Last week, Google DeepMind announced they would be releasing access to Genie 3, a multi-purpose generative world model first announced in August last year, to Google AI Ultra subscribers in the US. That means our resident American and AI Expert Mike has been able to try it out, and here’s some of the insights our team found after testing:
Video Created using Genie 3 showing a toy racecar driving around a track set in a children’s bedroom.
You can describe any world and character, and it puts you right into it.
For those unfamiliar with Genie 3, the concept is fairly straightforward – the model generates an output based on a user-given prompt in the same way one would with Gemini or Suno. However, instead of the output being a static image or piece of music, the end result is a full three-dimensional world that you can navigate in real time. Google DeepMind states that the model accounts for world consistency and stability, meaning exploring a previously visited location will recall the details of the first visit instead of regenerating something new.
Video created using Genie 3 showing a solder walking around a destroyed village.
This will change things for procedural gaming development.
It’s important to be cautious of new technologies as they emerge, as we’ve already discussed. But Genie 3 shows great potential for game development. Once the technology gets a little more polished – we encountered a bit of input lag during testing – it wouldn’t be far fetched for a development studio to include this model in an upcoming roguelike or open world game.
Video created using Genie 3 showing a figure walking towards a castle.
As for us? If you start seeing some killer open worlds in our upcoming creative, you’ll know who to thank.
Share this article
Related Articles
AI in Gaming: Navigating the Ethical Landscape
