and timeouts are also mentioned. A convincing face and natural speech do not remove the usual need to check important information when an AI system is handling a sensitive or complex request.
Access is another limitation for most users. Live Avatar is currently available through Gemini Enterprise rather than the standard consumer version. Custom avatar creation requires additional approval. That means the technology may appear first in services built by businesses rather than as a feature people can immediately activate inside their normal personal Gemini conversations.
It is a version of Gemini’s live conversational technology that produces animated video together with speech. The avatar can react visually during a conversation, match its mouth movement to spoken words and process what a user shows through a camera. It is intended for uses such as customer service and interactive assistance. Businesses can use preset characters, and approved enterprise customers can also create custom avatars from reference images.
The system can communicate across 97 supported languages. It can also detect a language change during an active conversation and continue without forcing the user into another session. The animation changes with the speech so that lip movement remains connected to the words being generated. This gives international businesses a way to use one conversational agent across different languages, though accuracy and speech quality can still vary depending on the request.
Yes. Gemini Live Avatar can call connected tools and retrieve information in the background during an active conversation. The avatar does not always need to stop talking until the outside process finishes. A hotel agent, for example, could start checking reservation details and continue speaking with a guest during that process. Once the information becomes available, the system can use it in the same conversation without requiring the person to begin again.
Businesses have access to preset characters and can also seek approval for custom avatars. The custom system uses a reference image to create an animated character that can speak during live conversations. Organizations can use this to keep visual elements connected to their existing identity or characters. Custom generation is not automatically open to every enterprise account. Approval is required before a business can create and deploy an avatar from its own reference material.
The current release is aimed at enterprise users rather than the general public. Organizations can access Live Avatar through Gemini Enterprise and connect it to business services. This makes customer support and interactive assistance some of the most likely places where ordinary users may encounter the technology first. Custom character generation is even more limited because it requires additional approval before a company can create an avatar using its own reference image.
