This approach helps preserve visual coherence over extended interactions. The generated worlds remain visually and physically consistent for "multiple minutes," marking a major technical step forward over previous models.
Genie 3 enables real-time navigation and interactive world-building, maintaining consistent environments at 720p for several minutes. | Image: Google Deepmind
According to Deepmind, Genie 3 is the first model to combine real-time interactivity with longer-term physical consistency in its environments, and the team sees Genie 3 as a foundation for developing more general AI systems (AGI). This release builds on Deepmind's earlier projects, including Genie 1 and Genie 2, as well as the Veo 2 and Veo 3 video generators.
Simulating physical and imagined worlds
Genie 3 can create a wide range of scenarios, from realistic landscapes with dynamic weather effects like lava, wind, and rain, to fantastical settings featuring portals, flying islands, or animated creatures. The model can also reconstruct historical locations such as Venice or ancient Knossos.
