Google Genie 4 is a new AI "world model" from Google DeepMind, announced on October 27, 2026. It generates interactive worlds you can move through and change in real time, and it is the first of its kind from Google to generate audio natively, including music, ambient sound and dialogue. It is in research preview for trusted testers. Details below come from Google DeepMind's announcement.
What is a world model?
A normal video model makes a clip you watch. A world model makes a scene you can act inside. You move the camera, interact with objects or characters, and the model generates what happens next, frame by frame. That makes it useful for games, film planning and training robots.
What is new in Genie 4
- Native audio. Worlds carry their own ambient sound and music.
- Dialogue. Characters hold conversations that react to what you say and do.
- Real-time steering. Type changes such as new weather, a camera move or an extra person, and the scene adapts.
- Better physics. It simulates aerodynamics and collisions more convincingly.
- More natural people. Characters move, react to their surroundings and can be steered live.
- More photorealism and smoother movement.
How Google Genie 4 uses Gemini
Genie 4 is built on Gemini capabilities. Gemini Omni keeps visuals and behaviour grounded in the real world. During generation, Gemini 3.8 Flash acts as a guiding agent. It watches the scene, breaks your goal into steps, checks each new frame and adjusts the plan when you step in.
DeepMind also showed experiments pairing Genie 4 with Gemini 4 Argon. In one demo, Apptronik's Apollo humanoid robot plays checkers inside a simulated world.
What Genie 4 could be used for
| Use | Example from DeepMind |
|---|---|
| Film and creative work | Running several "takes" of a scene in a live simulation |
| History and journalism | Turning one archival photo into an explorable scene, or visualising a flood |
| Game design | Going from an idea to a playable prototype with plain language |
| Business simulation | Generating test scenarios quickly |
| Robotics and AI agents | Training environments for robots and agents |
Limits to know about
DeepMind says details and object permanence can drift over time, especially in long sessions, after many edits or while moving around. Complex physics, accurate real-time audio, readable text, spatial accuracy and long-term consistency are all still being worked on. Treat it as an early research tool, not a finished product.
Safety and watermarking
Genie 4 follows Google's safety policies for its other models and was tested internally before the preview. All video and audio it generates carries SynthID watermarking, so its output can be identified as AI-made.
How to get access to Genie 4
- Now: research preview for trusted testers only.
- Coming soon: Google AI Ultra subscribers through Project Genie.
- Coming soon: early access partners in the Gemini API and Google AI Studio.
- Waitlist: join the Project Genie waitlist through the form linked in DeepMind's post.
Frequently Asked Questions
Is Genie 4 available to the public?
No. It is a research preview for trusted testers, with Google AI Ultra subscribers and API partners next.
Does Genie 4 generate sound?
Yes. DeepMind calls it its first real-time world model to generate audio natively, including dialogue.
Is Genie 4 output watermarked?
Yes. All Genie 4 video and audio includes SynthID watermarking.
Summary
Google Genie 4 creates interactive worlds with sound and dialogue that you can steer by typing, guided by Gemini models. It is aimed at film, games, simulation and robotics. It is still a research preview with consistency limits, and wider access is coming through Project Genie and the Gemini API.
Related articles
We cover gaming, software, tech, AI and trading. Every story links to its primary sources, and rumours are labelled as such. Spotted a mistake? Let us know and we will correct it.
Comments
Loading comments…