Folia Tools
Why your PDF is so large, and why compressing it sometimes does nothing
A PDF is not one thing, and its size comes from whichever part of it happens to be heaviest. Knowing which part decides whether compression will help you or waste your afternoon.
Three things take up the space, and only one of them compresses well
Almost every oversized PDF is oversized for one of three reasons, and they behave completely differently. Images are usually the culprit: a page scanned at 300 dpi holds around nine million pixels, and thirty of those pages is a file measured in tens of megabytes before anything else is counted. Embedded fonts are the quiet one — a document that embeds four weights of a typeface in full, rather than only the characters it uses, can carry several megabytes it never displays. Vector content is the surprising one: a map, a CAD export or a chart with fifty thousand plotted points stores every one of those points as drawing instructions, and there is no such thing as a low-resolution version of a line.
Text, by contrast, is almost free. The words of a three-hundred-page novel occupy a fraction of a megabyte. If you are looking at a large PDF that is mostly prose, the size is not the prose, and no amount of compression aimed at prose will find anything to take.
This is why the same setting produces a triumph on one file and nothing at all on another. Compression does not work on "the file". It works on whatever the file is made of, and half the time that is something it cannot touch.
What compression actually does to the bytes
There are two genuinely different operations that both get called compression, and it is worth knowing which one you are getting.
The first is a lossless repack: the document is reloaded and its objects are rewritten into object streams, which is a tighter way of storing the same information. Nothing is re-encoded. Text stays as vector text, and links, form fields and bookmarks all survive. The saving is only whatever the tighter structure buys — often ten or twenty per cent on a file that was written carelessly, and close to nothing on a file that was already written well.
The second is a rasterisation: every page is rendered as it appears and stored as a JPEG, and a new document is built from those pictures. This is enormously effective on scans and enormously destructive on everything else. Folia's three levels are simply three points on that trade — Light renders at 150 dpi and JPEG quality 0.72, Balanced at 110 dpi and 0.60, Strongest at 72 dpi and 0.50.
Run both on a 300 dpi scan and the raster route wins easily, because Balanced keeps roughly a seventh of the pixels the scan was carrying. Run both on a text-and-vector report and the raster version frequently comes out larger than the original — a page of crisp type is cheap as instructions and expensive as a photograph of itself. Compress PDF tries both and keeps the smaller one; where neither beats what you started with, your original bytes come back untouched and the note under the result says so.
How to tell which route your file took
Open the result and try to select a line of text. If you can, the lossless repack won and nothing was re-rendered. If you cannot, every page is now a picture of itself, and the text layer, links, bookmarks, annotations and form fields went with it.
That test takes two seconds and is worth doing every time, because the difference does not show up in the file size and does not show up on screen. A rasterised page looks identical at 100% zoom and falls apart when someone prints it, searches it, or feeds it to a screen reader. If the document is going to be read once and thrown away, that may be a fine trade. If it is going into an archive, a court bundle or a contract file, it is not.
The merged-file problem nobody expects
Merging ten reports produces a file that weighs roughly the sum of its parts, and people are often surprised it is not smaller. It is not smaller because identical resources in two different documents are not recognised as the same object. A company logo that appears in all ten sources is embedded ten times. A font embedded by all ten is embedded ten times.
Nothing in Merge PDF strips those duplicates afterwards, and compression will not find them either — Compress PDF brings the size down by downsampling images or rebuilding pages, not by de-duplicating objects. If a merged bundle is uncomfortably large, the honest fixes are to merge fewer things, or to accept the rasterisation trade knowingly.
When the answer is not compression at all
The usual reason anyone compresses a PDF is an attachment limit. Held up against that, crushing a 40 MB document to 9 MB by turning it into photographs of itself is often the worse of the two available moves. Split PDF sends the same document as three intact parts, each of which is still searchable, still selectable and still prints properly.
Two other approaches are worth knowing. If the file is large because it is a scan, running OCR PDF re-encodes every page at around 1100 pixels wide, which usually returns a much smaller file and makes it searchable — the size reduction is a side effect of a job worth doing anyway. And if the file is large because it embeds fonts and images you no longer need, converting to Word, deleting what is genuinely surplus and exporting again will beat any compressor, because you are removing content rather than degrading it.
The one thing worth avoiding is running the strongest setting reflexively. There is no target-size box on the compression page on purpose: pick a level, read the number the tool reports, and step down only if you actually need to.
Questions
Why did compressing my PDF make it bigger?
Because the rasterised version of a text page is bigger than the text page. Crisp type stored as drawing instructions is very cheap; the same page stored as a JPEG is not. Folia compares both results and keeps the smaller, so you will not actually receive the larger file — but on a text-heavy document the honest answer is often that there was very little to remove.
How much can I expect to save?
On a 300 dpi scan, usually most of it — Balanced keeps roughly a seventh of the pixels. On a report of text, tables and vector charts, frequently under twenty per cent, and sometimes nothing at all. The variation is not a defect in the compressor; it reflects what the two kinds of file are actually made of.
Will compression ruin the quality?
The lossless route changes nothing you can see. The raster route changes everything: pages become JPEGs at 150, 110 or 72 dpi depending on the level, and text stops being selectable. The reliable test is to open the result and try to select a line — if you cannot, your file took the raster route.
Is there a way to compress only the images and keep the text?
Not on this site. Folia either repacks the document losslessly or rebuilds the pages as images; it does not re-encode individual image streams or subset fonts the way a file-level optimiser does. Where you need that specific operation, a desktop optimiser is the right tool, and splitting the document is usually the better answer to an attachment limit anyway.