Google Genie 4 Explained: Worlds You Can Walk Through and Change as You Go

Google DeepMind announced Genie 4 on October 27, 2026. It is a world model, which is a different kind of AI from a chatbot. Instead of only writing an answer, it builds a scene you can move through in real time, with sound, movement and characters that react.

What is a world model, in plain words

A world model predicts what comes next in a scene, frame by frame. You give it a starting point, then you move with keyboard controls while it keeps drawing the world around you. A chatbot predicts the next word. A video generator makes a clip you watch. A world model makes a place you steer. That is the core difference, and it is why people in games, film and robotics are paying attention.

What Genie 4 adds

DeepMind says Genie 4 is more realistic than its predecessor, with more natural human movement and physics for things like collisions. It also generates sound with the scene: background noise, music and characters that speak, instead of adding audio later. You can also steer a scene while it runs — change the weather, move the camera or add a character by typing what you want. Gemini models help here: one part grounds the scene in how things look and behave, while other models help keep multi-step events coherent as the scene goes on.

Who can use it right now

Not everyone, yet. DeepMind says Genie 4 is rolling out as a research preview for trusted testers first. After that, it is planned for Google AI Ultra subscribers in Project Genie and for early-access partners through the Gemini API and Google AI Studio. If you see a waitlist or demo link, check that it comes from DeepMind or Google before you sign up anywhere.

What it still gets wrong

DeepMind is open about the limits. Details can drift over time, long scenes can lose consistency, and text inside a scene may not read clearly. A generated world also is not a finished game: there is no built-in scoring, saving or multiplayer, and the rules are whatever the model happens to keep consistent. Treat the demos as a research preview, not a product you can buy today.

Why it matters even if you never play a game

The bigger use may be training. Robots and AI agents can practise in a simulated place where mistakes cost nothing. Filmmakers can sketch a scene before a shoot. Teachers can walk students through a place that does not exist. That is a long way off for most people, but Genie 4 shows the direction: AI that builds places, not just text.

Genie 4 is not something to download today, and it is not a replacement for games made by people. It is a research step worth watching, because it changes the question from “what can AI write?” to “what world can AI build around you?”

Sources: DeepMind: Genie 4 and Google Blog: Introducing Genie 4. Related: Free AI Image Generators Compared and What Are AI Agents Explained.