I asked ChatGPT to describe each state of the United States as a woman, and then used the description as a prompt for Stable Diffusion. I generated the images in the style of digital paintings.
The base model I used was the Copax Timeless SDXL Model (version 7). On top of those I used a bunch of LoRAs as well to add detail and to impose certain styles.
I found that the model was especially sensitive to mentions of the clothing, the background, and the state flower. In most generations the state flower would be over-represented in the image. I am not sure why. I did edit the prompt in some cases to make things like the state's natural features more prominent.
The model tries its best to get some features of Native American clothing right, but because it doesn't "know" any of the actual details, I am sure it gets many things wrong. I have seen similar things with Indian clothing. What is also interesting is that some ChatGPT prompts didn't contain any mention of race or ethnicity, but I would consistently get generations that included a particular ethnicity. For example, in the southwest states, especially Arizona, New Mexico, and Texas, I would get Mexican or Native American. For states in the South, I would usually get African American as the race.
The pictures are ordered alphabetically by state; hope you like them!