Google’s Project Genie opens the door to AI-generated playable worlds

Google

• Google unveils Project Genie for AI-generated worlds, limited US access.
• Users describe worlds, choose movement, and interact in 60-second sessions
• Project Genie to enhance AI world models for research and future uses.

Did our AI summary help?

Google has begun rolling out access to Project Genie, a new experimental experience that allows users to generate and explore playable AI-created worlds. The feature is powered by Genie 3, described as a general-purpose world model capable of producing diverse, interactive environments in real time. Access is currently limited to AI Ultra subscribers in the US aged 18 and above, with wider availability planned for the future.

At its core, Project Genie is built on the idea of a world model. Unlike traditional AI systems that operate within narrowly defined environments, a world model attempts to simulate how an environment behaves, how it changes over time, and how user actions affect it. According to Google DeepMind, such models are a key step towards building systems that can handle the complexity and unpredictability of the real world, which is central to its long-term AGI ambitions.

Project Genie begins with a simple but flexible setup. Users are prompted to describe the world they want to create, including how they wish to explore it. Movement options range from walking and driving to flying or riding, and users can choose between first-person or third-person perspectives. They are then asked to define their character, which can be a person, an animal, an object, or something more abstract.

Before entering the world, Project Genie generates a preview image through a process Google calls World Sketching. This preview is created using Nano Banana Pro and allows users to fine-tune the look of their environment before committing to it. Once satisfied, users can create the world itself, although sessions are currently limited to 60 seconds at a time. The generated environments run at 720p resolution, with interaction rates of around 20 to 24 frames per second.

As users move through the environment, Genie 3 generates the path ahead in real time, adapting the world based on their actions. Google says the system simulates physics and interactions with what it calls breakthrough consistency, although it also acknowledges clear limitations. Generated worlds may not always look fully true to life, may not strictly follow prompts, and may sometimes behave inconsistently with real-world physics. Character control can also feel less precise, with occasional latency during interaction.

Project Genie also includes a Remix Worlds feature, allowing users to build on existing worlds by reusing and modifying prompts. Users can explore curated worlds through a gallery or use a randomiser for inspiration, then remix those ideas into new interpretations. Walkthroughs can be recorded, with videos available for download.

Google says Project Genie is an important research tool that will help it better understand how people interact with world models across AI research and generative media. Beyond entertainment, the company sees practical applications in areas such as robotics, animation, fiction, and even the exploration of historical or inaccessible locations. While still clearly experimental, Project Genie offers an early glimpse into how interactive, AI-generated worlds could play a role in the future of intelligent systems.