Here's the tl;dr from the makers themselves:

They contains 16894 line entries, which correspond to every frame from the original file. Every line is composed of several colored vector drawings in the form of cubes. Each one has the size of 1x1 pixels. These lines are aligned with the top left corner, and contain internal newlines to create rows and columns. Every single of these pixels is colored. It is a lossless transformation from the pngs produced to embedding them, so you could recover the original easily. TL;DR: every line contains one frame of video. In the subtitles.