Almost six months after launching Genie 3, a successor has arrived with more control, more detail and longer runtimes.
Running on a combination of Genie 3, Nano Banana Pro and Gemini, the new model produces 60 second (up from 30) fully generated, playable 3D worlds.
All the user needs to do is write a prompt for the world to build and then describe the character they want to navigate it, or just upload pictures. The rest is handled by the model in no time at all.
Generates its own physics
As the model is played, Project Genie generates the path ahead in real-time based on your actions, simulating physics on the go.
This is of course not made for simply fooling around in crazy 3D worlds, it is fundamental technology for developing robots and give them an understanding of their environment in training — or as Google says «to support our AGI mission.»
Project Genie is only available for AI Ultra subscribers for now, but the hope is to expand it at a later date.
Read more: Google’s announcement, writeups on Ars Technica and Mashable. Discussion on r/Singularity.