Google DeepMind Creates 3D Worlds! Last week, Google DeepMind introduced Genie 2, a new video model capable of generating believable, coherent, and playable 3D environments from a prompt image.
DeepMind states that Genie 2 possesses a variety of new and exciting emerging features that enhance lighting and reflections, and it can even produce videos from real-world images. In these generated environments, the model is capable of creating animated characters that serve as embodied agents for training. These characters can interact with the environment by performing actions such as popping balloons, opening doors, and engaging with non-playable characters.
The DeepMind team appears optimistic that these AI-generated video games will aid in training agents. "Genie 2 demonstrates the potential of foundational world models for generating diverse 3D environments and advancing agent research," they stated in a recent post.
Video games serve as a valuable resource in AI research. They offer interactivity and a distinctive mix of challenges. Additionally, they provide a secure environment for training, testing, and evaluating agents that could eventually be integrated into real-world products.
Comments