Subscribers to Google's premium AI tier can now access Genie 4, the company's latest generative model for crafting and exploring interactive digital environments. The rollout brings the technology to eligible Google AI Ultra members through Project Genie, a web-based interface that lets users build and navigate simulated worlds. Google AI Pro users have the option to join a waitlist for access.
The new version fundamentally changes how users interact with generated content. Rather than creating a static scene, Genie 4 allows real-time modification of environments as exploration happens—a typed prompt can alter weather patterns, introduce new characters, or steer narrative direction. Sound design is now native to the system, encompassing everything from ambient noise to spoken dialogue by characters within the world.
Google DeepMind first demonstrated Genie 4 to test audiences in October. This latest release introduces two significant additions: an AI agent capable of constructing worlds through conversational input, and a community gallery enabling users to publish, discover and adapt each other's creations. The platform already supported some users in exploring simulated urban environments derived from Street View imagery.
The previous iteration, Genie 3, remains in service with an extended runtime. Sessions can now stretch to three minutes per generation, a substantial increase from the original 60-second limit. However, Google acknowledges that both versions remain research prototypes with notable limitations. The company notes that visual coherence degrades over extended sessions, and the systems struggle with rendering legible text and simulating complex physical interactions.
All content generated within Project Genie receives SynthID, Google's imperceptible watermarking system designed to identify AI-generated material. Meanwhile, the broader world-model sector continues attracting substantial investment. AMD announced in September its acquisition of World Labs, the startup founded by Fei-Fei Li, for $8.2 billion. Google DeepMind suggests these models could serve as training environments for robotic systems and autonomous agents before real-world deployment. Competing research organization Odyssey has already demonstrated that its Odyssey-3 model can direct the operation of robots and vehicles.
The gaming industry has maintained cautious distance from generative AI tools. Take-Two Interactive stated that Grand Theft Auto 6 was developed without relying on any generative AI technology.
Source: The Next Web



