It's a consistent thing I am seeing when generating things like "new hair color" or altering anything about the image I am working with - the new generation is great, but is almost always different size from the original image, either stretched a little bit, or squished, or larger, or shifted in space, but never exactly perfectly blended with the original image - so it takes a bit of effort to blend the new generation back into the original image.
Seems like a bug vs. a feature to me 🙂
27.0.0 release
Mac OSX Sequoia 15.6.1