Genie by Deepmind

Genie by Deepmind is a foundation model for playable worlds, designed for creators and researchers looking to generate interactive environments from images or text. It learns to control without action labels, making it a stepping stone for generalist agents. Genie can be prompted with unseen images, such as real-world photographs or sketches, to create new interactive worlds.
Genie works by learning fine-grained controls exclusively from internet videos, without requiring action labels. This allows it to infer diverse latent actions consistent across generated environments. With Genie, users can create entire new interactive environments from a single image, opening doors to new ways of generating and interacting with virtual worlds.
Creators and researchers in the field of AI and virtual worlds will get the most value from Genie by Deepmind, as it enables them to generate interactive environments for training AI agents, testing new ideas, and exploring the possibilities of generative virtual worlds. This can lead to breakthroughs in areas like AI research, game development, and simulation-based training.