oogle is widening the world of world models.
On Thursday, the company’s AI unit released Project Genie, an “experimental research prototype” that lets users create and explore virtual worlds and video-game like environments. Currently available to Google AI Ultra subscribers in the US, the system is powered by Gemini, Nano Banana Pro and Genie 3, its most powerful world model yet, released in August.
Project Genie offers a number of features for users to play around with, including:
- World sketching, in which users can prompt the model with text and images to create a “living, expanding environment;”
- World exploration, allowing users to navigate these developed worlds by generating paths ahead in real time based on user actions;
- And world remixing, letting users build on top of existing prompts to edit and curate the world to their liking.
Google admitted that there are still some bumps to be ironed out with Project Genie, noting that generated worlds might not be completely true to life or adhere entirely to real-world physics. Generations are also limited to 60 seconds, and characters aren’t always entirely controllable.
However, the experimental model is the latest step in its broader journey to “support [it’s] AGI mission,” the company said. “Building AGI requires systems that navigate the diversity of the real world.”
Google’s release comes as world models receive more and more attention. Last week, Bloomberg reported that World Labs, a world model startup founded by Fei-Fei Li, Ph.D., is in talks to raise hundreds of millions at a valuation of $5 billion. AMI Labs, a world model firm founded by Yann LeCun, is also rumored to be discussing a funding round that would notch the company a $3.5 billion valuation.
Capable of reflecting the real world physics and consequences, many are eyeing world models as the key to achieving AGI, rather than simply scaling large language models to as many parameters as possible.