Developed by Google DeepMind, Project Genie is built on the company’s latest “world model” research, including the Genie 3 system first previewed last year.
Unlike traditional generative tools that produce static images or pre-rendered scenes, Google has said Genie generates environments dynamically in real time, predicting how worlds evolve as users move and interact within them.
Access to Project Genie is currently limited to U.S.-based Google AI Ultra subscribers aged 18 and over, with Google describing the release as an early research prototype rather than a finished consumer product.
The tool is delivered through a web app and combines Genie 3 with other Google models, including Gemini and Nano Banana Pro.
At its core, Project Genie offers three main capabilities - those being world sketching, world exploration and world remixing.
Users can sketch worlds using text prompts or images, defining characters, environments and perspectives before entering the scene.
Once inside, those worlds are fully navigable, with the system generating new terrain and interactions on the fly as the user explores.
Existing creations can also be remixed, allowing people to build on shared prompts or curated examples, and export short videos of their worlds.
Google positions the project as part of its longer-term push toward more general-purpose AI systems that can model complex, real-world dynamics.
World models like Genie 3 are seen as a potential foundation for applications ranging from robotics and simulation to game design, animation and historical exploration.
The company has been careful to stress the technology’s current limitations, as generated environments may not always follow real-world physics, character control can be imprecise, and individual sessions are capped at 60 seconds.
Some features demonstrated in earlier research, such as prompt-triggered events that alter the world mid-exploration, are not yet included.
Project Genie sits within Google Labs and is intended to gather feedback on how people might use interactive world models in creative and research contexts.
Google has said access will expand to additional regions over time, with the longer-term goal of making the technology available to a broader audience once the underlying models mature.
