> extracting ISBNs from the actual book scans themselves (in the case of Z-Library/Libgen).

OpenLibrary also uses book scans in Archive.org to extract ISBNs (and a few other bits of metadata, like urls in the text):

https://blog.openlibrary.org/2021/08/23/gsoc-2021-making-boo...

And have a software pipeline for that kind of thing available.