Project Genie

Project Genie is a Foundation World Model developed by DeepMind that automatically generates interactive virtual environments by learning from unlabeled visual data such as gameplay videos and real-world scenes. Its core capabilities include inferring dynamic scenes with physical interaction logic from a single image or video frame, and supporting user control of character behavior within generated environments via action commands. The project aims to provide a scalable virtual interaction platform for training general-purpose AI agents, representing a significant research achievement at the intersection of generative AI and reinforcement learning.