Genie 4 draws on Gemini models: Gemini Omni grounds how scenes look and behave, and Gemini 3.8 Flash watches each world as it renders, breaking goals into steps and checking that they happened. With Gemini 4 Argon planning the steps, it simulated a humanoid robot playing checkers. Trusted testers are using it to preview film scenes, turn archival photos into explorable spaces and prototype games. All output carries Google’s SynthID watermark. DeepMind says details can still drift over longer sessions, and text, spatial accuracy and complex physics remain hard.
Why it matters
DeepMind calls world simulation a crucial step toward AGI, and safe, endless virtual worlds could train robots and agents without real-world risk. Adding sound and speech makes the simulations far more useful for film and games. For now it is a limited preview; access for Google AI Ultra subscribers and Gemini API partners is promised but not dated.
Original source
Genie 4: A new generation of world simulation
Google DeepMind
deepmind.google
Summary by Second Order AI. Full reporting is at the original source.