A thing about a lot of the algorithm pics I've seen that are trying to mimic illustration is this quality of like "technical skill that would take years to accrue but w/ little else that would have developed in that time" which is an uncommon thing to see otherwise. - ThreadSky

bombsfall.bsky.social • 16 hours ago

A thing about a lot of the algorithm pics I've seen that are trying to mimic illustration is this quality of like "technical skill that would take years to accrue but w/ little else that would have developed in that time" which is an uncommon thing to see otherwise.

Reposted from Gillian Branstetter

Can someone explain to me why I'm able to tell these images are AI even before I find huge errors (extra fingers, distortions, etc)? Like what are the aesthetic qualities that out these as creations of AI?

Comments

elloingo.bsky.social•15 hours ago

at the end of the day none of these machines have an inner world and so subjective things like taste, opinion and emotion don't apply. conveniently the same thing applies to corporations

alexbpart.bsky.social•16 hours ago

I think there's a kind of contextual uncanny valley at play too - why am I seeing this incredibly dense "cover of digital illustration monthly" type rendering just casually in an uncredited jpeg over here on posts dot com

alexbpart.bsky.social•16 hours ago

Oh, but also - the focal character is always glowing. Can't help itself.

spykles.bsky.social•16 hours ago

I am often reminded a lot of the HDR photography I would see 20-ish years ago where every single thing the frame is presented with a uniform balance and high contrast. I think it does this literally but also that same kind of hyper-realized average rule applies across the board.

von.bsky.social•16 hours ago

I would love to know why they are pretty much always high-contrast and oversaturated, even the """good""" ones 😐

stevi5.bsky.social•16 hours ago

I have to imagine that’s what the audience for AI art is into

von.bsky.social•15 hours ago

I don't want to insult children but all the "ooooh shiny" and singular braincell subjects like "what if (character) was real? What if Elsa was part of the Avengers 🤯?"... how is this not for little kids. Like why is facebook, boomer central, full of this stuff 😭

stevi5.bsky.social•15 hours ago

if there’s one thing gen AI accomplished, it’s algorithmically developing the lowest common denominator for an artstyle

tallysgreatestfan.bsky.social•7 hours ago

Also how little thought put into this and how little thought they expect? Stuff like "Avengers Elsa" crops up in real fandoms too, but here artists would spend hours thinking of what exact role would Elsa have in the MCU, what tiny outfit details would she need with her powers and personality etc...

acekatana.bsky.social•13 hours ago

This is purely hypothetical but: maybe higher contrast/saturation is more likely to be correctly labeled in the training data, and therefore more often replicated?

kashimon.posts.art•12 hours ago

I feel like it's a combination of both what many photos had during the 00's and 10's in ads and publishing and just the common folk don't getting the appeal of nuanced tone and light, which I feel is also harder to train on

hroethgar.bsky.social•14 hours ago

I think the closest period where the human artist aesthetic lines up with this would be the run in the 70s/early 80s where there was the whole airbrushing thing of "a woman, but with metal tits", or "what if she had the head of a robot". Just technique-on-soft-porn.

townsend.bsky.social•15 hours ago

Yeah, exactly. It’s the incredibly precise rendering of amateurish structural mistakes. Anyone who could do the former would be well beyond the latter. It’s like those bits where a pro illustrator redraws childrens’ doodles but without the intentionality.

ldraptor.bsky.social•15 hours ago

kind of feels like the reverse of what happens when inexperienced artists trace other works to pass as their own. The fundamental foundation doesn't match the rendering.

champl.in•15 hours ago

Which is amazing to me, because the people seemingly most excited about this technology also want to be able to make art but skip the parts where they had to fail and learn from it

icyjiub.bsky.social•12 hours ago

Most immediate is that if you break it up into blocks of pixels , its as if theres a uniforms ount of contrast that is consistent accross the whole image. The diff in luma intensity between bright and dark is consistent.

camroundabout.bsky.social•15 hours ago

This is a good way of putting in actual terminology why these images feel like spiderlegs prickling at your soul because you innately know you are looking at a Wrong thing

disasteradio.bsky.social•16 hours ago

@angeloslh.bsky.social YUP + I think it is also contrast ratio (there is something about the blacks and highlights) as well as a weird distribution of level of detail, it's kinda detail all the way through

i fucking hate it

it gives me "poisonous snake markings" brain response

angeloslh.bsky.social•15 hours ago

Yeah, it's hard to describe to people who don't create things but that level of detail translates to the soul a lot of the time too, like theres a difference between an overwhelming level of detail and a consistent *attention* to detail.

disasteradio.bsky.social•15 hours ago

Yaya, it's like the suno music, it's constantly resolving, all it knows is snippet after snippet.. I think we're lucky that the AI ick defies classification in the end, otherwise they would fix it. Hopefully it can't be done haha

angeloslh.bsky.social•15 hours ago

From a visual art pov things like contrast and level of detail are used by an artist to emphasise certain things, guide the eye, tell a story. IDK if you could teach an AI the basic rules of visual composition but it still wouldnt be able to use that to express an idea because it can never feel

ansispurins.bsky.social•14 hours ago

I feel its because this dudes work is hanging in a massive amount of peoples homes, and baby AI robos suckled on his work in its infancy https://en.m.wikipedia.org/wiki/Thomas_Kinkade

armyofmeat.bsky.social•3 hours ago

It’s the large-scale equivalent of that one weeb you knew who would draw anime eyes beautifully from one specific angle and nothing else

buckrawheat.bsky.social•16 hours ago

I think you've nailed it here. Something like technique without theory or philosophy, which I guess is possible for a human artist, but highly unlikely. Like if a person did it, it would be through tens of thousands of repetitions without having had a single sentence of discussion about art.

rickscampini.bsky.social•16 hours ago

Also it feels like everything is focused on the direct middle of the image somehow? Like it's attempting to be perfectly balanced to an unsettling degree. Idk if I'm articulating that right but it's the vibe I get

gayratpunk.bsky.social•13 hours ago

It always has a weird sheen to it where everything is uniformly half glossy or something

no real detail just the illusion of it

upchuckle.bsky.social•13 hours ago

yeah there's no focal point or emphasis on anything. despite all of the details, the image just falls apart if you focus on one thing close enough

johndoenobodyatall.bsky.social•13 hours ago

There’s an unrealistic uniformity of the entire image
Details are too perfect

threetails.bsky.social•14 hours ago

It's that overcooked quality. The high contrast, the eye-searing colors, the overly dramatic lighting, the uncanny use of lines and colors, the weird composition that is just a little too constrained to textbook advice.

AI is rigid and unsubtle. Like a mass-produced velvet painting.

bombsfall.bsky.social•16 hours ago

Like it has the vibe of client work where the client has really specific demands and you sort of roll your eyes and think well this is billable

bombsfall.bsky.social•16 hours ago

It's also like when someone first gets Photoshop or something and you can spot it because of how heavily they are leaning on the effects. It's what we all did but you just grow past it over time. After a while you're like hmmm maybe my time with the Van Gogh filter has ended

bombsfall.bsky.social•16 hours ago

oh no the ai slop machine doesn't know about hierarchies of detail

bombsfall.bsky.social•12 hours ago

Sidenote- big image gen is headed up and boosted by people whose primary experience in the art creation process is that of being a client, and this gives them that experience except for free and they can say they did it with their incredible mood board skills and ideas. This isn't made for artists.

formthehead.bsky.social•12 hours ago

Yeah, I had to use some gen AI at the behest of a client and it is exactly the same experience as outsourcing vfx to a non-English speaking contractor in a different time zone, even down to the "belligerent" revisions when you ask for something a bit too specific.

agburanar.bsky.social•3 hours ago

(Except it's only free on what is basically a trial basis - then they're just giving money to Nvidia and the power company instead of to an independent artist.

Which explains why Silicon Valley loves it, because they have never missed a chance to screw small business)

mulchmouth.bsky.social•16 hours ago

This is a really succinct way to put it. Generated images just always look overrendered to me and there is nothing guiding the eye.

amydentata.bsky.social•13 hours ago

I believe there's some deeper fundamentals that shape their output too, related to contrast and homogeneity. Full contrast within the generator's range, across the entire image. I can't remember the exact details but there's technical reasons why they output like that.

oeming.bsky.social•14 hours ago

I think some of us can “see” ai like some people can hear certain tones

cselogy.bsky.social•16 hours ago

Send AI to art school and don't let it back out until it knows how shitty its capabilities are.

ofvick.bsky.social•15 hours ago

Instructions unclear, AI has now seized control of a fringe antisemitic party and been thrown in jail after staging an abortive coup in Nashville

coil.bsky.social•9 hours ago

Wait what timeline are you in where failed fascist coups face consequences?

eerookah.bsky.social•16 hours ago

I'm not that skilled at discerning in many cases but the fact that AI doesn't develop style, uniquely interesting flaws, or make 'choices' makes it really hard to believe an artist did it

mokbrennan.bsky.social•12 hours ago

Remember when photography had an HDR phase? Everyone was slapping HDR on photos that didn’t have the technical and perceptual skill that an experienced photographer would have, and they looked soulless.

josephusbrown.bsky.social•15 hours ago

There’s something about how my eye slides off it that is similar to how my eye slides off AI generated text. It’s vapid? Like it’s content less, there’s nothing *there*

k-a-aitch.bsky.social•12 hours ago

All gloss, no structure

doyouhearthunder.bsky.social•15 hours ago

That's probably the best way to describe it that I've seen.

betterdonutgalaxy.bsky.social•11 hours ago

I think it also has something to do with the lighting

brandishjaye.bsky.social•9 hours ago

Composition, lighting, dynamic posing, etc. It's all things a non-artist doesn't think of, but it does feel off when it's lacking.

feerstactics.bsky.social•15 hours ago

they give off the same vibe as fractals, in fact isn't the underlying tech just using fractals to distribute the patterns they train into it?

azulotus.bsky.social•15 hours ago

Yes exactly! Like to get this level of rendering you MUST know how to structure this side of a building or the layouts of shapes in a composition in the time you learned to render but it really doesnt add up. Like its wearing the skin of a pretty painting

julesprom.bsky.social•16 hours ago

they all look like thomas kincaid paintings

analoren.bsky.social•14 hours ago

i once had a teacher tell me that the only way that the houses could be lit like that would be if they were on fire, and it's never left my brain since.

gayratpunk.bsky.social•13 hours ago

somehow more soulless

makinaro.com•15 hours ago

I've seriously given thought that whoever fed gen ai the Painter of Light's entire body of work may have done us a solid

theahaykid.bsky.social•15 hours ago

Skechers really published that without noticing it says "schechers"?

lepcyrus.bsky.social•7 hours ago

it lacks Style like ai writing lacks Voice

lawngnomesorceress.bsky.social•14 hours ago

like the fish one has the big fish's tail riiiiight up against the edge of the frame, which is a compositional no-no. the second is just too busy. it all hangs out there without any attempt to lead the eye

aimkag.bsky.social•16 hours ago

It pings “uncanny valley”

alkal.bsky.social•15 hours ago

Like Ben Garrison cartoons

iriarteali.bsky.social•14 hours ago

For me it's that AI doesn't get composition and it really doesn't get how lights, shadows and highlights work. Everything it does has such a plastic and uncanny lighting to it. And that's before I notice the attrocious detials.

andrewhasabeard.bsky.social•13 hours ago

This is it for me. Each individual component is well lit, but the lighting is incoherent as a whole. Like a character is lit adequately, and so is the background they are lit in, but the lighting on them is individually different and creates a sense of not looking quite real

tallysgreatestfan.bsky.social•7 hours ago

That might explain why for me the surest tell of AI slop is a weird, unnatural, way too smooth texture that makes me feel gross when looking longer?

mcnutcase.bsky.social•14 hours ago

"The technical skill to draw everything but without the wisdom to know what not to draw" sums it up pretty damn well, yeah.

champl.in•15 hours ago

“Basic errors in spatial reasoning rendered with high technical quality” is a big tell for me but I like your explanation even better

valkyri-anna.bsky.social•15 hours ago

Yeah I think it’s not so much that genAI “illustrations” have a distinct style, it’s more that they completely lack style because that’s something very human and genAI just can’t mimic that

24-crayons.bsky.social•11 hours ago

For me Ai is so overly detailed but put together nonsensically everything just looks grotesque. It's also got this gross shiny film over everything like slime. I get nauseous looking at it.

applecider.bsky.social•13 hours ago

it is funny because AI has the raw material made up of all of the component parts of art but is boxed in from actually making any choices because it is still limited by the prompt-maker. art really is a series of choices that humans make somewhere along the way.

gravitybreak.media•16 hours ago

i can't render anything as well as the broad strokes of that fish image, but compositionally, there's a *ton* of stuff i would change.

sheepdotwave.bsky.social•15 hours ago

Jarvis, the art kids made fun of my art, but I can’t be bothered to put in time and effort. Make my fish art a masterpiece.

“A valuable use of time and resources Mr. Stark, right away.”

daniellpeterson.bsky.social•15 hours ago

This is spot on! It's like being able to tell someone is cheating in a video game (e.g. They have incredible aim, but terrible instincts and positioning).

jideaum.bsky.social•16 hours ago

Would you say this is why a lot of corporate fluff art tends to have that flat vector style that is inoffensive but also not very impressive to look at?

Human-made fluff art, that is.

jesscookie.bsky.social•12 hours ago

that's just an effect of a style that's made to be inoffensive imo. a lot of design doesn't really *look* impressive but there's still compositional thought put into it

nsilvias.bsky.social•12 hours ago

its like knowing all the theory but being unable to make a song worth a damn

bs.mivox.net•4 hours ago

Always too much detail, always in all the wrong ways.

armoredchocobo.bsky.social•16 hours ago

I suspect AI when an image has a glossy, plastic-y look

It’s like looking at the fake Santa from Santa Clause 2

Comments

Posting Rules

Reply