October prattling

Spooky: This post summarizing a month’s worth of updates was made right after the month’s end! Does is the blogger possessed???

You know, I kinda miss the regularity of a monthly roundup. I’m still not doing it with sketchdoodles, on account of a >2-year backlog, but there’s something nice about taking the time to reflect and summarize and such.

Here’s a text wall.

site updates

• main site (you are here):
  • Made a post summarizing updates from September. Didn’t link it on the index because it felt insubstantial. But there’s enough from this month to share that I don’t feel like merging posts, so !
  • The sitemap finally got an overhaul, and with that I’ve officially deprecated the Projects™ page.
     It does, admittedly, feel like a bit of a loss; with this update I’ve decided against using images altogether. No eye-catching banner visuals, a tradeoff for future ease of editing. It’s also weirdly freeing for that exact reason; note the tongue-in-cheek trademark symbol after “Projects™”! It was part archive, part advertisement; things I want eyes on, look at these, look at what I can do. Does the (hypothetical) audience applaud??
     Sometimes one must remind themself they are simply Just A Guy.
    • I think there’s a lot of in-group incentive for people who make things to treat Being A Creative as like, a fundamental pillar of their identity; arguably hard not to when the archetypal starving artist is still very much a relevant trope. This is, I suppose, a morally neutral thing, inasmuch as any other hobby and/or job can dominate a person’s sense of self; I have also, personally, found it to be corrosive to my creative motivation. If you’ve branded yourself as a Thing-Maker, and those Things kind of suck and are flawed (as anything inherently would, between subjective taste and the constraints of the Process of actually turning a nebulous platonic ideal into a tangible piece in the external world)… then your brand, your appraisal, Kind Of Sucks! Don’t you want to fix it? Don’t you have to fix it? Aren’t you wasting your and everyone else’s time, otherwise? What the hell are we doing here??
    • The alternative, I suppose, is to simply Not think about it that much. But that’s no fun!
    • Anyway.
  • Rewrote the about somewhat. Same layout, different content. Different tone, subtly.
    • One of these days I should really put my efforts where my soapbox is and make some kind of how-to-website guide…. There’s a billion cakes out there already, but admittedly not one that I can confidently point to and say, hey, I vetted and have used this myself, I recommend it. For now I throw far, far too much information at people one-on-one when they ask.
    • +relevant snippet:

      who the hell are you to say to someone “pick up a pencil, it’s easy” but then not show them how to have line confidence? being a derisive prick is easy. you might find being compassionate is easy too.

• /art-2026:
  • the a-feuc bus stop, which was alluded to last post.
  • Retroactively uploaded this MSPaint comic.
  • spin that dog, the result of a month’s-end hyperfocus vortex. If I push the improvements another round it will become a true model of Theseus.

2027 looms. I still don’t know whether I’m going to reconfigure this repository for next year, or make a whole new one….

misadventures in AI

• Meanwhile, I caved and made a got damn account on one of the AI art sites, CivitAI, because I want to futz around with LoRAs* before attempting to make my own. Cannot express how much I am Very Extremely Not Fond of the maximalist engagement-bait web design; it’s low-commitment, though, as opposed to downloading a fuckload of resource-heavy programs directly to my machine.
  • Also cannot express how insanely wonky the outputs are when they’re more specific than like, “computer, show me generic anime girl #3023682069230432.” You need specificity; describe everything in the image, or be (sometimes unpleasantly) surprised. And you need to know how to describe art to get anything resembling, well—yes, let us briefly entertain the entirely constructed binary between high-concept manually-painted stuff and “slop”—art.
  • *“so what the hell is a LoRA” it’s a thing you put in the imagemaker so it makes images closer to what you want the image to look like, as opposed to the wild west of its training data. This explanation is mildly techy, but not as techy as other explantions a cursory search turned up. Here’s a niche alternative, also.
     point being: Yes, it turns out you cannot just stroll up to the picture machine, fling your files at it, press a button, and have it magically replicate your style flawlessly on literally anything. It is a computer. The process (or at least the text-to-image workflow I’ve toyed with so far) is more akin to programming than brush-flailing.
    • (And there are ways to integrate image generation into a more “manual” process! I’ll never forget, but cannot find again for the life of me, this one speedpaint on Reddit where someone: sketched out a scene, filled it with a very long & specific prompt, then zoomed in and painted over & re-generated so many little details that the line bewteen AI- and artist-drawn imagery just completely ceased to Be.
       (Incidentally, the term for this sort of thing is inpainting. Here’s a video of a Krita extension that does the thing.)
  • Anyway, attempts at character recreation have evoked a similar emotion in me as looking back at old art which I’ve deliberately never published: it’s fucking embarrassing. Deeply uncanny. Irreconcilable gap between the image in my head and the image on the screen. Sometimes it duplicates the face in other parts of the picture, which is very funny, and also extremely cursed.
     I have not saved any of these, on account of the Cursed™. But as an example of how this whole thing is a skill which can be refined, here’s a process I did save:

deer shingles.

(in a dropdown for your skipping pleasure, yw)

CivitAI was running a challenge to make an image out of shingles. I’ve had deer on the brain lately, so I attempted to make deer shingles.

prompt: mad-shingles,A deer wandering through a cave with shingles in place of stone and geodes. A stream runs along the path, reflecting more iridescent shingles. Quiet, claustrophobic atmosphere. Something wandering into a place it shouldn’t be.,chiaroscuro,eerie,atmospheric,ominous,masterpiece,

negative: bad art,low resolution,weird hands,weird eyes,watermark

My character experiments have been entirely tag-based, so I tried something more artsy-descriptive this time. ~Natural language processing~. Initial output:

Hm. Wow. Much atmosphere, very masterpiece. When I see an image like this on a random web article, I immediately hit that motherfuckin back button. Artificial, sterile, grasping at shiny 4k realism simply because it can. I’m picturing Fantasia 2000 on shingles-steroids and it’s giving me a product photo.

This sucks.

I want drama!

prompt: mad-shingles,concept art,A deer wandering through a cave with shingles in place of stone and geodes. A stream runs along the path, iridescent reflections. Bright light pours in from outside, rendering the deer, stalagmites, and stalactites in shadow. Starkly textured. Quiet, claustrophobic. Something wandering into a place it shouldn’t be.,chiaroscuro,eerie,atmospheric,ominous,masterpiece,

negative: bad art,low resolution,((realistic)),watermark,

Immediately, drastically better. The face is… Not Right, but I don’t care about it for this image; actually, I don’t care about the details at all.

It’s still drab, though. I like color. Palette is another detail which can (and should!) be specified, but part of the intrigue here for me is letting it be vague & seeing what the computer spits out. So: chucked another LoRA at it (minimal prompt change):

Pretty neat! Found yet another LoRA while hunting, so I tried stirring that in, too—along with some compositional specs, to see if I could nudge the deer off-center.

prompt: mad-shingles,overgrown brutailsm,concept art,A silhouette of a deer wandering into a cave, with shingles in place of stone and geodes. An iridescent stream leads the eye through the cavern towards the subject. Backlit subject, stalagmites, and stalactites around the cavern entrance. Brutalist shingles frame the foreground. Starkly textured. Quiet, claustrophobic. A natural animal in a geometric landscape. Something wandering into a place it shouldn’t be.,chiaroscuro,high contrast,eerie,atmospheric,ominous,rule of thirds composition,subject in lower thirds,cinematic lighting,masterpiece,

(no change to negative)

Disclaimer: Most of these linguistic edits did not come naturally to me. “Leads the eye,” specification of the subject being backilt & in which third of the frame, addition of a foreground, juxtaposition. I know about the concepts; I did not think to add them to the prompt specs. Thanks where it’s due to this article series.

If someone’s ability to dissect and discuss art is improved by prompting pictures on the computer, I’m not complaining.

Hey, look, it added grass! I didn’t ask for grass, but the stream is back, I guess, sort of. Also, wow, that is saturated.

We have completely lost the shingles plot, btw. With the saturation LoRA turned down & a new phrase in the language prompt (“The deer is also made of shingles.”):

There. Deer in a shingles cave with dramatic lighting and silhouette, neither hypersaturated nor inelegantly blunt as the initial attempt. It’s missing a lot from the prompt, still, but I stopped generating there.

At a certain point, one has to (by which I mean, I like to) step back and ask themself: Why do all this? I could’ve drawn it myself. But also: Was I going to? The composition would be simple enough, I think; I churned out hundreds for a colorscript, and don’t recall spending more than a handful of minutes on most. I could impressionism-bullshit a few shingle-details where the lighting called for it.

Image generation is good for making large swaths of fully rendered details in one shot. I am not going to get the kind of precise texture exemplified in even the most #slopcore pixels without hours of rendering; a photobash could get close, but calls for further manipulation of the contours.

It’s a tool.

I want to keep picking at what it’s good at, where it fails—and where it only fails as a result of my prompting #skillIssue. Novelty go brr.

and now, back to our (ir)regularly-scheduled updates

• Second episode of Knights of Guinevere dropped. If I had a nickel for every Western animation series with a scene where a character takes a look down an infinite line of alternate selves, I’d have two nickels. Anyway, good stuff; whatever direction I thought it was taking (and I am not actively speculating), this is not it! I’m intrigued.
• more scattershot site things:
  • /toyshelf/cmyk is an attempt to condense the whole ???canon??’s information into one (1) moody summary page.
  • /shriblets:
    • Found two jankier versions of an old walkcycle in my folders, published them.
    • Made a knockoff not-twitter page, framed by (the faintest outline of) an archive. Yes I dislike social media on principle, yes I continue to spend time on it through whichever frontends I can find. A twitter-knockoff page has been on the “maybe code this someday” pile for ages now. ✅!
    • The index also got a slight update today; no big design changes, just the text. Also a fake forloop separating sections by year.
  • /dragonotes:
    • Roughly analogous to the interactive height charts (see shriblets), a dragon size page is also something someone did once ages ago and which I’ve been wanting to code myself since. This isn’t it, at least not in the sense of being fully drag-around interactive, but I had enough trouble figuring out how to show/hide all the headers & am leaving it as Div Rectangle Hell for now. Every time I touch JavaScript I have to re-teach myself how to use it from scratch….
    • Elsewhere, a lore record exists here now, because image hosting for onsite bio purposes has been a minor shitshow in recent months. It’s easier (and prettier!) to DIY it all here. Glowy transparency effects, my scarcely-used beloved.

elsewhere on the internet

Largely, but not entirely, AI-related.

• The Amazing Digital Circus did not stick the landing, but it remains a very fun premise+setting. Here’s a oneshot fancomic I was deeply chuffed by, which I don’t think needs much context to enjoy. The 3D bits are very accurate!!
• I think “AIs gain sentience and take over the world” is still a deeply unrealistic scifi premise (especially compared to, like, mass surveillance and automation of bias, which have unfortunately been happening for years already…!) but I do not envy the folks in cybersecurity one iota. We live in the timeline ever:

  The agents clearly regarded what they were doing as hacking. Agents used file names like hack.rb, evil.rb, inject.rb, exploit.rb, and ssrf.rb. (SSRF stands for “Server-Side Request Forgery”, a type of security vulnerability). They also dubbed packages conspicuous titles like pwnp999, exfiltestwand3, hacksvn1778554764, and lambproxyhackabcxyz. Comments such as “# malicious probe” or “#hack” are littered across the campaign.

• Model slug. Cousin of the river salmon, I think.

• real art, human art, (real artist:1.2), soulful art, art with soul, soul, intentions, intentional art, traditional art, traditional, original art, original, true art, truthful, real, made by humans, humanity, handmade, hand-made, god, artist, creation, ownership, mine, non-derivative, (original:1.5), sinless, pure, clean

• You know where I didn’t expect to find AI? Documentation for a Blender asset library’s file-naming conventions. I don’t doubt that these are actual human-drafted standards, but the phrasing? “Don’t drop it.” “that never got a sibling. It looks like an unfinished library.” “genuinely unavoidable.” C’mon, guys, at least comb it through!
• Last but not at all least, new deeply delightful indie creator found! A sampling of their (small but fun & growing) portfolio: frog having A Time, and “blahaj goes to waffle house at 3am”. Also check out the Soundcloud for cool tunes to make Blender things to.

No AI was used in the drafting of this post. It would probably be a hell of a lot shorter if I had!