Article

Google's Genie 4 world model is now available to Ultra users: the AI world can be played with and modified on the fly, and shared with others

Beating AI Flash News: Google DeepMind has integrated the next-generation Genie 4 world model into Project Genie and begun rolling it out to eligible Google AI Ultra subscribers. Project Genie is a web app launched by Google this January that lets users create virtual worlds via text or images, then control characters with a keyboard to freely explore AI-generated scenes in real time.

Genie 4 adds native audio support, generating ambient sounds, music, and character dialogue in real time as the visuals change. Users can also input text during exploration to alter the weather, add characters, or trigger events at any time, allowing the virtual world to evolve according to their vision.

Project Genie has also introduced a new AI creation assistant. Previously, users had to manually describe scenes, adjust initial visuals, and set viewpoints; now they can design worlds directly through chat, letting AI automatically generate initial scenes and background music. Once inside a world, Gemini can continuously analyze the visuals, recommend next steps for exploration, and even automatically trigger events within the scene.

The previous Genie 3 version has also been upgraded—the maximum exploration duration per session has been extended from 60 seconds to 3 minutes—and multiple virtual worlds can now be linked together for continuous exploration across different settings. Users can also upload 360-degree panoramic photos, enabling AI to generate explorable spaces based on real-world environments.

Project Genie has simultaneously added a community asset library, where users can share the worlds they've created. Others can directly enter to explore, or modify scenes, music, and rules to create their own versions.