How to compress a PDF here
The starting setting, Images at Balanced, suits most scans and photo-heavy documents.
- Add the PDF, or several. Try with a sample PDF loads a photo report made for this site.
- Choose a Level: 1 · Clean-up, 2 · Images or 3 · Rasterise. Or switch on Target size and type the most the file may weigh, with quick buttons for 1, 2, 6, 10 and 25 MB.
- For Images, pick a preset under Image quality; the line beneath it spells out the JPEG quality, the resolution limit on the page and the long-edge limit in pixels.
- Open More options for Greyscale, Remove metadata, Remove attached files and, for a file opened with a password, Keep the password.
- Press Compress PDF. The progress bar names each step, and Cancel stops it.
- Read the result, compare a page, and download. One smaller file saves by itself; several arrive as compressed-pdfs.zip.
Copies are named after the original with -compressed added, so statement.pdf comes back as statement-compressed.pdf and the file on your disk is never overwritten.
What happens to your file: it stays on your device
A PDF worth compressing is usually going somewhere official: passport scans for a visa, a medical report for an insurer. Many online compressors upload the file and keep it for a period you have to trust. Here the PDF is read into this tab’s memory, rewritten by code running in the tab, and handed back as a download. No request carries the file, its name, its size or its password.
The 60-second check is especially clear for compression, because the files are big. Open the Network tab in your browser’s developer tools (F12, or Cmd+Option+I on a Mac), add a 20 MB scan and compress it. An upload would show as a request of about 20 MB; you will see the tool’s own code (pdf.js and pdf-lib, loaded once), the site’s consent and analytics requests, and nothing near the size of your document. Are online PDF converters safe? explains the same test for any site.
What makes a PDF big
Text is cheap: in our tests a 200-page PDF of plain text weighed 234 KB, little more than 1 KB a page. The weight nearly always comes from one of these:
- Images. A phone photo often runs to several megabytes, and a PDF usually keeps every pixel even when it is printed the size of a postcard.
- Scan resolution. Each scanned page is one picture; at 300 DPI an A4 page is about 8.7 million pixels. Our 17-page letter scanned that way weighed 40.7 MB.
- Fonts embedded whole rather than cut down to the letters used, above all Chinese, Japanese or Korean ones.
- Duplicates: the same logo or font stored again for every page.
- Attachments, and old objects left behind by repeated editing.
For the longer story, including how the settings used to make a PDF decide most of its size, read How to reduce PDF file size.
The three levels: clean-up, images, rasterise
Level 1, Clean-up, is lossless: every page looks identical, pixel for pixel. It packs small records into compressed object streams, drops embedded page thumbnails and named destinations no link uses, keeps one copy of any stream stored twice (compared after decompression, so an image packed two ways still counts), compresses streams stored raw or as ASCIIHex or ASCII85 text with Flate, and deletes objects nothing refers to. Expect 5 to 30%. In our tests a 200-page text PDF went from 234 KB to 197 KB, a 2-page form from 48 KB to 35 KB, and a page of uncompressed photos from 11.2 MB to 5.0 MB; a 40.7 MB scan of JPEG pages did not change, since clean-up cannot shrink JPEG pictures.
Level 2, Images, is where scans shrink. Each image is handled once, however many pages use it. JPEGs are decoded by your browser, reduced in pixel size if they exceed the preset’s limits, and saved again at its quality. Other pixel images are sorted by sampling 4,096 pairs of neighbouring pixels: photos become JPEG, while charts and screenshots, mostly flat colour, are re-packed losslessly so edges stay sharp. A new version replaces the old only if it is at least 3% smaller. Text, fonts and vector drawings are not touched. Our 17-page scan came to 13.9 MB with Light, 8.1 MB with Balanced and 4.7 MB with Strong, each in about 2 seconds; a 3.1 MB photo report came to 406 KB, 100 KB and 52 KB.
Level 3, Rasterise, redraws every page as one JPEG picture at the DPI and quality you set. It runs only if you choose it or allow it in Target size, because the text stops being text, and it is not automatically smallest: at 150 DPI the same scan came to 5.9 MB, more than Strong’s 4.7 MB. It earns its place on pages the image level cannot help, such as those full of the skipped image kinds below, or dense vector maps and plans.
Target size: under 2 MB, 12 MB or 25 MB
Type the most the file may weigh and the tool searches for the gentlest setting that fits, instead of making you guess a preset:
- Clean-up runs first. If the lossless result fits, you get that, with nothing else changed.
- If not, it uses a ladder of ten image settings, from quality 85%, 300 DPI and 2,400 px on the long edge down to 40%, 72 DPI and 800 px, and tries the strongest step first. If even that is too big, it stops there.
- Otherwise it halves the distance, trying the middle of the ladder, then the middle of whichever half still holds the answer, closing in on the gentlest step that fits. Each try is a full compression weighed as a real file, with at most six tries in about 150 seconds.
- It rasterises, from 150 DPI down to 72, only if images cannot get there and you switched on Allow redrawing pages as images if needed. It cannot subset fonts, so a file whose weight is in its fonts has a floor.
Three examples with our 40.7 MB scan. For a portal that allows 2 MB per file, a 2 MB target gave 1.6 MB in 7 seconds, at quality 45%, 85 DPI and 900 px on the long edge; check the fine print at 200% before submitting. For the 12 MB per-file limit on USCIS online filing, a 10 MB target already stopped at the Balanced step, 8.1 MB, so 12 MB is met at that step or a gentler one. For email, where Gmail, Outlook.com and Yahoo Mail allowed 25 MB of attachments when we checked in September 2026, the ladder’s gentlest step equals the Light preset: 13.9 MB.
Some visa services are reported to accept only 2 MB or 6 MB per file, figures we could not confirm on the providers’ own pages, so follow the portal’s instructions. The box counts 1,048,576 bytes to the megabyte; for a limit counted in thousands, type 1.9 for 2 MB.
When a target is out of reach, the result says so rather than handing over a ruined file. Asked to get our 234 KB text PDF under 100 KB, it returned 192 KB, noting that the file has no images left to shrink and that text, fonts and drawings take about 192 KB, which only a different export can reduce. For files with images, it says how much is images and how many could not be recompressed. Split PDF cuts by size, and How to send a large PDF covers other routes.
Images that are skipped, and why
Some images cannot be re-encoded here without risking how they look, so they are left exactly as they were and listed with the reason. In our tests a transparent PNG, a CMYK JPEG, a 1-bit scan and a JPEG 2000 image came through unchanged while the photo beside them was recompressed.
- Transparency. A see-through image is a picture plus a separate mask that must line up with it. JPEG has no transparency, and shrinking one without the other risks fringes.
- CMYK and special colour. Print files store images in printing inks, palettes or spot colours; a browser works in screen RGB, and converting would change the colours a printer was sent.
- JPEG 2000, JBIG2 and CCITT fax. Browsers cannot write these. JBIG2 and CCITT are black-and-white compressions already far smaller than JPEG would be.
- 1-bit and 16-bit images. A fax-style black-and-white scan would grow as a JPEG; 16-bit samples hold more precision than a browser canvas keeps.
- Custom Decode arrays, which remap stored values; re-encoding could turn the picture into its own negative.
- Small and inline images: under 64 pixels there is nothing to gain, and inline images sit inside a page’s drawing instructions, which the tool does not rewrite so that nearby text cannot be disturbed.
If skipped images carry the weight, go back to the source: export again with a lower image resolution, ask a designer for a screen version of a print file, or rescan black-and-white paperwork in greyscale at 200 DPI. Otherwise Rasterise turns everything, skipped kinds included, into plain JPEG pages, at the cost of the text. PDF vs JPG vs PNG covers which format suits which scan.
Quality, DPI and the long-edge limit
Two limits decide how many pixels an image keeps, and the smaller wins. The DPI limit is about the page: the tool reads each page’s drawing instructions to learn the size every image is printed at, works out its effective DPI, and reduces anything above the preset’s figure (300 for Light, 200 for Balanced, 150 for Strong). The long-edge limit (2,400, 2,000 and 1,600 px) caps pixels whatever the placement. Reductions under 5% are skipped, and the report says which rule applied.
The DPI rule matters for photos placed small: in our photo report, 3000 × 2250 photos printed about 6.6 inches wide (roughly 450 DPI) came down to 200 DPI at Balanced, about 1,320 pixels wide. For full-page scans the long edge usually decides:
- 150 DPI: A4 1,754 px on the long edge, US Letter 1,650. Light and Balanced keep the pixel size; Strong trims A4 to 1,600.
- 200 DPI: A4 2,339 px, Letter 2,200. Light keeps them; Balanced caps them at 2,000, about 171 DPI on A4.
- 300 DPI: A4 3,508 px, Letter 3,300. Light reduces to 2,400 (about 205 DPI on A4), Balanced to 2,000, Strong to 1,600 (about 137). Our scan’s pages became 1,414 × 2,000 at Balanced.
Quality is separate: it sets how much detail JPEG keeps per pixel, and lower settings add faint smudges around letters, so small print, signatures and stamps are what to inspect. Custom lets you set all three.
Greyscale helps a colour scan of black-and-white paperwork; on a scan already grey it changes nothing: our grey scan came to 8.1 MB at Balanced either way. Skipped images keep their colour. Lossless only never reduces pixels or uses JPEG: pixel images are re-packed losslessly and JPEGs left alone. To start smaller, JPG to PDF sets size and quality while the PDF is being made.
Text, fonts and searchability
In a PDF exported from a word processor, text is characters in a font, placed by drawing instructions. Levels 1 and 2 never change those instructions or the fonts, only image objects and packing, so every character stays selectable, searchable, readable by a screen reader and sharp at any zoom. Links, bookmarks, form fields and annotations stay too. A scan with text recognition (OCR) keeps its invisible text layer the same way, so it remains searchable.
Rasterising replaces each page with one picture: characters, OCR layer, links and form fields are gone, and a screen reader has nothing to read. Where searchable text is required, stay at level 2.
Fonts are the honest limit. Most programs embed only the letters a document uses, called subsetting. When a file carries whole fonts, the library that writes the PDF here cannot subset them, and the tool never removes fonts, since that would change how text looks. Exporting again from the original document, where subsetting is normally on, is the fix.
Checking the result
Each file gets a result card: before and after bars, the percentage saved, and What was changed, which says in plain sentences how many images were recompressed or reduced, how many duplicates merged, and which images were left alone and why.
The text check extracts every page’s text from the original and from the copy and compares them page by page. A tick means all pages match; a warning means compare the pages before relying on the file. Rasterised files get a note instead.
Compare pages lays the same page of both files over each other: drag the slider, original to the left of the line, compressed to the right. At 200%, scroll inside the box to inspect small print, account numbers, signatures and stamps, which JPEG softens first. Then open the download in a normal viewer and search for a word you know is there.
No smaller than the original means the file was already compact at that setting. A single file in that state is not downloaded automatically: keep your original, or try Strong or a target size.
When not to compress
Compression makes a new file. For some documents the original is the point:
- Digitally signed PDFs. The signature covers the exact bytes signed, so a compressed copy no longer validates; the tool warns and waits for you to tick I understand. If a signed file is too large, ask the signer for a smaller signed version.
- Legal and evidential originals. A contract, deed or court exhibit may be expected exactly as issued. Keep the original as the record.
- Print-ready files. 200 DPI can look soft on paper, and RGB images are not skipped the way CMYK ones are, so use Clean-up or Light at most.
- Archival PDF/A. The result is not checked against PDF/A rules: it uses compressed object streams, which PDF/A-1 forbids, and Remove metadata deletes the XMP that declares a file PDF/A. Compress a copy, not the master.
- E-invoices. Some formats carry machine-readable data as an attached file, which is why Remove attached files is off by default.
Your data
Only settings are remembered, in this browser’s local storage under dtc.filetools.v1: level, preset, custom values, target size and the toggles. The PDFs, their names, the results and any password stay in the tab’s memory and are gone when you close it. There is no counter, no account and no record of what you compressed.
General information about compressing PDF files. Compression changes the file, so check the result, especially small print and signatures, and keep the original. Email services, portals and offices set their own size limits and change them; follow their current instructions.