It is impossible to 1:1 replicate the input as an output because the images are not stored. It isn't a database. It's basically aggregating summaries/abstractions/generalizations of a bunch of tags.

In other words, it is transformative by default.