Make a PDF smaller without wrecking it

A page of typed text is a few kilobytes of drawing instructions. What fills a PDF is pictures — and a scanner or a phone scanning app will write one 300 dpi photograph per page at a quality setting meant for printing. On the documents people bring here, the images are usually between 90 and 99 per cent of the bytes. Shrinking those and leaving everything else alone is the whole job, and it is why the default mode does not touch a single other object in the file.

In the default mode nothing is re-rendered. Every text object, every vector drawing, every hyperlink and every form field is copied through untouched, so the compressed copy is still a real document you can search, select from and fill in. That is not true of most compressors, which flatten first and ask questions later — and flattening cannot be undone.

Some images cannot be rewritten without changing how they draw: a palette-based image whose bytes are positions in a colour table, an image the page inverts as it draws it, a one-bit stencil mask, a CMYK photograph meant for a printing press. Rather than guess, this page leaves those exactly as it found them and tells you how many it left and why. Nine images shrunk out of ten, said plainly, is worth more than ten shrunk and one of them quietly ruined.

If the result comes out larger than what you brought — which happens when a PDF is already compressed, or is pure text with nothing to squeeze — you do not get a download button and a green tick. You get told, and you get told which of the two it was. A compressor that returns a larger file and calls it done is the most common dishonesty in this category.

Every service that caps you at 10 or 25 MB does it because it is paying to receive your file. This page never receives it — the work happens in the tab you are reading this in. A 400 MB scan costs this site exactly what a 400 KB one costs it, so there is no cap. Here is how that is enforced.