With Genie 3, Google offers a generative AI capable of recreating an interactive photorealist world.
“A new border” is how Google describes its new Genie 3 model, capable of generating interactive worlds.
The presentation video makes your eyes full and allows you to quickly understand the concept. Genie 3 allows you to generate a world in which you can move using the arrows of your keyboard, or move the view with a controller or the mouse.
A technological tour de force
So we still gain a dimension after the generation of image, then of video, by adding the interaction, we pass in the field of video games.
Finally a great flow, without superfluous
No phone and TV decoder needed? With B & You Pure Fiber, pay only for the main thing, without obligation. A 100 % fiber offer, 0 superfluous, for those who want the best … and nothing else.
The tour de force here is in particular in the ability of the model to memorize the environment already drawn, even out of scope.
The worlds generated also offer realistic physics, where it is possible to paint, to drop objects or simply to walk.
Unlike Veo, there is no question here of generating a video of a few seconds. The world generated by Genie 3 can extend for several minutes and offers real -time interaction.
Google does not only have a use to generate video games, but beyond, allowing training videos or simulations to create a scientific purpose.
GENIE 3 also allows you to create “events” in a generated world. For example, making a character or a reversal of the situation appear.
Still very real limits
Obviously, everything is not perfect, despite the progress demonstrated. Google notably notes the small number of minutes of the demo, while it would like to create worlds capable of interaction for several hours.
On the other hand, if Genie 3 can create photorealist worlds, they are not geographically correct. In other words, the engine does not know how to generate a New York or a Paris that corresponds to the real.
Finally, the model does not allow everything to do everything in events and Google stresses that the possibilities will have to be extended in the future.
The fact remains that the progress made by the Deepmind teams in a few years remains stunning on the technological aspect. As for creating a real artistic work as the video game offers today, this is another subject.