Compress PDF

How this tool works and its limitations

The page loads no PDF code until a file is added. Adding one downloads pdf-lib, an open source PDF library, with a progress bar, and runs it in a background worker, which checks the %PDF- marker in the first 1,024 bytes, whether the file opens without a password, and the size and rotation of every page. The lossless cleanup runs in that worker.

Balanced lists the pictures in the file in that worker, which hands them to this tab one at a time. The browser decodes each one, scales it down when it exceeds the resolution limit (a picture with a mask keeps its size), saves it again as a JPG and hands it back, and the worker replaces the picture only if the new one is at least 10 percent smaller. Then it writes the file. Balanced does not need pdf.js.

The strong mode also downloads pdf.js, Mozilla's open source PDF reader. It draws one page at a time in this tab, the browser saves each drawing as a JPG and the canvas is emptied, and the worker places each JPG as it is on a page of the same size, then saves the new file. Both libraries and their licenses are listed on the open source licenses page.

• Three modes: a lossless cleanup with small savings on most files, Balanced, which saves again the pictures it can make smaller, at a lower quality, and keeps the text, and a strong mode that turns every page into a compressed JPG picture, which makes the text unselectable and unsearchable and stops links and form fields from working.
• The cleanup does not touch pictures, so a file that is mostly photos or scans saves little or nothing in that mode.
• Balanced saves pictures again as JPG at the quality you choose, which loses detail that cannot be brought back, so keep your original. A picture is replaced only when the new one is at least 10 percent smaller; pictures under 20 KB, and pictures with a side under 64 pixels, are left alone, and a file with no picture Balanced can make smaller comes out at the size of the lossless cleanup.
• Balanced handles only 8 bit gray and RGB pictures stored as JPEG or Flate (zip compressed). Black and white scans (such as CCITT or JBIG2), JPEG 2000, CMYK and palette (indexed) pictures, stencil masks, pictures with a color key and pictures with a color profile other than an RGB or gray one are kept as they are.
• Balanced measures the resolution limit against the longest page side of the file, so a picture is not scaled down because it is drawn small on a page; only a picture with more pixels than the limit allows gets fewer. A photo that the page crops can end up below the chosen resolution, because the limit is measured against the longest page side, not against the part of the picture you see. A picture with a transparency mask keeps its pixel size. A picture with an RGB color profile keeps that profile; a gray picture, with or without a color profile, becomes an RGB picture when it is replaced.
• Balanced leaves text, links, bookmarks, document scripts and attached files as the file has them, and form fields keep working; a picture used inside the look of a form field or a note can be saved again like any other.
• A result that is not smaller than your file is never presented as an improvement: the page says by how much it is larger and suggests keeping the original.
• Any change to a digitally signed file breaks its signature, in every mode.
• The lossless cleanup keeps document scripts, links that open other files or programs, and attached files as the file has them; the strong mode leaves all of them out.
• A file with a drawing that unpacks to more than 64 MB is refused by the strong mode, because drawing it could fill the memory of the browser tab; the lossless cleanup and Balanced still work on it.
• One PDF of up to 100 MB and 500 pages per job; in the strong mode each page picture holds at most 16.8 megapixels and the new PDF at most 150 MB.