"Uniqlo x Akamai sells another design of shirt in the same range which is plainly incomplete"
Imagine having to return a t-shirt because that malfunction!
— I don't understand why are you returning this, was the size wrong or you didn't like it?
— No, there is a syntax error at line 37 that makes it impossible to run, and I'm concerned people on the street may think I promote unsafe bash scripting.
ATTN, Uniqlo customer service employees: If someone tries to return a shirt claiming an error in the printed computer code, decline the request and reply, "A real script kiddie would've spotted that error before buying."
There's something else a lot stranger going on, though. It is a proper monospace font, but the typesetting on the shirt is not. There's some kerning going on (I noticed it especially in the 'Iy' pair), and also it appears that narrower characters such as 'i' take less horizontal space. If I had to guess, I would say that it was set with a tool such as "optical kerning" in InDesign.
Has anyone ever made a monospace font with dynamic kerning? Which is a silly thing I never thought of until I read the above comment. This sounds nonsensical at first glance(and it may be) but hear me out.
We use monospace fonts for a reason, they stack in a grid nicely. But within the confines of that grid there is room to shift a character left or right a bit which may lead to a nicer to read monospace. (it is equally likely to lead to a hideous mess, every time a letter would shift left it would leave a larger space right)
Monaspace has a feature called "texture healing" that does something similar: it allows bigger letters to "steal" space from adjascent smaller letters, to make it easier to read. The result is that the letters are still in a grid, while still allowing for bigger letters to "breathe".
I'm open to trying it, but my gosh, having one 'w' offset slightly by a few pixels from the one right above it feels like it would drive me bonkers eventually.
And: doesn't this result in text that "jumps around" as you type?
iA Writer has Mono, Duo and Quattro fonts, with the latter two being almost monospaced. They concede some size variations for specific characters (Duo has 150% width characters, Quattro also uses 50% and 75% for narrow characters).
It's a fun subtle adaptation to keep close to the typewriter-like experience of the app.
Shifting a letter left or right a bit can break the grid. What if the user writes the text that keeps triggering left shifts? A better solution is to use ligatures, so that specific character combinations look better while the ligature can maintain the overall width correctly.
I love this shirt! Here's a nice video from the actual designer about the process of making this shirt (including intentionally making it hard to OCR): https://youtu.be/jocGLiecpjU?t=526
Author here. Thank you so much for the link which I hadn't seen! I'm very happy to see this and I'm gratified that it was deliberately difficult to OCR, not just me.
But it's not hard to OCR? And I don't know why the article dedicates an entire section to it.
On a Samsung S24U I held down the "circle to search" homescreen button which brings up the AI tools interface (I don't know what it's officially called), held down on the text and copied the whole thing in one shot.
OCRing this is a nightmare and is a good benchmark to any self-proclaimed good OCR/vision model.
I think though it could likely be easily OCR'd if you give the image to any decent agentic harness with a good vision model, e.g. newest Claude/GPT ones, and tell them to split the image per lines, and then just OCR each line individually.
I wonder if the script itself was written by an LLM before obfuscation? There seem to be a lot of comments in it, but in this case it's still ok :)
> OCRing this is a nightmare and is a good benchmark to any self-proclaimed good OCR/vision model.
It's not that difficult, our industrial OCR model read it correctly on its first attempt with default parameters. The characters are easily separable, there is no structured background (think expiration dates on yogurt aluminum lids) that confuses the reader, there is no almost-text-like texture anywhere that would clutter the result. The font is also nice and standard.
I don't think it was written by an LLM, some things stand out:
The congratulations text is both in English and Japanese. Contains a single heart emoji.
There was an intention to have a cyan to orange gradient, but the range starts in an ANSI block, ends halfway through the 256 color block and 256 terminal colors are not arranged like a gradient at all.
There's no sleep at the end of the loop where I feel like an LLM would add that defensively.
The last time Internet people were obsessed with OCRing some base64 was a few months ago when the DoJ released tons of emails from some guy who died, but they were released as rasterized PDFs.
Can't remember his name now, there's been so many distractions...
Safari's copy-text-from-image feature manages the entire base64 part of the string, except for the first character (I instead of a T). Weirdly, it gets much worse performance if you try to copy the entire string, including the hashbang part.
I wonder what it's doing under the hood to get such good performance?
I gave the photo to Opus 4.8 and it reconstructed the same script in one shot. Although it did say it had to correct some parts of it based on context where it suspected OCR mistakes.
> I wonder if the script itself was written by an LLM before obfuscation?
From the prototype shown here [0], and the way they talk about their process, I sincerely doubt it. Especially as they mention trying to make it hard for AI to handle the output.
I watched that whole video link - thank you for that - and he doesn't really say. In fact, he spends much more time on the beige color harkening to computer case plastics of the 80s & 90s.
The AI not handling the output relates to the final base64 output on the T-shirt (which other comments in this thread mention manually keying in or TFA discusses in the context of OCR). So, that is just not relevant to the question.
What made me start to wonder, personally, was that the output seems identical if you use "♥PEACE♥FOR♥ALL" instead of the version with internal repeats. IF there is any point to that "manual expansion of the cycles", IMO that deserves a comment much more so than "# Calculate length of text; text_length=".
Also, that `echo -n ...` followed by `echo ""` instead of just plain `echo` in the first place seems like the kind of copy-pasta code LLMs generate. Then again, regular devs also write pretty bad copy-pasta code.
There is also this the weirdly "broken down" calculation with 3 `bc` invocations not 1 as if it was translated from a language with more arithmetic/special function power than bash.
There is also the color scale stuff done in the loop instead of outside (except the one color=$(..)) which seems very unnatural and also very like machine translation.
Also, at least for me, on my bash-5.3.15(1), `char="${text:t % text_length:1}"` does not work to slice out the multi-byte UTF8 heart symbols, but it sure does look like the kind of thing an LLM would do translating from a python3 script (such as something like https://news.ycombinator.com/item?id=48830669) into bash.
Another thing is, as others here have observed, there is nothing "gradual" about the xterm-256 color cube. So, "gradient" is a misnomer and exactly the kind of weird things LLMs do when they cobble text together.
Finally, all the tput stuff the script does instead of just "print x spaces" really smells like a human description of the side scroll in the video game graphic he shows inspired him somehow LLM-corrupted/complexified into the vertical scroll terminals do.
None of this is conclusive, but the video mentions 2023..2025 as when he did it and given that he was a designer and his concerns more visual than code-oriented, I'd have to say I disagree with your sincere doubt and I do strongly suspect the decoded script was very likely LLM-circa2024-generated, possibly with light post-edits by hand.
> All the smells you pointed out, just look like a Python dev approaching bash without fully understanding it.
also, referring to Linux as "the language of the internet" when bash isn't particularly suited for internet tasks also smell like "excited windows Python dev"...
FWIW, his screens looked a lot like OSX to me (which tracks with graphic design users in my experience).
Anyway, he seems like a very nice fellow and I wish him and almost all T-shirt designers well. That bash script just gave me a lot of pause. (And even that seems possibly downstream of him being nice and doing it himself to spare his team from what he called a "FrankenProject".)
Yeah. The Flask web-page prototype was indeed in Python. (The prequel shirt was Go.)
{ Also, it was my own Py version which I mostly did in case anyone wanted to actually run the thing after such interest was expressed on this thread. :-) }
I already said regular devs and LLMs can both gen copy-pasta. That said, being "mostly" a Python dev, asking some LLM to translate to bash for him seems even more likely to me. Only he or those close to him knows for sure. You & I cannot settle it here conclusively (as also said).
I also noted from the video that the ♥s (hearts) worked on whatever version of bash he tested with though it failed for me (which is why I wrote that Python). And his terminal title bar is switching between `tput` and `bc` and such meaning that what he was demoing was not some Python script. ¯\_(ツ)_/¯
EDIT: Ah..another resolution of the hearts is to not run in an LC_ALL=C environment. Oops! `LC_ALL=en_US.UTF-8 bash ..` fixed it. Oh well, I think the Python script is nicer in almost every way. E.g., you could |head -n60 and send it to a line printer/dot matrix reminiscent of the 1980s computers he shows in the video, although your printer driver would have to strip the color escapes with a `sed` or maybe https://github.com/c-blake/bu/blob/main/noc.nim. ;-)
Those of us who grew up in the 8-bit era would have just typed it in, carefully, in silence, with no-one allowed to enter or leave the room until we were done ;-)
>I wonder if the script itself was written by an LLM before obfuscation?
I seem to recall seeing an Akamai-branded base64'd shell script on a white shirt pre-2021(?), so unless they've changed the code since then, I doubt it...
My code usually clocs at 50/50 (or thereabouts)[0]. Has, since my very first real engineering project (in 1987)[1]. I discuss in detail, here[2].
But one reason that I like LLMs, is that they help me to write even more documentation. I have found that I can instruct an LLM to revise my documentation, and make it even more effective.
[0] https://github.com/ChrisMarshallNY (My GH profile. Pretty much everything there, is like that -has, since long before LLMs were a broken rubber on the drug store shelf).
I didn't downvote. There's nothing wrong with your comment. It's just a bit silly, because humans definitely write that much. I learned from Apple and Adobe headerdocs.
I worked for a Japanese company that accreted comments.
Human could write that many comments to get enough base64 text for a design. Maybe to even get some of the highlighted characters in places they want (roughly equally spaced apart).
Especially in a case like this, I would definitely write a lot of comments to aid in understanding, thus increasing trust so people would try it out and tinker with it.
Honestly it's a bit of a shame. I checked and they could've shortened their base64 payload by 304 chars by removing all comments except the top two congratulatory ones, or by 524 if they removed those too.
Would they still get the highlighted "PEACE FOR ALL" text throughout the shortened string? It looks like the length, and presence of those characters, was an explicit design choice.
I thought it was funny that the author used a variety of OCR tools with mixed success before spending a lot of time manually fixing up the output from the best one, rather than just typing it in
That was also my thought… but I grew up mashing rubber keys for hours copying “games” out of magazines and books! Then hours after fixing all the typos!
I ran it through paddle paddle OCR and it flawlessly did it.
Google's OCR through my phone's Google lens had also worked at getting a very good extraction but not 100% correct. Definitely would spend less time fixing it than hand copying.
IDK what the author was using but I feel like he could have shared how his OCR attempt went, but I am thinking he tried some naive OCR tools.
Author here - that's a good idea actually, it shouldn't be too hard to compare the various attempts. The tools I used were whatever my Android built-in is (likely Google Gemini, but I can't tell whether this is something Samsung has replaced in OneUI); tesseract; tesseract with various tweaks and charsrt restrictions; Claude; and finally, manual fixes based on disagreements between all the previous.
Took me almost 2 minutes for 4 lines (and I missed a character in one of them!).
I would opt for OCR too, obviously so I'm prepared for the next bash t-shirt I'd come across...
I think this is a case where two people can successfully complete the task manually faster than one attempting to automate it. Get a ruler, read five centimetres of characters to your colleague, have them type it in as you go, then repeat that five centimetres back to you. Correct as you go. Format your string with the same line-breaks as the t-shirt, and remove them at the end, so you can be sure you've got the correct length on each row. Trial-and-error adjust the five-cm distance depending on your success rate as you go along
All in, you should have a non-corrupted string in 10-15 min.
Feels like my experienced reality of task automation in corporate environments. We routinely have engineers spend 40+ hours automating tasks that an entry level person can do manually in 10 minutes and only need to be done weekly. Automation at all costs seems to be the future
It’s certainly not a new phenomenon. I appreciate this XKCD [1], with a chart of “How long can you work on making a routine task more efficient before you’re spending more time than you save”
It’s not the final word, since automation has other benefits: documenting the procedure’s steps, reducing human errors, increasing consistency, etc.
Gemini3.5 Flash didn't have a problem OCR'ing and base64 decoding it, despite the OCR step having errors, it just fixed them in the base64 decoding step.
Not really. Transcribing long sequences of nonsense is annoying but quite easy to do without error as long you're patient enough to follow a simple process of reading, typing, and double-checking character-by-character.
Basically it just clusters same characters and asks the human to find the problems, which is easy when you're looking at a series of pictures like ssssss5sss.
The UI is kinda least-effort. Should ask a modern AI agent to make it look nice and intuitive, sometime maybe.
while base64 can be considered obfuscation in this context and its inverse as decoding I can't help but feel this title is overselling and catering to a rather cyber-cheesy marketing campaign at that.
Yeah, its a bit of a cheat. The best obfuscated C programs have the source looking like a Christmas tree (or something) and then play an xmas song (or whatever)
the base64 thing they did feels like a cheap version of that green-obscure-symbols-raining-on-a-terminal animation in The Matrix. should have gone with "Hack the Planet" instead ...
That Matrix visual was actually specifically mentioned as an inspiration in the video by the designer being linked to/discussed elsethread (e.g. https://news.ycombinator.com/item?id=48830326 )
For a non English locale that use comma instead of dot for decimals (in my case, Spanish), this script is partially crashing. Run using something like `chmod +x shirt.sh; LC_NUMERIC=C ./shirt.sh`.
Super cool, especially that the code is annotated!
In case the author is reading: The decorative feather images are between 2MB to almost 5MB in size. Compression might be in order to save users time and bandwidth, and make the site look less broken while the images are partially loaded :)
For anyone that cares, this is a slightly less stupid Python version:
#!/usr/bin/env python3
from os import environ; E = environ.get
from math import sin
from time import sleep
text = "♥PEACE♥FOR♥ALL" # The text to sine-scroll animate
nText = len(text) # Number of utf8 chars
freq = 0.2 # Frequency scaling factor
color0 = 12 # xt256 Color cube segment 12..<208
color1 = 208; nColor = color1 - color0
(w, h) = (int(E("COLUMNS", 80)), int(E("LINES", 24)))
t = 0
while True:
x = (w/2) + (w/4)*sin(t*freq) # x pos via sine value
x = max(0, min(w - 1, int(x + 0.5))) # bound to tty width
color = color0 + ((nColor*t)//h)%nColor # cycle colors
ch = text[t%nText] # Get char & Use xterm-256 color escs
print("%*s\033[38;5;%sm%s\033[m\n" % (x, "", color, ch))
t += 1
sleep(0.1) # original used bc shell outs to rate-limit
Great post! It's interesting, detailed but concise, and well-written. Also, I appreciate the "no cookies or tracking" and attractive, functional and performant site design.
> I guess Uniqlo is run through Windows though: one thing that struck me was the font, which I’m almost certain is Consolas,
Surely this would use whatever font the virtual terminal profile was set to? I don’t know of any method to choose a virtual terminal font from bash and don’t see any code that addresses it?
Ah, I see. I’m sure the clothing designer that actually made the design couldn’t care less about technical consistency and was just looking for something ‘tech’ looking that also read well in that design context in that medium.
Author here. All hallucinations are my own. Now you point it out, I see why the jump in context from the terminal back to the tshirt font would give the wrong impression.
Honestly it was quite a whiplash to go from what looked like a good article to something that seemed completely made up. But I would chalk that up more to my reading comprehension than your writing.
On one hand it's nice how it's clean and commented, but on the other hand some golfing could have made the encoded block a lot more reasonable to actually manually enter.
They should have just had the base64 block and forced you to decode and read it before running it, rather then having the `eval` bit at the beginning...
But that wouldn't have looked like a bash script, only a random sequence of characters. The shebang at the start definitely contributes to the geek factor.
Might have to do something like that for a verse on the next Carolina Code Conference shirt. Been trying to figure out a good way to pull in cybersecurity.
and use Consolas on Linux, but it's not available by default.
What they're suggesting is that a lot of their users are logging into an Akamai Linux box from a Windows machine, and therefore aren't "real Linux geeks".
Cool! I bought one a few months ago as soon as I spotted it at a Uniqlo store, and later ordered a larger size online—I really love wearing them. But it never occurred to me to look into the story behind them.
It’s encoded not obfuscated. It’s even commented, which is the opposite of obfuscated. Plus it’s not really an Easter egg that was found: it is literally printed in a shirt. Easter eggs are supposed to be hidden and either only found by insider knowledge or deep investigation. This was neither.
> I ran OCR in a few ways: First, using the built-in OCR of the circle-to-search feature on Android, which is often very good. Second, by using Tesseract with a few options and tweaks. And third by running it through Claude. After diffing the three to look for mismatches and getting Claude to output a table of locations for quick scanning, it became trivial but time-consuimg to tidy up the remainder
I bet 10$ I'd spend less time typing it from the t-shirt. And I wouldn't boil two kettles of water in the process.
But hey, AI makes you 10x more productive, I suppose
(Author here) for unrelated reasons my typing is very slow at the moment, so I was keen to automate. I see that people are getting different results from Claude than I did though.
P
./cool.sh: line 31: bc: command not found
./cool.sh: line 34: bc: command not found
./cool.sh: line 37: bc: command not found
E
./cool.sh: line 31: bc: command not found
./cool.sh: line 34: bc: command not found
./cool.sh: line 37: bc: command not found