Interesting that all four models converge on such similar designs, for such short prompts.

They were trained pretty much on the same data.