How to Try Gemini 3.8 Live With a Real-Time AI Avatar

Google’s Gemini 3.8 Live can hold a spoken conversation through a real-time animated avatar, and you do not need to build an app to try the basic experience.

The feature is available through Google Cloud’s Gemini Enterprise Agent Platform rather than the consumer Gemini app. Using Google’s Agent Platform Studio, you can select a premade avatar and voice, start a live session and talk with it directly from the browser.

For developers, the same technology can later be integrated into websites, mobile apps and kiosks through the Gemini Live API. But if you simply want to see how Live Avatar works, the Cloud console provides a much easier starting point.

What you need

You will need a Google Cloud account and a Cloud project with billing enabled. Google also requires the Agent Platform API to be enabled for the project. Its Gemini Live API setup documentation walks through those prerequisites if you have not previously used Agent Platform.

Gemini 3.8 Live is generally available and supports standard pay-as-you-go usage. The model is currently available through U.S. and EU multi-region endpoints.

You do not need to install the Google Cloud CLI, Python or an SDK just to test a prebuilt Live Avatar in the Cloud console. Those tools become relevant if you want to build your own application around the model.

How to start a Gemini Live Avatar session

Google provides the following setup path in its Live Avatar documentation:

1. Sign in to the Google Cloud console (console.cloud.google.com) and select the Cloud project you want to use.
2. Open Agent Platform >> Studio >> Stream realtime.
3. Select “Switch model” and choose gemini-3.8-live.
4. In the main panel, select “Live Avatar” and choose one of the available avatars.
5. Select a voice for the avatar.
6. Optionally: you can add system instructions describing how you want the AI to behave. If you want Gemini to see what your camera sees, enable camera input.
7. Select “Start Session” to begin.

Once the session starts, you can speak naturally to Gemini while the selected avatar responds with synchronized speech, facial expressions and lip movements.

Gemini 3.8 Live generates avatar video at 24 frames per second and is designed for continuous voice and video conversations. Google says the model can also react to speech characteristics such as pauses and vocal tone and can filter out some background conversation when deciding whether the user is addressing it.

Try giving the avatar a role

The optional system-instructions field is one of the easiest ways to make the demo more useful.

Instead of starting with a completely general assistant, you could instruct it to act as a hotel concierge, product guide, tutor or customer-service representative. System instructions affect how the model responds without requiring you to write an application.

This is also closer to how Google expects businesses to use Live Avatar. The company is positioning the technology for conversational agents that can appear on websites, mobile apps and interactive kiosks.

Gemini 3.8 Live also supports function calling, allowing a full application to connect the conversation to external services. A hotel agent, for example, could potentially look up reservation information while continuing the conversation rather than going silent while a backend request completes.

That kind of integration requires development work and is separate from simply experimenting with the avatar in Agent Platform Studio.

Can you make a custom Gemini avatar?

Google supports custom Live Avatars generated from a reference image, but this part of the feature is not generally available to everyone.

Custom avatar creation is currently restricted to select customers and requires approval through a Google Cloud account team. Organizations using the feature are also responsible for obtaining the necessary rights and consent for any face or voice samples they provide.

Google’s documentation prohibits using images of minors or celebrities for custom avatars.

For most people testing Gemini 3.8 Live today, the practical option is therefore to use one of Google’s built-in avatars and pair it with a supported voice.

How to build your own app with Live Avatar

Agent Platform Studio is useful for testing, but an avatar created there is not automatically something you can embed on your own website.

For that, developers need the Gemini Live API. It uses a persistent WebSocket connection so audio, video and other information can move between an application and Gemini continuously during a session.

Developers can configure gemini-3.8-live to return video, select an avatar and voice, send microphone or camera input and connect the model to outside tools or business systems.

That makes the basic Studio experience a useful way to test Live Avatar before moving into a custom implementation.

Google also says audio and video generated by Live Avatar carry its SynthID watermark, which is designed to make AI-generated media detectable.

For someone who simply wants to experiment with Gemini’s new visual conversation interface, Agent Platform Studio is the simplest place to start. Building that same experience into a customer-facing service is where the developer tools — and considerably more setup — begin.