The model incorporates major visual fidelity upgrades over its predecessor, Genie 2. Users can navigate Genie 3's simulated worlds using keyboard input at 720p resolution and 24 frames per second. The best part is that, where Genie 2 struggled to retain world details for more than 10 seconds, Genie 3 can maintain visual consistency for several minutes.
On Tuesday, Google DeepMind unveiled Genie 3, a generative AI model that can create interactive, 3D worlds in real time from a simple prompt or image. Users can add objects, change the weather, or introduce new characters, which DeepMind calls "promptable events." It's turned everything into an open-world game.
Genie 3's worlds are continuously generated, and users can alter them anytime. While some folks in the gaming industry doubt that such tools will be practical for mainstream game creation, DeepMind says Genie 3 is as much a research asset as it is a creator of virtual worlds, as reported by Ars Technica.
Genie 3 still has its challenges, however. Its memory isn't long enough for persistent worlds. Current retention lasts minutes instead of hours, and the model can't reproduce real-world locations. Sometimes, it produces odd results, such as awkward human movement or garbled text. AI agents in these worlds can move but can't interact with or modify their environments.
For now, Genie 3 is best described as a platform for research and play, though it's currently only available to a small population of scholars and creatives. DeepMind aims to widen access after obtaining early users' "crucial feedback and interdisciplinary perspectives."
Genie 3's 3D worlds are continuously generated and can be changed anytime.
Post Your Comments