We’ve all seen AI generate images and progressively improve at five-fingered hands. We’ve also all seen the incredibly realistic videos that AI produces, too, but how good is it at making games?
The team behind Google’s DeepMind have created a new interactive world model that can now produce 2D, playable world maps.
Called Genie, the model can be prompted to generate an endless variety of action-controllable virtual worlds with prompts from text or images. Taught with over 200,000 hours of unlabelled internet videos, Genie can generate these worlds and allows actionable input on a frame-by-frame basis. This means Genie will generate a new frame depending on the action.
Research Scientist at DeepMind, Tim Rocktäschel, posted on X the other day, showing off some of the worlds that Genie has been able to create.
I am really excited to reveal what @GoogleDeepMind‘s Open Endedness Team has been up to 🚀. We introduce Genie 🧞, a foundation world model trained exclusively from Internet videos that can generate an endless variety of action-controllable 2D worlds given image prompts. pic.twitter.com/TnQ8uv81wc
— Tim Rocktäschel (@_rockt) February 26, 2024
Rocktäschel explains a bit more about how they taught Genie to learn actions, showing that Genie is not only restricted to generating 2D worlds but can possibly be used to train robotics in the future.
For now, Genie is still in its “early stages”, and it’s not likely that we will be able to test it out for ourselves, as it was designed for research and not for public use.
With AI picking up so much speed so quickly, though, I’m pretty sure we’ll be seeing something similar quite soon. Perhaps even a fully AI-generated game with cutscenes, custom audio and playable worlds.