Compress PDF
Compress PDF makes a document smaller in your browser. The default, balanced mode recompresses the photos and scans inside it and leaves text, fonts and links untouched, so the file stays searchable. A lossless pass only clears out what an edited file leaves behind, and rasterising turns every page into a picture — the biggest saving for a scan, at the cost of its text.
Drop a PDF here
or drop anywhere on this page · Choose a file · 100 MB
Accepted: PDF
How does Compress PDF work?
“Compress PDF” is not one operation. A PDF's size can come from very different places, and the fix for one does nothing for the others — which is why a single magic button so often returns a file the same size as the one you gave it, or a file whose text has silently stopped being searchable.
Where the bytes actually are
In a scanned document, essentially all of them are in page images. In a report exported from a word processor, they are usually in embedded font subsets and in structural overhead. In a file that has been edited and re-saved a dozen times, a surprising share is in material that is no longer referenced at all: PDF supports incremental saving, where each edit appends to the file rather than rewriting it, so previous revisions can still be sitting in the bytes.
The balanced pass — the default
Most large PDFs are large because of their pictures: a phone photo pasted into a report keeps its twelve megapixels, a scanned page keeps the scanner's 300 or 600 DPI. The balanced pass finds every JPEG image in the file, scales it down to a size that still prints sharply — about 160 DPI across an A4 page at the recommended strength — and encodes it again. Text, fonts, vector drawings, links and bookmarks are not touched, so the document stays searchable and selectable. An image that would not shrink by at least a tenth keeps its original bytes, and CMYK or spot-colour images are left alone, because quietly converting a print file to RGB would be a defect, not a saving.
The lossless rebuild
This copies every page into a brand-new document. Copying walks the object graph from the page tree, so only what is actually reachable comes across — old revisions, orphaned fonts, images belonging to deleted pages and abandoned annotations are simply not visited. The result is then written with object streams, which pack the many small dictionaries describing pages, fonts and links into compressed streams rather than plain text.
Nothing is re-encoded. Text stays selectable and searchable, vector art stays vector, images are untouched. On a heavily edited file this can remove a large fraction; on a file that was written cleanly in the first place it removes almost nothing, and this page reports the real number rather than a flattering one.
The rasterising pass
This renders each page with pdf.js — Mozilla's engine, the one Firefox uses — and rebuilds the document from those images as JPEGs. For a scan the difference is enormous: a 200 MB scan at 600 DPI becomes a few megabytes at 120 DPI with no visible loss, because the original resolution was far beyond what the content justified.
It is also destructive in a way worth stating plainly. The text layer is gone. The document is no longer searchable, no longer selectable, no longer readable by a screen reader, and links stop working. For a scan there was no text layer to lose and this costs nothing. For a text document it turns a good file into a bad one, which is why it is not the default.
Choosing a resolution
120 DPI is legible on screen and prints acceptably. 150 DPI is comfortable for printing. 200 and above rarely earns its size for a scan of ordinary text. The JPEG quality matters just as much: 0.6 is the sweet spot for scanned text, where artefacts sit in the paper texture and nobody notices them.
What this tool will not claim
It will not promise a percentage. Any site advertising “up to 90% smaller” is quoting the rasterising case for a high-resolution scan and applying it to everything. If your PDF is already efficient, the honest answer is that it cannot get much smaller without losing something, and that is what you will be told here.
Everything happens in this tab
The file is processed in this tab. Required libraries or models may be downloaded when needed; the file itself is not uploaded by this operation.
Input
scan.pdf — 48.2 MB 12 pages at 600 DPI mode: rasterise, 120 DPI, q0.6
Output
scan-compressed.pdf — 2.1 MB -96% text layer: none (there was none)
What options and edge cases does Compress PDF support?
| Parameter | Type | Default | Behaviour & edge cases |
|---|---|---|---|
| Balanced | mode | default | Recompresses the JPEG photos and scans inside the PDF and leaves text, fonts, vectors and links exactly as they were. The usual saving on a document with pictures is half or more. |
| Strength | Light · Recommended · Strong | Recommended | How far photos are scaled and re-encoded: up to 2400 px at quality 0.82, 1800 px at 0.72, or 1200 px at 0.55 on the longest side. Recommended is about 160 DPI across an A4 page. |
| Lossless rebuild | mode | — | Copies reachable objects into a fresh document and repacks with object streams. Text and vectors untouched. |
| Rasterise | mode | — | Re-renders pages as JPEGs. Huge saving on scans; permanently removes the text layer. |
| 120 DPI | resolution | default | Legible on screen, acceptable in print. The right default for a scan. |
| 150 DPI | resolution | — | Comfortable for printing. Roughly 1.5× the size of 120. |
| 200 DPI | resolution | — | Rarely worth it for ordinary text. Nearly three times the size of 120. |
| JPEG quality | 0.3 – 0.9 | 0.6 | 0.6 suits scanned text — artefacts hide in the paper texture. Raise it for photographs. |
| Strip metadata | checkbox | on | Drops title, author and producer. A few hundred bytes, and one less thing leaking who made the file. |
| Page size | preserved | — | The rebuilt page keeps its original size in points, so an A4 page prints as A4 whatever DPI was used. |
Frequently asked questions
Why did the lossless mode barely change my file size?
Because your PDF was already efficient. The lossless rebuild removes objects nothing references and repacks the structural dictionaries — a large win on a file edited and re-saved many times, close to nothing on one written cleanly by a modern exporter — and it never touches images. If the bytes are in photos or scans, the balanced mode is the one that moves the number while keeping the text; rasterising goes further, at the cost of the text layer.
Will compressing make my PDF text unsearchable?
Not in the default balanced mode, and not in the lossless one: both leave the text layer, fonts and links exactly as they were and only rewrite pictures (balanced) or structure (lossless). Rasterising is different — it replaces each page with a picture of itself, so text stops being text: no searching, no selecting, no screen reader, no working links. For a scan that costs nothing, because there was no text layer to begin with; for a report it is a serious loss, which is why it is a deliberate choice.
What DPI should I use for a scan?
120 for something that will be read on screen and occasionally printed — it is legible and it is usually a 90%+ reduction from a 600 DPI scan. 150 if printing matters. Above 200 you are storing detail that the scanner's optics and the paper itself did not really contain. Pair it with JPEG quality around 0.6: scanned text hides compression artefacts in its paper texture better than almost any other content.
Can it compress a password-protected PDF?
Not while it is encrypted. A PDF with a user password cannot be read at all without that password, so there is nothing to rebuild. Use the PDF unlocker first with the password you have. If you do not have it, no tool can help — and any site claiming otherwise is either uploading your file to a cracking service or simply lying about what it does.
Is there a size limit?
100 MB per file, which is your browser's memory rather than a paywall. A PDF is held in memory several times over during a rebuild — parsed graph, new graph, serialised output — so a larger file would exhaust the tab and lose the work. Rasterising a very long document is the heaviest case, and each page's canvas is released as soon as its image is encoded specifically so that it remains possible.
Is my document uploaded?
No. The rebuild runs in a Web Worker in this tab using pdf-lib, and the rendering uses pdf.js, also locally. Open DevTools and watch the Network panel while you compress — nothing carries your file.