The "blurry" look is almost certainly antialiasing (often called Font Smoothing).

Do you mean like this:

Yes, that's it. Is there anything I can or should do about it?

The word "Antialiased" is meant to look better (cue argument about which is better) than the word "Aliased".

If you want it to look like the latter on Windows XP, then go to the Start menu, then Settings submenu, then Control Panels. When that opens, select Display and you will get a box like the top most box below. Make sure you are clicked onto the "Appearance" tab, then click on the "Effects" button.

The second box below will appear - simply uncheck the "Use the following method to smooth edges of screen fonts:" box, and then click the "OK" button.

That should do it!

Thank you, thank you!

Fonts have to have certain sophisticated information embedded in them to render well in b&w (no anti-aliasing). Mrs Eaves does not have this information, so it looks better when anti-aliased (to me at least). Unfortunately, this also means "blurry" (unless the anti-aliasing is done by hand, which is very rare).

hhp

This is also probably related to the different paths fonts with TrueType and fonts with PostScript outlines take in Windows. Fonts with TrueType outlines are rendered using the TrueType rasterizer licensed from Apple but maintained by Microsoft. Fonts with PostScript outlines (including OpenType CFF fonts) are rendered using the PostScript font rasterizer supplied and maintained by Adobe - this comes built in to Windows 2000 and XP, but was an add-on (Adobe Type Manager) for earlier version of Windows.

True, and the important thing is to realize that Postscript fonts cannot have clean b&w rendering on any platform, at least not until somebody writes a supremely intelligent renderer - don't hold your breath.

However do note that Mrs Eaves is available in both flavors (although the TT version is probably blind-generated).

hhp

breathe

Interesting, and encouraging!

Although it still doesn't touch The Good Stuff:



hhp

BTW, about the FontFocus thing, I just realized that people have probably tried that before, but maybe thought it wasn't "revolutionary" enough? I think it might actually be a decent compromise solution.

On the other hand:
- "FontFocus leaves the shapes of the glyphs completely unchanged" is clearly false.
- I think in the last example CT is coming out better (and I'm not the biggest CT fan).

hhp

Word has spacing problems too (as it's trying to be WYSIWYG). Although still not perfect at these small sizes Word's 'Reading Mode' is giving better results than the sample on their site.

Hrant: thanks for the feedback. You're certainly right that the language about changing glyph shape needs improving; I'm trying to distinguish it from the more common technique of displacing individual knot and control points relative to the underlying grid. A small amount of squooshing, while subtle, does not count as leaving the shape completely unchanged.

Simon: yes, the constraints are pretty different when you're trying to be WYSIWYG than if you can widen both glyph shape and spacing at small sizes.

I hope to have a live demo up pretty soon so people can kick the tires extensively.

This is next generation ClearType (with sub-pixel positioning, as discussed in the book Now Read This.



This is 7pt type at 146 dpi, which is the native resolution of my Dell laptop. Obviously, when talking about how text looks and, more importantly, reads best on screen one has to deal with the difficulty that unless you are sitting next to someone looking at exactly the same screen -- taking it in turns, in fact, to sit directly in front of the screen -- the number of variables involved from native resolution, gamma, manufacturer make it very difficult to have a meaningful discussion. And that's just talking about mechanical variables, and not human variables such as colour sensitivity, which come in to play with ClearType, especially on lower resolution screens.* As noted on the graphic, the red line is 6.1 cm long on my screen, which will give you some idea of the impact of resolution on this discussion.

*As noted in this thread, which should really be moved to the General Discussion forum.

Simon: That sample is pretty good (the color fringing seems under control), but the spacing is a bit sloppy.

Raph: So that FontFocus stuff is yours? Cool. BTW, I've always thought WYSIWYG is a hoax.

John: That sample is pretty good, but even when I pull back to match your apparent resolution*, the colors are too strong (and it might be a tad dark too). Also, you say "7 point", but it's important to realize that the lc vertical span is 12 pixels. That's plenty of room for manual anti-aliasing to do a better job, including curve fidelity (like look at the overly aliased bottom-left of the first "e"). The other problem of course is that glyph rendering is not consistent (which I think is not a problem accross saccades, but sometimes might be within a single fixation). BTW, what font is that? Assuming it's not a[n existing] core font, how would Times and Arial look?

* Otherwise I think my display is probably very close to yours.

To my eyes, John's sample is zinging colours all over the place rather distractingly. Resolution seems to be the key - on my screen here at work, that line comes in at 11cm rather than 6.1.

Even fine tuning the ClearType settings (see below) only went so far to avoid the colour flashes - and how many 'normal' computer users are going to do that? For me at least, I'd keep it on handhelds where the actual dpi resolution of the screens is sufficiently high enough.



I wonder if some people are particularly sensitive to it?

>Simon: That sample is pretty good (the color fringing seems under control), but the spacing is a bit sloppy.

True - probably a combination of TNR and the fact that Reading Mode is probably optimized for the larger sizes.

David, in fact the graphic you put up initially came up towards the bottom of my LCD screen, which means I saw it at a (moderate) angle, which gave it a distinct shade of mauve!

> actual dpi resolution

Note however that more important than "face" resolution is perceived resolution, which depends of course on how near to the eyes the given screen typically is.

--

There are some good things about CT (and other subpixel methods), but I think the key thing to remember is that it's simply The Next Big Thing*, which means it gets more positive mind share than it deserves. Just like hinting, which was all the rage a few years ago, but is no longer fashionable - even though effective screen resolution hasn't gone up (in fact it might have gone done on average if you count the spread of mobile devices).

* Something consumerism vitally needs.

hhp

John: That sample is pretty good, but even when I pull back to match your apparent resolution*, the colors are too strong (and it might be a tad dark too).

Well, you know I like my type dark. David Earl's example of ClearType tuned to reduce colour artifacts is a good demonstration of the relationship of colour and contrast in CT rendering: I wouldn't use that kind of tuning, because it reduces the contrast, which has an impact on what you like to call notan.

Of course, I don't normally read text on screen at 7pt and, obviously, at that size the pixel patterns get quite coarse. That fact that CT makes this recognisable as an individual typeface at that size is itself remarkable. The typeface is Constantia.

The other problem of course is that glyph rendering is not consistent (which I think is not a problem accross saccades, but sometimes might be within a single fixation).

I've not seen any evidence that this presents a problem, even in a single fixation, so long as contrast is maintained. At this very small size, the differences are more noticeable than they would be typically, and I think the improvements in letterspacing from subpixel positioning benefit more than the variation in rendering harms.


There are some good things about CT (and other subpixel methods), but I think the key thing to remember is that it's simply The Next Big Thing*, which means it gets more positive mind share than it deserves. Just like hinting, which was all the rage a few years ago, but is no longer fashionable...

If I thought this stuff was to do with fashion or consumerist vitality, I wouldn't be very interested in it. Background technologies in font manufacture or raster imaging are not driven by consumer trends: they're driven by the need to achieve goals efficiently. Even if one demonstrated that hand-tuned greyscale bitmaps were the best thing going in terms of readability -- even ignoring the spacing defects they share with b/w bitmaps --, no one is going to sit down and make a range of sizes of them for a family of 3000+ glyph fonts. TrueType hinting for b/w rendering was developed because manually making bitmaps is too labour-intensive; manually making hand-tuned greyscale bitmaps is that much more labour-intensive. It simply isn't feasible as a model for producing all the fonts that the users of the world's many languages need. ClearType rendering further reduces the font production overhead by substantially simplifying the kind of hinting that is required. So from a font developer perspective, ClearType rendering is part of a continuing effort to make the production of readable type for screen less labour-intensive. [Which is not to say that this is the impetus behind the development of the technology: the MS Advanced Reading Technologies team are not primarily involved in font development.]

> That fact that CT makes this recognisable as an
> individual typeface at that size is itself remarkable.

Not when you compare it to hand-made anti-aliasing. Then it's ho-hum.

> I've not seen any evidence that this presents a problem

But I think you and I agree that empirical testing of readability has been a joke so far.

> the improvements in letterspacing from subpixel positioning
> benefit more than the variation in rendering harms.

Yes, but you can have both.

> they're driven by the need to achieve goals efficiently

I think they're driven by the need to keep people in jobs.

> ignoring the spacing defects they share with b/w bitmaps

When you use grayscale properly you can space just fine, and the small problems remaining are more than enough to make the gain in crispness worthwhile. CT is still too fuzzy.

> no one is going to sit down and make a range of sizes of them for a family of 3000+ glyph fonts.

I wouldn't say "no one", and -as I find myself pointing out to you too often- it's not all-or-nothing: the people who only need a few dozen glyphs (like readers of English, the most widely used language) can still benefit a lot.

> manually making bitmaps is too labour-intensive

No, superhinting is much more so.

> manually making hand-tuned greyscale
> bitmaps is that much more labour-intensive.

Yeah, sort of like making outline fonts. Who needs 'em?

Different people have different needs. For some people, having a consistent and easy way to render fonts is the most important thing, even if it means there's still room for improvement at the glyph (not to mention bouma) level. For some other people (the type of people who think making new text faces is an activity inherently useful) that remaining room for improvement is highly significant, even if it can't be equally useful to everybody all of the time.

hhp

Yes, but you can have both.

No, actually, you can't. The improved spacing is specifically a function of allowing widths to align to subpixels rather than to whole pixels. This means that the stem of a letter may begin to be rendered on either a red, green or blue subpixel, and that will not be consistent because it is determined by allowing the widths to fall where they will on the subpixels. If you want totally consistent glyph rendering, then you have to space to whole pixels. This is the biggest single problem in b/w and traditional anti-aliasing (including the current generation of CT rendering, which does not use subpixel positioning): the smaller the type, the worse the spacing becomes because of the crudity of the full pixel grid.

In the graphic below, I've used what I call a diagramatic display to compare the spacing and rendering of type without subpixel positioning (top) and with subpixel positioning (below). It's diagramatic because I'm not showing the actual sub-pixels, only the cumulative affect on full pixel colour, which makes it easier to see the variant rendering and spacing.




If you think that superhinting is more labour intensive than making bitmaps, then you simply do not know what you are talking about. Why don't you talk to Sampo Kaasila about why the TrueType hinting model was developed? Or to people who have worked -- in manufacturing conditions with deadlines and costs -- on both bitmap making and TT hinting. Laurence Penney, who knows a lot more about TT hinting than either of us, defines TT hinting a method for the efficient creation of bitmaps. Yes, applying deltas to turn on an off individual bits is labour intensive, but the point is that this is something that only needed to be done after 95% of the hinting was complete and you already had largely correct bitmaps at a greater range of sizes than you could ever hope to obtain manually.

Hrant, if you want to spend your time making handtuned greyscale bitmaps, I'm the last person who will say you should not. But I wonder if you will ever ship anything other than illustrations to Typophile discussions? I think this is simply a practical issue, and I don't see the benefit as in any way justifying the costs, not when one considers relative impact of devoting one's time to other things, e.g. some good Armenian text faces.

Looking at the FontFocus website to which Raph directed us, I'm particularly struck by the image comparing FontFocus with ClearType, near the bottom of the page. It isn't a very fair comparison, because as Si has noted there are spacing problems that are the result of Word and not of ClearType; also the text size chosen in all the FontFocus examples is below normal reading size, and presumably this shows the rendering to advantage. But the real kicker is that, apart from the spacing problems, the ClearType rendering is better. The FontFocus has a nice even typographic colour, but that colour is a pale gray. Try actually reading the two passages, and you get a really good empirical understanding of the importance of high contrast to readability. To use Hrant's terminology, the ClearType rendering has better notan. The comparison of FontFocus with ClearType is a good example of the difference between making on-screen type look nice and making it readable.

As John Berry quipped in one of his recent CreativePro articles, there is a reason why printing was known as 'the black art' and not 'the grey art'. One of the few things on which legibility studies to date provide really conclusive evidence is the importance of high contrast; of course, it is also one of the easier things to test.

> The improved spacing is specifically a function
> of allowing widths to align to subpixels rather
> than to whole pixels.

You're talking about CT, I'm talking about the general case. You can most certainly have both good spacing and consistent glyph rendering, if you design your glyphs properly.

What you can't have -in conjunction with those two things- is WYSIWYG across sizes - but that's the least of worries.

> you have to space to whole pixels

But "grayspacing" allows the perception of spacing (what really matters) to be carefully controlled, effectively extending spacing beyond the pixel boundary. Look at the "v" in Mana-16.

> Why don't you talk to Sampo Kaasila about
> why the TrueType hinting model was developed?

Would he (and others) tell me what I'm supposed to hear, or what's really in his head? Duh.

I have done both superhinting and bitmap design (both under deadline), and to me, in the end the latter is better. Furthermore, I have done grayscale bitmap design (which you have not). Lastly, the orphaning of superhinting reinforces my view that The Next Big Thing is what's really going on.

> I wonder if you will ever ship anything other than illustrations

Hmmm. Try asking the people who have purchased Mana 16 and 13, and have been clamoring for the 11*. The reason my stuff isn't in Windows for example is not least due to my inability to bend over for the powers that be, to gayly adopt their vision of The Next Big Thing.

* BTW, there will be a 9 and an 8 as well, and the whole family will share a style, as well as crisp rendering with high fidelity. There's your range of sizes. You want greater than 16 PPEM? Use an outline font. You want less than 8 PPEM? Riiight. You want a 10, 12, 14 and 15? Pay me extra and you'll have them.

Grayscale bitmap fonts are simply not sexy enough for shareholders; but they deliver superior readability. And just you wait for my serifs to kick in...

> good Armenian text faces.

Talk about stuff that doesn't sell...

hhp

Yes. As I like to say, ideally I'd want two different modes: a "print preview" mode (maintains page color and WYSIWYG glyph positioning better) and a "readable" mode (sacrifices some WYSIWYG in favor of better contrast and somewhat sharper glyph edges).

T

Would he (and others) tell me what I'm supposed to hear, or what's really in his head? Duh.

Oh, sorry, I forgot: when you hear something you don't like, or someone disagrees with you, it must be because they're being dishonest. I'm not even going to pursue that line of discussion any further, because obviously it will just end in a spitting match. I continue to be interested in conversation, so instead I'll follow this line of thought:

I'm not denying that there are circumstances in which handtuned greyscale bitmaps might be useful. However, I don't find the Mana example you showed above anything like as readable as the CT text that surrounds it in my browser: like all bitmap fonts it suffers from the fact that the effective stem weight must be either one or two pixels, while the CT text can render to between one and two pixels, which is where you actually want it for typical text sizes on a 146ppi screen. Of course, you must also take into account that the pixels on my screen are very small, which means that your Mana single-pixel stem weight is really light for me: much too light and sparkly to be easily readable. Like specifying type in pixel sizes in CSS, using size-specific bitmap fonts fails to keep pace with varying screen resolutions. This is why I think the circumstances in which bitmap fonts continue to be useful are all device-specific, e.g. embedded fonts in mobile phones. If you live in a world in which you know the resolution of the output device then yes, you can design effective bitmap fonts for reading. But most of the text we read on screen does not exist in such a world: we don't know the resolution of the output devices for which we are making fonts and rendering systems, which is why a flexible, scaleable technology is much more desirable than bitmaps, however good. Of course, you could decide to inhabit forever the Steve Jobs world of 72ppi, but I'd rather not.

> the effective stem weight must be either one or two pixels

Nope.
There are two ways (that I know of) to make stems non-integer weights. And they force no more compromises than CT does in any situation. Also: there's more to good color than stem width, as you can see by comparing Mana 13 and 11 to Arial.

> the pixels on my screen are very small

Right. For your setup Mana-16 would be much better. If 16 is too big, please wait for Atanasia.

It remains, however, that if you want crisp type, 14 PPEM is about the cutoff between 1 and 2 pixel stems. The relevance of this cutoff changes however depending on effective dip.

> we don't know the resolution

But such uncertainties affect CT as well. For example in the other thread Simon defended the "only 5% don't like CT" statement, but between the lines it's clear that the test conditions that could possibly produce such low complaint levels have to be draconically and minutely enforced. On the ground the truth is very different.

> you could decide to inhabit forever the Steve Jobs world of 72ppi

It's not up to us, it's up to the behemoths who make our hardware and software. They've promised hi-res screens for ages, but the best they've managed to deliver is large low-dpi screens you have to put far away! And on top of that these are "brittle" systems, in terms of things like OS integration.

hhp

You're talking about CT, I'm talking about the general case. You can most certainly have both good spacing and consistent glyph rendering, if you design your glyphs properly.

If that were true, it seems that good typefaces could be designed on a coarse grid system, whereas we know that the higher the resolution the more subtleties in proportion and spacing can be used and that these enhance readability. Compare, for example, Verdana with the closely related Meiryo Latin in the Now Read This book. Verdana was designed around bitmaps, and the proportions of letters and spacing are determined by the bitmap grids. It is a great typeface for its purpose and original target rendering technology, but Meiryo is a better typeface with better proportions and spacing. As Matthew notes in the text, this is precisely because he was no longer constrained by having to draw around pixels, because the ClearType rendering allows much more subtlty in the proportions and spacing.

I notice that your Mana is quite condensed: well, in bitmap fonts you have to make a choice at some sizes between too narrow or too wide, and you probably made the right choice.

There are two ways (that I know of) to make stems non-integer weights [in bitmap fonts].

I would like to see example of these techniques.

One last note, before I go and do some real work, regarding 'the truth on the ground'.

One of the reasons why the ClearType tuner is a web-based interface is so that MS can conduct research on user preferences, i.e. they can track how many people select each of the different options for tuned display. Greg Hitchcock relates that they were surprised that the majority of users of the tuner were actually selecting the fuzzier, heavier options, rather than the cleaner but lower contrast options. Interesting, although certainly inconclusive in any way.

> the higher the resolution the more subtleties in proportion
> and spacing can be used and that these enhance readability

Well of course, but I fail to see how that precludes using grays for exactly that effect.

And there's no question that subpixel positioning helps increase resolution, the issues are that:
- it also causes color fringing;
- automatic rendering still can' t match the human "hand".

Ideally I'd like to make CT bitmaps by hand!

> Mana is quite condensed

Which actually makes it harder to scale down.
But horizontal space is (still) very valuable onscreen.

Atanasia will set about as wide as Mana-16, but be 2 pixels shorter (hence the name), and will have a 1-pixel stems (and serifs).

> I would like to see example of these techniques.

You can actually see traces of both techniques in the already released Mana fonts, but eventually I might use one of the techniques for Atanasia's Bold (which might end up a Demi instead).

hhp

> One of the reasons why the ClearType tuner is a web-based interface

I checked with Greg, and we're not actually counting user preferences. The observation is based on unscientific 'looking over shoulders' of people using the tuner.

On the subect of the tuner we now have a standalone version - http://www.microsoft.com/typography/ClearTypePowerToy.mspx

And there's no question that subpixel positioning helps increase resolution...

Careful with the terminology: subpixel positioning is specifically the technique of allowing glyph widths to fall on subpixel boundaries. This is different from subpixel rendering, which is what you are referring to in this comment.

Ideally I'd like to make CT bitmaps by hand!

Well that's technically impossible, because you can't paint individual subpixels. The best you could do is create colour bitmaps like the g on the left below, which would mimic the impact of ClearType on the colour density of full pixels, but this is quite different from what ClearType actually does, which is to rasterise to actual sub-pixels (as shown in the g on the right).


As you can see, there is still an overall reduction in contrast in the colour bitmap vs. the ClearType, and I don't see how this could be addressed in manual editing without making the letter bolder: the reduction of a horizontal grid of e.g. nine subpixels to three full pixels simply reduces the rendering options.

I checked with Greg, and we're not actually counting user preferences. The observation is based on unscientific 'looking over shoulders' of people using the tuner.

Thanks for the clarification, Si. I guess I'll have to go look over some shoulders too.

> that's technically impossible

As a font, sure, but you could always make images with text formed with any color pixels you want. Like I could take some CT screengrabs and improve them manually.

On the other hand, you don't want to do that in tandem with subpixel positioning, since you'd need multiple versions of each glyph! But that's where you could bring in consistent glyph rendering (while dumping WYSIWYG however).

hhp

All: thanks for the feedback. My goal in making the FontFocus samples (link is to more extensive comparison, btw) was to make the renderings consistent in all respects other than underlying font technology. So, among other things, they're all tuned to be as closed to WYSIWYG as the platform allows. I'm not going to argue that Times at 11ppem with default spacing is in any way optimum for readability, aesthetics, or what have you. In particular, it's easy to believe that just about anything that makes the text heavier is an improvement. I'm greatly reminded of the problems of trying to compare photo image quality when the color profiles are not carefully matched; most people will choose the better color rendition no matter what other artifacts are present.

It would be most fascinating to see what would happen if a good type designer were to design fonts for optimum screen reading with FF. I am a type designer of middling competence, but maybe I should get busy :)

Hrant: it turns out you can shift subpixel rendered graphics by exact 1/3 pixel increments with almost no change in quality. For a great many applications, 1/3px is Good Enough�.

I think hand grayscaling is great, but I'm not sure how practical it is on a larger scale. I'm not sure if the sum total of all hand grayscale fonts is enough to fill, say, the needs of a basic Web browser.

As a font, sure, but you could always make images with text formed with any color pixels you want.

You are missing my point, Hrant. I am not debating whether you can colour pixels however you want; I'm pointing out that you cannot colour individual subpixels. What you see in the diagramtic images I post showing coloured full pixels is an approximation of the impact of ClearType on pixel colour, which is a useful way of showing some things, but it does not represent the actual resolution benefit of ClearType, which is only apparent if you look at the actual subpixels. Note, for example the righthand, outer edge of the upper bowl in my image of the two g's, above. In the left g, you can see that the approximation of the colour impact on the full pixel is a pale blue. But if you look at the actual subpixel rendering you'll see that only two of the three subpixels that make up that pixel -- the red and the green subpixel -- are actually turned on by in the supersampled outline. So if you try to mimic or improve on this by colouring full pixels, you are going to affect the weight of the letter, usually by making it heavier.

> you cannot colour individual subpixels.

Not directly at the font outline level, but why couldn't I manipulate the three distinct color channels directly, in effect addressing the subpixels? Again, combining this with subpixel positioning makes it perhaps too tedious* - although still possible. But anyway I'd rather dump WYSIWYG and get consistent glyphs.

* Can you say "OpenType handmade grayscale pixelfont"? :-)

> if you try to mimic or improve on this by colouring full pixels, you are
> going to affect the weight of the letter, usually by making it heavier.

?
Why can't I for example take that italic "h", notice that there's a "twitch" in the stem just above the arch, and change the colors -minding the channels- to make it better?

--

It seems that you might be assuming there's simply no way to improve on the automatic rendering. Considering all the times hand-tweaking has helped in virtually every human endeavor, that doesn't seem tenable.

hhp

...but why couldn't I manipulate the three distinct color channels directly, in effect addressing the subpixels?

Yes, it is possible to address individual subpixels in this way. Someone would need to a) define a colour-bitmap font format (can PhotoFonts do this, I wonder? I've just never seen any PhotoFonts designed for size specific bitmap text), and b) make a colour-bitmap font editing tool that makes it graphically easy to address the individual colour channels. Without the latter, I think you would go mad trying to make a single font.

So are we talking conceptually, here, or are you actually going to do this?

> I've just never seen any PhotoFonts designed for size specific bitmap text

?
What else would it be used for?

Do you mean that it's display-centric? Does it have to be?

> are you actually going to do this?

Mayyybe. :-)

Has anybody here actually made or used PhotoFonts?
When is that amazing new generation of MS core fonts going to be released?

hhp