Reduce PDF size without quality loss

Why this works

PDFRun's Lossless mode strips embedded fonts you don't need, deduplicates repeated images, removes unused objects and re-encodes with Flate compression. Pages look pixel-identical to the original, but the file is meaningfully smaller.

Most PDF compression is lossy: images get downsampled, JPEG quality drops, fine detail softens. For documents going to the web or email that\'s usually fine — nobody zooms into an email attachment looking for pixel-perfect fidelity. But there are documents where every pixel matters. Architectural drawings, engineering schematics, design portfolios, scientific papers with detailed figures, anything destined for print or archival storage — these can\'t afford the softening that conventional compression introduces. You need smaller files and identical visual output.

Lossless compression is possible because most PDFs carry significant non-visual overhead that can be safely removed without touching pixel data. Embedded fonts are the largest offender: PDFs commonly embed entire font families when only a fraction of the characters are actually used in the document. A document using just one paragraph in Helvetica might still embed the full Helvetica glyph table — thousands of unused characters. Subsetting strips the unused glyphs, often saving 200 KB to 2 MB per font. Duplicate images: a logo or page-element that recurs across many pages is often re-embedded by the PDF generator each time rather than referenced once. Deduplication consolidates these. Unused objects: form fields that were drafted then deleted, orphan annotations, obsolete optional content groups — all dead weight that survives in the file. Inefficient stream encoding: many PDF generators use suboptimal compression on text and vector streams; re-encoding with Flate often picks up 10–20% additional reduction.

What lossless mode does, in order. Inventory the file: count fonts, images, and objects; identify which are referenced and which are orphaned. Subset fonts: for each embedded font, keep only the glyphs that actually appear in the document. Deduplicate images: hash each embedded image and replace duplicate copies with references to a single instance. Remove unused objects: drop orphan form fields, dead annotations, abandoned optional content groups. Re-encode streams: re-compress text and vector content with optimal Flate settings. Verify visual parity: render every page of the original and the compressed output, compare pixel-by-pixel, and abort the optimisation if any page differs visibly.

Typical savings. Born-digital PDFs from design tools (InDesign, Affinity, Figma exports): 25–40% reduction common, occasionally more if the source embedded duplicate placed graphics. Word and Pages exports: 10–20%, less because these tools already produce reasonably efficient PDFs. PDFs from accounting and ERP software: often 30–50% — these generators tend to over-embed fonts and waste space on logo repetition. Scanned PDFs: 5–15% only — most of the file size is raster image data which lossless can\'t touch. For scans, lossy compression is the right call.

When lossless isn\'t enough. If your file is 50 MB and the receiving system caps at 5 MB, lossless won\'t close that gap — it might trim 15 MB at best, leaving you well over cap. For aggressive size reduction, lossy compression is the right tool. Lossless is for documents that need to stay essentially identical to the original (print-ready, archive-ready, design portfolio) where 25% reduction is welcome but quality loss is unacceptable.

A workflow tip for design portfolios and high-fidelity documents. Always run lossless first on any document you might want both a high-quality version and a smaller version of. Lossless gives you a smaller-but-identical version that\'s often suitable for general use; if you still need a smaller version for email or upload caps, run lossy compression on the lossless output rather than the original. The two-step approach (lossless then lossy) often produces better lossy results than starting from the original, because the lossy pass has cleaner input.

How it works