Why figures fail at production, not at review

Two different checks, weeks apart

A reviewer looks at a figure on screen, at whatever size their PDF reader chose, and judges whether it supports the claim. Production looks at the file: its resolution at printed size, whether its fonts are embedded, its colour space, and whether the text is legible at the column width the journal will actually use.

Nothing a reviewer does tests any of that, which is why the failure arrives after acceptance — at the point where you are waiting for a DOI and the fix now means remaking figures you last opened months ago.

The four that cause most queries

• Resolution measured at the wrong size. A figure is not “300 dpi” in the abstract — it is 300 dpi at a stated physical width. An image that is 300 dpi at 20 cm is 150 dpi at 10 cm, and most journals print at a single or double column width well under what authors assume. Check the journal’s stated column widths and size the figure to them.
• Fonts not embedded. A vector file referencing a font the printer does not have substitutes another, which moves text, overlaps labels and occasionally changes a minus sign. Embed fonts on export, or convert text to outlines where the journal allows it.
• Colour space. Screens are RGB; some print workflows are CMYK. Saturated RGB blues and greens shift noticeably on conversion, and a figure whose meaning depends on distinguishing two similar colours can lose the distinction.
• Text too small at final size. Axis labels set at 6 point on a figure that will be reduced to half width end up at 3 point. Set the figure at final size and read the labels at that size before exporting.

Choosing the file format

Use a vector format — EPS, PDF or AI — for anything drawn: line graphs, bar charts, diagrams, flow charts. Vector scales without loss and prints crisply at any size. Use a raster format — TIFF, or PNG where the journal allows it — for anything photographed or captured: micrographs, gels, scans, clinical images.

The mistake that causes the most avoidable loss is exporting a drawn chart as a screenshot. Once it is pixels it cannot be recovered, and a reviewer will not notice while production will.

Colour that survives more than one reader

About one in twelve men of northern European descent has a red-green colour vision deficiency, and a substantial number of readers will print your figure in greyscale regardless of what you intended. Both problems have the same solution: never let colour be the only thing carrying the distinction.

Add a second channel — line style, marker shape, direct labelling, hatching. Then check the figure twice: convert it to greyscale and look at it, and run it through a colour-blindness simulator. Palettes designed to be safe under both conditions are freely available and worth adopting as a default rather than a fix.

Image integrity, which is checked and enforced

Journals increasingly screen images for manipulation, and the rules are stricter than most authors assume 1. Adjustments to brightness and contrast must be applied to the whole image, not to part of it. Splicing lanes from different gels requires a visible divider and a statement in the legend. Nothing may be added or removed.

Keep the unprocessed originals. Editors ask for them, sometimes long after publication, and an author who cannot produce them is in a difficult position whether or not anything was done wrong.

Before you submit

Open the journal’s figure guidelines &mdash 2; they are usually a separate page from the general instructions for authors — and check five things: accepted formats, minimum resolution and the width it is measured at, colour space, maximum file size, and whether figures are submitted separately or embedded. Then size each figure to the stated column width, read the smallest text at that size, and export with fonts embedded.

Common questions

@misc{figures_that_fail_at_production_2026,
  author       = {{Pubrica Academy}},
  title        = {Why figures fail at production, not at review},
  howpublished = {Pubrica},
  year         = {2026},
  month        = {mar},
  url          = {https://pubrica.com/insights/figures-that-fail-at-production/},
  note         = {Accessed: <span data-cite-acc="iso">2026-10-15</span>}
}