Hey Tim et al,
Regardless of what font I’m working with in KO, I’m finding that with essentially any letters that have extremes amounts of natural white space (e.g. L, X, A, T, etc.), if I don’t set them as model pairs with lots of other glyphs, KO will either auto-kern them way too tight OR way too loose compared to what’s consistent everywhere else in the family. BUT, setting those "extreme" model pairs almost always breaks all my other, more standard model pairs.
Tim, I’ve done my best to keep your “looking” vs. “thinking” heuristic in mind here. And still, I’m finding that I’m consistently disagreeing with KO in these same instances.
In other words, these “extreme-to-standard” model pair issues don’t seem to me to stem from being unwilling to change how I think about spacing / kerning a la Kern On. It feels to me more like a genuine disconnect between what KO considers consistent and what actually looks consistent given the other models that have been set. And this is happening seemingly regardless of how many models I’m setting or what type of font I'm working on.
I’ve tried doing minimal model pairs, and I’ve tried adding in a ton of model pairs. Both approaches yield the same result: combinations such as L-H and T-H are inevitably auto-kerned much tighter than looks right to me, while pairs like X-X, E-T, T-T, etc. are auto-kerned much, much looser than what looks consistent to me. And when I try to set those various models to a place that I’m happy with, they break the models I’ve set for lots of controls like n-o, o-n, H-O, H-A, etc.
I don’t want to be too dogmatic about any design process. That's what drew me to Kern On in the first place. I didn't want to overthink. I want to trust my eyes and only pay attention to whether or not the spacing is consistent in a given family. And for the most part, KO is incredible at finding that consistency. But I’m having a difficult time fully trusting this process with these instances. It seems like KO disagrees with how I like to, for example, space the n-o, H-O, and L-H relative to each other. I’ve included some screenshots for some examples of the kind of things I'm seeing on a regular basis.
I feel like there are only possible 3 causes for what's going on, or maybe some combination thereof. I would love any thoughtful perspectives / feedback on what is most probable:
Possible Cause #1: My spacing isn't right on those characters...DUH!
I’d of course like to think that how I space these things isn’t actually so inconsistent as to be more than 30 points off. But am I missing something? Does that L-H combo in the light weight example really look more consistent with the H-H spacing and I’m just going crazy with how I’ve spaced the T-H?
Possible Cause #2: Kern On still doesn’t have enough information (models).
If that’s the case though, and assuming I’m willing to give Kern On the benefit of the doubt on side-bearing adjustments within around 10 points in either direction, roughly how many Standard Kerning models should I expect to set before I can expect that there be no auto-kerning suggested between a Standard Kerning glyph with the H?
Possible Cause #3: There’s something going on under the hood with how Kern On assesses these characters.
I assume this is the least likely case but have to ask in case there’s something actually going on behind the scenes causing this issue if #1 and #2 aren’t easily identifiable as the culprit(s).
I apologize for the lengthy post. Wasn’t sure how to make it shorter and this has been on my mind for awhile.
Fighting with KO over the same kinds of model pairs regardless of font. What am I missing?
- brianbrubaker
- Posts: 28
- Joined: 21 Jun 2021
Fighting with KO over the same kinds of model pairs regardless of font. What am I missing?
- Attachments
-
- KO-example-3.png (107.62 KiB) Viewed 2121 times
-
- KO-example-2.png (368.96 KiB) Viewed 2121 times
-
- KO-example-1.png (362.74 KiB) Viewed 2121 times
- Tim Ahrens
- Site Admin
- Posts: 416
- Joined: 11 Jul 2019
Re: Fighting with KO over the same kinds of model pairs regardless of font. What am I missing?
Hi Brian,
Thanks for your detailed feedback. Some of this is impossible to comment on without having the file or seeing more details, some of which I will try to answer here.
I’d say LH as well as TH are sensible models, also XX and TT. The latter is a really interesting question, isn’t it? Since using Kern On, I noticed that it is usually much too tight if it doesn’t have positive kerning. If you have few models and it gets overly wide, just set it as a model to tell Kern On about your preferences.
Btw, I don’ think 38 is a lot of models. You should expect to add a few more.
Agreed, this warning should be ignored. I am aware that Kern On is not always “right” about these warnings. It is really just a hint to check one pair against another and it’s not necessary to remove all warnings. Sometimes the warnings can be very helpful to notice inconsistencies and sometimes they are wrong guesses. That’s the nature of warnings.
Also, there is always an interplay between all models and it might be that the real near-inconsistency is between another pair of models and Kern On didn’t pick the most appropriate comparison as a warning. Again, it’s difficult to say without seeing the whole file.
37 models is not many. So, don’t be surprised to find pairs that are wrong and that you need to add as models.
This particular case is a good example to explain an aspect of the engine’s inner workings:
As a general principle, many people (type designers and typographers alike) feel that pairs that “open upwards” should have a somewhat tighter physical spacing than pairs that “open downwards”. Do you know Jost Hochuli’s Detail in Typography?
He explains that the “light from above” is stronger than the “light from below”, this is why the A typically has somewhat smaller sidebearings than the V (AH and VH look consistent if VH has a slightly larger physical spacing). Same for LH vs TH, of course. You could also say, the reader tends to look at the top of the letters more than at the bottom, this is why we need to emphasize the top more. This is a different description of exactly the same thing as Hochuli’s light from above and below.
Kern On supports this principle but it doesn’t enforce it (and it doesn’t allow to apply it in reverse). It determines how strong this effect is supposed to be by looking at – you guessed it – the models! Therefore, it is perfectly natural that you need LH as well as TH (or AH as well as VH) so as to tell Kern On your preference, whether you want this effect to be strong, or just a bit, or not apply it at all.
In your screen shot it seems that Kern On has picked a very strong value for the effect. This is why the LH is so much (physically) tighter than TH. It may have guessed this from some of the other models (impossible to tell without seeing them but there may be model comparisons that suggest that you want the effect to be strong).
If you set the LH to zero, Kern On will have a very clear indication as to how strong you want this effect to be. I can’t see the RSBs of T and L in your font. Are they the same? That would imply that you don’t want this effect at all. After you set LH as a model (in addition to TH), Kern On will understand what you want.
Thanks for your detailed feedback. Some of this is impossible to comment on without having the file or seeing more details, some of which I will try to answer here.
Sorry, this is difficult to say without seeing your file.brianbrubaker wrote: ↑13 Mar 2024 Regardless of what font I’m working with in KO, I’m finding that with essentially any letters that have extremes amounts of natural white space (e.g. L, X, A, T, etc.), if I don’t set them as model pairs with lots of other glyphs, KO will either auto-kern them way too tight OR way too loose compared to what’s consistent everywhere else in the family. BUT, setting those "extreme" model pairs almost always breaks all my other, more standard model pairs.
How much do you consider minimal and how much is a ton?I’ve tried doing minimal model pairs, and I’ve tried adding in a ton of model pairs. Both approaches yield the same result: combinations such as L-H and T-H are inevitably auto-kerned much tighter than looks right to me, while pairs like X-X, E-T, T-T, etc. are auto-kerned much, much looser than what looks consistent to me. And when I try to set those various models to a place that I’m happy with, they break the models I’ve set for lots of controls like n-o, o-n, H-O, H-A, etc.
I’d say LH as well as TH are sensible models, also XX and TT. The latter is a really interesting question, isn’t it? Since using Kern On, I noticed that it is usually much too tight if it doesn’t have positive kerning. If you have few models and it gets overly wide, just set it as a model to tell Kern On about your preferences.
If you place the text cursor between L and O you will see the labels under the value slider. They are the models that the define the upper and lower limit of he possible span for the value (quite often it’s the same model for minimum and maximum). This also implies that the pair(s) given by the labels are the ones that have most impact on the autokerning value. There may be other models that have impacted the kerning value but this information cannot be accessed. The labels are a bit simplified.What is KO looking at in those other 38 Standard Kerning models that would lead it to think that's consistent?
Btw, I don’ think 38 is a lot of models. You should expect to add a few more.
Agreed, this warning should be ignored. I am aware that Kern On is not always “right” about these warnings. It is really just a hint to check one pair against another and it’s not necessary to remove all warnings. Sometimes the warnings can be very helpful to notice inconsistencies and sometimes they are wrong guesses. That’s the nature of warnings.
Also, there is always an interplay between all models and it might be that the real near-inconsistency is between another pair of models and Kern On didn’t pick the most appropriate comparison as a warning. Again, it’s difficult to say without seeing the whole file.
37 models is not many. So, don’t be surprised to find pairs that are wrong and that you need to add as models.
This particular case is a good example to explain an aspect of the engine’s inner workings:
As a general principle, many people (type designers and typographers alike) feel that pairs that “open upwards” should have a somewhat tighter physical spacing than pairs that “open downwards”. Do you know Jost Hochuli’s Detail in Typography?
He explains that the “light from above” is stronger than the “light from below”, this is why the A typically has somewhat smaller sidebearings than the V (AH and VH look consistent if VH has a slightly larger physical spacing). Same for LH vs TH, of course. You could also say, the reader tends to look at the top of the letters more than at the bottom, this is why we need to emphasize the top more. This is a different description of exactly the same thing as Hochuli’s light from above and below.
Kern On supports this principle but it doesn’t enforce it (and it doesn’t allow to apply it in reverse). It determines how strong this effect is supposed to be by looking at – you guessed it – the models! Therefore, it is perfectly natural that you need LH as well as TH (or AH as well as VH) so as to tell Kern On your preference, whether you want this effect to be strong, or just a bit, or not apply it at all.
In your screen shot it seems that Kern On has picked a very strong value for the effect. This is why the LH is so much (physically) tighter than TH. It may have guessed this from some of the other models (impossible to tell without seeing them but there may be model comparisons that suggest that you want the effect to be strong).
If you set the LH to zero, Kern On will have a very clear indication as to how strong you want this effect to be. I can’t see the RSBs of T and L in your font. Are they the same? That would imply that you don’t want this effect at all. After you set LH as a model (in addition to TH), Kern On will understand what you want.