Perhaps a few examples will be helpful. Please put down your coffee (or other tasty beverage), swallow that microwaved burrito, and take a look at these gorgeous transformations. Let's assume that the process involves a photographic morph from two copyrighted images, and that the morph-artist has not obtained (or even requested) permission or a license to do so. Where does that leave the copyright holder, the morph-artist, and the public? And, just to make things fun, notice that each transformation in the set consists mostly of one picture, with features from another one morphed in (with a very high degree of skill). Are the answers any different if there's a permission/license problem with only one of the source photos? Does it matter which one?
The tangled answer to this question is one of the reasons that I think transformation has little value in actually evaluating whether a specific use is a fair one. It is instead a frequently misapplied bit of rhetoric used as a post hoc rationalization for reuse, and does not actually provide us with part of a principled, replicable framework for evaluating fair use. This is not just because transformation is not one of the statutory factors under § 107, but because the degree of transformation is a purely subjective test in the eye of the beholder. In short, it's the Jacobellis test for obscenity applied almost directly to copyright infringement—and, considering the mess in First Amendment law that descends from Jacobellis and similar decisions that "obscenity" is not "speech" for First Amendment purposes, that's not a result we should attempt to reach.