Google DeepMind Unveils Genie 3 for Human-Like AI Development
Google DeepMind has introduced Genie 3, an upgraded version of its AI-powered “world” model that creates 3D environments. Unlike traditional video game design, where artists build assets manually, Genie uses artificial intelligence to generate interactive spaces based on a simple prompt. This marks a significant step in building environments where both humans and AI agents can interact more naturally.
Importantly, Genie 3 brings a major improvement in realism and duration. Users can now engage with its environments for several minutes a notable increase from the previous version’s 10–20 seconds. Additionally, it can remember visual details, such as objects or markings, for about a minute. This means when users return to a previously viewed area, the visual elements remain intact. These enhancements aim to create smoother, more immersive digital experiences.
Smoother Performance and Real-Time Interaction
Genie 3 operates at a resolution of 720p and maintains a steady 24 frames per second, resulting in fluid motion and sharper graphics. A new capability called “promptable world events” adds even more flexibility. With this, users can type commands to alter conditions inside the generated world from weather changes to the appearance of new characters. This interactive feature helps make the environments feel dynamic and responsive.
Previously, Genie 2 could generate playable scenes from a single image, but it lacked stability and had shorter interaction times. Many similar efforts in the industry have also struggled to produce stable, immersive visuals. Genie 3 addresses these issues by improving memory retention and visual consistency.
Gradual Rollout and Research-Focused Access
For now, Genie 3 will be available only as a research preview. DeepMind is limiting access to a small group of academics and creators to assess potential risks and test safety measures. Moreover, the interactive features will be restricted, and readable text will generally only appear if specified in the initial prompt.
Though the system is not yet open to the public, DeepMind may expand availability in the future. The updates introduced in Genie 3 suggest growing potential for AI-generated environments across education, entertainment, and robotics training.
