A PDF is big for one of two reasons, and they need different treatment. Text-heavy documents carry duplicate font subsets, loosely packed object data and resources no page references any more. Scans and photo-heavy documents are almost entirely image bytes — in one of our test files, 99.7% of it. Compress PDF offers a choice between the two: rebuild the internal structure and leave every pixel identical, or also re-encode the images inside, which is the only thing that meaningfully shrinks a scan. The second is lossy and the tool says so before you run it; it is never the default.
How to use Compress PDF
- Choose your privacy mode. Sensitive keeps the file in your browser tab; Normal uploads it to our sandboxed server for larger files.
- Drop in a single PDF (this tool works on one file at a time — for combining several files first, use Merge PDF).
- Choose how much to compress. "Keep quality" rebuilds the structure and changes nothing visible. "Smaller" also re-encodes the images. "Smallest" additionally reduces their resolution — the biggest saving, and the only setting where the difference is visible.
- Click Process. Nothing is uploaded in Sensitive mode; the work happens in the page.
- Download the result. If a file is already tightly packed, you'll get it back unchanged rather than degraded — Compress PDF never hands back a file larger than what you gave it, and never trades quality for nothing.
Worked example
What to expect, by what's in the file
The spread is wider than most compressors admit, so here are our own measurements rather than a promise. On a 30-page scan of 5.8 MB: "Keep quality" returned 0.1%, because structural compression cannot touch the image bytes that are 99.7% of that file. "Smaller" returned between 12% and 55% depending on how well the scan was already encoded. "Smallest" returned 74-83%. On a 300-page text report the picture inverts — "Keep quality" gave back 11% and the image settings changed nothing at all, because there are no images to re-encode. The rule of thumb: if your file is mostly text, the lossless setting is all there is to get; if it is a scan, the image settings are the only ones that matter.
Processing limits & specifications
| Mode | Location | Max file size | Retention |
|---|---|---|---|
| Sensitive mode | Your device (WASM) | Up to 75 MB (typical device) | Never stored / zero egress |
| Normal mode | Server sandbox | 300 MB | Deleted when you close the page |
Frequently asked questions
Will compressing change how my PDF looks?
With "Keep quality", no — that setting only restructures how the file is stored, and every page comes back pixel-identical. "Smaller" re-encodes the images inside, which is lossy, though it is tuned so the difference is hard to see. "Smallest" also lowers their resolution, and there the difference is visible: text inside a scan can look softer. The tool tells you which of these you have chosen before you run it.
Which setting should I use?
Start with "Keep quality" if the document is mostly text, or if it is something you may need to reproduce exactly. Use "Smaller" for a scan or a photo-heavy document you want to email. Use "Smallest" only when hitting a size limit matters more than fidelity, and keep the original — reducing resolution cannot be undone.
Why did my scan barely shrink on "Keep quality"?
Because a scan is mostly image data and that setting never touches images — it reclaims redundancy in how the document is stored, and on a scan that is a fraction of a percent of the file. This is the case the other two settings exist for.
Why did it say "No further structural reduction possible"?
Some PDFs are already efficiently packed — ones already compressed by another tool, or that are mostly text and vector graphics with little redundant stream data left. The same check runs image by image: if re-encoding a picture would not actually make it smaller, that picture is left exactly as it was rather than quietly losing quality for no gain. Rather than hand back a larger or byte-identical file and call it "compressed," we tell you directly and let you keep the original.
Does this remove any content from my PDF?
It removes unreferenced resources — objects like unused fonts or images that no page in the document actually points to — but never removes visible page content, text, or annotations that are in use.
Can I compress a password-protected PDF?
Not in V1. Encrypted PDFs are detected and returned with a clear error rather than a silent failure; you'd need to remove the password with the PDF's original software first.
Is there a file size limit?
In Sensitive mode, the limit depends on your device's available memory (shown on this page before you start). In Normal mode, files are capped at 300 MB and processed within a 120-second time limit.
How is this different from a "compress PDF" tool that rasterizes each page?
Some online tools shrink file size by converting every page into a compressed image, which quietly destroys selectable text, searchability, and print sharpness. Compress PDF here never rasterizes pages — it recompresses the PDF's existing internal objects and streams, so any text that was selectable before stays selectable, and vector graphics stay vector graphics.
Can I compress a PDF with embedded forms or annotations?
Yes. Form fields, annotations, and their appearance streams are preserved — compression touches how the underlying objects are stored, not which objects exist in the document.
Why would I choose Sensitive mode over Normal mode for compression, if the result is the same either way?
The result is structurally equivalent either way, but Sensitive mode means the file is never transmitted anywhere — everything happens inside your browser tab using a WebAssembly build of the same underlying engine. Normal mode is there for files too large for your device's available memory, or when you're on a device where running the WebAssembly engine isn't practical.
Why did my scanned document barely shrink?
Because a scan is mostly image data, and this tool does not touch images — it recompresses the file's structure around them. A text-and-vector PDF with redundant object data has plenty to reclaim; a page that is one big photograph has almost none. If the saving is small, that is the honest result for this kind of compression rather than a failure.