Google has introduced Genie 3, the latest version of its AI world model, designed to create 3D video game environments in real time. The model comes from DeepMind, Google’s advanced AI research unit, and is a major step in the company’s efforts to merge AI and immersive digital spaces.
Genie 3 allows users and AI agents to interact with these virtual worlds as they explore them. Unlike traditional video game environments built manually with 3D assets, Genie 3 generates its spaces on the fly, using only a text prompt.
This version of the model significantly improves on earlier versions. Previous models, like Genie 2, could only support interactions for around 10 to 20 seconds. Genie 3 now allows for a few minutes of real-time engagement, making the worlds feel more alive and continuous.
Another major update is memory. The AI can remember what was in the world, even when it is no longer in view. For example, if a user turns away from a chalkboard and then turns back, the writing remains in the same spot. This memory lasts for about a minute and brings a new sense of consistency and realism.
The visual quality has also been improved. Genie 3 runs at a resolution of 720p and a frame rate of 24 frames per second. This ensures smoother and more natural visuals, enhancing the user experience.
DeepMind has also added a new feature called “promptable world events.” This means users can change things inside the world by giving simple commands. For example, you could make it rain, switch to sunny weather, or introduce new characters with a text prompt. This gives users more control and adds depth to the interactive worlds.
World models like Genie are designed to create environments for various uses, from training robots and AI agents to developing educational tools and entertainment experiences. By generating a space with just a prompt, these models remove the need for hand-built 3D assets. This saves time and allows for more flexible and creative development.
Google has been focusing heavily on world model research. Genie 2, shown in December, could create basic interactive scenes from just an image. The world models team at DeepMind is now led by a former co-lead of OpenAI’s Sora, another advanced video generation tool.
Still, AI-generated worlds are not without challenges. Genie 2’s short interaction time and limited memory made its worlds feel static. Other similar attempts in the industry have produced strange and unclear visuals. One recent effort backed by a Pixar co-founder felt like a blurry version of Street View where objects changed unexpectedly.
With Genie 3, Google aims to fix many of these issues. The longer interaction time, improved memory, and better visuals create a more stable and enjoyable experience. The addition of world events also opens the door for more dynamic storytelling and creative design.
As AI models become more advanced, they are starting to change how we create and explore digital environments. Tools like Genie 3 offer a glimpse into the future of game design, virtual training, and even interactive learning.
Genie 3 shows how AI can not only build worlds but also respond to our ideas in real time. It represents a new way of thinking about interactive content—one that is driven by language, powered by memory, and shaped by user choice.
As Google continues to refine these systems, the line between human-made and AI-made worlds is getting thinner. Genie 3 may be the clearest sign yet that the future of virtual environments is being written by AI.
