Engine update: mirrored pictures, steadier text, and print layouts that behave
We launched our own conversion engine in mid-July. It hasn't sat still since. Today's update rolls several dozen targeted fixes into the live converter, each one found by comparing our output against what Publisher itself produces for real documents. Here's what got better.
Mirrored pictures come out mirrored
Publisher lets you flip any picture, and people used that button constantly: a clip-art figure turned to face the text, a photo mirrored to fit a layout. Until now those pictures converted un-flipped, facing the wrong way. Flip states now carry through, horizontal and vertical, including the tricky case where a flipped picture is also rotated. Nearly a thousand documents in our test library had at least one mirrored picture. All of them straightened out.
Text sits where Publisher put it
Three quiet fixes that add up to pages looking right:
The first line of a text box now starts at the exact height Publisher used. This sounds minor until every heading on the page rides three points too high and nothing lines up with anything. Over ten thousand documents in our test library moved closer to the original from this change alone.
Hyphenation now follows the same exception list Office uses, so words break at the same spot they broke in Publisher, and paragraphs wrap the same way.
And when a font family ships several faces, the engine picks the one Publisher picked. Fewer lines coming out bold that shouldn't be, and the reverse.
Tables keep their colors
Publisher stored table cell backgrounds at least four different ways across its versions. The engine now decodes all of them, plus cell border colors, plus tables that run at a quarter turn. Calendars look like calendars again, and the shaded header row on a price list stays shaded.
More fills survive
Two-color shades, tinted scheme colors from Publisher 2000-era files, and gradient center-shading all render now. These are the little recipes that make a 2004 newsletter masthead look like itself, and when they're dropped the page goes flat and gray.
Print layouts stop guessing
Label sheets put the right number of copies in the right positions. Booklet covers come out in reading order. A card designed for a print shop renders on its full sheet, the way Publisher exports it, instead of as a lonely cropped rectangle. A couple of odd paper sizes now normalize to what your printer actually expects.
Old clip art draws completely
The arcs, pies, and chord shapes inside 90s-era Windows metafiles now draw instead of vanishing. WordArt stored as a metafile recovers cleanly. And WordArt sizes itself to its frame the way Publisher does, so the arched headline fills the space it was given.
Fewer blank pages
Around eighty documents in our test library that used to convert to blank or nearly blank pages now render their content. Another forty stopped emitting duplicated content. Every one of those was somebody's real file that simply failed before, and this category matters more to us than any single feature.
All of this is live on the converter now, across every export format: PDF, Word, and IDML. It will ship in the desktop app too. If you tried a file a month ago and something looked off, try it again.
And if a file still comes out wrong, that's the most useful thing you can show us. Wrong conversions are how the engine gets better: each one points at the next fix. The story of how we got here is in why we built our own engine.