ORIGINAL THREE-FIXTURE BROWSER STUDY · 16 SEPTEMBER 2026
When does browser PDF compression actually help?
PaperGrid materially reduced the synthetic image-heavy and mixed PDFs, but made the already compact text-first PDF much larger. The tool's byte-size safeguard accepted only genuinely smaller candidates.
The strongest result was 94.1% smaller on the two-page image-heavy fixture. Balanced compression reduced the mixed fixture by 84.5%. On the text-first fixture, every candidate was rejected because none was smaller than its 3,936-byte source.
Both lossy modes accepted
Both lossy modes accepted
Candidates were equal or larger
| Fixture | Mode | File size | Change | Pages reopened | Decision |
|---|---|---|---|---|---|
| Image-heavy | Source | 474,438 B | — | 2 | — |
| Balanced | 56,476 B | 88.1% smaller | 2 | Download | |
| Smaller | 27,778 B | 94.1% smaller | 2 | Download | |
| Lossless | 474,438 B | No change | 2 | Reject | |
| Text-first | Source | 3,936 B | — | 3 | — |
| Balanced | 325,073 B | 8,159% larger | 3 | Reject | |
| Smaller | 154,411 B | 3,823% larger | 3 | Reject | |
| Lossless | 3,936 B | No change | 3 | Reject | |
| Mixed | Source | 474,846 B | — | 3 | — |
| Balanced | 73,520 B | 84.5% smaller | 3 | Download | |
| Smaller | 36,783 B | 92.3% smaller | 3 | Download | |
| Lossless | 474,846 B | No change | 3 | Reject |
METHOD
Exactly how the test was run
- The image-heavy fixture contained two deterministic 1,400 × 1,900 pixel canvas images encoded at JPEG quality 0.98. The text-first fixture contained three pages of selectable Helvetica text and vector rules. The mixed fixture combined one text cover with the two image pages.
- All sources were processed in one isolated Headless Chrome 152 session on macOS using the same self-hosted PDF.js and pdf-lib files as the live tool.
- Balanced rendered pages at scale 1.35 and JPEG quality 0.78. Smaller used scale 1 and JPEG quality 0.58. Lossless rewrote PDF structure with object streams.
- Every candidate output was loaded again as a PDF and its page count checked. A candidate qualified for download only when its exact byte size was below the source.
- The downloadable JSON includes per-fixture byte counts, page counts, timings, reduction percentages and the resulting download decision.
What the study supports
- Image recompression materially reduced both fixtures containing large images.
- Rasterization was counterproductive for the compact text-first fixture.
- The byte-size safeguard prevented equal or larger outputs from being called compressed.
What it does not prove
- That every image-heavy or mixed PDF will shrink by the same amount.
- That image quality is acceptable for every use.
- That processing will take the same time on another device.
Important tradeoff
Balanced and Smaller modes rasterize pages. Selectable text, links, forms, accessibility structure and digital signatures do not remain interactive. Lossless preserves more document behavior but produced no saving on these already optimized fixtures.
REPRODUCE IT
Run the safe fixture yourself
Open Compress PDF, choose “Try safe sample PDF,” select Balanced or Smaller, and process it locally. Browser versions, font rendering and device hardware can produce slightly different byte counts or timings.