Compress PDF

Where the Bytes Go: What Makes a PDF Large

A PDF is a container format — it bundles images, fonts, vector graphics, metadata, and content streams into a single file. Understanding where the size comes from helps you choose the right compression profile and set realistic expectations for the output.

Scanned documents are the biggest offenders. A flatbed scanner captures each page as a full-resolution raster image — typically 2-5 MB per page at 300 DPI. A 20-page scanned contract can easily reach 60 MB, even though it contains nothing but black text on white paper. Font embedding adds up too: each embedded font consumes 200-400 KB, and a document that uses four font families (regular, bold, italic, bold-italic) may carry over 1.5 MB of font data alone. When you merge multiple PDFs, duplicated fonts compound — three Word documents using the same Arial font will embed three separate copies.

Compression Profiles Explained: Choosing the Right One

The compressor offers five profiles, each tuned for a different use case. The right choice depends on what you plan to do with the file and how much quality trade-off you can accept.

Web is the default and works well for most situations. It downsamples images to 150 DPI — plenty for on-screen viewing but noticeably softer if you zoom to 400% or print on glossy paper. Color images are recompressed as JPEG with a quality setting that eliminates fine noise without introducing visible blocking. Fonts are subsetted to include only the glyphs present in the document. This profile typically reduces file size by 40-70%.

Print keeps images at 300 DPI — the standard for offset and laser printing. Image quality is higher than web but the files are correspondingly larger. This profile is the right choice when the compressed PDF will be sent to a commercial printer or when the document contains detailed diagrams that need crisp edges at full zoom. Expect 20-40% reduction.

Archive is the lossless profile. It applies stream compression (Flate/zlib) to content streams, subsets fonts, removes dead objects and old revision data, but never recompresses images or converts color spaces. The output is visually identical to the input, bit-for-bit for image data. Reductions are modest — typically 10-25% — but there is zero quality risk. Use this when the PDF is a legal record, a signed original, or contains medical imaging.

Scanned images (MRC) uses Mixed Raster Content layering, a technique that separates each scanned page into layers: a high-resolution foreground for text and line art, and a lower-resolution background for photos and gradients. The text layer is compressed with JBIG2 (a codec optimized for recurring character shapes), while the background uses JPEG or JPEG2000. The result is dramatically smaller — a 3 MB/page scanned contract can drop to 150-200 KB/page — while keeping text sharp and legible. This profile is specifically designed for scanned documents and has limited benefit on native-text PDFs.

Max applies the most aggressive settings from all the above techniques. Images are downsampled to 100 DPI, JPEG quality is pushed lower, and every available optimization is enabled. Files shrink the most — often 70-90% — but photographs will show noticeable compression artifacts and fine gradients may band. Best for internal drafts, quick-reference copies, or files where size is the only priority.

Beyond profiles, two additional settings let you fine-tune the output: colorspace conversion converts color images from RGB to grayscale when color is unnecessary (halving image data), and flatten signatures removes digital signature fields that would otherwise be invalidated by the compression anyway.

How the Optimization Engine Shrinks Your File

PDF compression is not a single operation — it's a pipeline of targeted optimizations, each addressing a different source of bloat. The engine analyzes every object in the PDF and applies the most effective technique for that object type.

Image optimization is where most of the savings come from. The engine inspects each embedded image and selects the best codec: CCITT Group 4 for pure black-and-white images (like faxes or line drawings), JBIG2 for scanned text pages where recurring character patterns can be stored once and referenced across the page, JPEG or JPEG2000 for photographs and continuous-tone images, and Flate (zlib) for images that must stay lossless. Images that exceed the target DPI are downsampled using bicubic interpolation, and images whose visible area on the page is smaller than the embedded resolution are cropped to their visible bounds — a surprisingly common source of waste in PDFs exported from design tools.

Font subsetting strips each embedded font down to only the glyphs the document actually uses. A full Arial Unicode font file is roughly 1.2 MB and contains over 50,000 glyphs. If your document uses only Latin characters, the subsetted font might be 40 KB. The engine also deduplicates font data — when multiple pages or merged files embed the same font independently, the redundant copies are consolidated into one.

Stream compression targets the content streams that define page layouts, vector paths, and text positioning. These streams are compressed using Flate (the same algorithm behind gzip). The engine also performs dead object removal: PDF editing software often leaves behind orphaned objects from previous revisions — deleted images, replaced fonts, outdated metadata — that remain in the file but are never referenced. Removing them is pure savings with no quality impact. Learn more about lossy vs. lossless compression techniques and how they apply to different content types.

Expected Results: Before and After Compression

Compression results vary widely depending on what's inside the PDF. A scanned document with uncompressed TIFF images will shrink dramatically. A text-heavy PDF exported from Word with already-optimized JPEG illustrations may only shrink 15-20%. Here's what to expect for common document types:

Already-optimized PDFs show diminishing returns — if the images were already JPEG-compressed and fonts were already subsetted, the engine can only squeeze out another 2-5% through dead object removal and stream recompression. This is normal and not a sign that the compressor isn't working.

When Compression Changes Your Document

Compression is not always invisible. Understanding the edge cases helps you avoid surprises and keep your originals safe.

Lossy image recompression (web, print, max profiles) permanently reduces image data. Fine details in photographs — texture in fabric, grain in wood, subtle gradients in a sunset — may soften or show blocking artifacts, especially at the “max” setting. Text rendered as vector outlines is never affected, but text in scanned images is subject to the same recompression as any other image. If you need to preserve every pixel, use the “archive” profile.

Font subsetting is irreversible. Once the engine strips unused glyphs from an embedded font, you cannot add them back. If you later need to edit the compressed PDF and type characters that weren't in the original document, the font won't contain those glyphs. Always compress a copy — never your only editable version.

Digital signatures are invalidated by any structural change to the PDF's internal structure. Compression modifies image streams, font tables, and the cross-reference table — all of which break the cryptographic hash that validates the signature. If a PDF is digitally signed, do not compress it unless you plan to re-sign it afterward. The “flatten signatures” setting removes the signature fields entirely rather than leaving them in an invalid state.

Color space conversion is permanent. If you enable the grayscale conversion option, color information is discarded and cannot be recovered. This is useful for text documents and forms where color adds no value, but it should never be applied to marketing materials, photographs, or any document where color carries meaning.

The general rule: keep your originals, compress copies for distribution, and never compress a document that serves as a legal record unless you're using the lossless archive profile.