This was trained on pictures I took of/through a sculpture in the lobby of a medical building. It can take some fidgeting with the weight and CFG scale on some checkpoints and with some prompts, but it leans towards concepts like buildings, windows, and vertical lines.