Amazing. I zoomed in with my phone on the generated image and was surprised to see I could just make out the text! https://storage.j0.hn/subpixelfont.png - Also wow my screen is dirty af
The 5x5 is very nice, and the 3x5 isn't bad either. Unfortunately neither of them have all of ASCII. The size is also a bit misleading because you need to add spacing so really they need a 6x6 or 4x6 grid.
It has a 5x8 font which has all of ASCII, but most glyphs are actually 4x8 and include horizontal spacing. I modified it to reduce the rest for a project I'm doing so all glyphs are 4x8. The result can be rendered on a 5x9 grid with a guaranteed line of horizontal and vertical spacing between all glyphs. It's very nice.
Using multi-level grayscale instead of just two pixel states, on and off, can produce readable text at even smaller font sizes. The catch is that I have to say "text", not "letters", i.e., rely on humans inferring the too-blurry letters from their context. And I do not even need a specially designed font for that.
> 4x4: Not enough to draw "E", "M" or "W" properly.
However, 5x5 isn't enough to draw "e" properly if you also want lowercase letters to have less height than uppercase, so you need at least 6 vertical pixels. And then that isn't enough to draw any character with a descender properly, so you need at least 7 vertical pixels (technically you should have 8 in order to allow "g" and "y" to have a distinct horizontal descender while still sitting on the baseline, but this is probably an acceptable compromise). And remember that in practice this means you will still need at least 8x6 pixels to draw each character, to allow for a visible gap between letters below and beside them.
I think the `e` looks better in the 'real pixels' example they gave; I find my tends to 'fill in' the space of the top part of the letter, and I suspect in the context of a longer sentence it'd be pretty easy to parse.
(but yeah, it's not quite right, and is especially jarring in the nice, clean, blown up pixels in the top example)
It definitely looks better in the second screenshot than the first, but you have to be very, very close to the screen in order to see individual pixels like that. And on low-res displays, which this sort of font might be necessary for, it's going to look somewhat different because low-res screens tend to be chosen for cheapness, and cheap screens tend to be monochrome, so none of that artistic fuzzy subpixel coloring.
If the author sees this. I think the lower case t would benefit from a pixel above the cross, similar to how the lower case k goes up one more pixel. It looks a lot like the capital T with how it is now. It is very well done though. Thanks for sharing.
EDIT: realized a better demo for this discussion would be the photos I took of all the tiny bitmap fonts I had converted to the C array style at the time. Thread has more but here are my favorites: https://bsky.app/profile/janiczek.cz/post/3mh25atboz224
I actually thought of this (or a previous similar project? The one posted here seems more recent...) just a few days ago while watching the announcement video for this new DJ device, since it seems to use a 5x5 font: https://driftdj.com/dj-hybrid
yo! thanks for sharing, this is a wicked cool device. I've always wanted a "pocket" mixer to take with me on planes and trains. I know there are smaller devices that can perform that duty but their more like repurposed EQ or audio interfaces that can be used to mix.
This is cool, but my problem is the opposite: I have a modern laptop screen with tiny pixels. What I need is then a huge bitmap font... does anybody have any?
I developed a font with a similar resolution that was not nearly as legible for my Pi Zero with an e-ink screen many years ago. It allowed for similar tooling such as the flipper zero and esp32 marauder projects. I should fire that project up and implement this font…
The 3x2 is fascinating, it's the same resolution as braille, albeit rotated 90 degrees. I wonder if this could become a braille-like system that's both visually and finger-readable.
Note: there are repeat glyphs here like c and o, though the example actually uses a different c somehow. But perhaps repeats are ok given context.
In the 2x2 resolution (which can't support all of Braille, but does support the first 10 characters), A, B, C, E, and F are all the recognizable Braille shapes for those letters (though in offset positions).
At first, it seemed like an Easter egg, but it's probably just a natural happenstance of two people centuries apart deciding to represent the first ten letters of the alphabet in a 2x2 grid with a general idea to use fewer dots at the start than at the end.
From using Okidata printers back in the day, a "tall" lower case g really does look bad in running text. (The similarity to the e is a little troublesome, but I don't have a fix for that...)
Here is how I would personally modify lowercase g:
xxx
x x
xxxxx
x
xxxx
It looks a bit like a 9 but the descender is different.
(Could play around with how many x's to put on the first and third lines, particularly whether to set the first/last bits on those lines as "corners" or make it more rounded.)
I would also modify the top of lowercase e in a similar fashion.
A little awkward, but very distinct from the 9. The reason to narrow the head of it is to allow the end of the descender to rise upwards without turning it into an 8.
One nice use for these tiny fonts is large text in terminals. Unicode now has 2x4 (from Kaypro), 2x3 (from Teletext, TRS-80), and 2x2 mosaic characters. Unicode also has 3x3 large text (from HP terminals) but font and terminal support is limited.
The 3x3 are the large text blocks - intended to use in 3x3 groups to form letters and numbers from the fragments in the range. There are 2x4 mosaics as well, separate from the Braille alphabet. These are symbols coming from legacy systems (such as the HP terminals and Kaypro CP/M machines).
Too bad "tiny screens" pretty much do not exist anymore. Screens with hundreds of pixels on each side are very cheap already.
It reminds me people who research "colorizing grayscale photos", which do not exist anymore either (if you want a color photo of someone you met in your life, there probably exists a color photo of that person).
> Too bad "tiny screens" pretty much do not exist anymore. Screens with hundreds of pixels on each side are very cheap already.
Find me a 0.66" OLED display for ~$1 that has hundreds of pixels on each side then.
> It reminds me people who research "colorizing grayscale photos", which do not exist anymore either (if you want a color photo of someone you met in your life, there probably exists a color photo of that person).
What train of thought led you to think people are primarily researching colorising new B&W photos? As opposed to historical ones, or those of relatives taken when they were young? You can take a colour photo of granddad today but most likely the photos of him in his 20s are all in black and white.
If you know a person who is 70 years old, they were 20 in 1975 - color photos existed back then.
Every grayscale photo of someone famous has already been colorized during the past 50 years. If there are only grayscale photos of you, you were probably born before 1900, and all your friends or your children (who might want to colorize your photo) are probably dead, too.
1. Improving the colourisation algorithms has value, it might be that the available colourised photos of celebrities have inaccurate colours or are of poorer quality than say, one done with a diffusion model that can be instructed about the colours of certain objects
2. Don’t forget about B&W films! Getting automatic methods to be consistent over a long length is still not 100% solved. People are very interested in seeing films from WW1 and WW2 in colour, for instance.
3. Plenty of people (myself included) have relatives in their 80s or 90s. Or maybe someone wants to see their ancestors from the 19th century in colour for whatever reason?
Color photos existed but color film and processing was very expensive (and while mono film development "middle school student can do at home" for a generation, home color work wasn't a thing until late 80s/early 90s as far as I recall.) So in practice, I personally have childhood pics of my dad with his mom and sister - that were shot black and white but colorized by being hand painted, and this was pretty common...
I wish that were the case. I'm trying to make a tiny emulated z80 computer, and to fit a screen of 64x16 test on a smartwatch sized screen, I have to use a 4x6 pixel font, because the highest res, most available screen I can get in that size is just 240x280. High-res 400+ px smartwatch screens like the apple watch has - you can only get those if you buy 10000 at once and sign an NDA.
Actually, the 4x6 doesn't look half bad if viewed at wrist-level.
128x64 monochrome screens are very common in both LCD and OLED format.
Quick browsing at adafruit.com (or any other similar vendor), reveals plenty of displays that are 128, 240, and 320 pixels wide. At 6 pixels of width per character, that's only 21, 40, and 53 characters wide. Seems quite useful to me.
There are also several 32x32 led panels, which one could imagine needing some text.
Also, this kind of thing is just interesting, regardless of the usefulness.
0.27mm dot pitch, so each letter would be 1.35mm square in a box of 1.62mm square. I expect I could read it just fine at the distances I'd expect to look at such a screen.
I tested this on a phone, and was able to read it without much difficulty at roughly 18-30 inches.
There exist plenty of reasons to colorize grayscale photos in 2026.
* a huge corpus of historical imagery
* cheaper grayscale cameras + post processing will surely enable all sorts of uses we haven't imagined yet.
* a lower power CCD and post-processing after the fact or on a different device allows for better power budget in cheap drones (etc).
* these algorithms can likely be tuned or used as a stepping stone for ones that convert non-visible wavelengths into color images.
And that's just off the top of my head as someone who doesn't really work with that stuff. I'm sure there are plenty of other reasons I can't think of.
Grayscale cameras are not that much cheaper than color cameras. And if you decided to use a grayscale camera on purpose, you probably do not care about the color information (which would be totally "made up" by the colorizing algorithm).
Also, if there are only grayscale photos of you, you were probably born before 1900, and all your friends or your children (who might want to colorize your photo) are probably dead, too.
What does the existence of a color photograph of my grandmother as an old woman have to do with my desire to colorize a grayscale photo of her as a child? Or colorize the photos of her wedding?
It's a very strange argument to make: there exist some photos therefore other photos may not be colorized!
I have not yet, because my uncle hasn't scanned those photos yet. I have colorized the pictures of my grandmother as a child, and some previously unmentioned ones of the farm my grandfather grew up on. I've also colorized some photos of ancestors that no one alive this century has ever met.
Just because you don't want to use a tool, it doesn't mean others also won't.
You can get nicer 5x5 fonts amd it was not that uncommon back in the day. 4 wide is not too bad if you make the center of M and W just two pixels inset from top or bottom respectively or borrow the spacing column.
Plenty of systems did it like CP/M on the Spectrum +3 and it looks pretty decent.
I think on my 6.8" WQHD+ 3200 x 1440 phone screen (70.5mm width), an actual 5x5 pixel square might be the size of a period → .
Doing the math, 70.5mm / 1440px = 0.05mm per pixel, x5 = 0.25mm wide. Just a guess though, I'm not exactly sure how big a period is without a microscope.
You could do a bit better with a 4x5 font for every characters except M, W, m, and w which would be 5x5 but use the pixels normaly used to separate them from the next character, so every caracters still use the same width.
A 3x5 font does not sacrifice the M or W. H,M,W end up as similar looking characters, but the M has the center pixel one higher, and the W has the center pixel one lower.
IIRC the really cheap Casio Organizers/DataBanks of 90's used 5x5 font. And then my ex used something like that on linux in order to fit a ridiculous amount of xterms onto 14" CRT (somewhat absurd feat with her congenital vision defect).
Many decades ago it was anti aliasing with MS clear type that made small text possible. Yet, before that, for regular TV, this had been worked out, so lots of text could be shown on things like business TV, in glorious analogue PAL or never twice same colour for Americans.
Small text is an interesting problem, but we have moved on from pixels as useful units.
I'm quite fond of Spleen:
https://github.com/fcambus/spleen
It has a 5x8 font which has all of ASCII, but most glyphs are actually 4x8 and include horizontal spacing. I modified it to reduce the rest for a project I'm doing so all glyphs are 4x8. The result can be rendered on a 5x9 grid with a guaranteed line of horizontal and vertical spacing between all glyphs. It's very nice.
Example: https://imgur.com/a/text-80-characters-per-line-240-pixels-w...
That's 3 horizontal pixels per character on average, including inter-character spacing.
However, 5x5 isn't enough to draw "e" properly if you also want lowercase letters to have less height than uppercase, so you need at least 6 vertical pixels. And then that isn't enough to draw any character with a descender properly, so you need at least 7 vertical pixels (technically you should have 8 in order to allow "g" and "y" to have a distinct horizontal descender while still sitting on the baseline, but this is probably an acceptable compromise). And remember that in practice this means you will still need at least 8x6 pixels to draw each character, to allow for a visible gap between letters below and beside them.
I think that's the least of the properties I'd be willing to sacrifice to have a font that tiny.
(but yeah, it's not quite right, and is especially jarring in the nice, clean, blown up pixels in the top example)
https://chinese.stackexchange.com/questions/16669/lowest-pix...
The extra 1 pixel of height for the text in green, in particular, allowed for some cool "italic" styling, especially for letters like E, D, J, U, V
Yowsa. For those playing at home, that monitor is over 20 years old:
https://everymac.com/monitors/apple/studio_cinema/specs/appl...
https://archive.org/details/zx_Tasword_2_Tutor_1983_Tasman_S...
EDIT: realized a better demo for this discussion would be the photos I took of all the tiny bitmap fonts I had converted to the C array style at the time. Thread has more but here are my favorites: https://bsky.app/profile/janiczek.cz/post/3mh25atboz224
but wowza!! $900 for that lil guy, dang!
Note: there are repeat glyphs here like c and o, though the example actually uses a different c somehow. But perhaps repeats are ok given context.
At first, it seemed like an Easter egg, but it's probably just a natural happenstance of two people centuries apart deciding to represent the first ten letters of the alphabet in a 2x2 grid with a general idea to use fewer dots at the start than at the end.
(Could play around with how many x's to put on the first and third lines, particularly whether to set the first/last bits on those lines as "corners" or make it more rounded.)
I would also modify the top of lowercase e in a similar fashion.
https://tonypai.itch.io/3x3-pixel-font
It reminds me people who research "colorizing grayscale photos", which do not exist anymore either (if you want a color photo of someone you met in your life, there probably exists a color photo of that person).
- https://github.com/akavel/clawtype#clawtype
- mandatory "Bad Apple" vid (not mine): https://youtu.be/v6HidvezKBI
(for the "splash screen" linked above I used font u8g2_font_3x5im_te: https://docs.rs/u8g2-fonts/latest/u8g2_fonts/fonts/struct.u8... and a multilingual u8g2_font_tiny5_t_all: https://docs.rs/u8g2-fonts/latest/u8g2_fonts/fonts/struct.u8...)
Find me a 0.66" OLED display for ~$1 that has hundreds of pixels on each side then.
> It reminds me people who research "colorizing grayscale photos", which do not exist anymore either (if you want a color photo of someone you met in your life, there probably exists a color photo of that person).
What train of thought led you to think people are primarily researching colorising new B&W photos? As opposed to historical ones, or those of relatives taken when they were young? You can take a colour photo of granddad today but most likely the photos of him in his 20s are all in black and white.
Every grayscale photo of someone famous has already been colorized during the past 50 years. If there are only grayscale photos of you, you were probably born before 1900, and all your friends or your children (who might want to colorize your photo) are probably dead, too.
2. Don’t forget about B&W films! Getting automatic methods to be consistent over a long length is still not 100% solved. People are very interested in seeing films from WW1 and WW2 in colour, for instance.
3. Plenty of people (myself included) have relatives in their 80s or 90s. Or maybe someone wants to see their ancestors from the 19th century in colour for whatever reason?
Bloody hell, warn people before you post things like that.
Actually, the 4x6 doesn't look half bad if viewed at wrist-level.
128x64 monochrome screens are very common in both LCD and OLED format.
There are also several 32x32 led panels, which one could imagine needing some text.
Also, this kind of thing is just interesting, regardless of the usefulness.
https://www.crystalfontz.com/product/cfal12856a00151b-128x56... - 128x56
https://www.crystalfontz.com/product/cfag12864u4nfi-128x64-t... - 128x64
There's a whole world of embedded devices with wide varieties of screen resolutions.
I think you will not be able to read 5x5 pixel letters on that display (a letter would be about 1 mm tall).
I tested this on a phone, and was able to read it without much difficulty at roughly 18-30 inches.
* a huge corpus of historical imagery
* cheaper grayscale cameras + post processing will surely enable all sorts of uses we haven't imagined yet.
* a lower power CCD and post-processing after the fact or on a different device allows for better power budget in cheap drones (etc).
* these algorithms can likely be tuned or used as a stepping stone for ones that convert non-visible wavelengths into color images.
And that's just off the top of my head as someone who doesn't really work with that stuff. I'm sure there are plenty of other reasons I can't think of.
Also, if there are only grayscale photos of you, you were probably born before 1900, and all your friends or your children (who might want to colorize your photo) are probably dead, too.
It's a very strange argument to make: there exist some photos therefore other photos may not be colorized!
Seeing that a "neat tool" exists and using that "neat tool" are two diffrent things. Google Glass was neat, too.
Just because you don't want to use a tool, it doesn't mean others also won't.
https://knowyourmeme.com/memes/colorized-history
Plenty of systems did it like CP/M on the Spectrum +3 and it looks pretty decent.
Doing the math, 70.5mm / 1440px = 0.05mm per pixel, x5 = 0.25mm wide. Just a guess though, I'm not exactly sure how big a period is without a microscope.
I would have loved to have seen a sample of the 4x5, not just the 5x5.
VIP Term, and others, had 3x7 fonts on 4x8 grids to connect to 80-column mainframes.
I haven't done the pixel-by-pixel deviation checking, but they may be comparable and independently derived!
I can't at all.
But I'm backfilling a lot of information from context, the same way that this works: https://www.dictionary.com/articles/typoglycemia
It would have been much harder to read a series of random words, or another piece of text with a less predictable structure.
Small text is an interesting problem, but we have moved on from pixels as useful units.
https://fontstruct.com/fontstructions/show/1656341/tom-thumb