Google DeepMind releases interactive world model Genie3

On August 5th, according to the DeepMind blog, Google DeepMind announced the launch of Genie 3, a general-purpose world model.

The model can generate diverse interactive virtual environments in real time based on text prompts, supports dynamic world navigation at 24 frames per second at 720p resolution, and can maintain environmental consistency for several minutes.

Genie 3 not only excels in physical property modeling and restoration of geographical and historical scenes, but also achieves controllable generation of complex world events.

As a key step towards general artificial intelligence (AGI), Genie 3 provides a broader simulation space for the training and evaluation of AI somatosensory agents.

Currently, the model is open to some scholars and creators in a limited research preview, and is expected to be expanded to more testers in the future.