Genesis is a physics engine that combines its own simulation capabilities with a Language Model, opening up a new approach in video generation.
Unlike previous efforts that use text-to-video generative AI to understand and simulate the physical world, Genesis leverages the concept of Generative Simulation proposed by Zhou Xian in an earlier paper. This paradigm is a way to interact with the engine through language to generate scenes, character motions, trajectories, camera motions, and physically-accurate videos.
Project page: https://genesis-embodied-ai.github.io/