Three compression levels to reduce PDF file size safely
Shrinking a PDF for email or portal upload limits should never turn crisp contracts into blurry bitmaps. Under the PDF Association ISO 32000-2 Specification, documents store vector text and fonts separately from raster image XObjects, allowing targeted image downsampling with 0 bytes sent to external servers.
Comparing Light, Medium, and Strong DPI and JPEG settings
Light rewrites the file losslessly using compressed object streams and removes unreachable objects without touching pixels. Medium downsamples oversized photos to 150 DPI at 0.72 JPEG quality, while Strong targets 96 DPI at 0.55 JPEG quality for tight upload portals.
Images that are already smaller than the target DPI on the page are never upscaled or re-encoded unnecessarily. Each recoded image replaces the original stream only when the resulting JPEG bytes are strictly smaller.
| Level | Target image DPI | JPEG quality | Page content impact | Best use case |
|---|---|---|---|---|
| Light | Unchanged | Lossless | Prunes dead objects, packs streams | Text-heavy files and archives |
| Medium | 150 DPI | 0.72 | Resizes large photos, keeps text vector | Email attachments and client sharing |
| Strong | 96 DPI | 0.55 | Aggressive photo shrink, keeps text vector | Strict web upload limits |
| Skip threshold | 99% of input | Auto-guarded | Aborts save if no meaningful gain | Already-optimized PDFs |
Reading the live byte breakdown before you compress
As soon as you drop a file, a background analyzer checks what share of total bytes comes from embedded images. If images make up under 20% of the file, the estimator recommends Light; for a complete object-by-object breakdown, inspect the document in the PDF size analyzer.
Checking that estimate before running compression prevents guesswork on multi-megabyte packets. You immediately see whether shrinking raster photos will cut the file in half or whether fonts and vector diagrams account for most of the byte count.
Why did my compressed PDF only shrink by a small percentage?
A PDF shrinks very little when most of its bytes live in fonts, vector graphics, or already-compressed streams rather than high-resolution photos. Because modern PDF generators already apply lossless DEFLATE compression to text and structural streams, re-packing them yields modest gains compared to shrinking raw camera scans.
Under IETF RFC 1951 DEFLATE Compressed Data Format, FlateDecode streams already eliminate repetitive byte patterns in text content. When a multi-hundred-page text document still exceeds an email gateway cap, you can split oversized PDFs to meet email attachment caps instead of degrading readability.
Common causes of small compression ratios include:
- Mostly text and fonts: Embedded font subsets and vector CAD drawings cannot be downsampled like raster photos.
- Already low-DPI scans: Images already at or below 96 DPI or 150 DPI are never upscaled or re-blurred unnecessarily.
- Skipped color spaces or masks: CMYK images, 16-bit channels, and transparent cutouts are left intact to avoid color shifts.
How worker-thread image recoding protects fonts and transparency
Running image decoding and scaling inside a background Web Worker keeps the browser responsive while enforcing strict visual safeguards. Using the MDN Web Docs OffscreenCanvas API, the worker decodes eligible 8-bit grayscale and RGB images, scales them proportionally to the largest page side at 72 pt per inch, and keeps any replacement only if it is smaller.
Skipping SMask and Mask transparency to avoid JPEG halos
Images with /SMask or /Mask entries rely on crisp pixel edges for transparent logos and signatures. Because lossy JPEG encoding introduces ringing artifacts around sharp alpha boundaries, the compressor skips every masked image and every ImageMask bitmap automatically.
Leaving those masked graphics untouched preserves clean company letterheads, watermark seals, and scanned ink signatures against colored backgrounds. Photographs and full-page scans without transparency masks still downsample normally to hit your chosen size target.
Keeping vector paths and selectable text untouched
Only Image XObjects are touched during Medium and Strong passes, leaving every glyph, hyperlink, and vector line intact. Even FlateDecode images using PNG row predictors 10 through 15 are unpredict-decoded cleanly before downsampling.
If your document also carries bulky interactive form fields, you can flatten interactive form fields and comments in a separate pass.
Lossless object streams, color spaces, and minimum savings checks
Understanding when lossless cleanup beats lossy re-encoding helps you pick the right setting for archival and print workflows. Checking the live estimate before clicking Compress PDF saves unnecessary trial-and-error runs.
When Light lossless cleanup is the safer archival choice
Light mode enables PDF 1.5 object streams and prunes unreachable objects while avoiding lossy DCTDecode recompression described in the Library of Congress JPEG JFIF Format Description. For color scans that still weigh too much, switching to grayscale scan conversion can further reduce byte weight.
Why the tool stops if output stays at 99% or higher
To prevent wasting time on an unshrinkable file, the compressor throws a clear notice if the output reaches 99% or more of the input size. When Medium hits that threshold, switching to Strong or inspecting embedded fonts in the analyzer clarifies where the remaining bytes reside.
Every compression run preserves the original PDF filename with a -compressed.pdf suffix so your source document stays untouched on disk. You can compare visual clarity side by side before deleting the larger original.
Tip: Start with Medium for mixed reports and scanned forms. Check the live byte estimate banner before clicking Compress PDF so you know whether Light or Strong will deliver better results.