What Are World Models?
World models are AI systems that generate a dynamic visual environment on the fly, creating video that reacts to user inputs as if you were inside a virtual world.
From Genie 3 to Project Genie
Genie 3 introduced “memory” that could retain world details for a few minutes. Project Genie builds on that foundation, plugging into newer models such as Nano Banana Pro and Gemini 3 and adding a “world sketching” workflow.
How Project Genie Works
The workflow starts with a text prompt or a reference image. Nano Banana Pro produces a still “sketch,” which you can edit before handing it to Genie. The system then renders a 720p video at roughly 24 fps, allowing you to move a character with WASD keys while the AI generates the path ahead in near‑real time.
Key Features
- Instant world generation from text or images
- 720p video at ~24 fps
- Real‑time‑ish navigation for up to 60 seconds per run
- Ability to remix pre‑built worlds with new characters and visual styles
- Downloadable video of your exploration
Current Limitations
- Input lag and a hard 60‑second exploration window
- No “promotable events” to insert new elements during a run
- Occasional physics or visual glitches
- Content‑restriction filters that can block fan‑made game prompts
Pricing and Availability
Project Genie is currently offered only through a dedicated web app and requires an AI Ultra subscription ($250 per month). Google says broader access will be rolled out over time.
What to Expect
While still a research prototype, Project Genie demonstrates how generative AI can approach interactive video experiences. Expect short, slightly different runs each time you prompt, and be prepared for occasional delays or content blocks.