That seems to be an incredibly fringe use case to justify the "wrong" implementation (i.e., modeling images as text when your protocol already has affordances f

That seems to be an incredibly fringe use case to justify the "wrong" implementation (i.e., modeling images as text when your protocol already has affordances for images).