Google's new Genie 4 AI builds worlds you can walk through, with sound and talking people

Google DeepMind says you can steer each scene live with your keyboard and plain words.

October 28, 2026 · 2 min read

Google DeepMind has announced Genie 4, a new AI world model. A world model is an AI program that builds a scene you can move around in. Think of a video game that is made up as you play.

Google says Genie 4 is its first real-time world model to make sound on its own. Each world has its own background noise, from wind in the trees to waves at the beach. People in the scene can also talk and react to what you do.

Users move through a scene with normal keyboard controls. They can also type plain words to change the scene as it runs, like changing the weather or adding a new person or object.

Genie 4 works with Google's Gemini AI. Gemini watches the scene, breaks goals into steps and checks that each step worked.

Google says testers are trying it in film, news, schools, gaming and engineering. Filmmakers can step into a live scene and shoot many takes. Game makers can go from an idea to a test version in seconds.

Every video and sound made by Genie 4 will carry SynthID. That is Google's hidden watermark that helps spot AI-made content.

For now, Genie 4 is a research preview only for trusted testers. Google says it is coming soon to Google AI Ultra subscribers in Project Genie. Early partners using the Gemini API and Google AI Studio will also get it. An API is the connection that lets other software use the model.