Corrupt was first built with Proce55ing. The corruption process start by reading the binary of an image file [JPG or GIF], then some bytes are swaps [the number of replacement is a random value from 1 to 20]. The file is then "saved as" a new document.
Depending on the number replacement and of the original compression, the image will have a completely different and unpredictable aesthetics.
So from a single image the program can generate millions of corrupted versions. And because it is a real corruption system that damages the binaries of a file, some of the results can't be showed because they are too damaged.
Related post: Glitch video (old skool way) from Ben Hanbury