Presented as a "new frontier" by the firm, Google's latest feat transforms artificial intelligence into the architect of living universes. The Genie 3 model no longer simply produces images or videos: it generates entire, photorealistic, interactive worlds, in which the user can move around in real time. In a spectacular demonstration, Genie 3 invites the user to move through an AI-generated setting using keyboard keys, a controller, or a mouse. It's no longer just a video that we're watching, but an environment in which we interact. The system is even capable of memorizing off-screen elements, managing realistic physical effects, and producing dynamic events, such as the sudden appearance of a character or a change of scenery. This advancement opens up dizzying possibilities, going well beyond video games. Google plans to use Genie 3 in areas such as immersive training, scientific simulation, and automated audiovisual production.
Limits that are being pushed further and further
The technology is impressive, but still incomplete. The demonstrations last only a few minutes, and the generated worlds, while visually accomplished, are not modeled on real-life locations. It is currently impossible to visit a true-to-life Paris. Furthermore, the variety of simulatable events remains limited, even though Google promises rapid developments. With Genie 3, the DeepMind teams are crossing a new threshold. After mastering image and video, they are tackling the spatial and interactive dimension, the very one that constitutes the foundation of video games and virtual worlds. This is a first step toward an AI capable not only of creating sets, but also complete, vibrant, and customized experiences. The video game, professional training, and immersive cinema industries could well be disrupted by this type of tool, still in its infancy, but already eminently promising.
