It still doesn't seem to have an understanding of the "logic" of certain structures. I asked it for a photorealistic image of a Vekoma Boomerang roller coaster,

It still doesn't seem to have an understanding of the "logic" of certain structures. I asked it for a photorealistic image of a Vekoma Boomerang roller coaster, which has a very specific layout and type of track. Not only did it generate images with far more intricate sets layouts than a boomerang would ever have, it also still generates pieces of track that don't connect to anything. Like, it has a general sense of what a photo of a roller coaster looks like, but none of the specific details that make a coaster work. I next asked for a diagram of a diverging diamond interchange, and it came up with 4 images that looked like 3D models of interchanges on a white background, but none of them were diverging diamonds. I'm going to experiment with describing the thing I want to see in more detail, to see of Dall-E 3 can make a nice image out of it.