Last week, Google DeepMind announced Genie 2, a new video model that generates plausible, consistent, and playable 3D environments based on prompt images.
According to DeepMind, Genie 2 comes with a host of new and exciting features that improve lighting and reflections, and can even generate videos from real-world images. Within these generated worlds, models can also create animated characters that act as embodied agents for training purposes. Characters can interact with the world by popping balloons, opening doors, and even interacting with non-playable characters.
The DeepMind team seems to be hoping that these AI-generated video games will help train agents. “Genie 2 demonstrates the potential of fundamental world models to create diverse 3D environments and accelerate agent research,” they wrote in a recent post. . Video games are useful tools for AI research. These are interactive and have a unique combination of different challenges. They are also a safe playground for training, testing, and measuring agents that may be used in real-world products.