Google's fake worlds can now talk back — and rewrite themselves live

Google DeepMind announced Genie 4 on October 27, a world model that generates real-time interactive worlds you can walk through with a keyboard — and for the first time, the worlds come with sound built in: music, ambient noise, and characters that talk back and react to what you do, all generated live alongside the picture.

You can rewrite the world while it runs — change the weather, move the camera, drop in a new person — and Gemini adapts your prompt to fit the scene, steering the world between your inputs so it keeps progressing. A second model, Gemini 3.8 Flash, watches the frames and breaks multi-step events into steps, so a world can play out a chain of causes and effects instead of just a pretty scene.

The catch: almost nobody can try it yet. Genie 4 is a research preview for trusted testers, reaching Google AI Ultra subscribers in Project Genie and early Gemini API partners next, and scenes hold together for about two minutes before details start drifting. DeepMind says every clip it makes carries its SynthID watermark.