Google has taken its generative art capabilities to the next level with the release of Genie 2, a system that can create playable 3D game worlds entirely from text prompts. The technology builds upon its predecessor, Generative Interactive Environments (Genie 1), which could generate 2D environments.
To create a 3D environment, Genie 2 requires a starting image prompt, which can itself be generated by an ordinary text prompt. This means users can ask for specific types of games or environments, such as a cyberpunk Western or a sailing simulation. The system uses this prompt to generate a world with animations, physics, and object interactions.
During gameplay, the user or AI player can interact with the environment, and Google’s demo showed a traditional WASD setup with arrow key alternatives. However, the model loses coherence after around 20 seconds due to its ability to generate “counterfactuals” – alternative paths and actions within the environment.
Genie 2 can accommodate different perspectives, including isometric views, driving videos, or first-person experiences. The system also models complex interactions with the environment, such as water effects and object physics. However, details on computing resources, release plans, and potential commercialization remain unclear.
Source: https://www.pcworld.com/article/2545516/google-can-generate-playable-3d-game-worlds-from-ai-prompts.html