DeepMind launched Genie 3 which they say is a world model that can produce three-dimensional worlds that not only look realistic but are also interactive. By simply entering a prompt, the 3D world can be generated by this generative AI and by using the keyboard the user can move around in it.
Genie 3 also has a memory that remembers the actions performed by the user. In the video above, the user can paint a wall and when moving to another area the painted part of the wall will remain. According to DeepMind this allows the production of consistent 3D content for video games, entertainment and even emergency training purposes.
It also has the unique ability to insert new elements into the world that has been generated in real time by simply typing additional prompts. Each world generated by Genie 3 is at 720p 24FPS resolution with interactions that can be done for several minutes. This is an improvement from Genie 2 which produced worlds with 360p resolution and interactions that lasted up to 20 seconds only.
According to DeepMind, Genie 3 is another step towards the development of artificial general intelligence (AGI). The model is now accessible to researchers, with applications first being made by DeepMind.