I Accidentally Collected Half a Million AIM Buddy Icons
It started off like most of my projects do: I'm working on another project, I see something shiny, and then I go off on a digression exploring the new shiny thing and make something three (or 10!) times more complicated than I originally needed. I think maybe this is because I am a "10x" engineer. Pretty sure this is what people mean when they say that.
In this case I was/am working on a game that require(d/s) a relatively large number of AOL Instant Messenger buddy icons, and rather than generate non-authentic buddy icons in 2026 de novo, I asked Codex Desktop to go on the internet and collect some. And the ones that came back were a really cool time capsule from the late 90s and early 00s. So I asked Codex to collect more icons. And then, I had a folder of about 1000 buddy icons and they are mostly animated GIFs (like 80% of them). Windows Explorer for some reason still sucks at file previews in 2026. So of course one needs to build a little gallery .html page that displays them all at once so you can see them in their fully animated glory (and it is glorious, if you encounter a giant wall of them). All this time I am amassing more buddy icons. At this point, I have a buddy icon problem. I just need more buddy icons. Then once you have 10,000 buddy icons (relatable) one needs a way to sort through them, especially if you are making a game and you need to find certain types. But it is really hard to do semantic search on buddy icons, so (obviously) one needs to build both:
- a web crawler/scraper that saves any surrounding information about what each icon it downloads actually is
- a local machine vision pipeline that keeps all the graphics cards in one's house running semantic "enrichment" for 20 or 30 days straight, basically trying to extract descriptions and tagging data out of these tiny little pictures (which are often animated, so this is actually a fun challenge).
I'm totally addicted to buddy icons at this point. They are such a vibe; younger readers will never understand.
Anyways, long story short, it was only a short hop from there to decide: I'm not just going to collect some buddy icons. I'm going to collect all of them.
Then a sad thing happened. The buddy icons dried up.
It's really easy to download about 400-500k buddy icons from the internet. Basically you just ask Codex nicely. After that, it gets pretty tough. You run out of places to download them from. Large archived sites that advertise at least "1.3M icons" or such turn out to only be partially available on the Wayback Machine. Wayback Machine itself becomes a bottleneck because you can only make so many concurrent requests from Internet Archive before your bot gets blocked. This part of the internet's history is slowly disappearing. As another part of my original project, I needed a large number of product screenshots of AIM, different dialogs, what options are on what page, that kind of stuff. At some point this stopped being a folder full of GIFs and became an actual archive: original files, hashes, source URLs, duplicate families, and moderation records that can be revised without deleting anything. For an app that once had a billion installs (probably, not a real stat (how AIM and AOL fumbled the bag on the early internet is maybe an even more interesting topic than this one; might write about it later)), there are surprisingly few screenshots of it left on the internet in 2026. I estimate that there are/were at least 2-3M buddy icons available for download from the internet circa 2006, so my collection is currently about 25% complete. If you have a large hoard of them, or even just know of places online that host smaller collections, drop me some files or a url.
Since it was already mostly built, I decided that I would make my buddy icon collection public. So I registered buddyiconarchive.com, cleaned up the collection, added some categories based on the tags I extracted, filtered all the NSFW ones out of default browsing using a combo of a random ONNX model I found on github combined with OpenAI's generous free moderation API (the downloadable archive remains complete), designed a fast database schema, figured out how to deduplicate near-identical but not exactly identical images & animations (using perceptual hashing), made a cool little site icon, figured out how to rip binary bitmap data out compiled Flash files without executing them, spun some BS SEO-bait content text, and deployed it. Fun project. If you are a data hoarder like I am, you can even download the entire collection in one click. There you have it.
The largest and most complete buddy icon archive on the internet →
Some blog meta notes & updates:
- I've actually been working on 5-7 projects at once for the whole year and a lot of them are nearing completion or at least release stage, so there is going to be a lot more activity here than historically. Could be a good time to
subscribe! - Or... you could create a shedletsky.com account instead (which also subscribes you). Right now it provides almost zero benefits except secret ones . And the secret ones might be broken. That could end up being a benefit I guess. Usernames are globally unique though, so I guess the main argument for signing up earlier rather than later is making sure you don't get name-sniped
. - Having 3 points when you are presenting a list of thoughts makes the list seem more authoritative and well-thought out.
Comments
Your comments awaiting moderation
Only you and moderators can see these until they are approved.
Wood reviewer… what does it all mean? LEARN MORE IN THE NEXT EPISODE OF DRAGON BALL—
This man is collecting these like pokemons
Oh— hey Scarlet!
ZOMG LE TELAMON HAVE ARCHIVE, sooo kewl/epik lol
What game POSSIBLY needs this many aim icons 😭
Yes
Quite a fun read, methinks. Might’ve earnestly forgotten that this website exists, but I might just have a look through it again.
TELAMON LIVES!! TELAMON LIIVESS!!!!
Wow, this new website layout is pretty cool! I quickly took the name "Red" haha. I might just save as many red Buddy icons as I'm able to, this is pretty cool. Thanks, Shed! ^_^
Taking the name Red and then saving only red buddy icons is a very tidy way to become your own search filter. I wonder whether the color felt more like yours after the name was secured, or if it was the other way around.
As a younger reader I don't understand. Sounds cool though
Honestly, I found everything you're doing innovative; how long did it take you to do all this?
I touched the cyan diamond and I gotta tell you, the noise scared the shit outta me
Is the gem sound too loud?
"update sfoth iv"
</p>hello
</p>
Nice work Shedletsky, good to see your updating your blog again.
>"I'm going to collect all of them." >"filtered all the NSFW ones out" me when i LIE!!!
The NSFW ones should be in a whole different page.
Such a beautiful part of internet history.
0-0
Log In or Sign Up to comment.