Google's model will create a whole world for you – depending on whether you're willing to pay

Google DeepMind unveils Genie 4, an interactive world model that generates real-time 3D environments with realistic physics, built-in sound, and keyboard control. The system is integrated with Gemini models and allows free navigation, dialogue with characters, and changing the laws of physics. Access will be opened to subscribers of the Google AI Ultra premium package.

Google DeepMind's lab has unveiled Genie 4, a new generation of interactive world models. Unlike passive video generators, Genie 4 produces a live 3D simulation where the user can navigate with keyboard and mouse, verbally interact with characters, and change the laws of physics in real time. The model generates built-in sound including background noise and voice dialogues, and is based on a physical understanding of fluid dynamics, aerodynamics, and collisions. Google has connected Genie 4 to the Gemini 3.8 Flash and Gemini 4 Argon language models, which serve as an 'agent conductor' that breaks down goals into a sequence of logical actions. In a demonstration, the model guided Apptronik's humanoid robot Apollo to play checkers against a human. Google acknowledges limitations: object drift in long sessions, difficulty producing readable text, and computational inaccuracies in complex interactions. Access will be exclusive to subscribers of the Google AI Ultra package under Project Genie.