Google DeepMind released Genie 4, its newest AI world model, in research preview on Tuesday. According to The Verge and TechCrunch, the model generates interactive, photorealistic environments that users can steer with natural language prompts while a simulation is running, and it can now produce matching audio as well.
The launch puts Google at the front of a fast-moving world-model race that includes Runway's GWM Worlds 2. Genie 4 is not a finished product: it is going first to "trusted testers" and a waitlist, with a wider rollout to Google AI Ultra subscribers in Project Genie promised in the coming weeks.
Genie 4 specs and new capabilities
Google says the update brings four main changes over Genie 3, which arrived as a research preview a little over a year ago. The model is more photorealistic, supports freeform edits mid-simulation, generates synchronized audio, and relies on Gemini 3.8 Flash under the hood, per The Verge. TechCrunch adds that Gemini acts as a director, planning steps and checking frames while Genie builds the visuals.
Simulations can now hold a consistent scene for about two minutes, double the one-minute limit of Genie 3, with better physics and more human-like characters.
- More realistic, photorealistic worlds than Genie 3
- Real-time edits via text prompts: weather, camera angle, new people or objects
- Audio generation: music, ambient noise and interactive dialogue
- Gemini 3.8 Flash powers the simulations; Gemini directs the scene
