Google DeepMind unveils AI that builds worlds in real time
Google DeepMind has introduced the third version of its world model, Genie 3, an AI system that can generate interactive virtual environments based on text prompts, Kazinform News Agency correspondent reports.
Unlike previous versions, Genie 3 allows users to move through the generated world in real time at 24 frames per second and 720p resolution. The environments remain visually and logically consistent for several minutes.
The model is part of a class known as world models, designed to predict how an environment will change in response to user actions. This type of technology is seen as a key step toward building general-purpose artificial intelligence, as it allows agents to train in simulated settings with an unlimited range of scenarios.
Genie 3 shows progress in several areas, including the simulation of physical processes, animated scenes, natural ecosystems, and historical settings. One of the new features is the ability to trigger events using text commands in addition to navigation. For example, users can change the weather or add objects.
Despite its advancements, the developers note that Genie 3 still has several limitations. These include a restricted set of actions, the inability to accurately recreate real-world locations, and a limited interaction time, with consistency maintained only for a few minutes.
Google DeepMind has acknowledged the potential risks associated with this kind of technology. For now, access to Genie 3 is limited to a small group of researchers and creators. The company plans to use this phase to gather feedback and assess possible implications before considering a broader rollout.
Earlier, Kazinform News Agency reported on Google's flagship tech conference of the year, Google I/O 2025.
