@elaine1helen @cstross Yes, one key point of failure for asking "simple" questions of genAI is that they are designed to generate something original, combining and randomizing their training inputs instead of just copying one.
And for something like this, there's the issue that they don't really have a 3D model of the world, so every different angle of an image is like a different variation on the diagram, and these then get combined in ways that don't make sense from a real-world perspective.