PDF compression guides
PDF Compression Levels Explained: File Size vs. Image Quality
A level name is only a shortcut; the underlying operations and the source PDF determine the result.
Last updated
A PDF compression level is a bundle of choices, not a universal quality score. Lossless cleanup can reorganize data without intentionally discarding content, while lossy image compression and downsampling trade pixel detail for fewer bytes. Embedded fonts, redundant objects, and content that is already compressed also change the outcome. Compare raw input and output bytes, then inspect the document instead of treating a preset name as proof of quality.
What the compression terms actually mean
| Term | Meaning | Practical consequence |
|---|---|---|
| Lossless compression | Re-encodes data so it can be reconstructed without deliberately throwing information away. | Safer for masters, but it may save less than changing image detail.2 |
| Lossy image compression | Permanently removes some pixel data to reduce image streams. | Can save substantially more, but text edges, photos, and marks can visibly change.1 |
| Downsampling | Combines source pixels into fewer, larger pixels and lowers image resolution. | Reduces image data while limiting how far a page can be enlarged or printed sharply.1 |
| Embedded fonts | Font programs or subsets travel inside the PDF so text can reproduce as designed. | They add bytes; removing them can trigger substitution and alter appearance.1 |
| Redundant objects | Unused, obsolete, duplicate, hidden, or application-specific structures remain in the file. | Cleanup may save space, but aggressive removal can break tags, forms, layers, or links.1 |
| Already-compressed content | Images and streams have already been encoded compactly before the new pass. | Another pass may save little or require quality loss to move the number. |
Measure two percentages that answer different questions
remaining-size ratio = output bytes / input bytes × 100A 6 MB output from a 10 MB input has a 60% remaining-size ratio.
reduction percentage = (input bytes − output bytes) / input bytes × 100The same 10 MB to 6 MB result is a 40% reduction.
Calculate one fixture from its raw byte counts
The text-heavy record starts at 11,762 bytes and ends at 6,077 bytes. Its remaining-size ratio is 6,077 / 11,762 × 100. Its reduction percentage is (11,762 − 6,077) / 11,762 × 100 = 48.33%, matching the published record after rounding. Both the methodology page and JSON preserve the raw values and boundary.3, 4
Compare the three fixed files before generalizing
The same test set reports 48.33% for the text-heavy fixture, 91.26% for the scan-style fixture, and 81.74% for the image-heavy fixture. The recorded input/output page counts are 4/4, 2/2, and 2/2; the readable-PDF fields are true, true, and true. Those checks do not measure every visual detail or predict a new document.3, 4
| Evidence ID | Fixture | Recorded mode | Input bytes | Output bytes | Reduction | Input pages | Output pages | Readable-PDF check | Boundary |
|---|---|---|---|---|---|---|---|---|---|
| geo-compress-text-v1 | text-heavy-pdf-v1 | standard | 11,762 bytes | 6,077 bytes | 48.33% | 4 | 4 | true | Fixed PDFStay fixture; not a promised compression rate or visual-quality score |
| geo-compress-scan-v1 | scan-style-pdf-v1 | scanned | 2,141,091 bytes | 187,107 bytes | 91.26% | 2 | 2 | true | Fixed PDFStay fixture; not a promised compression rate or visual-quality score |
| geo-compress-image-v1 | image-heavy-pdf-v1 | image | 1,845,799 bytes | 337,064 bytes | 81.74% | 2 | 2 | true | Fixed PDFStay fixture; not a promised compression rate or visual-quality score |
Fixed PDFStay fixture; not a promised compression rate or visual-quality score
Choose by use case, not by the strongest label
| Use case | Starting choice | What decides acceptance |
|---|---|---|
| Digital office PDF with selectable text | Smart compression | Text selection, links, forms, layout, and page count |
| Photo-rich report or portfolio for screen viewing | Image optimization | Faces, gradients, captions, and acceptable enlargement |
| Scan-heavy sharing or upload copy | Scanned PDF compression, beginning with Clarity first or Balanced | Fine print, handwriting, codes, color meaning, and page order |
| Archival, signed, or print-production master | Retain the original and follow the governing specification | Format compliance, signature validity, preflight, and an approved proof2, 1 |
Why one level cannot promise one result
A preset encounters a different mix of pixels, fonts, object streams, metadata, layers, and prior compression in every PDF. It may leave one file almost unchanged and substantially reduce another. A page-count match and successful open are necessary checks, not a visual-quality grade. Save the original, record the raw bytes, and approve the output for its intended use.
FAQ
Is a 90% reduction always better than a 50% reduction?
No. The percentage measures bytes removed, not fitness for use. A strong result can reflect a scan with abundant image data, but it can also hide unacceptable detail loss.
Why did a high compression level barely change my PDF?
The file may already contain compact image streams, subset fonts, and few removable objects. Forcing a much smaller result can require downsampling or lossy image changes that are not appropriate for every document.
If the reduction is 48.33%, how much of the file remains?
Subtract the recorded 48.33% from 100%, or divide the 6,077 output bytes by the 11,762 input bytes and multiply by 100. Both calculations describe the share that remains.4
Sources
- Adobe Acrobat — PDF Optimizer settingsofficial technical documentationSource reviewed:
- FADGI — Technical Guidelines for Digitizing Cultural Heritage Materialsgovernment technical guidanceSource reviewed:
- PDFStay reproducible compression checksPDFStay first-party methodologySource reviewed:
- PDFStay compression benchmark data (JSON)PDFStay first-party dataSource reviewed: