login-required

[login-required]

Self-referential.

Got a hold of a cheap hand microscope and predictably spent several days scanning stuff. Only now did it occur to me that I can plug it to a computer and not just look at things on a big screen, but also record stills and video. I inaugurated this by recording a couple words in the manual.

[login-required]

This Shire weed ain't shit.

They don't call them High Elves for nothing.

Warning for methed up elves and slutty elves.

So I had silly thoughts about an Elf becoming overwhelmed by the passage of time and seeking solace in drugs. Of course being an Elf, it would also be a cultured individual surrounded by works of art.

My first genAI attempt specified an elvish hierophant and a crack pipe. I specified art-déco surroundings to match the entre-deux-guerres themes of drug use and the last stages of imperial sensual decadence.

Ah. This looks too much like a decadent lord, with all that gold, the subtle ring, and the laying-on-ground. It did get the pre WW2 languid eroticism just right though -- and it rendered my requested plate of Elvish bread (you know, for the munchies) pretty damn well! Let's make him more of a hierophant though (perhaps asking for sensuality was a mistake?)... Needs to be a bit older / unkempt, not very well slept, a not too venerable beard, and some scholarly items such as, y'know, books, ambiguous trinkets and (OF COURSE) unidentifiable clutter.

OK, that's definitely not Shire weed. Morgulmeth, not even once.

Let's try modifying the initial image again, but lose the haggardness and go for more of a frat elf / gym goblin vibe.

Ha! You add muscle tone and it immediately starts showing more skin. Quite something. Surprised that this isn't getting blocked as 'too risqué'.

To be fair, a couple other versions that came out alongside the terminally methed up scholar look a bit better, still ever so slightly erotic but not too appealing to the necrophiliacs:

These are old men one could shag. Still not quite my original vision. I need that elf to look more tragic, languid and pretty. Perhaps have him lean forward and let a solitary lock hang off his forehead almost brushing the pipe. Perhaps need to provide some mood indicators. And what if I was to change the drug description, would this carry some implicit associations that come through in the character's attitude? Ummm...

Ok. Let's start off slow. How about I ask for opium and some chinoiserie.

Nice and sensual, great. The markings under the eyes which are a mix of bags, after cry blush, and edgy tattoos definitely sell it. The yonic jewel on the forehead is odd but fits the environment (probablly a degeneration of the previous iterations' crown/diadem). How about we make amp the sensuality and add some fourth wall breaking -- looking at the observer as if trying to seduce them?

Oh.

Alright, can we simply ask "stop short of breaking ToS and don't show TOO much skin"?

Looks like it!

At this point I came up against a bit of a Pareto front of melancholy/sadness and sensuality, whereby it was hard to push much farther in one direction without losing the other. But along the way, before I realised that this was the case, some interesting stuff happened. That's for another post...

[login-required]

Vaporwave Must Die

(Not really, I am quite fond of the thing.)

A while back I read this terrible, terrible book, "The death of truth" by some NYT high-ranking goon. I found a copy spanking new in a discount shop and the binding was lovely, so I couldn't resist; luckily it was a short read heavy with references but light in ideas. It was basically a case study of Trump Derangement Syndrome complete with plenty of Orwell references, focussing on a 'takedown' of evil Russian/commie/postmodern epistemology (no no guys and gals, you don't get it, it's not a repackaging of antisemitic ramblings around cultural Marxism I swear). Of course, it sidestepped even the most basic "mistakes were made" acknowledgement of complicity of the Western press in the criminal invasion of Iraq under false pretenses -- nor in any other fault of the "free world" for that matter. Overall, an excellent bit of circumstantial evidence on how the "liberal" establishment ditched and alienated its constituencies and prepared things for DJ Trump's ascension and the descent of Clinton and Obama adjacent criminals into irrelevance.

(note two details in page above: one, just how little text fits in one page; two, the absence of any allusion to any role the press may have had in selling the Halliburton Bush Jr administration's war)

There was, however, a neat aesthetic detail: a reference to the drawing "The death of truth" -- showing a corpse which I would've sworn was supposed to be infested with crawling snakes. In reality if you look at the Goya (presented below, extremely hi res available here) snakes are not particularly prominent. And sure enough Kakutani does not mention snakes. Talk about post-truth!

Anyhow I liked the idea of snakes on a corpse. And since I was playing around with death of vaporwave imagery / "vaporwave is dead" music, the combination came naturally.

What's more, you have some very much not hypothetical classical statue with snakes - and this one has been replicated quite a few times! You could say it's been done to death:

Throw lots of additional references into the DALL-E sink and you get...

[login-required]

Quest for the best boy. (3/?)

You're a carnivore because you like sheep. I'm a carnivore because I dislike cabbage. We are not the same.

I a couple prior posts (1, 2), we tweaked our dapper werewolf to become more expressive, getting particularly good with Sora and ChatGPT. Now it's time to play with preset styles.

Our last output was a group of blep-facing execs. While I do love the pose and how we are left on a knife's-edge between adorable derpiness and amicable animality, we've departed from our logo intentions as we ended up with too much detail.

First, let's specify that we want no more than a single tone in the fur.

I also want to check if we can have a tradie. We are designing a factory, not an open plan office with pool tables and after work cocaine parties! How about we substitute "bumpcap and work clothes" for the suit:

...I'm actually liking the attire and how Sora tries (and fails imo) to resolve the tension between the bumpcap and work clothes, and the corporate side, by seemingly generating a suit but changing the cuffs and breast pocket. Also interesting that it seems to tip the background concept over, to once more show factory imagery. Can I make it more logo like though ? Let me include an ask for a "stencil or vector style in flat tones" and set the aspect ratio to 1:1.

Um! I'm not convinced. The description got long enough that the "no text" instruction is not reliably followed, and the stack snout-goggles-cap feels awkward. Maybe I just got the wrong PPE!

Let me experiment a bit with styles now. I'll revert to my exec description and simply change the pre-set. I loved the Wombo style options (especially playing with the festive specials), perhaps we'll get some wacky mix here.

Below are (each row respectively from the top) the results for "cartoonify", "archival", "SuperBowl commercial", and "whimsical stop motion" -- they don't look that much different (though the stop motion ones look reliably goofier), perhaps I've provided too much artistic direction and it's trumping the pre-set?

Some other presets do have a clear effect. Here are "cardboard & papercraft", "pixel art", and (somewhat cursed) "balloon world".

And then there's... Noir.

[login-required]

Adding concrete and colorful ceramic tiles to the sketty portfolio. with some prodding about the author (mentioning an anonymous soviet workers' collective worked great) you start to get striking results.

[login-required]

2KETTY: A Bridge Into Meat(ball)space

We have previously shown how DALL-E, presumably thanks to a training set with plenty of references for artwork and textures and media, gives us interesting compositions when given some 'vibe' artistic direction and allowed to get on with it. I used this to some effect in the Reproach of Hunger / Sketty series.

I recently came across an artwork in the real world, made of real materials by real hands, that conveyed many times the pathos and hunger I could manage with my clumps of pixels. This image is not super easy to find on the internet even if you search it by name.

I first asked ChatGPT/Sora to make me a version of the sculpture, using my usual Sketty series mix of media and style. I ran into some problems: one, it was clear that ChatGPT did not know the artwork even when allowed to perform a search, and so it filled in the blanks with a mediocre guess. Two, limitations are being enforced on imitating styles, so the poor chatbot was tying itself up in knots saying "this image goes against ToS" -- "apologies, your generic description of a style to imitate does not go against ToS after all".

Overall I was not too impressed. If you can't tell, the dot pattern that looks like kitchen roll is the result of my attempt to inject some "by Yayoi Kusama" into the style.

Here one can clearly see the evolution of my prompt. First I had to ask for no text, specify that my spaghetti meatballs be not sculpted, and tweak the pose... while making the beggar less fit and youthful. Then I needed some tweaks to the hood. By the last iteration, the Kusama-inspired pattern had almost been forgotten / drowned out by the other instructions. Also for some reason the shawl blends into pasta tendrils, which I don't really mind.

I figured that either I provide an excruciatingly detailed description of the original work myself, or find another way. Back when I started messing around with GenAI, I had a go at inputting images as a starting point. I had a brief play mostly with taking a relative's drawings and changing the style (problem: cherubs get snagged in NSFW filters because, y'know, cherub genitals), and some "LinkedIn profile pic into fursona" funposting. Presumably by now, those tools are much advanced.

The day after, I returned to the museum, took a few snaps of the wood carving from different angles, and gave Sora's "Remix" tool a go.

Not bad at all. We get a pleasant mix of types of marble and gold, the hood is doing more or less what it's supposed to do, and the plentiful spaghetti meatballs are a treat. I had specifically asked for tomato stains as a "blood in the hands of inequality" type marker, and it really shows!

Something that bothers me though, is the hood and hands. Those aren't quite imitating the original. Let me add some text in the remix prompt, specifying that

[the beggar] covers her head and face with a large shawl, which droops way over her face. She has her sinewy hands outstretched. The right hand is closed into a loose fist, palm down, emphasising the knuckles, and rests on the upward-facing palm of the left hand.

Eh well, the shawl is definitely forming more of a hood now, but the hands aren't quite there. Perhaps I should ask for a 3/4 view to give it a better chance? I did provide some reference pictures taken from the side, after all.

Points for effort. We are seeing some more variation of the hand position, and specifically the fourth picture is starting to get close -- together with some exquisite tomato splatter.

So what can I do next? It feels like I'm starting to run into the limitations of free tools hosted by third parties. Perhaps I need to acquire a more systematic understanding of their inner workings. Here are some things I would like to see:

Use of references to be followed closely, including local/detailed references that the tool can trace back to the 'big picture'. (Autodesk did this a decade ago, I've built 3D models of objects several m long from pictures taken on my phone!).

A way of surgically addressing key details such as the length of the hood, the mix of materials (which sometimes is a bit awkward - note in the last picture how both hands and shirtsleeves are black marble).

Directing proportions: spaghet thickness isn't quite right for the scale of the original work.

Assistants to acknowledge when they 'know' something with high confidence, and when they are filling the gaps.

A reliable way of capturing style (especially without randomly getting blocked-or-not over plagiarism concerns) that won't be slowly forgot as you tweak the image.

Capturing and saving striking aesthetic elements, such as the tomato-on-marble mess and the shawl-turns-into-pasta, for use on other pieces.

NUKING the male gaze -- the default woman is still a young smooth-skinned madonna whose attire allows breasts and hips to be noticed.

...And once I've mastered these, I'll finally be able to make a living producing furry porn.

[login-required]

[login-required]

[login-required]

Quest for the best boy. (2/?)

"Why, a bare bones proposal? You've got my attention."

In a previous post we managed to get a Gendo-ing corporate werewolf, conceiving future factories in his VR mind palace. Now we will have a go at spicing up the style with more instructions.

So our starting point is a dapper wolf man, stylised (ish), in three corporate colours (four/five if you count shades of grey and background colours), with some corpo "factory of the future" imagery behind. (Below is the version I got through Copilot - which surprisingly, after some instructions to drop unnecessary text, managed to not shit the bed this time... just don't look at the fingers for too long)

I would like to make this gentlewolf a bit wilder and hungrier, both to convey high-octane-exec vibes, and perhaps to ever so lightly lean into furry homoerotic associations. Let me find some references:

sorry, what??:

And of course the OG:

So my updated design bill will include something like, (1) making the werewolf's fur shaggier and rendering it in swirly twisted strokes; (2) adding some chop-licking and salivation; (3) perhaps makig the claws gnarlier. I wanna remind the assistant to stay in-palette, and specify the allowable deviations. Maybe I want the tongue to be dark red to match the other corpo tones, and allow a couple more shades of grey in the fur to help the swirly strokes pop and make it look even shaggier.

Copilot takes a decent shot at altering the image presented before, though not shaggy enough for my taste and more of a Chinese lion. For some reason, if I ask for alterations the suit tends to also be affected.

ChatGPT does a slightly surprising take on the Gendo we presented in the last post, with a pose change that better accommodates the salivation...

"what do you mean 'modest Q3 performance'?"

OK, let's reinforce the instruction by asking for the Gendo pose and chop licking again...

"Fine fine, I'm listening. Let's circle back to that last point, please explain it like I'm five" (that intern won't survive til the next earnings call)

Now let me try to generate a wilder werewolf from scratch, knowing that I need to doubly emphasise the pose, tongue, and flat rendering of the suit. What will Sora do?

Blep.

[login-required]

[login-required]

[login-required]

Quest for the best boy. (1/?)

The unintended consequences of MS Teams adding a werewolf react.

A couple weeks back I started some silly banter with a work colleague. This culminated in us deciding that we needed a werewolf logo for his project proposal about VR and factory design.

An initial attempt using the bog standard Edge Copilot didn't yield anything particularly exciting. I took to other tools in my own time.

A couple quick asks to ChatGPT yielded a decent first base for a logo-looking image. The porn-stache on the second is definitely a nice touch, but the factory thought bubble is a bit too literal.

I then went to Sora and asked for some corporate colours (including the absolutely vital concrete-grey) as well as a more abstract representation of the factory...

Not bad, not bad. Can we give it a smidge more character? Let's give him the Gendo Ikari pose, for the bants.

ChatGPT gets it right instantly, though requiring some nudging to get the colour patterns right (note how in two images, the corporate palette manages to creep into the werewolf's fur).

Sora smashes it out of the park with the Gendo pose, but accidentally loses the VR goggles and gets the project name (VVOLF) not quite right:

Let's do another remix to reinforce the goggles-and-name part...

Not too shabby! (One may say, even, not that shaggy...)

Next up: Finishing touches, playing with styles and having a go at DALL-E.

[login-required]

[login-required]

Dyiing Cybergoth (2)

When your AI overlord keeps getting it wrong and lying about it.

In a previous post we saw OpenAI's shot at generating a little pastiche. Now it's the turn of Microsoft's Copilot.

On receiving the initial set of instructions, the response sounds promising enough. Copilot correctly performs a search and describes a plausible enough image. (Copilot's response below is toped off by a lengthy description full of epithets.)

First hurdle: Copilot claims to have produced an image, but the image is nowhere to be found. After repeated prodding and asking for different ways of serving an image, we finally get there -- though confusingly, a promised download link does not materialise. Does Copilot not double check for what it's actually able to do?

...So we've got a classic looking statue with some cyber chintz slapped on top, though rather than lay dying, the pose is more "lounging whorishly". Let's ask for something closer to a cybergoth getup and insist that we want a Dying Gaul pose.

Ehh, it's definitely more of a Cybergoth (though a bit more steampunk than I'd hoped?), but the whorish lounging hasn't subsided and we have a skin tone with more perceptible nipples. The blemishes on the skin suggest rot or the veins on marble, so in a way the death and statue instructions haven't completely go away?

Perhaps we can get a version of the image that de-emphasises the upper torso and captures the full body of the 'goth, with the pathos conveyed by the limp lower body?

OK, so we managed to lose the skin tone (though the nipples still stand out through a metallic sheen). We clearly are not getting a full body, and we also are getting the order of the reply wrong -- with the image coming before the text response announcing it.

Can I be more specific? Let's ask for a head-to-toe statue...

Sure thing buddy.

To be fair, some aesthetic elements are delivered well enough -- dreads, respirator, goggles, smooth leather and bright neon plastic, 'classic' sculpture elements (especially the material), and of course the bits of waste suggesting rave enhancing substances.

Perhaps we need to accept that this one will generate vibes well enough, but fail at following halfway specific instructions.

As the millenials who want to stay hip say, "mood".

[login-required]

Bitch I might be.

[login-required]

Some kind of statement about tech.

First picture: electronic ad screen acting up on my way to work.

Second picture: interactive cyber-installation. Enter the pod and place your hands on the glowing bulb, you will feel the breathing and heart beat of someone standing on another pod.

[login-required]

Dyiing Cybergoth (1)

That's right: ii because it's a remake!

Creation of this was not super involved, and really goes to show how far some tools available to the public have gone. I merely instructed ChatGPT to search both the Dying Gaul statue and references on cybergoth aesthetic, then generate an image integrating both -- substituting hangover for the agony of death. Some follow-on prompts were needed, asking to emphasise specific aesthetic elements. The corrections were very much surgical and the rest of the image remained almost the same as the original attempt.

Here is the initial image.

My main concern was getting that big respirator in there. The veins in the marble have shifted slightly, the drink can (notMonster?) has rotated and the use of leather has expanded to a more complete harness and boots (which I didn't explicitly ask for but actually works) but otherwise the pose and finish are the same. Interestingly, the initial image retained a stub of a sword, which eventually vanished completely.

Part 2 will show the Microsoft attempt which didn't pan out quite as impressive.