What is best practise in variable font development?

Igor Petrovic

I am working on my first type family. It has one axis (weight) and for now two masters (thin and black). I draw about 80% of the set (draft quality) and before stepping into the finalization phase I would like to check a few things. 

First is that I draw in 1000upm, I thought it's a standard these days, but a friend told me that 2000upm is maybe a better option for the variable font and that Google Fonts variable projects use 2000upm. What's your opinion? Sounds logical to me, especially because my font has softened corners, and I feel that the 1000upm integer grid sometimes offers two ugly options to choose between. (I also see the kink near the top terminal in the regular instance) 

Secondly, I feel that just thin and bold masters are not enough. Here is the example of the lowercase "a" as it looks currently. While it's not perfect, I generally like the look in thin and black, but when I check interpolation at 400 (Regular) horizontals and lower bowl-to-stem joint are too thin. Is it usual to add intermediate master, or you rather adjust thin and black masters? 

Also, I would highly appreciate any other recommendations besides these two I mentioned.

Thanks!

Peter Constable

The masters you use in a design tool are one thing. How the data gets compiled into a font may be another—depending on how the tools go about generating the data.

Some key things to understand about how the variation data gets organized within the font are:

- For each axis, the font will have default, plus min and max values. The default for a given axis could be the same as either the min or the max, but doesn't need to be.

- Combining axes together, there will be a default instance, which corresponds to all axis values set to the default for each given axis.

- The font will have one set of outlines, which get used for the default instance, plus delta data that gets used to interpolate all of the non-default instances.

So, in deciding whether to have additional masters, it may be useful to consider what you would want as the default instance. More and more apps are supporting variable fonts, but if your font happens to get used in a context that doesn't support variations, then only the default instance of the font will work in that app. What would you want it to be? This? Black? Or something in between, like Regular?

If you want to have Regular as the default instance, with Thin and Black as min/max on the wght axis, then you might want to have a Regular master in your design-tool sources. If you only had Thin and Black design masters but then had the tool generate the font with Regular as the default, then the outlines that end up in the actual font will be an interpolation generated by the tool, and not what you actually created as masters. That may or may not be what you want.

Of course, you could have just Thin and Black masters, and then use one or the other as the default instance. In that case, the generated font should have the same outlines as what you drew in that master, plus delta derived as the diff between the two masters.

One other thing to be aware of, in terms of the OpenType format itself. You can add additional delta data for some intermediate position on an axis on a glyph-by-glyph basis. So, if (say) most glyphs interpolate just fine between Thin and Black masters but there are issues with "a" or certain other glyphs, then the font could have additional delta data for just those glyphs. (Of course, that's talking about the OT format; how you get some tool to do that will depend on the tool.)

These comments are solely from a perspective of familiarity with the OT format. You should still get someone who actually designs variable fonts to provide recommendations.

Alex Visi

Peter Constable said:

More and more apps are supporting variable fonts, but if your font happens to get used in a context that doesn't support variations, then only the default instance of the font will work in that app. What would you want it to be? 

Should we really care about it? Most people still don’t know about variable fonts, so people who do know about them also know how to use them. So I wouldn’t make the file size bigger with the whole Regular master, just because someone might (mis-)use the variable font in MS Word.

Also, in Adobe apps, the default instance will appear on top of the list, meaning if you make your Bold the default, the list is going to be Bold-Light-Regular-Bold, and that’s where most people will use the v-font.

Rafael Saraiva

Is it usual to add intermediate master, or you rather adjust thin and black masters? 

Hi Igor, besides the technical information Peter shared, which I found excelent and very valuable, on the design perspective in general, yes -- it's pretty common to add an intermediate master. That's because the Black weight has a lot of weight compensation to work. By that your Regular will have thin strokes in  the most complex shapes as you already noticed. There are other tricks you could use, like have bracket / adjustment layers in the Black weight (if you are using GlyphsApp) where you can draw the outlines without compensation in a separate layer and use them for inteporlation. However, as Peter mentioned you'd loose the ability to have the Regular as the default instance which is still important nowadays.

Georg Seifert

Adding an intermediate Regular master gives you a tiny bit more control to the expense of much bigger files and more work to actually do the drawings and kerning.

That doesn't mean that you can’t have intermediate drawings for some glyphs. Usually tha a, e, and g are good candidates. In my experience, most of the other glyphs interpolate quite well.

With rounded corners it might be a good idea to switch to 2000 units. But the kinks in the Regular you mention are not caused by rounding do to a too coarse grid. That happens because the proportion and angle of three consecutive points are different. So even without a grid, you will get the kink. (there is a section on our website that discribes this: https://glyphsapp.com/tutorials/multiple-masters-part-2-keeping-your-outlines-compatible)

Alex Visi

Are there any cons to using high UPMs, over 2000? In complex and detailed variable fonts even 2000 isn’t enough to keep everything in-between smooth.

Thomas Phinney

Further to what Georg mentions, one trick to reducing kinks and interpolation problems in variable fonts is to avoid excess points, and especially avoid having on-curve points that are not at extrema.

Igor Petrovic

@Peter Constable @Rafael Saraiva

Many thanks for your valuable comments, they shed light on the aspects I wasn't aware of. Just to check one thing. If I understand right, it's not possible to have Regular as the default instance if it's not actually the master? Or it's possible, but it will use outlines which are interpolation between Thin and Black masters. 

If later, why that would be an issue in any scenario, because I anyway have to tune interpolation (this way or another) in order that Regular looks good. In other words, I am more worried if my Regular looks problematic generally, than if it's default or not.

And the other question related, the interpolation based on masters I defined will be used regardless of what is pointed as a default instance, right?

@Georg Seifert @Thomas Phinney

Thanks, this solves my big concern! I was worried about the kink issue, but thought to raise that question later. And yes, now I remember this advice from your excellent article. I read all the articles in the MM series before started to draw a family, but forgot the equalizing proportion trick because I was like: I am not gonna have points anywhere except extremes. Then later I realized that is very tricky to define terminals of the rounded corners typeface just using extreme points, and I thought I would somehow get away with that additional on-curve point. 

And just to check about @Alex Visi question about higher UPM's. What are things to consider in the context of having UPM larger than 2000? Does it make files much bigger?   


  

Andreas Stötzner

Without knowing much about variable fonts, actually, I do know, as a matter of fact, that an automated interpolation of a Regular cut out of a Thin and a Bold can never produce a satisfactory result. Never ever.
The issues outlined in this thread confirm my believe not to touch ‘variable fonts’ ever. The very thought of handing over the moulding of my Regular, the core cut of every typeface, to some sort of automatism behind my back, is just insane i.m.h.o. Never, never ever would I consider for one second to admit this. Interpolation functionality may be a handsome tool during the production process to generate a variety of cuts, but my eye has to have the final plea about everything. Therefore I think the VF concept is hardly more but a revenant of the good old Multiple-Master concept, which has been buried with dignity many years ago, with good reasons. If one is fascinated by technique in the first place, I understand that VF is a lovely playground. But if you are serious about the very design of your outlines in the first place, with no compromise admitted to that, than you’re probably better off the old way, to choose a discrete range of cuts deliberately and concentrate on the direct control of every single outline you are going to deliver. And I hardly believe anyone out there needs a typeface with 50 or 89 weights generated automatically … maybe it’s just me.
Sorry to speak as a heretic here. Good luck with your work, anyway.

Mark Simonson

You do know you can use intermediate masters, right?

Jens Kutilek

Andreas Stötzner said:

Without knowing much about variable fonts, actually, I do know, as a matter of fact, that an automated interpolation of a Regular cut out of a Thin and a Bold can never produce a satisfactory result. Never ever.

It's not Thin to Bold, but I was surprised to learn that TheSans uses no intermediate masters between ExtraLight and Black.

(For the variable font version we introduced intermediate masters for a few glyphs to avoid the need to keep overlapping contours in the shipped font. This concerns glyphs where curves are intersecting, like 8, &, …)

Peter Constable

@Igor Petrovic: "master" is a concept that best applies to your source data in a font tool, not to the generated font. In your source data, there are multiple sets of outlines for different "master" designs, and the tools can interpolate instances between the masters. In the font, there is exactly one set of outlines, which are for the default instance, plus sets of deltas, and the runtime rasterizer can interpolate instances using the outlines plus some subset of the delta sets, depending on the instance coordinates. In many cases, the default-instance outline plus delta sets for certain extremes may correspond to source masters outlines, but the relationship between source and output may be more complicated.
 

Alex Kaczun

Multi-master fonts & QuickDraw GX fonts have been introduced since the early 1990s. Nothing new here. Interesting that it has been introduced today again as OpenType variable fonts.

It's an interesting concept and gets everyone excited, but has any of this technology ever taken off, and has it ever been utilized by the general public? No, it has not. And IMHO it probably never will. In the past, there were too many printing errors associated with this technology. Many print shops refused to accept these fonts for large commercial print jobs. As a result, many still will shy away from this technology today. It's very complicated to manage.

In addition, the amount of time spent correctly creating all of these instances is enormous and difficult to do. Special care needs to be taken with every step of the way. Looking, testing, looking, testing... on and on—it never ends. Trust me, I've created a few of these fonts for Linotype-Hell in the early days of this technology. After more than a year developing just one axis for weight— it was never fully implemented.

A simple grotesque is somewhat doable, but a serif font with complex shapes is extremely tedious. You will soon lose all your hair. Sorry to rain on the parade but I'm just being honest. Good luck to all those who care to try.

May the force be with you. Amen.

Mark Simonson

I've said this before when variable fonts first started making a comeback: Things are quite a bit different now than in the nineties when they failed to catch on.

First, the tools and hardware we have now for building variable fonts are vastly better and faster than what was available then.

Second, OS/app support is much broader now than it was then. In the nineties, GX fonts only worked on Macs. Adobe MM only worked as variable fonts in Adobe apps. Modern variable fonts are backed by all the major players--Microsoft, Apple, Adobe, and Google, and all the major browsers.

Third, in the nineties, the main use case for fonts was for creating documents for print. Being able to vary characteristics of fonts didn't have much attraction to designers, and it also didn't work that well. I found myself having to convert a MM font to outlines in order to get it to output at the printer for a project I did back then.

Today, the web is arguably the main use case for fonts, and variable fonts have several advantages for web design. First, since you can get an entire family into a font the size of a couple of static fonts, it reduces the bandwidth of using more than a few styles. It basically gives you the same flexibility for style as existed for size. Second, it allows for responsive typography, so you can adjust font characteristics to fit displays of different sizes and shapes.

Just because they failed over two decades ago doesn't mean they will fail now. They still could, or they may not be as big as some imagine, but I think chances are better they will stick this time. 

Daniel Benjamin Miller

I think they can be good, but I think right now they may be most useful in being able to produce multiple preset sizes.

Alex Kaczun

Mark, you are absolutely right that the technology has improved dramatically over the last decade. But, these fonts are still very difficult to make. I give anyone credit for trying their hand at it.

I'm convinced that at some point they may catch on—especially as you say for the web. Maybe when artificial intelligence & apps improve to the point where all the tedious shaping, testing, and point placements are automated completely. Then, I'll jump in again. As I said, I really like the concept, the implementation is still tedious and difficult.

However, when AI takes over we may all be out of a job!

But, then again, our creativity can be unleashed in other ways. Looking forward to tomorrow...

Andreas Stötzner

Jens Kutilek said:

… glyphs where curves are intersecting, like 8, &, …)

What??

Thomas Phinney

Variable fonts allow for overlapping (including self-intersecting) contours, as they make interpolation easier in some cases. Due to some particular rasterization issues, some of us try to avoid using this feature, but it certainly exists. I think there are only two or three glyphs out of ~1200 in Science Gothic that have overlapping contours, but there are some.

Peter Baker

For Glyphs users anyway (I can't speak about the other editors), the process of making a variable font is so very much like the process of making a conventional font family that the effort involved won't be wasted, even if variable fonts fail to take off (as I don't think they will, at least on the web, for reasons outlined by @Mark Simonson). The very same source file can be used to build either the variable font or the font family.

Thomas Phinney

That’s entirely true for other font editors as well. Structuring a variable font gives you a later shortcut to creating a substantial number of instances, relative to the number of masters you put in. @Alex Kaczun’s experience with the early days of predecessor technologies is doubtless useful for a bunch of reasons. But the tools are better now. A lot better. Maintaining master compatibility is usually pretty easy if you understand the principles well. Doing a family with a weight axis today would take a lot less time than designing a bunch of weights separately.

Andreas Stötzner said:

Without knowing much about variable fonts, actually, I do know, as a matter of fact, that an automated interpolation of a Regular cut out of a Thin and a Bold can never produce a satisfactory result. Never ever.

Bullshit.

I mean, you are welcome to have an opinion about what you find acceptable, but that sure as heck isn’t a “matter of fact.” Most people think Myriad Regular is quite well crafted—and its Regular is not merely interpolated between a Thin and a Bold, but between ExtraLight Condensed, ExtraLight SemiExpanded, Black Condensed and Black SemiExpanded.

 Not to mention that if you don’t like the results in your example (with just a weight axis), you could put a master in for the Regular, so as to get a more controlled result. Three masters for five or six weights is still a significant savings of time and effort.

Alex Visi

Andreas Stötzner said:

And I hardly believe anyone out there needs a typeface with 50 or 89 weights generated automatically … maybe it’s just me.

Everyone needs only a couple of fonts, but can pick exactly the right ones from those 89 options.

Alex Kaczun

@Thomas Phinney—Just because the tools are better today & allow you to create fonts quicker doesn't make the results better. We all here have varying degrees of experience and talent.

In 1979, Mike Parker purchased for the Mergenthaler Linotype Type Design NYC office a 3/4 Million PDP-1135 digitizing system using Ikarus software for font interpolation & font development. As I said, font interpolation is nothing new—the tools existed even in the early days while we still did letter hand drawings. We did extensive testing, using original hand drawing, marking up the drawings, digitizing the outlines & converting the outlines to vectors. Usually, we had a regular weight and we had a bold weight to work from. Sometimes, we even had a Black weight to use as a guide. Regardless, When we interpolated between the regular & the Black we did not get a good Bold result. It wasn't bad, but the interpolation did not match the Bold hand drawings we had. The reason being, the Black hand drawings had subtle proportional & aesthetical changes incorporated to make a successful Black weight. You could use the interpolated Bold but you had to adjust almost every letter form and clean things up. Now, if you tried using a light weight with a Black weight—the resulting interpolated weights in between would require even more work to clean up. It's not just a matter of interpolation and you are done. Many experienced type designers will confirm what I'm saying. Someone like Matthew Carter doesn't rely on digital interpolations to create his typefaces. He probably still draws his letterforms by hand and has a technician digitize his drawings and create interpolations if needed of maybe 1 or 2 intermediate close instances. But, I'm sure he looks at the results and makes visual adjustments afterwards as needed. That's why Carters' typefaces are so beautiful and successful. His talent speaks for itself.

I'm not saying that we all need to be like Matthew Carter. As I said, we all have varying degrees of experience and talent. But, do not try to convince me that creating digitally two vast extremes of font weights, interpolate the in between master instances, and you are done. And, on to the next new font design. Some may work this way and find this acceptable. But, the results speak for themselves. There are a lot of fonts being generated today—some better than others. Some are really bad.

Why are we all rushing to make so many new fonts? Do we really need so many fonts?

How about a few really good fonts, carefully crafted...

Peter Baker

Alex Kaczun said:

But, do not try to convince me that creating digitally two vast extremes of font weights, interpolate the in between master instances, and you are done.

No one is saying that. It has been explained in this thread that current tools give you various ways to intervene between the extremes. The kind of craftsmanship you argue for so eloquently does not have to be sacrificed.

As for why there are so many fonts about, it is simply that the availabiity of affordable tools has made the creation of fonts like the other arts. As anyone who can afford brushes, paint and canvas can have a go at painting, so anyone who can lash out a few hundred on a font editor (or simply download FontForge) can have a go at making fonts.

As a not-terribly-talented amateur who nevertheless feels the itch to make fonts (and to sell them for precisely what they're worth!), I of course think this is a very good thing. But whether it's a good thing or not, there's no going back now. But I don't think the truly great work will be drowned in the flood of amateur stuff, any more than a Monet is diminished by the thousands of imitations by weekend painters.

John Hudson

Someone like Matthew Carter doesn't rely on digital interpolations to create his typefaces. He probably still draws his letterforms by hand and has a technician digitize his drawings and create interpolations if needed of maybe 1 or 2 intermediate close instances.

Matthew doesn't draw letterforms by hand. When I last worked with him, he was still preferring to use Fontographer, because it's what he's used to: that's how long he has been making his own digital outlines.

Matthew often collaborates with people on production, which may indeed involve interpolation. We built the Sitka fonts for Microsoft, and the sources provided by Matthew were master extremes for weight (regular to bold) and optical size (6pt to 36 pt). Everything else was interpolated.

The point I would make, though — apart from the obvious one that a weight interpolation range of regular to bold is easier to manage than one of light to black —, is that Matthew knew that some form of interpolation or outline deltas would be used in the font production (MS were initially looking at using pre-variable TTF outline instructions to implement the optical size adjustments), and designed accordingly. He knows letterforms and how they behave well enough to be able to anticipate interpolation, even when working in tools that have limited capabilities in that regard. We had to modify the point architecture of some outlines to get them to interpolate without kinks, but the actual shapes of the glyphs were perfect, and the interpolated intermediate instances worked beautifully without the need for intermediate masters because that was the design brief.

From my own experience, I would say that designing for static fonts and designing for variable fonts involve different kinds of decisions, and I don't see any point in fetishising either. Personally, I usually like to have an anchor somewhere in the middle of the design space, but at the moment I am also working on a project in which the default instance is in the corner between two axes (and part way along a third). Understanding how variations technology works is crucial to being able to anticipate what will happen when extrapolating into design space between the axes. That is design as decision making within a technical framework, which is exactly what Matthew is very good at.

Andreas Stötzner

Thomas Phinney said:

… Most people think Myriad Regular is quite well crafted—and its Regular is not merely interpolated between a Thin and a Bold, but between ExtraLight Condensed, ExtraLight SemiExpanded, Black Condensed and Black SemiExpanded.

I accept the invitation for re-visiting Myriad (of which I am fond of, like myriads of others), after what you say.

I hesitate to do this, but anyhow, here is what my eyes tell me about it.

Myriad Regular

Lowercase
The left half of the a looks a little leight-weight compared to the right side. Not much, but still. The long stems of b, f, h, k and l appear a bit too heavy. In comparison the thick parts of the bowls of b and d appear a little too light besides their stems, although they are actually quite right. The same occurs in g, p and q, although not so much obvious. The e is too light altogether, especially the left side. The centre part of s is too light. The e and s are too small (as in many typefaces), they would benefit from a 101% scaling. The downward leg of k is too light (or it looks being to light, perhaps it isn’t, considering the aforementioned thickness of the main stem). The three somewhat comparable curvy descenders of g, j and y are not well balanced. That of g appears to be rather light-weight, that of the y is considerably heavier. That of the j is inbetween. The overall stroke contrast pattern of the design, which is very moderate, seems to have not made it convincingly into the letters v, w, x and y. In v, x and y an evenness of strokes has been achieved, so there appears nothing of the sublime strong/light rhythmic pattern which guides most of the other glyphs. The strokes of the w have been kept deliberately lighter for to avoid that glyph coming out being too dark, OK. But the actual rythmic pattern of those four strokes – strong-light-strong-light –, are altered to strong-light-light-strong, which makes the glyph look a bit alien among, e.g., m, n, o, u or others. The stroke which appears as the heaviest is the 4th one, the outer right arm, that seems really illogical to me. Having said that, it occurs to me that also the central diagonal part of z is far too light.

Capitals
What has been observed with the v, w etc. minuscules above, is valid in a very similar sense for their uppercase counterparts. The upper left arm of Y is, although slightly thicker than the right one, still too light. Also the central part of S is too light (as in the lowercase), as is, again, the downward leg of K. It is a acceptable decision to keep the M very narrow, OK. But in that case I reckon, it is very narrow. The questionable distribution of stroke emphasis is the same here as has been noted for the w and W. Moreover, as that M is clearly modelled after the classical Roman design (slightly slanted outer stems, not straight ones as the overall Bodonian flavour of Myriad could have been suggesting), the departure from the light-strong-light-strong scheme isn’t satisfactory at all. In a very similar way the weight distribution of the N suggests, that the outer stems would do well with less, the central diagonal with more weight. What has been done very well, the weight balance between the horizontals of E and F, goes missing with C. The upper ending is too light. Also B, R and P appear rather pleasing to me, except that in all three glyphs the righthand-side bowl parts are too light-weight, again.

Numerals
As pleasing as they are in general, the tricky aspects which have been detected in the letters do occur here again. I can see no reason why the horizontals of 2, 5 and 7 have not been treated as strong parts. For instance the glyphs of 4, 3 or 6 come with a decent yet lively rythmic contrast, with 2 and 5 it is somewhat half-baked but with the 7 that idea has gone. The 8 is too small, a similar phenomenon as with s and e (and sometimes a) in many typefaces, but the 8 glyph is always a very tricky one.

_ _ _ _ _ _ _ _ _

All these points I do not consider being grave issues nor would I call them b… .

But I must confess I would not like to see one of my fonts being shipped in that state. Wether these things originate from interpolation or for any other reason, does not matter for me. But I question this typeface is a very good example for the promotion of creating fonts by interpolation and to handle final quality control as a secondary aspect, more or less.