How to reduce a PDF’s file size without turning pages into pictures

A PDF is a container. It can hold real text, fonts, vector drawings, and embedded pictures. Those pieces do not weigh the same. A ten-page letter that is actual text is often a few hundred kilobytes even with a couple of fonts embedded. A ten-page scan of the same letter, where every page is a photograph of paper, can be tens of megabytes. The page count is a poor guide to file size. The contents are the guide.

People reach for a “compress PDF” button and expect every file to drop by half. That happens when the file is full of photos or scans that were saved at a high resolution. It does not happen when the file is already mostly text. QuickConvert’s PDF Compress tool is built for the first case. It finds raster images inside the PDF and re-encodes those images. It does not flatten each page into a new screenshot. If you can select a sentence in the original, you should still be able to select it afterward, because the letters were never turned into pixels.

That distinction is the whole quality question. Flattening a page to an image makes a file that looks like the original from a distance and then falls apart when you zoom in or try to copy a paragraph. Re-encoding only the embedded photos leaves the text alone and spends the quality budget where the bytes actually are.

The tool reads the PDF in your browser with pdf-lib and walks the file’s image objects. Images it understands are drawn on a canvas and saved as JPEG. An image is replaced only when the new JPEG is smaller than about 97 percent of the original image stream. If the new version is not clearly smaller, the original image stays. You are not gambling the whole document on a single quality slider that might make every page worse for a one-percent saving.

Three presets set the two numbers you can also edit by hand.

• Strong uses JPEG quality 42 and limits the long edge of an image to 1000 pixels. Use it when the file must fit a tight email limit and the pictures are reference scans, not portfolio prints.
• Balanced, the default, uses quality 62 and a 1500 pixel long edge. Start here for a scan you still need to read.
• Light uses quality 80 and a 2200 pixel long edge. Use it when the photos matter and you only need a modest cut.

The quality slider itself runs from 30 to 90, and the max edge from 400 to 4000 pixels. Images that already fit inside the edge are not enlarged. Images that carry a soft mask keep their pixel dimensions so a transparent cutout is not resized, though they can still be re-encoded.

Some images are skipped, and the skip is deliberate. Masks, tiny images under 32 pixels on a side or under 8,000 pixels in area, images that are not 8 bits per component, JPEG 2000, JBIG2, CCITT fax encodings, hard masks, and non-JPEG images in a color space other than DeviceGray or DeviceRGB are left as they are. A fax-style black-and-white scan is often CCITT and will not shrink here. The summary after a run tells you how many images were found and how many were re-encoded, plus the before and after file size. Read that line. It is more honest than a promise of “up to 90 percent smaller.”

Each PDF must be 80 MB or smaller. The work happens in the tab’s memory. A larger file is refused up front rather than half-processed.

Open the original and note whether you can select text. If you can, you have a born-digital PDF and you should expect modest savings unless it also contains photos. If you cannot select text and each page looks like a picture of paper, you have a scan and Balanced is a fair first pass.

Run PDF Compress on a copy. The download is named with compressed.pdf on the end, so scan.pdf becomes scan.compressed.pdf. The original on disk is not overwritten. Open the download at the zoom you actually use. Look at small type that was part of a photograph — a pictured spreadsheet, a photographed whiteboard — because that type is pixels and JPEG will soften it. Look at ordinary text that was real text. It should be unchanged.

If the summary says no embedded photos were found, stop. Strong will not invent a smaller file. The bytes are in text, vectors, fonts, or an encoding this tool does not recompress. Export again from the program that made the PDF, or reduce the images before they are placed, if you still have that source. If you only need a few pages of a long scan, PDF Split can extract them so you send a shorter file even when each page stays large.

If the summary re-encoded images but the file barely moved, the pictures were already efficient. Try Strong only if you have looked at Balanced and can accept softer photos. If Strong is still not enough, the limit is the page content, not the button.

Photos that are not inside a PDF yet should be reduced with Image Compress before you place them. Shrinking a 4000 pixel photo to a sensible long edge, then exporting the PDF, beats hoping a later pass will undo a huge embed.

Lossless PDF compression, the kind that only repacks the same bytes more tightly, is real and usually small. The visible savings on a scan come from throwing away information: fewer pixels, and a JPEG that does not store every shade. “Without losing quality” in practice means “without losing the quality you use.” A receipt scanned so you can read the total does not need a 4000 pixel edge. A photo you will print at A4 might. Match the preset to that use, and keep the original whenever the download is for sending rather than for archiving.

Password-protected PDFs are refused. The tool does not ask for the password. Unlock a copy in a reader that already has it, then compress the unlocked file. Nothing in this process uploads the PDF. The page, once loaded, does the work locally. Google Analytics on the public site can still record that you opened the page; it does not receive the file. That split is covered in privacy with online file tools.