How many models is reasonable—can you have too many?

Post Reply
brianbrubaker
Posts: 14
Joined: 21 Jun 2021

How many models is reasonable—can you have too many?

Post by brianbrubaker »

I'm sure this is somewhat dependent on the font. As I've read different perspectives on this, it seems like anywhere from 50–100 can be a good starting place.

I'm testing out running KO with almost 200 models in a font and I'm wondering if there's a reason I should stick to fewer models.

My thought was just to train Kern On as much as possible to see what happens. And having more models seems to have helped me identify small inconsistencies in my spacing different places. Obviously there are diminishing returns at a certain point, but it doesn't seem to have had any unwanted side effects other than the extra time involved.

Just curious what other people's experiences are with few (~50) vs. many (100+) models has been so far.
User avatar
SCarewe
Posts: 71
Joined: 23 Apr 2021

Re: How many models is reasonable—can you have too many?

Post by SCarewe »

Hi, depending on the design of the master (script, glyph count, sans, serif, heavy, thin, detailed, simple...) I usually get very decent results with around 100 models per master for a Latin sans, most often with quite a few less for a heavy master, as the shapes will be a lot more similar in regards to white space.

What is important is that you cover all the special combination cases, e.g. number to caps, punctuation to caps/lc, small caps to uc/lc, parentheses, hyphens, etc.

Once you cover all these basic references, you can already bet on pretty good results overall, after that it's more and more fine-tuning. My "base" models for these categories (uc, lc, numbers, small figures, punctuation, parentheses, maths, I'm sure I'm forgetting some) usually cover around 40 models.

Hope this helps as a reference. Quite some time ago, I wrote a script that can set these base models for a new font, so that you don't need to go through the same starting process every time you kern a new project: https://github.com/eweracs/glyphs-scrip ... eModels.py

Note this somehow seems to break spacing groups in Kern On since a few version, I haven't properly investigated yet. It's quite an old script, I'll improve it some day, but I hope it helps for now ;)
brianbrubaker
Posts: 14
Joined: 21 Jun 2021

Re: How many models is reasonable—can you have too many?

Post by brianbrubaker »

That script is great, will definitely use for some reference. Thanks for sharing!

As I've gone through and toyed with more of the models, I do think it's helpful to not have so many. My first pass I just kept adding all the suggested models to see what would happen. After more review, it's seeming like KO finds more clarity / certainty with fewer but more varied models as opposed to more models across the board, just like you're saying with the special combos.

KO still suggests model pairs even when there are a ton of models, but as I understand it those are often simply based on the frequency of those pairs occurring out in the world according to inputs in the backend, not on what's necessary for better results.
Post Reply