PDF compressor

How to use it

1. Drop one PDF on the card, paste with ⌘V / Ctrl+V, or press “Choose PDF”. Processing starts automatically in Balanced mode; the before and after sizes appear in the same panel. There is no Compress button.
2. Pick Clean to remove metadata only, Balanced to re-encode large JPEGs while keeping text and links, or Strong to turn every page into an image. The line under the modes always says what the current mode keeps and loses.
3. When the build finishes, the status turns green and shows how much was saved. Preview opens the actual output in a dialog; download saves it as “{name}-compressed.pdf”. If the output is not smaller, no download is offered.

Example

Large JPEG scans, text and links on each page

3.1 MB · 83.2% smaller

These sizes are illustrative: the real result depends on the images and structure of your file. Clean may only save a few percent, and an already optimised PDF may not get smaller at all.

Rules, limits and known issues

• Nothing is uploaded: the PDF is read with the File API and compressed in this tab. The page sends no network request, writes nothing to storage, and keeps no history — a reload starts empty.
• A smaller file is not guaranteed. Text-only, already optimised or protected PDFs may show no gain; in that case the tool refuses to offer a download rather than handing you a larger file.
• Clean removes the Info dictionary and XMP metadata, then rewrites the file with object streams. Page content, images, fonts, links, forms, bookmarks and attachments are not changed. Savings are often only a few percent.
• Balanced only re-encodes JPEG images that a page references directly and that pass strict checks: 8 bit, RGB or grayscale, no mask, no Decode/DecodeParms, at least 120,000 pixels and 16 KB. Images nested in forms or patterns, PNG/Flate scans, CMYK/ICC and 16-bit images are left alone.
• Balanced is lossy: replaced images are capped at a 2200 px long edge and re-encoded at JPEG quality 0.75. A replacement is kept only when it saves at least 5%; zooming in may look softer.
• Strong renders every page at 120 DPI and rebuilds the PDF from images. Text is no longer selectable, searchable or copyable; links, forms, annotations, bookmarks, attachments and scripts are gone. Small text may look soft.
• All modes remove metadata: title, author, subject, keywords, creation/modification dates, Creator/Producer and XMP are not kept, and no new title is written.
• Any compression changes the bytes, so digital signatures become invalid. The tool does not verify or preserve signatures, PDF/A, PDF/X or tagged-PDF conformance.
• Password-protected PDFs are rejected; there is no “ignore encryption” option. Open the file with its password and save an unprotected copy first.
• Clean and Balanced do not strip scripts or attachments; they remain part of your document. Use a dedicated sanitiser if the source is untrusted. Strong naturally drops them because it rebuilds the pages as images.
• Strong output follows the page orientation a reader shows: rotated pages are written in their visual orientation, and the /Rotate marker is not kept. Page dimensions in points stay the same.
• Compression runs on the main thread. Large jobs can occupy the tab for several seconds; strong mode holds all page images in memory, so 400 pages may be slow or fail on low-memory devices. Keep the tab open; a browser may discard it under pressure.
• Preview uses your browser’s built-in PDF viewer, not a rendered preview. Inline display and mobile behaviour depend on the browser; if the preview stays blank, open the file in a new tab or download it. Preview never changes the file.
• Limits: one PDF, 50 MB, 2,000 pages; strong mode is limited to 400 pages, and each rendered page is capped at a 4,000 px long edge and 16 million pixels. Jobs over 20 MB or 300 pages show a hint. Output names replace reserved characters with “-”, trim the ends and cut long names; the source file is never modified.

Related tools

Other PDF tools that also run entirely inside this tab.