Google DeepMind has introduced Genie 4, a world-simulation model that generates an interactive scene from a prompt and lets you move through it, steer the camera, and change events with text commands such as adding an object, altering the weather, or introducing a character.

  • Sound: ambient noise, music, and dialogue are generated together with the world instead of added afterwards.
  • Physics: characters react to events, and the model reproduces collisions and other interactions with more realistic motion.
  • Gemini control: Gemini watches the scene, breaks a multi-step event into actions, checks whether each one happened, and adjusts the plan as the simulation runs.

DeepMind names game prototyping, film pre-visualization, and training environments for robots and AI agents as target uses. Genie 4 is an experimental model limited to trusted testers, with wider access planned for Google AI Ultra subscribers through Project Genie and for early-access partners in the Gemini API and AI Studio; a waitlist is open.

Genie 4: A new generation of world simulation

Related: Odyssey Releases Odyssey-3, a Real-Time Foundation World Model, Tencent Open-Sources Hy-Nomas, an Interactive Video World Model