Good point. What if we combined the best of both: more realistic images from tpdne.com and API-accessible metadata from your site?
Seems like a matter of crawling the former to get all the images and then tagging them and serving them via your API.
Anyway, I think your approach is great and only wanted to propose a possible approach to the original feedback above :)
