WenWare turns history into GeoGuessr

This is a clever proof that generative image models are moving from “make art” into “simulate a world,” but the hard part is consistency, not generation.

• –gpt-image-2 makes the supply side easy: you can spin up panoramic scenes at scale instead of hand-building every map
• –The “when and where” mechanic is stronger than a plain image demo because it gives the output a game loop and a score system
• –Historical fidelity will be the main limitation; iconic events and period details are easy to get wrong or homogenize
• –If the game is cheap enough to populate with daily maps, the real product becomes curation and challenge design, not raw image quality
• –This is a good showcase for developers thinking about model-driven content pipelines, especially anything that needs endless synthetic environments