Google published a new episode of the podcast "Google AI: Release Notes" where the Project Genie team presents Genie 3, a real-time interactive world model. Can you imagine stepping into a 2D image and having the world respond like a video game? That's exactly the kind of ideas they discuss in the episode released on January 29, 2026.
What is Genie 3
Genie 3 is a model that goes beyond generating passive videos: it creates simulated, playable environments where an agent can interact and remember what happens. It's not a visual trick, but a bet on worlds that remain consistent over time, with memory and internal rules that let actions have coherent consequences.
Why does this sound like science fiction? Because for years AI was great at producing images or isolated clips. Genie 3 aims for an environment that persists, where you can come back and find changes, and where that world can be used to train more capable agents.
