Google DeepMind is opening a new AI model called Project Genie to some users. Users can instantly generate a virtual world of any style through text prompts or uploading images, and can freely explore it by controlling characters or vehicles like playing a video game. This feature is currently available through a web app, and users who have a Google Ultra account in the United States and are over 18 years old can try it out.
Project Genie is based on the Genie 3 model that Google previously demonstrated to a small range of testers in August 2025, and combines the company's own Nano Banana Pro image generation model and Gemini multi-modal model to convert natural language cues into immersive interactive scenes. The types of worlds that users can build are extremely diverse, such as flying a spaceship over an alien planet, flying an airship over a 1950s-style European city, or letting a tapir run deep in the Amazon rainforest, etc., all of which can be quickly generated with simple descriptions or reference photos.
In terms of specific experience, Project Genie will generate frames in the virtual world in real time based on how the player moves the character and rotates the perspective, and supports interaction with objects in the scene and response to environmental changes. For example, in the official demonstration video, when a blue ball rolls in the white grass, everything it passes will be "painted" blue, forming a continuously changing trajectory effect; at the same time, the application also has a built-in world template library for users to create and modify for quick start.
Users can not only navigate these environments, but also upload their own drawn characters, photos of real-world objects, etc. as materials, and set interaction logic between different elements to build a more personalized virtual experience. In some scenes, you can even see dynamic feedback similar to that of a GPS navigation interface: when the player turns or moves, the map on the interface also updates, enhancing immersion and playability.
In an interview with Project Genie team members Jack Parker-Holder and Diego Rivas, YouTuber Bilawal Sidhu pointed out that the current system still has occasional bugs in actual demonstrations, and each experience is limited to 60 seconds, which is one of the biggest functional constraints at the moment. However, the development team stated that in the future, it will focus on exploring how to give more control over the environment and character behavior in user input, and study feasible solutions to extend the duration of the virtual world and break the 60-second upper limit.
As for application scenarios, the DeepMind team has not yet given a single clear direction, but is optimistic about its potential in multiple fields. Reports indicate that Project Genie is expected to be used to rapidly prototype video game ideas, help film creators visualize scenes and storyboards, and help students understand complex concepts or professional scenarios in a more interactive way in the classroom, such as simulating participation in post-disaster rescue work, without requiring users to have professional development skills.
The author is also concerned that the opening of this technology will inevitably be accompanied by content security and compliance challenges, including how to set up "guardrails" to prevent the generation of harmful or inappropriate content, and how to avoid potential copyright infringement issues during the creation process. Google has not yet fully disclosed its governance mechanism in detail, but it is expected to continue to adjust relevant strategies in future product iterations.
Currently, users who live in the United States and subscribe to Google AI Ultra can already experience Project Genie directly through the Google experimental project page. As it is further opened to more regions and users, this new AI interaction method of "generating and entering the virtual world in one sentence" may become one of the important tools in the fields of game development, creative design, education and training.