█

AI/TLDR

Google DeepMind · 2026-11-19 · major

Genie 4 in the Gemini API — developers can stream playable 720p worlds

In short

Genie 4, Google DeepMind's world model, is now in the Gemini API through the Interactions API and in Google AI Studio. It streams 720p, 24 fps worlds with native audio that users steer with keys and text prompts.

Google DeepMind · 2026-11-19 · 2 sources checked

Genie 4 social share artwork from Google's Gemini API announcement

Google opens its Genie 4 world model to developers, one day after the consumer launch.

Quick facts

Model idmodels/genie-4
AccessGemini API (Interactions API), Google AI Studio
Output1280x720, 24 fps, 48 kHz stereo audio
Session cap120 seconds
Billing240 video tokens per frame, paid-tier key
InputsText prompt, one 16:9 image, keyboard and camera actions
WatermarkSynthID on every output

What is it?

Genie 4 is now available to developers through the Gemini API's Interactions API and in Google AI Studio, one day after it launched in Project Genie. Apps can start a world from a text prompt, a reference image or both, then steer it live.

How does it work?

A backend creates the interaction and mints an ephemeral token; the client connects over WebRTC or WebSocket, sends a world_setup event, then streams keyboard actions and text inputs that change weather, lighting, objects or events. Google says Gemini reasoning keeps the world state consistent.

Why does it matter?

World models were only something people could try in Google's own apps. With per-second API billing, developers can now prototype interactive experiences, visualize spatial designs and build environments to train robots and AI agents.

Who is it for?

game, simulation and robotics developers

Frequently asked questions

How much does Genie 4 cost in the Gemini API?
Genie 4 needs a paid-tier Gemini API key. Google's docs say it bills 240 video output tokens per frame, which is 5,760 tokens per second at 24 fps, and put the cost at roughly $0.10 per second with stereo audio included. Calling complete() ends the session and stops billing.
How long can a Genie 4 API session run?
A Genie 4 session in the Gemini API has a hard cap of 120 seconds. Google's docs warn that visual stability may degrade after 60 to 90 seconds and that object permanence is short-horizon: looking away for eight seconds or more can change the scene.
How is Genie 4 in the API different from Project Genie?
Project Genie, launched on November 18, is a consumer experience for Google AI Ultra subscribers. The Gemini API release lets developers build Genie 4 into their own apps, connecting over WebRTC from browsers or WebSocket from Python and Node.js, with ephemeral tokens minted by their backend.
What can't the Genie 4 API do yet?
Google's Genie 4 docs list no portrait 9:16 output, no configurable frame rate, no live microphone or webcam input, no function calling, no Search grounding, no structured outputs and no session resumption. Prompts with real people, public figures or brand IP can be blocked.

Try it

https://aistudio.google.com/live?model=genie-4

Sources · 2 outlets

Tags

  • genie-4
  • world-model
  • gemini-api
  • google-ai-studio
  • google-deepmind
  • interactions-api
  • synthid

← All releases · Learn AI