Google DeepMind unveils Genie 2 for generating interactive 3D worlds from images

Google DeepMind recently introduced Genie 2, a new video model that creates playable 3D environments from prompt images. This model enhances lighting and reflections and can generate videos from real-world images. Genie 2 can also create animated characters that interact with the environment, performing tasks like popping balloons and opening doors. This development aims to support training for AI agents. The use of video games in AI research provides a safe and interactive way to test and measure agents. DeepMind highlights Genie 2's potential to advance research in diverse 3D environments.