Journal figure checker: resolution, colour, fonts and lines before submission

$$\text{effective dpi}=\frac{\text{pixels across}}{\text{width (mm)}÷25.4}$$$$\text{pixels needed}=\frac{\text{width (mm)}}{25.4}\times \text{dpi}$$$$\text{size at final width}=\text{size in file}\times \frac{\text{final width (pt)}}{\text{width in file (pt)}}$$

Resolution. A raster figure has a fixed number of pixels; the journal judges how many fall in each inch once it is printed at the column width. One inch is 25.4 mm, so a figure 89 mm wide is 3.504 inches, and 1200 pixels across it are $1200÷3.504=342.5$ dpi. Publishers set the minimum by the kind of figure: photographs and micrographs (halftones) about 300 dpi, combination figures (images with text or lines) 500–600 dpi, line art (graphs, drawings) 600–1200 dpi, because thin lines and small letters show their pixel steps first. The checker uses the journal’s own number; where the guideline gives none (Nature, for example, wants line art as vector files instead), it uses 600 dpi for combination figures and 1000 dpi for line art as benchmarks and can only warn, never fail, on them. ACS asks 600 dpi for greyscale images, which the checker applies when the image is greyscale.

Recorded DPI. PNG files store it in the pHYs chunk, JPEG in the JFIF header or EXIF, TIFF in the XResolution and ResolutionUnit tags. It changes no pixels, only the size programs place the image at: 1200 pixels recorded at 72 dpi are placed $1200÷72\times 25.4=423.3$ mm wide. The checker warns when it does not match the chosen width; recording the right DPI (without resampling) fixes it.

Colour mode, bit depth, transparency and profile. These are read from the file, not guessed from the pixels: TIFF photometric interpretation, ink set, bits per sample, extra samples (alpha), Photoshop layer data and the ICC profile tag; JPEG component count, the Adobe APP14 transform (CMYK or YCCK), sampling factors and ICC_PROFILE segments; PNG colour type, bit depth, tRNS, iCCP and sRGB chunks. Alpha is reported as “in use” only when some pixel is not fully opaque. Most journals now publish online first and ask for RGB; CMYK files are converted to RGB with the embedded profile where the browser applies one, otherwise with $R=255(1-C)(1-K)$, which is labelled as not colour-managed.

JPEG quality. A JPEG’s quality setting is not stored, but its quantisation tables are. Software built on libjpeg (Pillow, ImageJ, most cameras’ and plotting programs’ exports) scales the standard tables of ITU-T T.81 Annex K by $5000/q$ below quality 50 and $200-2q$ above it. The checker computes the tables for every quality from 1 to 100 and reports the one that matches; when no quality matches exactly (Photoshop and others use their own tables) it reports the nearest by average step size and says it is an estimate. Graphs and figures with text should not be JPEG at all; photos below quality 90 get a warning.

SVG and PDF. A vector figure has no fixed size in print, so the checker scales it to the chosen width and measures what is inside. In an SVG it reads each text element’s computed font size and each stroke’s width, multiplied by every transform above it. In a PDF it reads the page’s drawing instructions: font size (Tf) times the text matrix and the current transformation matrix, line widths (w) at each stroke, images placed with their pixel size and placed size (so their effective dpi), and the fonts, whether they are embedded and whether they are Type 3 (matplotlib’s default, which Nature asks authors to avoid). Text converted to outlines cannot be measured or edited by the journal, and Nature says not to outline text; the report says when no live text is found. Lines of width 0 are hairlines, one device pixel wide, which disappear in print. PDF previews and colour checks are not drawn on this page; EPS files get the format and size checks only, because reading them needs a PostScript interpreter.

Text in raster images. Pixels do not record which marks are letters, so the checker does not estimate font sizes from images: a guess from shapes would be wrong too often to trust. It gives the pixel size a minimum-size letter must have at that width and draws the journal’s smallest text and thinnest line on the preview, to scale, to compare by eye.

Colour-blind safety. The main colours of the image (or of the rendered SVG) are simulated for protanopia and deuteranopia with the Machado, Oliveira and Fernandes (2009) model; pairs that differ by more than 20 ΔE for normal vision but less than 12 after simulation are listed. Nature and ACS ask authors to avoid red/green combinations.

Fixes. Resampling uses a Lanczos filter (a = 3, separable, with premultiplied alpha), widened when shrinking so every source pixel contributes. Shrinking is safe; enlarging meets the number but adds no detail, and the tool says so. The TIFF is 8 bits per channel, one page, flattened, with LZW compression and the horizontal predictor and the DPI in its tags; PLOS requires LZW and PNAS asks for it instead of JPEG compression. A naive RGB to CMYK conversion ($K=1-max(R,G,B)$, no profile) is offered only if a journal or printer requires CMYK.

Sources. 14 publishers’ artwork guidelines, linked in the table below and checked on 10 November 2026. Science, Cell Press, RSC and MDPI refused automated access on that date: their values were checked through search results of their own pages or carried over from the journal-figure-size tool (checked 8 November 2026), and what could not be confirmed is marked. Individual journals within a publisher can differ; the target journal’s own guide is the final word.

Machado GM, Oliveira MM, Fernandes LAF. A physiologically-based model for simulation of color vision deficiency. IEEE Trans Vis Comput Graph 2009;15:1291–1298, doi:10.1109/TVCG.2009.113. Wong B. Points of view: Color blindness. Nat Methods 2011;8:441, doi:10.1038/nmeth.1618. Cromey DW. Avoiding twisted pixels: ethical guidelines for the appropriate use and manipulation of scientific digital images. Sci Eng Ethics 2010;16:639–667, doi:10.1007/s11948-010-9201-y. ITU-T Recommendation T.81 (JPEG), Annex K, w3.org/Graphics/JPEG/itu-t81.pdf.

1. A graph exported as a 1200 × 800 px PNG at 72 dpi, for a Nature single column (89 mm).

1. Width: $89÷25.4=3.504$ in, so the image has $1200÷3.504=342.5$ dpi at that width. That meets Nature’s 300 dpi for photographs, but a graph is line art: Nature wants it as a vector file, and against the 1000 dpi line-art benchmark it would need $3.504\times 1000=3504$ px. Result: warn on resolution, fail on format (Nature does not accept PNG for main figures).
2. Recorded DPI: at 72 dpi it is placed $1200÷72\times 25.4=423.3$ mm wide. Record 342.5 dpi instead (no pixels change) and it opens at 89 mm.
3. Text: 5 pt (Nature’s minimum) at 89 mm is $5÷72\times 342.5=23.8$ px of font size in this image; labels smaller than that are too small. The preview draws a 5 pt guide to compare.
4. Colours: the red line (#D62728) and the green one (#2CA02C) are distinct for normal vision, but with deuteranopia both become olive (the checker lists the pair). Use blue and orange, or label the lines.

2. A matplotlib figure saved as PDF with figsize = (7, 4.5), so 504 pt wide, for the same 89 mm column. The page is scaled by $252.3÷504=0.5006$: 10 pt axis labels become 5.0 pt, 8 pt tick labels $8\times 0.5006=4.0$ pt (below Nature’s 5 pt: fail) and a 0.4 pt line 0.20 pt (below Nature’s 0.25 pt: fail). With matplotlib’s default pdf.fonttype = 3 the fonts are Type 3, which Nature asks to avoid. Fix it at the source: plt.rcParams["pdf.fonttype"] = 42 and figsize=(3.5, 2.25) (89 mm is 3.5 in), so every size you set is the printed size.

3. A micrograph saved as JPEG. Its luminance table starts 8, 6, 5, 8, 12, 20, 26, 31: exactly libjpeg’s table at quality 75 ($\lfloor (16\times 50+50)/100\rfloor =8$ for the first entry), so the checker reports quality 75 and warns that 8 × 8 block artefacts may show; at quality 95 the table would start 2, 1, 1, 2. Re-saving as TIFF stops further loss but cannot undo it; export again from the original data.

4. A Cell Press full-width (174 mm) line drawing. 1000 dpi needs $174÷25.4\times 1000=6850$ px across. A 2400 px export is 350.3 dpi: enough for a photograph, not for line art, so it fails; export the drawing as PDF or EPS instead.