Overview
- DeepMind announced Genie 4 on Oct. 27 as a research preview that produces editable simulated worlds with native, time-synced audio and visuals.
- Genie 4 accepts natural-language edits during simulation so users can change weather, move cameras, add characters, and observe the scene continue to evolve.
- Google demonstrated integration with Gemini where Gemini supplies plans and code while Genie 4 renders those plans in real time for robotics and task validation.
- DeepMind warns of limits: scene details can drift in longer runs, generated audio can be inaccurate, and in-scene text can be hard to read.
- Genie 4 access is staged—first for trusted testers and then rolling out to Google AI Ultra subscribers, Project Genie participants, Gemini API users, and Google AI Studio partners—reflecting broader industry moves to combine simulation, pre-release testing, and governance for powerful models.