Background

Depending on the pdf creator/engine used, a .pdf file may include content not this much relevant for reading by a human. Examples include bitmap illustrations with a resolution greater than 300 dpi, or color images while an illustration in grayscale is good enough. A reprint of the .pdf again as a pdf with ghostscript may address this and yield a smaller file size easier to store / to transfer (for instance attached to an email). Text-only files print into postscript (e.g., with a2ps ¹), then distilled into a pdf equally can benefit from such an optimization.

By far, this bash script does not claim to be the first one collecting bits and bolts to address the issue. It rather serves as an aide-memoire of finds encountered earlier, and to moderate ghostscript in Linux accordingly. Within reason, the snippets were joined as provided; thus, the credit belongs to those already in the field.

Intended Use

The intended use adds the executable bit to the script (chmod +x pdf_reprint.sh) and accesses the functionality by an alias. Debian's default shell is BASH; thus, this configuration can be set either in file /etc/bash.bashrc for any user, in file file ~/.bashrc for the current user; or as a template for any (new) user's own file ~/.bashrc in file /etc/skel/.bashrc in a pattern of

alias pdf_rewrite="/path/to/pdf_rewrite.sh"

• Presuming the alias to pdf_rewrite.sh set is pdf_rewrite, run either one of the following commands to reprint the .pdf while retaining the color:

  pdf_rewrite --reprint input.pdf
  pdf_rewrite -r input.pdf
  pdf_rewrite --colour input.pdf
  pdf_rewrite --color input.pdf
  pdf_rewrite -c input.pdf

  If the reprint is smaller in size than the original file, the reprint replaces the original file. In addition, a brief note states the percentage of the savings achieved. If wanted, you may repeat the reprint; eventually, either the savings are insignificant in comparison to the remaining file size, or the script itself will report no change and thus retain the original file.

  The credit for the underlying approach and implementation belongs to Evan Langlois.²

• Often, a reprint in grayscale is sufficient. Use either one of the following commands

  pdf_rewrite --gray input.pdf
  pdf_rewrite --grey input.pdf
  pdf_rewrite -g input.pdf

  to overwrite file input.pdf accordingly. The credit for this approach belongs to user slm on the Unix stackexchange.³

• To process multiple .pdf, you may consider a for loop as in

  for file in *.pdf
  do
      echo "$file"
      bash ./pdf_rewrite.sh -r "$file"
  done

  This approach equally provides you a brief progress report, too.

Note, illustrations in a reprint in grayscale may render illustrations less intelligible. When preparing a document, a service like https://colorbrewer2.org/ may guide your selection for color palettes suitable for this kind of "photocopying" / identify a palette safe for the color blind.

Keep a backup of the .pdf to be processed. Though the script may report problems while processing the data (or even crash, which may destroy the .pdf), it is not a PDF validator such as e.g., veraPDF.⁴

Benchmark

Initially written for Linux Xubuntu 18.04.3 LTS and ghostscript (version 9.26), the script is known to work well for instance with Debian 13/trixie and GPL Ghostscript (version 10.05.1).

• File link2web.pdf included to the project was compiled with pdfLaTeX based on an example provided by www.texample.net. This .pdf contains a color figure and link to an external reference. Note, the simplification into half-tones (option -g) affects the document printed; depending on the pdf viewer used, the highlighting box around the link may remain colored for the display on screen.

• The performance of the utility was tested on a couple of recent publications in chemistry and gs (10.07.0) as currently provided in Debian 14/forky (currently branch testing). To ease a replication, only open access publications used for the bench mark were used.

The table below compares the difference of the file size prior and after the processing with either option after a single run of optimization.

journal	publisher	original	with -r	saved %	with -g	saved %
2023ACR3654	ACS	2.8 MB	1.6 MB	42.9	1.6 MB	42.9
2026ACR1414	ACS	9.1 MB	1.9 MB	79.1	1.8 MB	80.2
2023CrystGrowthDes8469	ACS	3.7 MB	0.7 MB	81.1	0.7 MB	81.1
2026CrystGrowthDes2939	ACS	8.2 MB	2.0 MB	75.6	1.8 MB	78.0
2023CRV13291	ACS	25.5 MB	3.5 MB	86.3	3.3 MB	87.1
2026CRV4375	ACS	12.6 MB	2.6 MB	79.4	2.5 MB	80.2
2023JCE4728	ACS	3.1 MB	1.0 MB	67.7	1.0 MB	67.7
2026JCE1723	ACS	3.2 MB	1.2 MB	62.5	1.2 MB	62.5
2023JOC16719	ACS	9.9 MB	2.3 MB	76.8	2.0 MB	79.8
2026JOC5520	ACS	5.4 MB	2.0 MB	63.0	1.8 MB	66.7
2023OL9243	ACS	2.2 MB	1.4 MB	36.4	1.3 MB	40.9
2026OL5021	ACS	2.5 MB	1.4 MB	44.0	1.4 MB	44.0
2026Tetrahedron135290	Elsevier	2.4 MB	2.2 MB	8.3	1.4 MB	41.7
2026Tetrahedron135286	Elsevier	1.1 MB	1.0 MB	9.1	0.9 MB	18.2
2026TL156046	Elsevier	498 kB	374 kB	24.9	286 kB	42.6
2026TL156057	Elsevier	2.1 MB	2.0 MB	4.8	0.5 MB	76.2
2024PCCP770	RSC	2.3 MB	1.3 MB	43.5	0.9 MB	60.9
2026PCCP9840	RSC	5.3 MB	1.9 MB	64.2	1.3 MB	75.5
2024TheorChemAcc4	Springer	1.8 MB	0.7 MB	61.1	0.7 MB	61.1
2026TheorChemAcc25	Springer	1.5 MB	0.9 MB	40.0	0.9 MB	40.0
2023Synthesis3777	Thieme	976 kB	949 kB	2.8	540 kB	44.7
2026Synthesis910	Thieme	759 kB	438 kB	42.3	436 kB	42.6
2024ACIEe202314446	Wiley	2.5 MB	1.3 MB	48.0	0.8 MB	68.0
2026ACIEe26144	Wiley	5.3 MB	4.0 MB	24.5	2.9 MB	45.3
2023HCAe202300110	Wiley	10.4 MB	1.2 MB	88.5	1.2 MB	88.5
2026HCA0:e00224	Wiley	4.1 MB	1.3 MB	68.3	1.3 MB	68.3
2023JApplCryst1639	Wiley	1.1 MB	0.7 MB	36.4	0.7 MB	36.4
2026JApplCryst291	Wiley	7.3 MB	1.3 MB	82.2	0.7 MB	90.4
2026BJoc672	Beilstein	1.6 MB	0.5 MB	68.8	0.3 MB	81.2
2026BJoc620	Beilstein	461 kB	461 kB	0.0	350 kB	24.1
2026Molecules1499	MDPI	771 kB	558 kB	27.6	328 kB	57.5
2026Molecules1495	MDPI	2.8 MB	0.8 MB	71.4	0.6 MB	78.6
2026JOSS0825	JOSS	213 kB	84 kB	60.6	84 kB	60.6
2026JOSS09890	JOSS	921 kB	392 kB	57.4	246 kB	73.3
arXiv:2605.00564v1	arxiv	3.1 MB	1.4 MB	54.8	0.9 MB	71.0
arXiv:2605.00149v1	arxiv	371 kB	232 kB	37.5	232 kB	37.5
link2web.pdf	pdflatex	38.9 kB	10.7 kB	72.5	10.6 kB	72.8

Disclaimer

By inspection with the utilities of pdfinfo and exiftools, the rewrite overwrites pdf metadata such as Producer (which can be an entry like LaTeX with hyperref), CreationDate, and ModDate, while others are lost for good. With metadata such as TITLE, SUBJECT, KEYWORDS, and AUTHOR typically retained, a reference manager like manager zotero (tested with version 9.0.1) still can collect complementary bibliographic metadata.

A conversion to grayscale is more likely to be successful if the pdf of interest is converted directly. This seems especially the case if the document includes ligatures like fl, fl, ae, oe, umlauts in the Latin script; if the (intermediate) color retaining reprint failed to properly define these, a subsequent reprint constraint to gray scale may yield a gap. This issue depends both on the version of ghostscript installed, and font / pdf-engine of the pdf to be processed because recent journal publications (like by ACS, member of STIX project⁵ tend to be less frequently affected by this. This pdf-reprinter is not tested on pdf about documents predominantly written in other scripts than Latin.

Footnotes

Footnotes

1. https://www.gnu.org/software/a2ps/ ↩

2. https://tex.stackexchange.com/questions/18987/how-to-make-the-pdfs-produced-by-pdflatex-smaller?rq=1 ↩

3. https://unix.stackexchange.com/questions/93959/how-to-convert-a-color-pdf-to-black-white ↩

4. https://openpreservation.org/tools/verapdf/ ↩

5. https://en.wikipedia.org/wiki/STIX_Fonts_project ↩