Google DeepMind’s AI Creates Video Games from a Single Image

Google DeepMind specialists have introduced a relatively small model with 11 billion parameters. After reviewing 200,000 hours of videos of people playing two-dimensional platform video games, it deduced the rules of the game and the physical laws of the flat world, despite the recordings lacking information on which buttons were pressed by the players. But the most intriguing aspect here is the trend. AI has learned to transform photographs into games.

The fully trained Genie model takes a single photo, sketch, or AI-generated illustration and transforms it into a two-dimensional video game that responds to button presses. The creation of graphics and a rudimentary interactive environment happens simultaneously, in one step.

“We introduce Genie, a base world model trained exclusively on videos from the internet, which can generate an infinite number of controllable 2D worlds on demand from images,” wrote Tim Rocktäschel, head of the Open-Endedness team, on X.

At a scrutinizing glance, the quality of Genie’s games is somewhat lacking, but it’s important to remember that this is merely a research project, not a final product. The AI was trained on videos with very low resolution and frame rate, so the games generated by the neural network lack sharpness and smoothness. However, Genie serves as proof of concept. Given enough high-quality video footage and computational power, the results of its efforts would also improve.

Actually, it’s not just about Genie, but about the trend of integrating AI models—transforming text into images, videos or music, images into videos—beginning to integrate into three-dimensional worlds. The implications could be colossal, not only for the video game industry but for entertainment as a whole: high-level visual effects, interactive virtual worlds, personalized musical accompaniment, and so on.

Game artists in China are losing their jobs en masse due to generative artificial intelligence systems. Game companies, using AI image generators, can create an illustration in seconds over which an artist would labor for a week. As a result, the number of illustrators in the industry has decreased by about 70% over the last year, which translates to thousands of jobs lost.

Leave a Reply

Your email address will not be published. Required fields are marked *