Google DeepMind Launches Project Genie For AI-Generated Worlds, Try It Here

Google DeepMind's Project Genie is a research prototype that allows users to "create, explore, and remix their own interactive worlds" based on text or image prompts. Regardless of what is used to generate the world, controls are limited to WASD, arrow keys, and the space bar, and the output will be capped to 720p resolution and 24 FPS, but the end result is still eye-opening.

While the limitations and long-term industry implications are worth scrutinizing, this is still an impressive technical achievement. This is especially true when compared to AI-generated Doom or Minecraft clones that have almost no persistent tracking of environmental spaces at all. In comparison, Project Genie has a much more convincing sense of space and consequence, and its high-fidelity graphics even allow it to generate playable facsimiles of AAA games like Elden Ring or Legend of Zelda: Breath of the Wild when potential copyright issues don't kick in and stop you.

While the gameplay of such experiences is limited by various factors, Project Genie seems to have been trained on a massive dataset of existing digital worlds and is able to mimic existing titles fairly well. When imitating Breath of the Wild and jumping off a cliff, for example, Link will automatically deploy a paraglider, even though that requires a specific button press in the original game.


While hardcore gamers and some developers are unlikely to look upon this technology favorably, there's huge potential here. Genie 3 can use the input of real-life photographs to create playable (or explorable) 3D spaces, which has interesting implications for 3D scanning, VR/AR gaming, and education. With the ability to modify or alter the world with additional prompts rather than just exploring them, Genie can be compared to OpenAI's Sora, while providing a much more interactive experience.

In the long run, Genie, Sora, and other powerful AI video, 3D model, and game generation tools raise all kinds of questions, from sustainability to legality. While the field is still relatively young, Google is making some impressive steps forward, now flaunting perhaps the most impressive GenAI world modeling system yet. If you want to try it out for yourself, you need only head to the official DeepMind Genie web page, though you'll need to pay for Google's AI Ultra plan to generate your own worlds. Free access is limited to interactive with the available demos.