Compress a PDF

1Drop the PDF inIt never leaves your machine.

2Choose how hard to squeezeEverything updates as you change it.

3Build and downloadAssembled here and saved from your own browser.

Compression that tells you the truth

The saving is measured on your actual file rather than promised in advance, and if the result comes out larger you are told instead of being handed a worse document.

How to use this tool

Drop a PDF on the box at the top. Before anything is changed, every page is checked for a real text layer, and you get a plain verdict on whether compressing this particular file is worth doing.

If it is, set a resolution and a quality and press the button. The result reports the actual size change, measured after the fact rather than promised beforehand.

The original file on your disk is never touched. If the result comes out larger, and on some documents it will, you simply keep what you already had.

What compressing a PDF actually does here

There is no general purpose way to make a PDF smaller in a browser without changing what is in it. What this tool does is specific and worth stating plainly: it renders every page as a picture at a lower resolution, compresses those pictures as JPEG, and builds a new document out of them.

That is a real technique and it is what most compression tools do. It is also destructive in a particular way. Whatever was on the page as instructions, text you could select, vectors that stayed sharp at any zoom, becomes a grid of pixels at one fixed resolution. There is no undo.

So it is superb on a scan and wrong on a report. A scanned page was already a picture, and re-rendering it smaller loses only detail you were unlikely to use. A typed report is instructions, and turning it into pictures throws away everything that made it a document.

Why a text PDF gets bigger, not smaller

This is the part that surprises people, and it is the reason the tool checks before it acts.

A page of text in a PDF is stored as a short list of instructions: use this font, put these character codes at these positions. The font itself is stored once for the whole document. A dense page of prose might cost two or three kilobytes. That is already an extraordinarily efficient representation, far better than any image format could manage, because it is not describing the appearance of the letters at all, only which letters go where.

Render that page at 120 dpi and you get roughly a million pixels. Even compressed hard as a JPEG that is tens of kilobytes. So you have taken something that cost three kilobytes and made it cost forty, while destroying the ability to search it. Both numbers moved the wrong way at once.

This is why any tool promising to compress any PDF is either doing something different from what this one does, or quietly making some files worse. When a document is already efficiently stored, there is nothing to remove, and the honest answer is that it is as small as it is going to get.

A worked check you can repeat

Take an eight page A4 scan totalling 12 MB, which is 1.5 MB a page. That is a typical flatbed scan at 300 dpi in colour. No page has a text layer, so all eight are image pages and the verdict is that it is a good candidate.

At 120 dpi the scale is 120 divided by 72, or 1.6667. An A4 page of 595.28 by 841.89 points becomes 992 by 1403 pixels, which is 1.39 megapixels a page and 11.1 across the document. At quality 70 that lands somewhere around 2.6 MB, a saving of roughly four fifths.

Now the opposite case. Eight A4 pages of typed text totalling 300 KB, which is 37 KB a page. Every page has a text layer, so the verdict is to leave it alone, and the projection confirms why: the same render produces about 2.6 MB of image data from a 300 KB document. Nearly nine times larger, and no longer searchable.

The maths behind this was checked against an independent implementation: forty three hand written cases and thirty thousand generated ones, covering render sizing at every resolution, the savings calculation including the case where the file grows, and the recommendation thresholds. That fuzzing found a real bug in the sizing code, where clamping an oversized page to the browser canvas limit could round back over the limit and produce a blank page. It is fixed here and in the other tools that shared the same function.

Mistakes that catch people out

Compressing a document you will need to search later. Once the pages are images, the text is gone as text. If the file needs to stay searchable, do not compress it this way at all.

Reaching for the lowest quality first. Below about 50 the JPEG artefacts around text become obvious, and on a scan of a document that is exactly where you are looking. Start at 70 and only go lower if the size still will not fit.

Using greyscale on a colour document. It helps enormously on a black and white scan, where the colour was noise. On anything with photographs or coloured markup you lose the colour and save comparatively little.

Expecting a fixed percentage. How well a page compresses depends entirely on what is on it. A blank scan collapses to almost nothing, a page of dense photography barely moves. The estimate before you run is a guide; the number afterwards is the answer.

Compressing twice. Running an already compressed file through again degrades it further for very little gain, because you are re-encoding JPEG artefacts as new JPEG artefacts.

Common questions

Is my PDF uploaded anywhere?

No. The file is opened, rendered and rebuilt by your own browser, and the download comes from your machine. Nothing crosses the network, so there is nothing on a server to store or delete.

Why does it tell me not to compress my document?

Because the pages carry real text. Compressing works by turning every page into a picture, which destroys that text permanently: it can no longer be searched, selected or copied. On a text document the result is usually larger than the original as well, so there is nothing to gain.

Why did my file get bigger?

Because it started as text. Text in a PDF is stored as instructions, which is extremely compact. Turning a page of instructions into a photograph of that page produces far more data. The tool reports the real figure after the fact rather than hiding it.

Will the text still be searchable afterwards?

No. Every page becomes an image, so any text layer is gone. That is why the tool is only recommended for scans, which never had a searchable text layer in the first place.

What resolution should I use?

120 dpi stays comfortably readable on screen and is a good default for a scan being emailed. Use 150 or 200 if the result will be printed. Below 96 dpi small text starts to break down.

Does greyscale help?

On a scan of a black and white document, a great deal, because the colour information was noise anyway. On a document with photographs or coloured drawings it saves less and costs you the colour, so it is rarely worth it.

Related PDF tools and reference

Reference: ISO 32000, the standard that defines how content is stored in a PDF.