AI generated art

[TranscendentThots]

If you used it, you would see that it just isn't ready for prime time right now. You can make one really pretty image, sure. Therefore you can make a ton of pretty images. But you'll never make the same character twice, no matter how hard you try. and that makes it utterly useless for production, animation, or anything that requires consistency.

Even if you train a LORA to force the character to be Wonder Woman or something, all the clothing details shift around every frame. Belt details. Hair chunks. Literally any detail that has ever been drawn differently by two different artists, it will flicker and jump around in a state of perpetual quantum entanglement.

Best case scenario, you can make an image uncovering game, with no broader story context. Maybe something like Sunless Sea or 80 Days, where it's a string of unconnected vingettes. Maybe something like Hearthstone where each character and ability is a seperate card depicting a dissociated story beat. That's it. Board games, basically. If you need narrative or animations or clothed CGs, you're SOL.

 

[XcentY]

Well, technically, it is not really art. He is right.
Now, if AI gained consciousness, then it'd be a different story, but we are not there quite yet

That's your point of view...
And you can add words like "technically"... It just makes no sense at all.

Here are all arguments I can think of

You must be registered to see the links

 

[Winterfire]

If you used it, you would see that it just isn't ready for prime time right now. You can make one really pretty image, sure. Therefore you can make a ton of pretty images. But you'll never make the same character twice, no matter how hard you try. and that makes it utterly useless for production, animation, or anything that requires consistency.

Even if you train a LORA to force the character to be Wonder Woman or something, all the clothing details shift around every frame. Belt details. Hair chunks. Literally any detail that has ever been drawn differently by two different artists, it will flicker and jump around in a state of perpetual quantum entanglement.

Best case scenario, you can make an image uncovering game, with no broader story context. Maybe something like Sunless Sea or 80 Days, where it's a string of unconnected vingettes. Maybe something like Hearthstone where each character and ability is a seperate card depicting a dissociated story beat. That's it. Board games, basically. If you need narrative or animations or clothed CGs, you're SOL.

It has actually more issues than the one you listed.
Not only most are actually cherry-picked, but those examples often still have many mistakes if you spend more than a few seconds looking at them, and the outline is never clean.

AI generated images are in no way going to replace artists anytime soon, however it is a good tool for artists to use.

 

[Winterfire]

That's your point of view...
And you can add words like "technically"... It just makes no sense at all.

Here are all arguments I can think of

You must be registered to see the links

Art is an expression of the soul, it requires emotions and therefore cosciousness.
I used "technically" because you could argue it becomes art once an image is cherry picked (assuming it is not edited by a human) and looked at, which evokes emotions... However, most of the times it just evokes questions such as "Why is there a floating piece of roof over there?"

 

[XcentY]

Art is an expression of the soul, it requires emotions and therefore cosciousness.
I used "technically" because you could argue it becomes art once an image is cherry picked (assuming it is not edited by a human) and looked at, which evokes emotions... However, most of the times it just evokes questions such as "Why is there a floating piece of roof over there?"

That's your point of view...
Adding that it's an expression of the soul is a subjective perception of art.
Also, the one prompting the AI to generate a content is conscious...


So, if I wanted to argue your way, I would come up with something like this :



behind this AI generated image, there is a person that prompted the AI...
Deep inside his soul, he was imagining something and then he prompted the AI trying to depict what came to his mind at this particular moment of his life...

Of course, he doesn't use any of the tool we had before... ( Graphite pencils, brushes, paint, painting palette, ... and so on ) but rather he prompted an AI which has the skill to produce what you have in mind...

 

[Winterfire]

View attachment 2564081

behind this AI generated image, there is a person that prompted the AI...
Deep inside his soul, he was imagining something and then he prompted the AI trying to depict what came to his mind at this particular moment of his life...

... And then came disappointment as he looked at the messy lsd-like image with half of the elements being plain wrong.


If anything, you are proving my argument right .
Also, it is not really "My point of view", art requiring emotion is a fact and currently it is pretty one-sided.

 

[TranscendentThots]

That's your point of view...
Adding that it's an expression of the soul is a subjective perception of art.
Also, the one prompting the AI to generate a content is conscious...


So, if I wanted to argue your way, I would come up with something like this :

View attachment 2564081

behind this AI generated image, there is a person that prompted the AI...
Deep inside his soul, he was imagining something and then he prompted the AI trying to depict what came to his mind at this particular moment of his life...

pink trees, blue sky, WOW-style house in a field D&D concept art in an oil painting style

negative prompt clouds, outlines, 3D

FEEL THE HEARTFELT EMOTIONS OF MY SOUL!!!111

 

[XcentY]

... And then came disappointment as he looked at the messy lsd-like image with half of the elements being plain wrong.


If anything, you are proving my argument right .
Also, it is not really "My point of view", art requiring emotion is a fact and currently it is pretty one-sided.

Look here

I prompted the AI to produce this image a time ago :



And If you don't see my soul and my emotions in this picture.. Because you so want art to be attached to something subjective which is not the definition of art...

I didn't use any tool you use to depict my emotions ... I used an AI...

 

[Winterfire]

Besides, AI can create prompts out of an image and it is not the AI that has to understand and adapt from the prompts you give (As an actual artist if you were to describe an image you want to be painted), but you have to be the one understanding the AI, adapt and give prompts that hopefully will give you a not-too-broken image.

Do not get me wrong, I am not against AI as you may have seen from my previous response, I totally embrace it.
However, AI is not an artist, and it can be considered art only one-sidedly in those rare cases where a cherry picked generation can actually evoke feelings and not questions.
You can be a fan of AI, while still calling it for what it truly is.

 

[XcentY]

Besides, AI can create prompts out of an image and it is not the AI that has to understand and adapt from the prompts you give (As an actual artist if you were to describe an image you want to be painted), but you have to be the one understanding the AI, adapt and give prompts that hopefully will give you a not-too-broken image.

Do not get me wrong, I am not against AI as you may have seen from my previous response, I totally embrace it.
However, AI is not an artist, and it can be considered art only one-sidedly in those rare cases where a cherry picked generation can actually evoke feelings and not questions.
You can be a fan of AI, while still calling it for what it truly is.

Okay so when an artist, deep in his soul, full of passion, full of emotions, ... wants to draw something, paint something, build something, write something, ... How much time will he spend on his creation ?
The same goes for the one prompting the AI... He hasn't the same tools as someone drawing, someone painting, someone building or someone writing,... He uses an AI and prompt it, changes key settings, learn how the AI react to his prompts, ... until he gets the skill he needs to produce something that correspond to the emotions he wanted to share....

So what's so different ???

 

[Winterfire]

So what's so different ???

One will get exactly what they wanted, the other will go "Meh, good enough."

 

[XcentY]

One will get exactly what they wanted, the other will go "Meh, good enough."

In the "Meh, good enough" you'll also see a part of the soul of the person prompting the AI...

And when you'll see a superb creation, you'll also see the time the guy behind the prompt took to tweak the result to get what he wanted...

 

[Meaning Less]

behind this AI generated image, there is a person that prompted the AI...

So by this logic if I use a prompt on google, and find an image similar to what I was looking for, I'm an artist now?

Because that's what you are doing, that image was not your creation, it already exists and is result of mixing and matching millions of images it was trained with, you just found that seed, but it can always be recreated by the same system.

At no point it is removed after you found it and there is always a chance someone else will stumble upon that image again.

You just browsed a massive library until you found something that you liked, that's why calling yourself an artist for just browsing that library is a huge leap.

 

[XcentY]

So by this logic if I use a prompt on google, and find an image similar to what I was looking for, I'm an artist now?

Because that's what you are doing, that image was not your creation, it already exists and is result of mixing and matching millions of images it was trained with, you just found that seed, but it can always be recreated by the same system.

At no point it is removed after you found it and there is always a chance someone else will stumble upon that image again.

You just browsed a massive library until you found something that you liked, that's why calling yourself an artist for just browsing that library is a huge leap.

You're the one taking the huge leap by confusing a search engine with a generative AI.
The way AI generate images isn't by mixing images from a library.
Let's say that you do a very basic AI not to generate but to recognize hand written digits.
When you are in the process of learning, the AI will take a bunch of data that will be back propagated to the neural network to change weights and biases on the neural networks neurons...
Once the dataset has been learned... You don't have any image in these neurons, only weights and biases... And yet these neurons will recognize a new handwritten sample..
Once you've your neural network, you can turn it into a generative neural network that will just create handwritten sample... and you can say that it's by mixing images but it's not the way it works...

 

[Winterfire]

You're the one taking the huge leap by confusing a search engine with a generative AI.
The way AI generate images isn't by mixing images from a library.
Let's say that you do a very basic AI not to generate but to recognize hand written digits.
When you are in the process of learning, the AI will take a bunch of data that will be back propagated to the neural network to change weights and biases on the neural networks neurons...
Once the dataset has been learned... You don't have any image in these neurons, only weights and biases... And yet these neurons will recognize a new handwritten sample..
Once you've your neural network, you can turn it into a generative neural network that will just create handwritten sample... and you can say that it's by mixing images but it's not the way it works...

His point still stands though, anyone can recreate it with the seed.
Also, neither searching on google or writing a prompt while crossing fingers make you an artist.

 

[XcentY]

His point still stands though, anyone can recreate it with the seed.
Also, neither searching on google or writing a prompt while crossing fingers make you an artist.

Of course writing a prompt while crossing fingers doesn't make you an artist.
But the art of writing prompts is something you can't deny...
The seed can only recreate the same image with all parameters having the same value.
And even with that, it appears that photoshop showed some differences between two images created with totally the same values.
However the main human parameter is his ability to prompt and tweaks... And so unless you share your prompt and tweaks, your image is unique.

 

[XcentY]

Here is an example :
The image :

The prompt :

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

The Negative Prompt :
((3d, cartoon, anime, sketches)), (worst quality:2), (low quality:2), (normal quality:2), lowres, normal quality, ((monochrome)), ((grayscale)), bad anatomy, out of view, cut off, ugly, deformed, mutated, EasyNegative, paintings, sketches, (worst quality:2), (low quality:2), (normal quality:2), lowres, normal quality, ((monochrome)), ((grayscale)), skin spots, acnes, skin blemishes, age spot, glans,extra fingers,fewer fingers, "(ugly eyes, deformed iris, deformed pupils, fused lips and teeth:1.2), (un-detailed skin, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, anime:1.2), text, close up, cropped, out of frame, worst quality, low quality, jpeg artifacts, ugly, duplicate, morbid, mutilated, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neck"
The seed :
2934841804
The steps :
50
the sampler :
pm++_sde_karras
the unique model hash :
a60cfaa90d
The Guidance scale :
7



So Unless You have all these same parameters, you won't have the same unique image.
So tell me, if the creator didn't share this information (his prompt, his negative prompt and all his settings), would you be able to create the same image like it's a search engine ?

 

[Meaning Less]

and you can say that it's by mixing images but it's not the way it works...

You just said the same thing with extra steps, during the training process you effectively disassemble all the training data into non-visual data that the AI can later replicate.

But after trained, once you press "generate" the AI isn't really "creating" anything as much as it is simply rendering a scene that it was trained to render when given that specific seed.

And I'm not saying you can't call that stuff art, I mean accidentally kicking a can of paint over a canvas can also be called art.

However I do feel like many people are overestimating the value of deep learning in general in its current state.

So Unless You have all these same parameters, you won't have the same unique image.

Sure, the odds are pretty low but there is always a chance depending on how big/small the training dataset was for it to happen.

However you don't even need the exact same image, that's the thing, even by just looking at it I feel like I've seen several similar characters to that one, in fact I just browsed that gallery and found this image:
Same nose, same mouth, same angle, very similar shading and face. You can see how both of those images aren't that far off even though the seed is completely different.

You must be registered to see the links

So yeah, the more people use it the more people will start noticing patterns and how limited it actually is.

 

[TranscendentThots]

Honestly, guys, this entire branch of humanity's tech tree is a dead-end. It's destined to be eventually superseded by something similar but with more precision and control.

Where hand poses are not only natural, they're deterministic. Where facial expressions are keyframed, not random. Where the specific details of all the ways this purple sweater is different from every other purple sweater are dictated by the seed projected against a meticulously indexed possibility space of all previously seen purple sweaters, not random noise, and posed using the un-transformed world coordinates of a 3D model representing the platonic ideal of the human form, not a 2D image of a color-coded stick figure.

But in order for OpenAI to ever have a motivation to develop free AI with those capabilities, their existing projects need to become unprofitable within the scope of their donation-based business model.

That basically can't happen as long as porn consumers value surface over substance/story, and fan works of copyrighted characters over original IP. As long as tech bros value what they imagine somebody way smarter than them might one day do with the technology, over the obvious existing limitations of the technology staring them in the face.

 

[XcentY]

You just said the same thing with extra steps, during the training process you effectively disassemble all the training data into non-visual data that the AI can later replicate.

But after trained, once you press "generate" the AI isn't really "creating" anything as much as it is simply rendering a scene that it was trained to render when given that specific seed.

And I'm not saying you can't call that stuff art, I mean accidentally kicking a can of paint over a canvas can also be called art.

However I do feel like many people are overestimating the value of deep learning in general in its current state.


Sure, the odds are pretty low but there is always a chance depending on how big/small the training dataset was for it to happen.

However you don't even need the exact same image, that's the thing, even by just looking at it I feel like I've seen several similar characters to that one, in fact I just browsed that gallery and found this image:
Same nose, same mouth, same angle, very similar shading and face. You can see how both of those images aren't that far off even though the seed is completely different.
View attachment 2564155

You must be registered to see the links

So yeah, the more people use it the more people will start noticing patterns and how limited it actually is.

First calculate the odds...
Then do the same (calculate the odds) between two human artists that could draw something similar in the world...

And if you think that the learning process of human brain is quite different... How do you remember an event and get an image in your brain from this event ??? Don't you think that your brain will reconstruct a distorded image of the event in your brain ? A blurred image of this memory because let's be honnest, what your eyes have seen, what you've heard, ... has not be engraved in your brain as a "real image" or a "real sound"... And when you try to remember, the brain does exactly the same construction as an AI... It reconstructs your memory based on pieces of data which are no more "sound" or "image".

You must be registered to see the links

If you think that Billions parameters give a odd that can be estimated as "there is a chance"... I tell you there is also a chance to get the same drawing from 2 different artists...

That's because you so want to dismiss this art for "technical" reasons... while it's craftmanship behind it is art by essence... Sorry but we'll have to agree to disagree because for me AI generative is well producing art ...