One example:
(sorry, there's a kind of mess in here while I'm adding multiprocessing) - an example of efficient misuse of Scale2x and Scale3x algorithms to improve scans with "historically poor" resolution before OCR and stuff. For IO, surely I'm using your PNG module (since I managed to understand how to deal with it there's no way I will deal with Pillow, full with features I don't need, and having it's own opinion on how data should look).
With low-res scans, OCR software tend to complain, and I found out that, say, Scale3x with its unusual interpolation often makes an improvements in OCR, so I've made these pure Python implementations of ScaleNx, to use them from time to time.
My OCR software uses input image resolution to assign page size. Therefore, it would be very convenient if, scaling images up with Scale2x, I will double the resolution (and triple for Scale3x) so it keeps original print size in meters and precisely fit the old page layout.
With your PNG module, I easily do that, with just several lines of code. Programs are happy to see new resolution, I'm happy that I don't have to set it manually, everything is fine.