Google launched Gemini 3.8 Live with Live Avatar on September 24, 2026. It gives a live voice agent a real-time animated face with lip sync, and it works in 97 languages.
What you need
- Access to Gemini Enterprise on Google Cloud. That's where the feature lives.
- It runs on US and EU endpoints.
- You can use preset avatars from a curated library. Custom avatars built from your own reference image need enterprise allowlisting and verification.
Try it in 5 steps
- Open the live studio in the Google Cloud console:
console.cloud.google.com/agent-platform/studio/multimodal-live - Pick a preset avatar from the library.
- Turn on your camera or share your screen. It takes in live video and audio together.
- Talk in any of the 97 languages. It detects the language automatically.
- Interrupt it mid-answer to see how it recovers.
Five things to test
| Test | What to look for |
|---|---|
| Switch languages mid-conversation | Lip sync and expressions stay aligned |
| Cut it off mid-sentence | It keeps the conversation context |
| Hold something up to the camera | It reacts to what it sees |
| Share your screen | It can talk about what's on it |
| Ask for data from a connected tool | It keeps talking while the tool runs in the background |
Build with the API
- Start with the Live API docs:
docs.cloud.google.com/gemini-enterprise-agent-platform/models/live-api - Avatar settings are on the Configure live avatars page in the same docs section.
- The Live API streams voice and video in real time over WebSockets.
- Check prices before you ship:
cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing
Know the limits
- Custom avatars are gated. Plan your first build around a preset avatar.
- All generated audio and video carries an invisible SynthID watermark, so viewers and tools can detect that it's AI-generated. Tell your users they're talking to an AI.
- Provisioned throughput is available if you need guaranteed capacity.