The scene isn’t copyrighted, anyone could go to the scene (theoretically) and take their own photo from a different angle. What’s copyrighted is the expression that went into staging the shot.

An AI tool is the one doing the creative expression when generating its images is the argument. The prompt is where the creative expression of the user ends, and copyrighting just a phrase seems ridiculous. I tend to agree with these sort of arguments, especially when models like this are often trained on other people’s copyright work.