#behind the bot

[samplerbot]

The temporary pause is over! Samplerbot is up and samplering again, and all the requests I saw in the notes have been queued up along with its regular posts.

Today's update has a couple parts to it: first, four new borders have been added; second, I've made it easier (on my end) to test new borders, new fonts, and new styles of sampler in bulk, and to tweak all those things if they don't look right at first; and finally, Samplerbot has a new style of sampler: Lasagna Samplers!

The lasagna samplers are inspired by Elizabeth Holland's sampler from 1737 (Samplerbot's 'border 66' was adapted from her work, too!). Along with a regular border, lasagna samplers have decorative dividers in-between the lines of text (the lasagna 'noodles', you could call them).

I'm currently working on making it so every existing border could make a lasagna sampler, but since Samplerbot just passed 150 borders (!🥳) it'll probably take me a while.

Also, not new, but I'm very happy with how the new line spacing function is working. I think this next batch of samplers is going to be very lovely!

Thanks for reading, and thanks for checking out the bot! Hope you all have been having a lovely summer,

-Bot Dev

[behindthebot]

Theres currently a kickstarter going on to fund Haven Social, a social media site meant to be an alternative to everywhere else enshittifying everything! It's meant to be like a better youtube, with no AI crap or anything like that.

The campaign ends in less than a month and they're less than halfway through their goal, which is sad because it's really something we need right now! So if you can, please donate anything. Even a dollar makes a difference.I know lots of people who follow me are also children who don't have a bank account or financial independence, but then you could reblog and spread the word about it so it can maybe reach the fyp of someone who can donate!

Heres the link to the kickstarter: https://www.kickstarter.com/projects/havensocial/haven-stopping-ai-art-theft-and-surveillance#reward-UmV3YXJkLVVtVjNZWEprTFRFd09UVTRPRFkz

And heres a video that also teaches you more about it: https://www.youtube.com/watch?v=AOW9sGpip6M

[samplerbot]

(guy who's always messing with the line spacing function voice): so I was messing with the line spacing function again today

by which I mean, I gutted all the old code and made something new! Here are the goals I had:

Equalize the line lengths

Stop making 'scroll' samplers-- a 1:2 height:width ratio is okay, but a 1:4 is not

And here's what I came up with:

Once Samplerbot has the text it wants to samplerify, and has chosen its font and border, have it run the numbers for what the sampler size and line breakdown would be for a wide variety of potential line lengths. For each of those possibilities, Samplerbot calculates:

The standard deviation of the lengths of all the lines of text. This is the most important-- a small standard deviation in length means they're all close to the same size. Samplerbot starts by finding the fifteen line lengths that would lead to the smallest standard deviations, then runs the other calculations over those fifteen.

The ratio of total sampler height divided by width. It eliminates any options that are much wider than they are tall, and also vice versa, though it's less restrictive for height. Samplerbot then finds the absolute value of the difference between height/width and the golden ratio (and width/height and the golden ratio, whichever is smaller)-- another calculation where smaller is better.

Whether or not the given line length would create any single-word lines; if it would, it give that a value of '2', and if not, a '0'. Small is once again the goal here.

Samplerbot then finds the average of those three numbers-- standard deviation of the line lengths, difference from the golden ratio, and the single-word penalty-- and whichever line length has the lowest average is the one it goes with!

I've had it set up to make batches of ~50 test samplers at a time to see how they'll look, and I think it's a good improvement! Good enough that I considered figuring out how to have Samplerbot re-generate some of the ones already in its queue, but-- not this time!

The first samplers from this update will go live in late June.

Have a good night, all!

-Bot Dev

[samplerbot]

Samplerbot has been online for one year! 🥳

One year online and 150 samplers created-- wow! And many of them even look good! 😄

For Samplerbot's birthday, it has received one of its biggest updates yet: it can now do two new types of samplers!

I am calling them 'midpoint' and 'rooftop' samplers, and the ones it's been making I call 'loop' samplers.

From left to right: a loop sampler, a midpoint sampler, and a rooftop sampler.

A loop sampler (the OG type!) is made out of only two blocks: one for corners and one to tile around all four borders. The math for a loop sampler is simple: you take however wide the border block is, make sure the sampler's height and width are both multiples of that number, and then tile the border blocks and corners all the way around (plus some extra calculations in cases where the corner block isn't a square).

A midpoint sampler uses three blocks: a corner block, a border block, and a midpoint block, which is placed in the center of each of the four sides. Its math has a few extra conditions: along with leaving space for the midpoint blocks, it also needs to size the sampler so that there can be an equal number of border blocks on both sides of the midpoint.

A rooftop sampler uses seven blocks: it has one set for the 'roof' (a corner block, a border block, and a midpoint block), another for the 'floor' (corner, border, and midpoint), and a separate side block to tile the 'walls'. Rooftop samplers take a little longer to design than midpoint samplers, but their math isn't much different.

I think the midpoint borders in particular have the chance to be very, very beautiful-- I am excited to see them pop up as Samplerbot makes the next batches of samplers!

There are still many, many things to improve on with Samplerbot. Some of them are major changes (having it generate its own PDFs), some are small but finicky (text is still not consistently vertically centered), and some are pie-in-the-sky dreams (could it ever run online?). But whatever comes next-- I'm excited to see what Samplerbot looks like a year from now!

Thanks for checking out the bot, and here's to another year! 🥂

-Bot Dev

[samplerbot]

Mini update: I've done a bunch of things to the bot!

First, major changes to the pattern-sizing function to make it leave less white space between the text and the border. Not the first time I've messed with that function, and it won't be the last.

Got a couple cool glitches out of it, too! Third one is giving me Think Quick!-- would be neat to see a game like that in a sampler style.

I also discovered that at some point I managed to break samplerbot's font attribution-- looks like it was during the December 8th update. I've fixed that error, and manually gone in and corrected the attribution for everything in samplerbot's queue, but haven't touched existing posts. Anything between Dec. 8th and now, assume that the font label isn't accurate. I'll try and get those fixed as I have time.

Also made a few little tweaks and added two new borders. One of them's fairly decent and the other is dope as hell. Excited to make more!

Have a good Wednesday, tumblr!

-Bot Dev

[behindthebot]

"This Black Person is Not Real" by Pillboy on youtube.

This is such an important video to be circulating around right now. Its about how AI works to reinforce racism and perpetuates anti-blackness--- not just in obvious ways like if you generate rancid shit like people saying the n-word or saying we should go back to being slaves or whatever, but subtly reinforcing narratives we've already cemented in our head from previous years and years of propaganda.

You cannot be pro-AI and pro-black liberation. If you support AI, you are supporting a tool of reinforcing and uplifting systematic racism and a machine that justifies the type of violence we see being brought upon racial minorities, but especially black people (as focused on in the video) like with what ICE is doing in America right now.

[samplerbot]

Update notes: no longer just a pixelartbot-- Samplerbot is stitching!

This is a BIG update, both aesthetically and in terms of functionality: the new code makes it possible for samplerbot to convert one of its pixel art samplers into a numerical matrix, and then use that matrix to generate a mockup of what the sampler might look like stitched.

Also it can do colored fabrics now! And I fixed the antialiasing glitch and added about a dozen fonts.

The mockup generator was VERY fiddly to get right. The cliff's notes: it has three large fabric tile sprites (black, white, and colorizable) and six stitch tiles. Everything in the samplers is aligned on a 15x15 pixel grid, but the stitch tiles themselves are 19x19, with 2px of overlap on all sides-- which makes the stitches look compact and cozy when in a big block, but full and properly X-shaped when they're standalone.

(stole that idea from Cross Stitch Master, Jeff Tullin's old Amiga patternmaker from '92. good stuff!)

Each of the six stitches has three parts: a high-contrast, high-saturation base image, a black-and-white overlay image, and a special mask for the overlay. There's also a generic shadow image that goes on as a multiply layer underneath the stitches.

The base images are from a scanned physical pattern, and are recolored by the program to match any given DMC color. The recoloring washes out the contrast, though, so the overlay (with its mask) adds those highlights and deep shadows (and textures) back in. The effect is still artificial-looking, especially up close, but it's beautiful, and that's what counts. :)

A comparison between Samplerbot, WinStitch, PCStitch, and FlossCross's stitched preview options:

You can see that Jeff's fabric mockup is still leagues better than mine (and I'm cheating a bit by emulating a vague 28ct instead of Aida), and FlossCross's is too, but if you don't look too close I think Samplerbot's still looks super. Straight up, not to sound boastful, but-- I think it looks fantastic.

Aight, that's enough chatter. I'm so incredibly proud of this update. Hope you guys like the pretty pictures. :)

[samplerbot]

Update notes: another batch of pattern PDFs is up! Regular versions + Pattern Keeper ones.

I also fixed a math error in the DMC color-calculating function.

Actually, math moment incoming:

The bot gets its colors from the PyPalettes library, which is designed primarily for people who want to make good-looking scatterplots and bar graphs. It has a couple thousand color palettes in it, and the bot takes one at random and uses its colors when it makes a sampler.

To convert those colors into DMC ones, it takes each sampler color's RGB colors (like '0, 0, 0' for black, or '255, 255, 255' for white) and treats each as essentially a 3D coordinate-- like with 'x, y, z' coordinates. Then it checks to find which of the existing DMC threads has color coordinates that are closest to it in that 3D space using the distance formula:

√((x₂-x₁)² + (y₂-y₁)² + (z₂-z₁)²)

(if you've made it to high school math, you should recognize this! it's the same as the regular 2D distance formula, but with three variables).

(also the bot doesn't actually bother with the square root, since it doesn't care what the distance is, just which possible distance is smallest).

Anyway I messed it up the first time, and had it cube the binomials instead of square them, which led to some kinda wack negative stuff happening. The effect on the samplers wasn't noticeable, except that some of the samplers had less-harmonious color palettes than they might have.

So-- the samplers might be a little prettier from here on out! Or possibly similarly-pretty, but the math that makes them is nicer, at least. :)

I have some big ideas for things I'd like to add next, but life keeps me busy~

And as always, feel free to @ the bot if you see any cool posts!