Anybody have any insight into what kind of trickery goes into this? I know a decent amount about both of the compression algorithms, but I've forgotten the structure of the file formats so I can't tell if this required any really funny business.
Basically the "top level" zip file section lives at the end of the zip file and contains pointers back into the file for the actual data. This is so that you can keep adding to a zip file without having to rewrite anything you have done previously - just append a few more files and then your updated directory at the end.
JPEG segments contain two special opening bytes, maybe a length, and then the actual data. So it would be trivial to have the last jpeg section in a jpeg file end with zip directory bytes. This directory could point back to data in other sections of the jpeg.
..my guess, anyway.
What's surprising is that it works on Twitter. I tried the same thing with image hosts and the data got scrubbed. I guess they don't re-encode small images but I dunno.
cat file.jpg file.zip > magic.jpg
(Although if they do something with this file and you still are able to do this trick that would seem even more .. interesting on their side than leaving images below certain size untouched)