DeepMind Genie

DeepMind Genie is a generative interactive environment that creates new, playable worlds from an image or text prompt. It's trained on a large dataset of public internet videos, learning to infer actions and understand game mechanics without explicit instructions. Users can prompt Genie with an image or text, and it will generate a unique, playable 2D platformer world. Users can then interact with this world, controlling a character and exploring the AI-generated environment. This is a significant step towards creating AI agents that can be trained in various simulated worlds. Genie serves as a prototype for future world models, setting the stage for AI that can learn and act in diverse digital realities.