Project Genie is an experimental research prototype by Google, rolling out to Google AI Ultra subscribers in the U.S. (18+), designed for creating, exploring, and remixing interactive, infinite worlds. It is powered by Genie 3, Nano Banana Pro, and Gemini, and represents a significant step in advancing general-purpose world models.

The core methodology revolves around Genie 3, a general-purpose world model developed by Google DeepMind. Unlike conventional systems that might rely on static 3D snapshots, Genie 3 is engineered to simulate environmental dynamics, predict their evolution, and understand how actions affect them. Its key technical capabilities include:

• Real-time Path Generation: As users navigate and interact within a generated world, Genie 3 dynamically generates the path ahead in real time, ensuring a continuous and responsive exploration experience rather than pre-rendered or fixed environments.
• Physics Simulation and Interactions: The model simulates physics and interactions within dynamic worlds, allowing for realistic environmental responses to user actions and the behavior of objects within the scene.
• Breakthrough Consistency: Genie 3 is noted for its "breakthrough consistency," enabling the robust simulation of diverse real-world scenarios, making it applicable to domains ranging from robotics and animation to fiction and historical simulations. This consistency is crucial for maintaining coherence and believability across expansive, dynamically generated environments.

Project Genie, as a web application, integrates these capabilities into three core user experiences:

1. World Sketching: This capability allows users to initiate world creation using text prompts and either generated or uploaded images. The system produces a "living, expanding environment." Users can define their character, preferred exploration mode (e.g., walking, riding, flying, driving), and camera perspective (first-person or third-person). For enhanced control, "World Sketching" integrates Nano Banana Pro, which provides precise control over the initial world generation. Nano Banana Pro enables users to preview the world's appearance and modify input images to fine-tune the environment prior to entering it.

2. World Exploration: Once a world is sketched, it becomes a navigable environment. As the user moves, Genie 3 continues to generate the surrounding path in real time, adapting to user actions and maintaining a continuous, interactive experience. Users can also adjust the camera during traversal.

3. World Remixing: Users can remix existing worlds by building upon their prompts, allowing for iterative creative processes and new interpretations. The platform also offers a gallery of curated worlds and a randomizer for inspiration, which users can build upon. Completed explorations can be downloaded as videos.

The prototype, hosted in Google Labs, acknowledges several limitations inherent to its experimental nature. Generated worlds may not always achieve complete photo-realism, adhere strictly to prompts or real-world physics, or characters might exhibit reduced controllability or higher latency. Current generations are limited to 60 seconds, and certain advanced Genie 3 capabilities, such as promptable in-world events, are not yet implemented in this prototype. Google aims to gather user feedback to improve the experience and eventually expand access to this technology.