tooldura
PDF Tools

Why Some PDFs Shrink a Lot and Others Barely Move

T
tooldura editorial
8 min readUpdated November 11, 2026Open tool →

Take two ten-page PDFs. One is a contract run through an office scanner, the other is a report exported from a word processor. The scan weighs 30 MB, the report 400 KB. Put both through the same compressor and the scan may come out at a few megabytes while the report hardly changes. Neither result means the tool worked well or badly. The difference comes from what each file is made of, and once you know where the bytes sit in a PDF, you can predict roughly how much any compressor will save before you run it.

What a PDF Is Made Of

A PDF page is a list of drawing instructions: put this text here in that font, draw this line, paint this picture in that rectangle. The instructions themselves are small. A full page of text is a few kilobytes, and most exporters already store it compressed with Flate, the same lossless method used in ZIP files and PNG images.

Around those instructions sit three other kinds of data. Fonts are embedded so the page looks the same on a computer that does not have them. Many exporters already include only the characters the document actually uses, a technique called subsetting. Images are stored as separate objects and drawn wherever the instructions say. And a layer of structure ties everything together: the page tree, bookmarks, links, form fields and the cross-reference table that tells a reader where each object starts.

In a text document, fonts and structure make up most of the file and none of it has much slack left. In a scan, each page is one large photograph of the paper, and the images are close to the whole file. That is why the Compress PDF tool shows the split between images, fonts and everything else as soon as a file is opened.

What Usually Shrinks and What Does Not

Typical content and what an image-focused compressor can do with it. The actual saving depends on how the file was made.

What the PDF containsTypical sourceRoom to shrink
Colour or greyscale scans stored as JPEGOffice scanner, phone scanning appLarge, mostly from lowering the resolution
Phone photos placed on pagesSlides, reports, photo booksLarge, photos are often far sharper than the page can show
Screenshots and diagrams stored losslesslyManuals, exported slidesModerate, they should stay lossless to keep edges clean
Text, tables and vector chartsWord processors, spreadsheetsSmall, already compressed
Embedded fontsAny document with custom fontsNone here, removing glyphs needs a font engine
Black-and-white scans (CCITT or JBIG2)Fax-style document scannersLittle, these formats are already very compact

Pixels the Page Can Never Show

The biggest saving comes from resolution. A photo from a modern phone is easily 4,000 pixels wide. An A4 page is 210 mm, about 8.27 inches, wide. Spread across the full width of the page, that photo carries about 484 pixels per inch, while a screen shows the page well at around 150 and a good office printer at around 300.

Scale the photo down to 150 pixels per inch at full page width and it becomes about 1,240 pixels wide. Because both sides shrink, the pixel count falls to roughly 9.6 percent of the original, and the stored image usually shrinks by a similar order. Nothing visible is lost on screen, because the extra pixels were never displayed at that size.

Ghostscript, the open-source PDF engine behind many compressors, names its presets after this idea: screen targets 72 pixels per inch for colour images, ebook 150 and printer 300. It also only downsamples an image whose resolution is more than 1.5 times the target, so images that are only slightly over are left alone. Compress PDF uses a related but more cautious rule. It measures every image against the largest page it appears on, as if it filled that page, and never takes it below the chosen resolution at that size. A small logo therefore keeps more pixels than it strictly needs, but no image can come out softer than the setting promises, whatever size it is drawn.

Re-Saving a JPEG and When Not To

JPEG is lossy. Each time an image is decoded and encoded again, a little more detail goes, an effect known as generation loss. A compressor that re-saves every JPEG at a lower quality setting earns a small saving at a visible cost, especially on scans that were already compressed hard by the scanner.

So the useful rule is to re-encode only when it buys something. Compress PDF re-saves a JPEG when the image is being scaled down, or when the new copy is at least 10 percent smaller. Otherwise the original bytes stay in the file untouched, and the report under the result says how many images were left alone and why.

Losslessly stored images need the opposite caution. A photo stored with Flate is often several times larger than the same photo as a JPEG, so converting it is a big win. A screenshot full of text and sharp lines compresses well losslessly and comes out of JPEG with grey smudges around every letter. The tool tells the two apart by result: a lossless image becomes a JPEG only if the JPEG is at most half its current size at the Recommended setting. Photos clear that easily, screenshots rarely do. The Print quality setting never converts lossless images at all.

🔄

Why the JPEG's own metadata is removed before re-encoding

A JPEG copied from a phone can carry an Exif orientation tag that says to turn the picture before showing it, and an embedded colour profile. A PDF reader ignores both: it draws the stored pixels unturned, in the colour space the PDF itself names. A web browser decoding the same JPEG applies them, so a sideways photo would come back upright and the page would show it rotated. Compress PDF strips those two segments before the browser decodes the image, so the re-saved picture keeps exactly the orientation and colour values the PDF already used.

What Lossless Cleanup Still Finds

Even with every image left alone, a PDF can carry weight that nobody sees. PDF editors often save changes as an incremental update: the edits are appended to the end of the file, and the replaced objects stay in it, unreachable. A file that has been edited many times, or had pages deleted, can hold old images and old page content that no reader will ever display. Rewriting the file from the objects that are still reachable drops them.

Some exporters embed the same logo or background once per page instead of once per document. Byte-identical copies can be merged into one, with every page pointing at it. Old or hand-made PDFs sometimes store page content with no compression at all, and Flate packs that losslessly. Finally, PDF 1.5 added object streams, which pack many small objects together into one compressed block, along with a compressed cross-reference stream to match. On a document with thousands of small objects, such as a long report with many bookmarks and links, that alone can save a noticeable share.

Compress your PDF

See what takes up the space, pick a level, and get a report of every image that was resized, re-saved or left as it was.

Open Compress PDF →

What a Browser-Based Compressor Leaves Alone

Some content cannot be re-encoded safely with the image tools a browser has. CMYK images, common in files prepared for commercial printing, would have to pass through RGB on a canvas, and the colours would shift. Images in spot colours or indexed palettes, and images whose colours are remapped by a Decode array or masked by exact colour values, would change appearance if their samples changed. Black-and-white scans in CCITT or JBIG2 and images in JPEG 2000 are formats a canvas cannot write. All of these are kept byte for byte and listed in the report.

Fonts are the other limit. Cutting unused glyphs out of an embedded font means parsing and rewriting the font program, which is the job of a font engine, not an image compressor. Because many exporters already subset fonts, the saving would often be small anyway.

When nothing in a file has room to shrink, the honest result is that it stays the same size. If the compressed copy comes out no smaller than the original, the tool says so and does not download it.

Compressing a Signed PDF

A certificate-based digital signature covers exact byte ranges of the file, recorded in the signature's ByteRange entry. Any rewrite moves those bytes, and a reader will then report the signature as invalid, even if every page looks the same. Compress PDF warns when a file it opens carries a signature. The workable order is to compress first and sign the compressed file, or to keep the signed original for anything that depends on the signature.

Frequently Asked Questions

Found this guide useful? Pass it on.

Related Tools

Keep Reading