Turning a PDF grey saved the most on the file you would least expect. The 226-page C2PA specification went from 10.0 MB to 2.74 MB, and almost none of that came from colour: the file stores most of its page content with no compression at all, and rewriting 194 coloured pages packed it. Three photos, one per page, went from 4.83 MB to 1.48 MB, and that saving was all colour.
Eight PDFs made grey
We ran the eight PDFs from our compression test through our Grayscale PDF tool on 24 September 2026, in Chromium 153 on a laptop, then drew every page of the result and counted the pages that still showed colour.
| Original | Grey | Change | Coloured pages before, after | |
|---|---|---|---|---|
| Three photos, one per page | 4.83 MB | 1.48 MB | −69% | 3, 0 |
| Three scanned text pages, 300 dpi | 1.82 MB | 1.21 MB | −34% | 3, 0 |
| NIST AI framework, LaTeX, 48 pages | 1.95 MB | 1.04 MB | −47% | 19, 0 |
| Attention Is All You Need, 15 pages | 2.22 MB | 1.51 MB | −32% | 6, 0 |
| C2PA specification, 226 pages | 10.0 MB | 2.74 MB | −73% | 194, 0 |
| NIST Cybersecurity Framework, Word, 32 pages | 1.52 MB | 974 KB | −36% | 32, 0 |
| IRS Form W-9, 6 pages | 141 KB | 126 KB | −10% | 0, 0 |
| One-page memo printed from Chromium | 42 KB | 39 KB | −7% | 0, 0 |
The savings come from three places, and only one is colour. Photos saved in grey lose two of their three colour channels. Pictures stored losslessly, like the charts and figures in the NIST AI framework, drop to one channel and stay exact; that file’s nine pictures took it from 1.90 MB, re-saved without metadata, to 1.04 MB. Then any page whose colours change gets its content rewritten and compressed, which is what shrank the C2PA specification. The arXiv paper is the clean control: saved again with only its metadata removed it was already 1.51 MB, and making it grey took off another 400 bytes.
The IRS form and the memo had no colour to begin with, so their 7 to 10% is the metadata our tools always remove.
Is the grey right?
We didn’t turn pages into pictures to do this. Our tool rewrites each colour in the PDF with the formula the PDF standard gives for turning colour into grey, 30% red, 59% green and 11% blue, and converts pictures pixel by pixel the same way. To check it, we drew every page of each original, turned that drawing grey with the same formula, and compared it with a drawing of our grey file. Across all 334 pages the average difference was under half a shade out of 255, and not one pixel was more than 24 shades off. The text read from each page came out character for character the same in all six files that had text, so it stays selectable and searchable.
We also built two test pages meant to be hard: one with print colours (CMYK), a named spot ink, a Lab colour, two smooth colour blends and a colour picture stored inside the page text, and one printed from Chromium with CSS gradients, a patterned SVG and a coloured shadow. Both came out with no coloured pages left.
Photos are saved again at about the quality they had, which is why the photo PDF shrank: a grey JPEG at the same setting is smaller. If size is the real goal, run Compress PDF on the grey file too.
Sources
- ISO 32000-1:2008 (PDF 1.7), section 10.3 Conversions among device colour spaces, Adobe’s published copy
- C2PA Technical Specification 2.1
- NIST CSWP 29: The NIST Cybersecurity Framework (CSF) 2.0
- NIST AI 100-1: AI Risk Management Framework
- arXiv: Attention Is All You Need (1706.03762v7)
- IRS Form W-9 (Rev. March 2024)
- pdf.js, the PDF renderer we checked the pages with