Why My Display Serif Died at 4 mm: A Packaging Post-Mortem
AI-assisted: Drafted with AI tools, edited and refined by HX Studio.
Neck band prototype, 24 mm deep-indigo cotton stock, wordmark set at 4 mm cap height. AI-generated scene and materials; the letterforms come from the HX FONT-001 font file.
The proof came back in a padded envelope and I opened it in the car, which is not how you should open a proof. The bottle was fine. The glass had that amber weight you pay for, the foil on the shoulder caught the light exactly the way the mockup said it would, and the wax was clean.
The name on the neck band was a grey smear.
Four letters, 4 mm cap height, on a band 24 mm tall. The hairlines had broken into a row of dots. The ball terminals on the S had fused back into their stems.
The counters inside the round letters had closed up, so that the o read as a full stop with ambitions.
I had drawn that typeface. I had also approved the proof on a 27-inch screen at 300 pt, where it looked like a page from an 1890s specimen book, and where nothing in the world is ever printed.
For two days I blamed the printer. On the third day I opened the font file and measured.
The failure was arithmetic, not taste
A high-contrast display face is a bet. You are betting that a specific press, on a specific stock, at a specific size, can hold your thinnest stroke. Nobody makes that bet explicitly, which is why so many designers lose it: they judge the drawing at a size where every hairline is gorgeous, then ship it at a size where the same hairline is thinner than a speck of toner.
In this face the thinnest horizontal strokes measure 24 to 27 units on a 2,048-unit em. That is 0.012 em. Set a name at 4 mm cap height on a 24 mm neck band and the stroke lands on paper at roughly 0.07 mm.
Now put that next to what presses actually hold. Coated offset holds about 0.05 mm; uncoated stock holds less, and less predictably.
A toner-based digital press starts breaking a line somewhere around 0.08 mm. Letterpress wants more than either, because the impression squeezes ink outward before it lifts.
0.07 mm sits in the gap. That is the worst place to sit, because it means the same file prints on one stock and disappears on another, and nothing in the purchase process tells you which one you are buying.
The contrast ratio, measured across the bowl of the lowercase o, is 7:1. That ratio is the reason the face looks expensive at poster size, and the reason it has no margin at label size. High contrast is a loan taken out against the thin end.
Three tests you can run before you spend anything
Test one is the photocopier test, and it takes about four minutes.
Print your label at 100% on the stock you will actually use, with the ink you will actually use, then photocopy it twice and look at the second copy. Photocopying is a crude halftone followed by a threshold, which is roughly what a bad print run does to a hairline, only faster.
If the counters of e, a and o survive two generations, the drawing has room at that size. If they fill in, no amount of retouching saves the run, because the failure is already in the curves.
Test two is the 2 mm test, and it is the one that matters commercially.
Get HX Studio | Type Design’s stories in your inbox
Join Medium for free to get updates from this writer.
Shrink the wordmark to a 2 mm cap height — roughly the size it becomes when someone photographs your bottle from across a bar — and look at three things. Does daylight still get into the eye of the e, or has the terminal closed against the crossbar?
Is there still a neck under the ball on the S, or has the ball melted into the stroke? Does the apex of the M look shorter than the flat top of the E beside it?
If it does, nobody gave the pointed letters any overshoot, and the word reads as if it is sagging in the middle even when the baseline is perfect.
Test three needs no equipment. Cut the proof out, tape it to the bottle, step back to arm’s length, and look at it in ordinary room light — not your desk lamp, not the light you use to check curves.
You are checking whether you remember the name afterwards. If you remember that the type was exquisite, but not what it said, the type is doing the wrong job.
None of these tests cost money. All three of them are cheaper than a print run, and considerably cheaper than the design fee you will spend building a label around a face that cannot survive its own size.
Four decisions that got this one through
Everything below was drawn against that 0.07 mm number, not against how it looked on screen.
First, the counters in the round letters. The white inside o, e and a is pinched into four lobes around a small diamond of ink. It is the most conspicuous feature in the face, and clients assume it is decorative.
It is not. A bowl begins to close as ink spreads, and a single soft oval closes into a blob. A counter that is already divided keeps reading as an opening long after the shape around it has started to fill.
Second, the ball terminals sit on a neck. Every ball grows out of a narrow waist rather than directly off the stroke, so when ink spread fattens the join, the ball stays a separate shape instead of becoming a club. On the bad proof, that join was where the letters died first.
Third, overshoot where the eye expects it. The apex of the M rises 10 units above cap height; the flat letters sit almost exactly on it.
Ten units, at 4 mm, is not measurable with a ruler, but it is the difference between a line that looks level and a line that looks like it is collapsing toward the middle. Pointed letters need more than flat ones to look equal, and that is a drawing decision, not a spacing one.
Fourth, a deeper e aperture than the weight would normally allow. The eye is cut back further than the stroke thickness suggests it should be, which looks slightly wrong at poster size and is the reason the letter still breathes at 4 mm. Almost every small-size rescue I have made in the last three years has been this same move: give the smallest white space more room than the drawing wants.
If you want to see whether that arithmetic holds up on someone else’s bottle, the specimen sheet prints at true size and the calculations above are all reproducible from it. It is on the studio page here: Print it on your own stock, run the copier test, and decide from the paper rather than from my screenshots.
What this face cannot do
It has 90+ glyphs, one weight, no small caps, no alternates, no Cyrillic, and kerning only where GPOS provides it. It is Latin, it is Regular, and that is the entire family.
That means it sets names, not sentences. One or two words on a neck band, a bottle, a box lid, a hotel door: fine.
A wine list, an ingredient panel, a legal line, anything longer than about seven words: wrong tool, and you will feel it inside a paragraph. If you need a family that carries a menu, buy a text face; this one is for being recognised, not for being read.
I would rather write that here than have you discover it after you have built a brand on top of it.
One last thing about labels
A label is rarely read. It is glanced at, at arm’s length, in bad light, by someone who has already decided whether to pick the bottle up and is looking for a reason to keep walking. The type has about a second and a half to look like it belongs on that shelf.
That second and a half is not won by taste. It is won at the drawing stage, by somebody measuring a stroke against a process window instead of admiring it at 300 pt. Which is a duller sentence than I wanted to write, and the only one that turned out to be true.
