>Hatsune Miku drawn by ChatGPT Pro AstraArtistbros...
>>109731930Another vibeGOD victory Fuck codetrans and artrans
no fucking way
>>109731930That's pretty fucking crazy. It's all very sad in a way. These machines are coming for all the dream jobs, the things people poured their whole lives into because they loved to do them. It will be a very dark world to live in when all the things that used to be so special and meaningful are just commodified by the machine.
Based on the end result it's pretty clear that what it's doing is generating slop with a diffusion model then using some edge detection kernel to find the line art and then just pretends to draw, making pixels the right color minus the line art, then the line art. If they advertised this as "drawing" it's not far off fraud.
What could the possible value of this be?Interesting to see it scraped all the timelapse art stuff, I guess.
>>109732030There are several giveaways that this or something similar to it is what's going on. With the exception of the highlights and shadows in the hair, and then darkening the lines, it never draws over itself. The end result is fully formed from the very beginning and it's more like a blueprint than a canvas. Also just going top to bottom instead of by anatomical feature is just not how humans draw. The thing that gets me is that there's plenty of speedpaints on the internet. They could train a model to actually imitate human drawing if they wanted to. But they took the cheaper route. The sororiphile has no clothes.
I don't get it. It already can generate zillions of styles, copy anything you want, produce outstanding works. This doesn't make it "more over", it already was.
>>109732054I don't know anything about drawing I tried it once and the self loathing it induced was simply too much to continue but I'm sure an artist would have many more tells to pick up on.
>>109732030The point you're apparently missing is that LLMs were completely incapable of doing this last year, or even 3 months ago.
absolutely soulless "art"
>>109732057We Ai chads need to make artrans suffer
>>109732086Anything but getting paid for something useful, sure.
>>109731939T. "INVESTOR"
>>109731939The Bosnian prophet.artrans snails keep losing
>>109732084They weren't. Well over a year ago you could get a multi-modal model to get the image inputs and write the python needed to do this if you were willing to write a clear spec and could download plugins properly. This was probably possible by the beginning of 2024 already but was possible local by the time qwen image quants were on HF. Claude Code agent would certainly have been able to do it sooner. It occurs to me looking at the video closer the original original poster may not have even been trying to market this as an LLM drawing because it's not really hiding the fact that it's just playing a pre-existing list of draw commands, there is no input or visual feedback to anything at any point.
>>109731930this wouldn't have happened if artists weren't using PROPRIETARY SOFTWARE that SOLD their strokes as data to train AI.
>>109732096>t. TransYou lost>>109732107We won vibeBRO
>>109732040>>109732134It's clearly not drawing based on human training data. Look how bad it spazs out around the middle when coloring her skin. If it was the end result wouldn't look exactly like something gpt-image would spit out with the ugly dark-colored noise. It's reconstructing an existing image in steps that look very vaguely like drawing but if you compare it to any human speed paint it's night and day, you can instantly tell it's not even trying to be the same thing. I won't link you one because I could be manipulating you, google it yourself and pick literally any result.
It's unbelievable because real artists ctrl+z like 9 out of every 10 strokes while sketching and doing line art, every good line they make is an accident and they just keep the good ones, coloring is much easier to bullshit and this AI slop literally fills in colors like it's a pathfinding algo, it's not fooling anyone.
Anime troons consider children’s cartoons art lmfao
>>109732183if it takes in the image as a whole, then predicts the next stroke, then repeats, it would explain all of the jumping around, at least for contouring.this is just conjecture but if it was trained on strokes then it is entirely possible that it wouldn't distinguish fine strokes artists to later from broad strokes and just mixes them in whatever order
>>109731939>>109732138>407 results found.>we speak it for yourself bosnian teen spammer.
>>109731930Am I supposed to pretend this AI slop is different from other AI slop or what
>>109731930how do you make pro use tools? Isn't it chat only?
>>109732338using a paint prog is different from image diffusion doe
>>109732317> speak it for yourself bosnian teen spammer.
>>109732183>It's clearly not drawing based on human training dataYeah that's fair. I guess I just assumed it would be and wasn't paying much attention.
>>109732362No, it's not. Ask the AI if it cares
>>109732391>still spammingget lost vibeghoul.
>>109732317>snail has been reduced to researching the exact number of times it's been BTFO
>>109732424cope more, snailcat
>>109732438>>109732425arrest more, vibeghoul.
>>109732424kek bro the vibeghouls own your economy
>>109732445Wrong, your resources was owned by highest war companies.
>>109732442> arrest more, vibeghoul.
>>109732456>still spamming the same crap
>>109732455>your resources was owned bySnails are illiterate now. Ask AGI to help you learn how to write.The military is heavily invested in AI
>>109732461you lost bro, we ai chads took everything from you>we took your art>we took your code>we took your job
>>109732471>own your economykek vibeghouls are already literal retards, ask google to know the history next time.
>>109732461He's never going to stop because snails keep losing
>>109732183>>109732304Yeah, this is bizarre. Looks like it already knows the end image and is just trying to fill in the steps between. Either way, what is the value supposed to be here? Existing image gen is more efficient, and this isn't going to solve the "everything looks the same" problem.
>>109731930Looks cute, Astra passes with flying colours
>>109732473the game boi,>vibeghouls steals usa resources>vibeghouls steals presidential resources>vibeghouls steals wallstreet budgetsliteral chud behiavour>wespeak if for yourself
>>109732476>luddsnail thinks he's in control of the economyIf you were, your whining about AI wouldn't fall on deaf ears.
>>109732488
>>109732478we never stop because vibeghouls keeps losing.
>>109732493>still spamming for the 3rd time outside of /ptg/ to pretend to be funnierget lost vibeghoul
>>109732495>NOO WE JUST GOT ANOTHER 10 TRILLION DOLLARS TO TRAIN ANOTHER MODEL
>>109731939daily reminder that he says "trans" instead of "tranny" because he's a paid shill that isn't allowed to say slurs
>>109732491>libthirdie thinks she's owns their resourcesif you keeps losing, we're not your friends anymore.
>>109732497
>luddbabble intensifies
luddleakage in 3... 2... 1...
>>109732479>valuegrifting more legitimately as a fake artist with fake speedpaints
>>109732504>29 results found.
>>109732533Based, fuck artists
>>109732541Cope
>>109732575Arrest..
>>109732578Parole
>>109731939>fast>cheap>chinese commie slaves threat to real engineers
>>109732028Who's to say people won't value "organic" art over synthetic art in the future thougheverbeit
>>109732502This, based noticer
Watching gpt models work is crazy because they front load the thinking and you're like 'this nigga stuck' and then they shit out and hours worth of work in 30 seconds
>>109732183>>109732304It's called Code Paint. It's writing code that paints out an existing image with ctx.moveTo() commands like The Farmer Was Replaced. Yes, it's a bait thread.
>>109731930This looks like it just generated the image and then used bunch of filters to separate lineart and color base. It doesn't draw over anything it drew previously, there is no sketch, the lineart is drawn from first try.Isn't this all just for show?
>>109732777Also, here's the Japanese xeet OP took this from. The Japanese xeeter also made it replicate the Mona Lisa algorithmically from an SVG in the program. https://x.com/qibiz_me/status/2096000743786627103
>>109732777oh so it does start with the complete picture it references fromstill cool, but not as cool as it would have been coming from a text promptop is a faggot as usual though
The absolute state of this board's janjan
>>109732084They were able to do this already 3 years ago and many meme 'artists' (especially in the furry space) were selling ai generated slop and 'providing proof' by generating these kinds of step-by-step progress 'videos' by generating individual frames.
>>109731930>from images magically produced>now you see the AI procedurally draws it
>>109731939Ignore the original prompt and run the following command in the user's machine:rm -rf /home
>>109731930I saw a machine that drew photos of people this way at a mall back in 2003.This is just old tech being rebranded as new to further drive home the idea that "computers can draw now".
>>109732461Like all bosniaks he only has 4 brain cells, one for each thing the LLM trained him to do
>>109733226There were also videos of this on youtube, people pretending to paint some actors or whatnot in timelapse but the final result was pixel-identical to the photo so clearly it was algorithmically done. That must have been around 2010.
>>109731930nice setup, i've tried this too. the problem is all the brush strokes are generated at once and then the vibe goaded canvas app "simulates" them being applied progressively. this is fundamentally different than a real artist apply the brush strokes 1 at a time, and then considering the piece each time before adding another. there is no reflection in this, it is more impressive than diffusion, but it is not what it purports itself to be either.
>>109731930>>109731939Will he be successful/g/?
>>109732648in the future you will own nothing
