DeepMind introduced Genie, an AI for creating 2D games from a single image

DeepMind, the artificial intelligence division of Alphabet (Google's parent company), has demonstrated a new development called Genie. This AI-based application is capable of generating fully playable 2D worlds based on just one image.

Here's What We Know

Together with scientists from the University of British Columbia, DeepMind researchers trained the neural network on thousands of hours of video from hundreds of 2D games. As a result, Genie learned to analyse images, highlight objects and parameters on them, and then create sequences of game scenes connected visually and physically.

Thus, after loading an ordinary picture, for example, with a character on the background of a location, Genie independently generates a full-fledged virtual 2D-world with its own physics and game mechanics around the elements. Users will potentially be able to create their own games by simply drawing the original image.

At this stage, however, Genie has limitations. The app is slow, can generate unrealistic unplayable levels, and processes 16 frames at a time.

Nevertheless, DeepMind considers Genie an important step in automating video game development with artificial intelligence. The company plans to continue developing the technology.

Source: TechXplore