A digital universe created from some words. In a blog note published on August 5, 2025, Google Deepmind, the department of the technological giant, has presented Genie 3. It is a unique generative AI since it does not create image, music or video, but a world in which you can move, as in a video game.
Released in December 2024, his predecessor, Genie 2, was relatively discreet. But in comparative videos, published by Google, the difference between the two models is obvious. In the eight -month space, Genie has become much more fluid, the resolution of the images is of better quality (in 720p) and the duration of the simulation has greatly increased.
Interactive environments
In the first or third person, demonstration videos published by Google show a variety of universes generated. You can see realistic, fantastic worlds, sometimes cartoons, in which it is possible to move with the keys of your keyboard.
Users can even interact in real time with the environment. By formulating a request to the model, they can even reveal elements or events in the digital universe.
But one of the main additions of Genie 3 is his memory. Previously, for AIS similar, as soon as an element of the Avatar field of vision came, it could disappear or become something else. A world could change at all simply by touring oneself. Genie 3, manages to remember the location of certain objects, or even modifications to the environment. Be witness to the painting, in one of the examples taken in the video, which persists on the wall, although it has left the field of vision.
There are still many limits
However, Google points to certain limits to Genie 3. starting with a limited action zone in the created environment, as well as a simulation period only a few minutes. The real generation of an interactive environment, with memory, requires monumental calculation power. Deepmind also specifies that Genie currently cannot answer real places and generate text regardless of the notice.
The uses of such algorithm can be numerous. We immediately think of the video game sector, but Google plans to implement the genius for educational or training purposes. But where I could play an essential role, it is in the training of other AI. These alternative digital worlds could become privileged reasons, since no risk can be incurred there.
Unlike other Google generators (I especially see for videos and image image), Genie 3 is not open to the public. At the moment, the company reserves the use of the model to a handful of researchers. However, it does not exclude “making Genie 3 available for other testers in the future.”
Source: BFM TV
