Koezu and PowerPoint: how faithful?

Measured 2026-11-07 by comparing every slide as Koezu draws it with PowerPoint's own PDF of the same file.

Opening PowerPoint files

81%

of 104 slides in 5 decks come out the same as in PowerPoint; 18% have small differences and 1% visible ones.

Decks made in PowerPoint (or by tools that write PowerPoint files), opened in Koezu. Pixel similarity 0.91 (1 is identical).

Saving to PowerPoint

75%

of 76 slides in 4 decks come out the same as in PowerPoint; 25% have small differences and 0% visible ones.

Decks made or edited in Koezu and saved as .pptx, as PowerPoint shows them. Pixel similarity 0.90 (1 is identical).

Open and save again

100%

of 156 slides in 6 PowerPoint decks come back byte for byte after opening them in Koezu and saving with nothing changed.

Slides edited in Koezu are rewritten; the others are not touched.

How it is measured

PowerPoint for Mac saves each deck as a PDF: that is the reference. Koezu opens the same .pptx and draws each slide. Both pictures are 960 px wide. A part of a slide counts as out of place when the other picture has nothing of a similar colour within 3 px of it (6 px on a full-HD slide), so letters drawn a hair apart by two programs do not count, but a missing, moved, recoloured or rewrapped object does. A slide is the same when under 2% of it is out of place, has small differences at 2–5% (typically a paragraph sitting a little lower, or a line breaking at another word), and visible ones above that. Pixel similarity (SSIM, 1 is identical) is shown too: it counts every pixel, letters included. Slides hidden in a deck are skipped, as PowerPoint's PDF skips them.

The decks are lecture decks from a university engineering course and test decks for equations, arrows, fonts and scripts. The round trip is checked in the files themselves: each slide's XML and its relationships, before and after.

Run it yourself: tests/fidelity/score.sh in ref results in the Koezu repository, with the decks in in and PowerPoint's PDFs in ref.

Each deck

Deck Direction Slides The same Small differences Pixel similarity
arrows Opening 2 100% 0% 1.000
extras Opening 2 100% 0% 0.995
lec04 Opening 43 77% 23% 0.898
lec05 Opening 47 85% 13% 0.900
omml-test Opening 10 70% 30% 0.926
lec05ed Saving 47 70% 30% 0.889
omml-test-ed Saving 11 73% 27% 0.930
sample-dk Saving 9 89% 11% 0.928
sample-dk2 Saving 9 89% 11% 0.928

Known limits