Compress a PDF
Shrinks the pictures inside the document, which is what makes a heavy PDF heavy. Every row tells you what it re-encoded and what it left alone, including when the answer is nothing.
click to choose · or paste from the clipboard
Nothing here is uploaded. See how the page enforces it.
What this actually does, and what it does not
A heavy PDF is almost always heavy because of the pictures in it. A scanner produces one JPEG per sheet at 300 DPI; the text around them costs a rounding error by comparison. So this finds those pictures, re-encodes them at the quality you pick, optionally shrinks the very large ones, and leaves the rest of the document exactly as it was.
| Document | What to expect |
|---|---|
| Scanned pages, photos, screenshots | Large reductions, often half to a fifth of the original |
| A report with a few illustrations | Modest, proportional to how much of it is picture |
| Pure text, a contract, an invoice | Almost nothing, and the row will say so |
That last row is the honest part. A text-only PDF has nothing here to compress. Instead of reporting a two percent saving as a success, the row states that it found no images it could safely re-encode. Fonts and vector art are left untouched, because rewriting them is how a document ends up with the wrong glyphs.
What is deliberately left alone
An image is only rewritten when the result is certain to look the same, just lighter. Anything else is passed through untouched and counted on the row:
- CMYK images. They decode unreliably in a browser and would come back with their colours inverted. A print-ready document is safer left as it is.
- Palette and stencil images: logos, line art and scanned masks. JPEG is the wrong format for them and would look worse at a larger size.
- Images with a remapped sample range. Re-encoding would change how they are displayed.
- Anything that came out bigger. Re-encoding an already-tight JPEG usually grows it, so the original is kept. A "compressed" file that got heavier is the one outcome nobody came here for.
Why it is not the 80% the other sites advertise
Tools that claim eighty percent on any document run Ghostscript on a server, which rebuilds the file from the ground up, resampling every stream and subsetting every font. Doing that in a browser means shipping Ghostscript compiled to WebAssembly: a ten-megabyte download that has to be fetched at runtime.
This site is served with a Content-Security-Policy of connect-src 'none',
which forbids the page from making any network request at all. That is the guarantee
everything else here rests on, and buying a bigger headline number would have meant putting a
hole in it. So the honest trade was taken: a smaller claim, kept completely.
Questions
+ My file barely got smaller. Is it broken?
Probably not. Check what the row says: if it reports no images it could safely re-encode, the document is text, fonts or vector art, and there is nothing in it that compresses. If it says it re-encoded images and the total still barely moved, they were already well compressed.
+ Will the pages look worse?
At the default of 70% the difference on a scanned page is very hard to see, and the file is typically a fraction of the size. Drag the quality up if you are archiving something and want to be safe; drag it down if you are up against an upload limit.
+ What does "shrink images wider than" do?
It caps the resolution of the pictures. A phone photo inside a PDF is often 4000 px wide, which is far more than any screen or printer will use. Bringing it to 1600 px loses nothing visible and is usually the single biggest saving. Set it to 0 to leave dimensions alone and only change quality.
+ Are my files uploaded anywhere?
No. The PDF is read and rewritten by your own browser. Open devtools, watch the Network tab while you run it, and nothing leaves. The page keeps working with your connection switched off, which is the part worth caring about when the document is a payslip, a contract or a medical letter.
+ Does it change the text or the page count?
No. Only image streams are rewritten. Text stays selectable and searchable, the page count is unchanged, and nothing is re-rendered into a picture.