Google has introduced the Genie 3 neural network, which can create interactive digital environments in real time based on simple text or image commands. The new model provides HD resolution video (720p, 24 frames/s). One of the most important advances of Genie 3 is the ability to expand memory. If the previous version, Genie 2, maintained the visual integrity of the generated environment for only 10 seconds, then in Genie 3, this figure reaches several minutes. This means that objects and places in the environment are not “lost” after leaving the field of view, but are preserved, and it is possible to return to them.
Genie 3 allows you to modify the environment in the current mode using text commands. The user can add new objects, change the weather, create characters, etc.
Despite the progress, the model still cannot reproduce real-world environments, and the reproduction of human movement and text is sometimes inaccurate.
Genie 3 is currently in closed access and will only be available to a limited number of researchers.
