Google DeepMind has released Genie 4 in research preview, calling it its most realistic interactive world model yet. Users can create photorealistic worlds from text or images, explore them live and alter scenes with prompts. Genie 4 also generates music, ambient sounds and dialogue, while maintaining scene consistency for up to two minutes without stopping the simulation.

Google DeepMind released Genie 4 in research preview, calling it its most realistic and interactive world model yet. Users type a text or image prompt and get a photorealistic world they can explore, and they can change it live while it runs.

What is new

Genie 4 is DeepMind's first real-time world model with built-in audio. Music, ambient sounds and interactive dialogue are generated together with the world. Users can type prompts mid-simulation to change the weather or camera angle, or to add a person or an object, while the scene keeps going.

DeepMind says the visuals are built on Gemini Omni and that the physics is better, with more convincing aerodynamics and collisions and more natural human movement. A scene stays consistent for up to two minutes, double the limit of Genie 3, which launched in August 2025.