Deepmind Introduces Genie – An Artificial Intelligence That Creates 2d Game Environments From A Single Image

Home AI Projects Deepmind Introduces Genie – An Artificial Intelligence That Creates 2d Game Environments From A Single Image

DeepMind experts in collaboration with researchers from the University of British Columbia presented Genie, a revolutionary artificial intelligence tool designed to transform a single image into a fully interactive two-dimensional gaming environment. This innovative project was detailed in a paper available on the arXiv preprints server, accompanied by a detailed update on DeepMind’s research website.


Classic side-scrolling games like Super Mario Brothers involve players guiding a character through various layers of a digital landscape. DeepMind’s initiative revolutionized this concept, giving Genie the ability to create entire game levels from just one image, such as the main character against a fictional setting. This achievement was realized through intensive AI training that included countless hours of gameplay from a wide range of 2D games.


The initial step in the development of Genie involved the creation of an artificial intelligence system capable of breaking down video frames into millions of individual elements that could be assembled into new scenarios. The team then integrated a “hidden action model” to predict potential future scenes derived from the initial image. A further improvement included the inclusion of a dynamic prediction module, improving the AI’s ability to make predictions about subsequent frame sequences based on its prior learning. The culmination of these efforts is a sequence of frames that seamlessly blend to create a virtual two-dimensional game sphere.


Despite the promising progress, Genie’s creators admit that their creations are still in development, facing several challenges. For example, the processing time is significantly slower – 20-30 times more than normal game speed. Additionally, the AI occasionally generates unrealistic scenarios that cannot be recreated and is currently limited to creating sequences of 16 frames at a time. Still, the DeepMind research team sees Genie as a significant step forward in video game production, offering players a unique way to modify their personalized gaming experience.