revids is used to determine the pageids/titles and prop=images can only return the files used by the current version, because only these are stored extra in the database to query by api module or special page (Special:WhatLinksHere)

To get the result of old revisions, the old revision needs to be parse

https://en.wikipedia.org/w/api.php?action=parse&oldid=817810268&format=json&prop=images

Oh nice, thanks for the quick and helpful response! This pretty much solves my problem.

Just one follow-up question please - I would also like to distinguish between images and videos in the edits, and likely disregard audio items. For that, my plan was to use the images as a generator to prop=imageinfo in the Query API, like so:

https://en.wikipedia.org/w/api.php?action=query&titles=Albert_Einstein&format=json&prop=imageinfo&generator=images&iiprop=mediatype

Now that I need to use the Parse API, it looks like I'll need the one Parse call and then a few Query calls to get the imageinfo of batches of the returned images. Is that correct, or is there perhaps some way to use the results of a Parse as a generator for a Query?

There is no way to use the output of parse API as generator. You have to query them. If first getting the current images than only the missing one needs a query (or none if nothing changes)

Sounds good, thank you!