This really needs to scan for images in a page rather than taking just image urls.
Really strange that it doesn't, given the description. They describe it as working on an entire page.
Hopefully it's just a bug and someone'll push a change because it's definitely interesting.
It also needs an API so you can replace images automatically as part of your build or upload process.
Archive.org shows it's been like this since at least Jan 1, 2011 (did't check any other or earlier dates), so it seems like fixing it must be really low priority.