Published , by Sam Chandler
Published , by Sam Chandler
Google is one of the many major companies behind a push toward utilizing artificial intelligence to do things that were traditionally human-led activities. The latest development comes out of its DeepMind wing from its Open Endedness Team. The team has managed to create an AI model that can generate 2D platformer worlds that are, somewhat, playable.
On February 26, 2024, Tim Rocktäschel took to X (formerly Twitter), to share that the Open Endedness Team (part of Google DeepMind), has created an AI tool called Genie that can generate 2D playable platformers.
According to Rocktäschel, Genie has been trained using a dataset of 200,000 hours of videos of 2D platformers. This has allowed the AI model to generate playable 2D worlds from images, be it sketches or full artwork. However, one user posed the question of whether the gifs shown in the thread are actually happening or spliced together, to which Rocktäschel responds, “The model is currently running at 1 FPS, so right now it's far away from real-time playable.”
While this might not be functionally playable this year, the advancements of artificial intelligence over recent years have been dramatic. Just recently, OpenAI revealed Sora, a text-to-video AI model that is able to generate some startlingly realistic videos, even taking into account the strangeness that always accompanies AI.
Genie might be in its infancy stage right now, but there’s no knowing how it might function a few months from now, let alone a year. How AI tools like Genie and Sora continue to impact creativity and the art of movies, photography, and video games remain to be seen. Take a look at our AI page for more insight into what the various companies are doing in this new sector.