DeepMind's Genie 3 Creates Real-time 3D Interactive Worlds

• Genie 3 generates interactive 3D worlds from text prompts, running at 720p, 24 FPS.
• Unlike Genie 2’s 10-20 second window, Genie 3 sustains interaction for several minutes.
• Visual consistency ensures locations remain unchanged when revisited, enhancing immersive simulation experiences.

Google DeepMind has introduced Genie 3, an advanced AI model designed to create interactive 3D worlds. By entering a text prompt, users can describe an environment, which Genie 3 simulates in real time at 720p resolution, maintaining 24 frames per second for several minutes.

This is considered a major invention towards the predecessor Genie 2 which was interactive only in 10 to 20 seconds. Genie 3 offers visual consistency, such that users visiting the same location in the simulation will not make out any changes, making it a very immersive experience.

Last year, DeepMind released Genie 1 and Genie 2, their initial "world models," alongside other AI video generation models such as Veo 2 and Veo 3, which demonstrate an understanding of physical environments. According to a blog post accompanying the release, “World models are also a key stepping stone on the path to AGI, since they make it possible to train AI agents in an unlimited curriculum of rich simulation environments.”

Also Read: Tech Data, HCLSoftware Expand Enterprise Software in APJ

These models enable AI agents to predict environmental changes and understand how their actions influence the surroundings, a critical step toward advanced AI development. Genie 3 is not yet available for public preview and is currently limited to a select group of creators for testing. This controlled rollout allows DeepMind to refine the model based on feedback before broader release.

The enhanced interactivity and consistency of Genie 3 position it as a groundbreaking tool for simulating dynamic, user-defined virtual worlds, with potential applications in gaming, training, and creative industries.