Lost in models influences

Post Reply
Posts: 4
Joined: 15 Jul 2023

Lost in models influences

Post by alex »

Hi Tim!
I’ve finally gotten my hands onto learning Kern On and it looks great! However, I’m having hard times understanding how models influence each other and therefore what to expect and how to proceed. After mostly following the tool’s model suggestions, it feels like Kern On starts linking pairs which are not really relevant to each other and ignoring pairs which are. Here’s a couple of examples:

— “Removed T, because it’s too loose compared to ‘G”
Why? Is that the closest connection it found? Empty vs round? I want T. to be a bit looser, but ‘G is fine. Ok, let’s “try again” — then it suggests removing HH, which seems to be even less relevant.

— Changing V. doesn’t affect T. right next to it at all.
I’d expect V. and T. to heavily influence each other, but apparently changing one affects some other pairs, such as g, becoming too tight or even removing :H. Sure, those are also related, but there are seemingly other higher priority pairs which are ignored.

Knowing about such misconnections, I feel uncertain about what I’m doing. I kern T. and it suggest forgetting HH, but ignores V. Where in the process does that put me? Should I check v. or y. or Y. or ‘A for that matter? How to know what else requires human attention?

Am I doing something wrong? Am I misunderstanding the process? :)
User avatar
Tim Ahrens
Site Admin
Posts: 421
Joined: 11 Jul 2019

Re: Lost in models influences

Post by Tim Ahrens »

Kern On “links” all models to all autokerned pairs.

It does not link any models to each other. Rather, it tries to assemble them all into an autokerning engine. Sometimes it doesn’t manage to “fit them all under one hat”, as we say in German, and it needs to remove one or several models. There is a certain built-in flexibility that helps KO fit them under the hat and if you tell KO about your priorities (“Try again”) it can sometimes fit these two models under the hat at the price of not being able to fit two other modes under the hat.

It’s not easy to comment on your specific examples without seeing the font. In any case:

> I want T. to be a bit looser, but ‘G is fine.

Why exactly is this? It seems that you say: if [‘G] was looser it would be too loose, it would be inconsistently loose compared to something else. Keep in mind that consistency is the only reasoning that exists in KO’s world. So, it would be interesting to determine in more detail what is actually behind the “I want”.

> Changing V. doesn’t affect T. right next to it at all.

It seems they are both models, [V.] and [T.]? The model values are fixed, set by you, they don’t influence each other. The model pairs influence the kerning values of the auto pairs. You can change the value of the model [V.] as long as it does not contradict the model [T.] or any other model.


It seems some of the model removals that cause you headache are related to special spaced glyphs. Keep in mind that you can, and sometimes need to split up special spacing groups in order to achieve what you want. Often it is helpful to unite them if you can’t really say that one group of glyphs is supposed to be spaced more loosely or tightly than another group.

In general, it makes life easier (for you and Kern On) if you don’t have too many special-spaced models. Sometimes, if you are getting non-zero autokerning for pairs that are typically un-kerned (i.e. zero) then it might be better to adjust the sidebearing instead of simply setting a zero model. Unless, of course, you are really sure this makes the pair visually too tight or too loose. I am kind of referring to the post I just wrote on another thread. If [:H] really Looks to tight if you switch it to auto, and the model causes problems then maybe it’s just in the wrong special spacing group? If you just Think the value should be zero then adjust the sidebearing so that the autokerning becomes zero. Again, this is difficult to say without looking at the font.
Post Reply