Google Genie 2

Google Genie 2 is a large-scale foundation world model designed for researchers and developers working with AI agents in 3D virtual environments. It enables the creation of diverse, action-controllable, and playable environments for training and evaluating embodied agents. With Genie 2, users can generate an endless variety of environments based on a single prompt image, allowing for human or AI agent interaction using keyboard and mouse inputs.
Genie 2 works by generating consistent worlds for up to a minute, with the majority of examples lasting 10-20 seconds. It can be prompted with a single image generated by Imagen 3, GDM’s state-of-the-art text-to-image model, enabling users to describe a world they want in text, select their favorite rendering, and then step into and interact with that newly created world.
Researchers and developers working with AI agents in 3D virtual environments can get the most value from Google Genie 2. It allows for rapid prototyping and the development of more general embodied agents, which can be trained and evaluated in a limitless curriculum of novel worlds. This can lead to new, creative workflows for prototyping interactive experiences and advancing AI capabilities.