Genie 4

Google DeepMind's next world model, built on Gemini, and the first real-time world model to generate its own sound: music, ambient noise and dialogue arrive with the picture, and the people in a scene move, react and talk back. You walk and turn the camera with the keyboard and steer the world in plain language as it runs, changing the weather or adding people and objects, while Gemini adapts each prompt to the scene and keeps it coherent between inputs. A Gemini agent can also work inside it, breaking a goal into steps, checking each one and replanning when the world changes.

DeepMind named filmmakers directly. Creatives were testing it as a "draft mode" for previz, stepping into a live simulation of a scene to run multiple takes and adjust camera angles, lighting, dialogue and action on the fly, and the launch prompts asked for 16mm documentary grain and a 1995 Hong Kong indie look. DeepMind was candid about the limits: details and object permanence drift over longer sessions or after several edits. It launched as a research preview for trusted testers, coming to Google AI Ultra subscribers in Project Genie and to early partners through the Gemini API, with every frame and sound watermarked with SynthID. A little over a year after Genie 3 made the world model a silent sandbox, Genie 4 pitched it as a soundstage.