Google's new image model "Nano Banana" can actually read text inside images and use that information to generate new pictures. I tested it with a historical criminal mugshot, and without me adding details to the prompt, it read the crime description and created an accurate 1920s-style scene of the guy committing that exact crime.