Google launches Genie 4, a world model you can control with words

Google DeepMind has launched Genie 4, a world model it calls a major research milestone. It upgrades AI scenes from one-shot clips into a dynamic world you can control, hear, and reason about in real time.

Users can change the simulated world with plain text commands, like shifting the weather, moving the camera, or adding characters. Instead of regenerating a whole video like other AI tools, Genie 4 updates the scene with context, working like an interactive sandbox.

It is DeepMind's first world model to create audio in real time, with sound and visuals generated together. Physics and character movements look more natural. Google CEO Sundar Pichai called it Google's most realistic and interactive world model yet. The team says Gemini will handle text and planning, while Genie 4 turns that into an interactive world.

Genie 4 is now open to trusted testers as a research preview, and will reach Google AI Ultra subscribers and Gemini API partners in the coming weeks. Engineers admit it is still early, and long simulations may show visual drift.