Project Genie: Google opens experimental AI world model to users
Google Genie 3 generates interactive 3D environments from text prompts and images. With Project Genie, US users can now try out the world model themselves.
From text input to playable simulation: An environment created by Project Genie, including controls.
(Image: Google)
Following the introduction of Google Genie 3 in August and a closed testing phase with selected users, Project Genie is now launching as the first public prototype. Usage is currently limited to the USA and requires an active subscription for Google AI Ultra.
Google Genie 3 goes significantly beyond pure video generators like Sora and Veo and creates freely explorable 3D environments that remain consistent over several minutes. Simple interactions are also possible, including realistically simulated physics.
Videos by heise
The public prototype Project Genie revolves around three functions: First, users create a 3D environment through text prompts and generated or uploaded images, with the world being adaptable and refineable through a preview. The character, perspective, and movement types, such as walking, flying, or driving, can be individually defined. The actual world exploration takes place in a freely navigable environment that Project Genie generates in real-time based on the user's actions. Finally, it is also possible to explore other users' worlds and modify or expand them via a remix function.
Use in AI Training, Video Games, and the Metaverse
According to Google, there are currently still technical limitations. The environments do not always look or behave realistically, or complex prompts are not implemented precisely. Furthermore, characters sometimes react with a delay to input or are harder to control, while the duration of simulations is currently limited to 60 seconds. However, the system is planned to be further improved and expanded in the future. For example, a function is planned that allows users to change the environment in real-time with text input.
Empfohlener redaktioneller Inhalt
Mit Ihrer Zustimmung wird hier ein externes YouTube-Video (Google Ireland Limited) geladen.
Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (Google Ireland Limited) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.
Obvious application scenarios for Google Genie 3 include prototyping for video games. The technology also shows great potential in the context of the Metaverse: AI world models like this could mature into a Holodeck machine in the future, enabling the creation and joint exploration of any world at the push of a button.
Google DeepMind and leading AI scientists such as Yann LeCun and Fei-Fei Li see AI world models as an important building block for general artificial intelligence. Instead of just working with rigid data, AI agents could gain experience with physical interactions based on these realistic environments. The goal is an AI that understands the cause-and-effect principles of the real world by testing different action options and their physical consequences in the simulation.
Project Genie is available immediately for Google AI Ultra subscribers in the USA and will be expanded to additional countries in the future.
(tobe)
