Compress PDF

Shrink a PDF so it fits under an upload limit, an email cap, or a slower connection — without uploading it to anyone. Compress PDF works in two passes: it repacks the file losslessly, removing duplicate objects, orphaned data and the metadata packet, then downsamples and re-encodes the images, which is where nearly all of a normal PDF's weight sits. Text, fonts and vector lines are never touched, so they stay sharp. Three levels set how hard the image pass pushes: Light keeps images close to their original detail, Recommended is the default balance, and Strong goes for the smallest file. If nothing can safely be saved, the original bytes come back unchanged rather than a worse file.

How to use Compress PDF

1. Click "Select PDF file" and choose the PDF to shrink.
2. Pick a compression level. Recommended is the default because it usually cuts a scanned document the most without visible damage; choose Light for image-heavy brochures or Strong when the file must get under a hard limit.
3. Click "Compress PDF". A text-only PDF compresses structurally; a scan is re-encoded, so it takes longer.
4. Download the smaller file and check the pages you care about — a high-contrast scan holds up better than a photograph of a page.

Who it is for

Compress PDF is for the moment a document will not go where it needs to: an application portal that rejects anything over 5 MB, an inbox with a 25 MB ceiling, a printing service that charges by upload size, or a product manual that has to load quickly on a phone. The usual case is a scanned document, because a scan is mostly photographs and photographs are most of the bytes. The other case is a PDF exported from design software that carries far more image data than the page actually uses. Both respond to the same treatment here. If you are sending a contract, a medical letter or anything confidential, the compression happens on your own machine, which is the reason people look past the big sites for this job.

How it works

• The tool builds two candidate files and keeps whichever is smaller. The first is a lossless structural re-save: pdf-lib repacks the document into object streams, which is the same packing a well-made PDF uses, and drops orphaned objects. That costs nothing in quality and is all some files need.
• The second candidate also downsamples and re-encodes the images. Every image above 3 KB is decoded, scaled so its longest side fits the level's ceiling, and written back as JPEG at the level's quality. This is where the real savings sit, because images are nearly all of a normal PDF's weight.
• The original file is always in the running as a third candidate, so the output can never be larger than what you put in. If neither pass beats the original, you get the original bytes back with a message saying so, rather than a "compressed" file that is worse.
• A re-encode is only kept if it saves at least five percent. A small image that would lose detail for a handful of bytes is left alone, which is why the loss is concentrated in the images that were actually heavy.
• Text, fonts and vector lines are never touched by either pass. They stay as instructions rather than becoming pixels, so they remain sharp at every zoom level and at print size, and the text stays selectable and searchable.

What it cannot do

• A text-only PDF barely shrinks. Once the structural waste is gone, the remaining bytes are the fonts and the page instructions, and discarding those would mean changing the page. Expect little movement on a letter typed in Word; the gains are in scans and image-heavy documents.
• Compression is lossy above Light. Recommended and Strong re-encode images at lower quality, and on a photograph of a page that shows as softness in fine print. Always open the result and look at a dense page before sending it on.
• An encrypted PDF is opened but not usefully compressed. The image pass cannot re-encode what it cannot decode, so you get the lossless pass only and usually a file identical in size. Remove the password first and compress that copy.
• Files over 100 MB are skipped with a message rather than attempted, because a large scan decoded in a browser tab will exhaust memory. Splitting the document and compressing the parts is the way to handle one that big.
• There is no "target file size" mode. You choose a level and see what it gives; getting under a specific number can take two attempts at different levels.

Frequently asked questions