What is your recommendation for kB in "limit kerning to"?

Post Reply
stephennixon
Posts: 13
Joined: 18 Jun 2021

What is your recommendation for kB in "limit kerning to"?

Post by stephennixon »

I am presented with a difficult question each time I use Kern On: how many kB should I limit kerning to?

I understand that it is a logical setting for you to include, as I'm sure that first challenge to any auto-kerning system is that "it adds junk data and too many pairs." However, I suspect that very few type designers have a strong knowledge of or opinion about how precisely many kilobytes of kerning should be included in a given font. In general, I suspect that we want as much kerning as is necessary for a high-quality result, and no more. How should we know how many kilobytes that will take?

I don't mean to be overly challenging, I am merely expressing that I feel like KernOn should help a user understand how to decide how much to limit kerning. I don't pretend it's an easy answer! But, I'm sure there are various tradeoffs made to reach a certain limit ... maybe those could be somewhat exposed to the user?

For instance, I could imagine a series of toggles, with options for categories like "kern UC-UC, kern UC-lc, kern lc-UC, kern UC-numerals, kern UC-lc diacritics" etc ... and toggling different categories on and off would update a predicted amount of kerning data.

Since there isn't currently that preference panel, can you help me to understand what limit I should be setting? I am happy to email my current Glyphs file to you, but here are some specifics on it:

- There are 8 masters (4 upright, 4 sloped-roman italics)
- There are glyphs to fill out the Google Fonts Latin Plus and Cyrillic Plus glyphsets, with a few additions and subtractions. In this particular font, there are 990 glyphs.
- 418 glyphs have at least one contour that is drawn/editable, and the other 572 are purely composed
- It's a "marker" font, so glyphs have relatively irregular sides. That is, the left side of the D is slightly different from the left side of the H, and so on (screenshot attached of the basic Latin in the light style, to give a general sense of it).

Thanks for any suggestions and insights!

As one extra point of feedback: it seems that my preference for "limit kerning" isn't stored in the font, so I have to consider it each time from the 36 kB default. Ideally, I think, once I come up with a limit, that would be suggested next time I run it.
Attachments
Screen Shot 2022-08-05 at 2.37.07 PM.png
Screen Shot 2022-08-05 at 2.37.07 PM.png (317.98 KiB) Viewed 4158 times
Screen Shot 2022-08-05 at 2.24.47 PM.png
Screen Shot 2022-08-05 at 2.24.47 PM.png (116.17 KiB) Viewed 4158 times
stephennixon
Posts: 13
Joined: 18 Jun 2021

Re: What is your recommendation for kB in "limit kerning to"?

Post by stephennixon »

Having now watched https://kern-on.com/under-the-hood, I see that the kilobyte limit isn't quite aligned to which categories are or are not kerned, but rather, which pairs are most likely in Wikipedia or Twitter text. It is a very clever solution to the problem of what to include in kerning (though I can't help but feel like it probably misses important cases that might pop up in e.g. company logos, where a numeral might appear between letters, or camel casing might occur...)

But, knowing the general approach, it is less clear how Kern On might best surface that kb limit decision to users.

Even so, I do think might be some middle ground. One idea...

Maybe KO could analyze the font’s glyph set, and determine how far down your pair priority list the font could go. Then, the limit decision could at least say something like, "36 kb limit will cover approximately 92% of real-world kerning pairs," essentially saying how far down the list of pairs a certain data cap will go.

This could potentially help give users a sense of how things are working under the hood (without necessarily watching the videos), and give them some understanding of potential diminishing returns. For instance, if going from 36kb to 50kb will get me to 98% coverage, and I can't get to 99% without boosing to 80kb, that would give me a reasonable level of certainty about where to cut of the data. I can see how it might be difficult to accurately predict this number before running the first time, but perhaps the estimate could just get more accurate on subsequent runs. After all, I would guess that almost all projects will run the Kern On process multiple times before being done.

Anyway, I'm sure there are many things here I still don't understand well enough to make a really good suggestion, but I'd still love a bit of guidance around what kb limit to set, and in the long run, I'd love for that guidance to be part of the tool itself.

Thanks!
Post Reply