Good one

I have something that does a similar job, however it works in a slightly different way. First of all the assumption is that the exact size of the image gives a quicker hint than the checksum so it keeps a hash that translates file sizes into a list of names and only looks at the contents when possible matches are detected.

This lets me specify a root directory, find candidates for cleaning up and interactively delete them using Tk.

To answer the other question I am running this under Windows (and I am not allowed to install a real environment) so diff is not available.