This requires downloading are parsing the file manifest and discarding the images marked "not in output format" and then indexing correctly.

This will have to be done at Toolforge, probably with https://pagelister.toolforge.org/ because of CORS/CSP.