Art is dead? Long live AI
🐴
IRL 🎠 stallion
@Victorious
All you said was copied was the plot, not the character names or cover art. Plots are a dime a dozen.
All you said was copied was the plot, not the character names or cover art. Plots are a dime a dozen.
Victorious
@PUBLIQclopAccountant
Oh no I’m sorry. I mean EVERYTHING. I took your plot, your characters, your sidequests, your weapons. I created your game and said it is mine.
Oh no I’m sorry. I mean EVERYTHING. I took your plot, your characters, your sidequests, your weapons. I created your game and said it is mine.
LiquidToast
@Starless
I sincerely doubt artists will be going anywhere, people will always demand hand crafted stuff. That and it’s getting easier, at least for me, to spot AI art.
I sincerely doubt artists will be going anywhere, people will always demand hand crafted stuff. That and it’s getting easier, at least for me, to spot AI art.
Buttlord Wayne
Protects Butters
@hugo231929
Yeah, but like, people do that too, all the time. Heck, people are the ones using this algorithm.
Yeah, but like, people do that too, all the time. Heck, people are the ones using this algorithm.
Background Pony #20CE
@hugo231929
Mate as an artist I can tell you I don’t understand what I’m doing either, and if not for the work of everyone else who came before me I would have nothing to learn from.
Mate as an artist I can tell you I don’t understand what I’m doing either, and if not for the work of everyone else who came before me I would have nothing to learn from.
absentia616
@Victorious
No that’s duplication, you don’t need a AI for that, just a couple of keys on the keyboard, so that entire argument is invalid, and just goes to show that you don’t know how the AI even works, to non artists the AI is a gimmick or a stand-in not a replacement, and to artists that except it, they can use it as a tool or aid.
No that’s duplication, you don’t need a AI for that, just a couple of keys on the keyboard, so that entire argument is invalid, and just goes to show that you don’t know how the AI even works, to non artists the AI is a gimmick or a stand-in not a replacement, and to artists that except it, they can use it as a tool or aid.
And as for the whole it’s stealing people’s art nonsense, the reason you show it an image is so it knows what the image is, it can’t draw a pony if it doesn’t know what one is, and the reason why you have to show it so many images is because despite having intelligence in its acronym, it is dumb AF, a human artist can create an image of a pony after seeing one image to accurately create an image of one, an AI needs thousands before it even knows what it’s looking at, and even then it’s still gets things very very wrong.
The whole it steals peoples art sits quite nicely with the world is flat, the world is run by lizard people and aliens built the pyramids when it comes to credibility.
NightLuna
@absentia616
Your words only confirm that AI is not powerful enough at the moment, but it gets better over time, so it’s possible that one picture of a pony in the future will be enough for it to draw a lot of similar ones.
Your words only confirm that AI is not powerful enough at the moment, but it gets better over time, so it’s possible that one picture of a pony in the future will be enough for it to draw a lot of similar ones.
Background Pony #A36B
@NightLuna
I;m not an expert at any of this or anything like that, but, zero-shot learning as it is believed to be achievable today is essentially just fancy CLIP. and anyone can tell you that there will always be some limitations in CLIP’s ability to categorize the content of a given image, there’s always more dimensionality then labels. Ponies become cats, how does it label something like a kirin, etc, etc. Whenever there’s a CLIP 3.0 or 4.0 people are going to call it zero-shot but really it’s going to just be zero-shot for base stable diffusion and more specialist models will be out to dry. If people are expecting a revolution in training practices they probably shouldn’t hold their breath, we’re in the era of ‘bigger datasets better model’ for a few more years I think.
I;m not an expert at any of this or anything like that, but, zero-shot learning as it is believed to be achievable today is essentially just fancy CLIP. and anyone can tell you that there will always be some limitations in CLIP’s ability to categorize the content of a given image, there’s always more dimensionality then labels. Ponies become cats, how does it label something like a kirin, etc, etc. Whenever there’s a CLIP 3.0 or 4.0 people are going to call it zero-shot but really it’s going to just be zero-shot for base stable diffusion and more specialist models will be out to dry. If people are expecting a revolution in training practices they probably shouldn’t hold their breath, we’re in the era of ‘bigger datasets better model’ for a few more years I think.
🐴
IRL 🎠 stallion
@NightLuna
The idea of encoding the concept of a “horse” to the AI is an older school technique. These days, the “dumb” statistical models actually produce results while the AIs designed with knowledge and intellect have been left in the dust.
The idea of encoding the concept of a “horse” to the AI is an older school technique. These days, the “dumb” statistical models actually produce results while the AIs designed with knowledge and intellect have been left in the dust.
Current models depend on a massive input corpus. With a single image, they’d have no idea which part is the horse and what is the background and what is statistical noise of the pose and shadows.
Beth
In digital purgatory
Yeah AI is just mashing together average horse crap. If it happens to make something beautiful, that’s because lots of beautiful art has already been made.
People will get bored of this current AI, even when it gets hands and faces right.
UnderwoodART
Just Another Pink Pony
Here’s a thought: What if you, as an artist, used a machine learning AI trained exclusively on our own work to create new art. Is that art? Are you the artist?
d1ckbitch
Doughnuts Worshipper
@UnderwoodART
This question might be nothing but another way of asking the eternity-old real question : “what is art?”, and maybe not really relevant to AI by itself.
This question might be nothing but another way of asking the eternity-old real question : “what is art?”, and maybe not really relevant to AI by itself.
Is putting a white canvas in a museum art? Is splatting some paint in a large canvas and selling it for a very expensive price art? What about destroying it, by the artist or someone else, is it still art? Is putting a huge amount of creations in a computer software to get a new creation, made of little pieces of all the others, art? Don’t artists already do it but with their brain instead?
TBH, I’d be curious to see what a model trained on exclusively one artist would output… This can be done, just need someone to do it now
🐴
IRL 🎠 stallion
@d1ckbitch
What about tipping a urinal on its side?
What about tipping a urinal on its side?
@Beth
This isn’t necessarily true. However, getting results outside the AI house style requires deliberately asking for “in the style of Mark Rothko,” for example.
This isn’t necessarily true. However, getting results outside the AI house style requires deliberately asking for “in the style of Mark Rothko,” for example.
Bigcheese
TBH, I’d be curious to see what a model trained on exclusively one artist would output… This can be done, just need someone to do it now
It would be mostly noise. A single artist cannot product enough images to train a model from scratch. StabilityAI is working on a public domain + opt-in model. Using that as a base you could train on a single artist.
d1ckbitch
Doughnuts Worshipper
@PUBLIQclopAccountant
As a trick that can never be pulled again… Fountain is one of the most important piece of art ever made.
As a trick that can never be pulled again… Fountain is one of the most important piece of art ever made.
@Bigcheese
IDK, I heard about this thing called LoRA that seems to be able to learn an artist’s style from a very low data set, like 50 pictures or so… And most profilient (good) artists can have like 1000 pictures under their name. Not enough to get decent generations, but it might be readable pictures at least…?
IDK, I heard about this thing called LoRA that seems to be able to learn an artist’s style from a very low data set, like 50 pictures or so… And most profilient (good) artists can have like 1000 pictures under their name. Not enough to get decent generations, but it might be readable pictures at least…?
-Craft-
@d1ckbitch
Loras are for training singular concepts or styles predominantly. They still require an underlying base model to understand concepts like ‘Apple’ or ‘car’ if those concepts were not in the lora.
Loras are for training singular concepts or styles predominantly. They still require an underlying base model to understand concepts like ‘Apple’ or ‘car’ if those concepts were not in the lora.
If that’s good enough for you then it’s already being done in anime spaces but is not as popular in pony.
Buttlord Wayne
Protects Butters
So are people angry at AI for being able to replace humans or are they angry at AI for sucking at replacing humans?
d1ckbitch
Doughnuts Worshipper
@WaynePonePone
Using human’s hard work for replacing humans
Using human’s hard work for replacing humans
@Ring Team
Art on top of the picture is plastered on every new Kickstarter campaigns, quite a few graphic designers were not hired for these jobs, actually… The placeholder pictures market might be suffering a lot these days.
Art on top of the picture is plastered on every new Kickstarter campaigns, quite a few graphic designers were not hired for these jobs, actually… The placeholder pictures market might be suffering a lot these days.
AI art is certainly not perfect, but being good enough is all that’s needed, sometimes (and I mean everytime)
🐴
IRL 🎠 stallion
@d1ckbitch
The caption on the bottom right of the first image seems to acknowledge this point without realizing it’s the central point. Looks good at thumbnail preview but is uncanny when viewed full-rez was one of the points of contention on AI art being posted here and people complaining about how many updoots it earned.
The caption on the bottom right of the first image seems to acknowledge this point without realizing it’s the central point. Looks good at thumbnail preview but is uncanny when viewed full-rez was one of the points of contention on AI art being posted here and people complaining about how many updoots it earned.
d1ckbitch
Doughnuts Worshipper
That’s right, AI art kinda revealed the existence of people who upvote without actually clicking the thumbnail… the new AI policy is good for solving that, in theory, no more crudely upscaled AI pictures hopefully
Interested in advertising on Derpibooru? Click here for information!
Help fund the $15 daily operational cost of Derpibooru - support us financially!