Kickstarter bans AI-generated art enthusiast group "Unstable Diffusion" and refuses to deliver their successfully raised $56k funds

[Avocado Diaboli]

2 minutes ago, NumLock21 said:

You can't really copyright shapes and such, that looks more like inspiration, rather than stealing like a direct 1:1 copy of that art style.

That doesn't answer my question. Where's the difference in me manually copying a style and calling it "inspiration" and the AI doing the same with my prompt serving as the inspiration?

 

3 minutes ago, NumLock21 said:

So the payment is given to the AI, does the AI have a bank account, Paypal, Venmo, etc where the payment goes directly to it, instead of the human?

No, the payment is given to the owner of the hardware the AI runs on. How did you get to anthropomorphizing AI from my response?

[NumLock21]

3 minutes ago, Avocado Diaboli said:

That doesn't answer my question. Where's the difference in me manually copying a style and calling it "inspiration" and the AI doing the same with my prompt serving as the inspiration?

With a human they can add their own twist on things based on the inspiration of the artist they got it from. With an AI, they just spit out the exact style of that particular artist.

 

5 minutes ago, Avocado Diaboli said:

No, the payment is given to the owner of the hardware the AI runs on. How did you get to anthropomorphizing AI from my response?

So you're paying the human for letting you use their AI to create the work based on the style of an artist, rather than paying the actual artist for their style of work?

[Avocado Diaboli]

9 minutes ago, NumLock21 said:

With a human they can add their own twist on things based on the inspiration of the artist they got it from. With an AI, they just spit out the exact style of that particular artist.

I can also add my own twist to an AI generated image. I can also not do that and instead copy it verbatim if I draw it myself. My father owns copies of oil paintings by Salvador Dali, commissioned to someone who meticulously studied his style. They are exact replicas of the original paintings, no added twist. He got paid for his efforts to create those oil paintings. If that's not theft, then claiming an AI doing so is neither.

 

9 minutes ago, NumLock21 said:

So you're paying the human for letting you use their AI to create the work based on the style of an artist, rather than paying the actual artist for their style of work?

Yes. I also paid HP to print my Word document instead of the 15th century scribe to manually write it down for me. Instead of trying to lay traps with your ill-considered attempts to goad a reply you can pounce on, I suggest you try to form coherent arguments yourself. To me it seems obvious that you have hard time coming up with arguments for why your arbitrary definitions and objections hold up to scrutiny.

[NumLock21]

26 minutes ago, Avocado Diaboli said:

I can also add my own twist to an AI generated image. I can also not do that and instead copy it verbatim if I draw it myself. My father owns copies of oil paintings by Salvador Dali, commissioned to someone who meticulously studied his style. They are exact replicas of the original paintings, no added twist. He got paid for his efforts to create those oil paintings. If that's not theft, then claiming an AI doing so is neither.

 

Yes. I also paid HP to print my Word document instead of the 15th century scribe to manually write it down for me. Instead of trying to lay traps with your ill-considered attempts to goad a reply you can pounce on, I suggest you try to form coherent arguments yourself. To me it seems obvious that you have hard time coming up with arguments for why your arbitrary definitions and objections hold up to scrutiny.

 

What does buying a HP printer have anything to do with hiring an 15th century scribe to manually write it down for you, you're the one that wrote the document, you can use whatever tool to help you print out your work, be it the HP printer or hire that 15th century scribe. So based on your logic, if you have a particular way of drawing things, then it's okay for me to pay the creator of the AI to create an artwork based on your style, rather than pay you for your style of work?

[Avocado Diaboli]

6 minutes ago, NumLock21 said:

What does buying a HP printer have anything to do with hiring an 15th century scribe to manually write it down for you, you're the one that wrote the document, you can use whatever tool to help you print out your work, be it the HP printer or hire that 15th century scribe. So based on your logic, if you have a particular way of drawing things, then it's okay for me to pay the creator of the AI to create an artwork based on your style, rather than paying you for your style of work?

Let me turn it around on you, since you like to loaded questions so much: Would it be ok for you to pay another artist than me to create them something in my signature style? If not, why?

[NumLock21]

4 minutes ago, Avocado Diaboli said:

Let me turn it around on you, since you like to loaded questions so much: Would it not be ok for you to pay another artist than me to create them something in my signature style? If not, why?

It's not their own style of work, it's yours, so you deserve the payment. If you're not interested in create the artwork for me, then I would have no choice to go out and find an alternative be it an AI or a person to draw based on your style and pay them for their service.

[Avocado Diaboli]

Just now, NumLock21 said:

It's not their own style of work, it's yours, so you deserve the payment.

So if I create a painting in the distinctive styles of Dali, Beksinski, Giger, Pollock, Basquiat, etc. if I hone my skill to perfection, do their estates deserve the payment for the images I create in their respective styles? We've really done a complete 180 here in this thread. You started out arguing that the mere act of creating itself is enough to make a distinction from real art to AI generated imagery. And now you're claiming that even if I personally pick up a brush, train the prerequisite 10'000 hours to master the skill and settle for copying the style of others, I don't deserve compensation for my creations? Do you finally see the cognitive dissonance in your stance here and why mine is intellectually congruent by comparison?

[NumLock21]

1 hour ago, Avocado Diaboli said:

snip

Where did I say you don't deserve the compensation for your work. Stop embarrassing yourself.

[Avocado Diaboli]

13 minutes ago, NumLock21 said:

Where did I say you don't deserve the compensation for your work. Stop embarrassing yourself.

You edited your post afterwards to add the disclaimer that someone painting in my style would deserve compensation for their work. And you transparently contradicted yourself with that edit. You can't say this:

 

1 hour ago, NumLock21 said:

It's not their own style of work, it's yours, so you deserve the payment.

and then immediately afterwards say this:

 

1 hour ago, NumLock21 said:

If you're not interested in create the artwork for me, then I would have no choice to go out and find an alternative be it an AI or a person to draw based on your style and pay them for their service.

and still maintain that "it's my style, I deserve payment" is true.

 

You are being patently dishonest here with your unmarked edits and your clearly unfocused way of arguing in order to not finally admit that I was right.

[NumLock21]

9 minutes ago, Avocado Diaboli said:

You edited your post afterwards to add the disclaimer that someone painting in my style would deserve compensation for their work. And you transparently contradicted yourself with that edit. You can't say this:

 

and then immediately afterwards say this:

 

and still maintain that "it's my style, I deserve payment" is true.

 

You are being patently dishonest here with your unmarked edits and your clearly unfocused way of arguing in order to not finally admit that I was right.

No point to continue to discuss with your way of thinking.

[mr moose]

18 hours ago, Avocado Diaboli said:

 

But it doesn't support your position at all. Your claim was that comparisons are moot because technology evolves too fast for the comparisons to stick. I gave you an example where the parameters are comparable, a disruptive technology that had severe economic implications for those it made redundant and their reactions to that. They are identical, regardless of how fast the development was. If you really want to make a point here, you need to be a bit more verbose why exactly my specific example of the printing press is ill-suited as an illustrative point for the current conundrum.

 

 

My position is that you can't take one event in history and extrapolate it out to argue against or for another event no matter how many things seem similar.

 

 

There are two issues pertaining to the discussion, the first is the concept that comparing the evolution of anything at any one given point in time is enough to argue against the evolution of another.  This is not as relative as you might think,  I only addressed the time scale as being a problem, but there is also economic conditions which have changed greatly since then, there is quality of life and access to services which have changed greatly since then also.    Introduction of the printing press literally took decades to become anything sizable enough to cause job losses.  AI has taken less than a decade and is occurring in a time where social welfare is orders of magnitude better. 

 

Put simple the poorest of advanced nations are living much better lives than even the richest people of the 15th century and mainly it's due to technology that reduces the need for workers.   We simply cannot make a direct comparison of Device A in 1440 versus device B in 2022.    

 

 

[Avocado Diaboli]

54 minutes ago, mr moose said:

My position is that you can't take one event in history and extrapolate it out to argue against or for another event no matter how many things seem similar.

 

 

There are two issues pertaining to the discussion, the first is the concept that comparing the evolution of anything at any one given point in time is enough to argue against the evolution of another.  This is not as relative as you might think,  I only addressed the time scale as being a problem, but there is also economic conditions which have changed greatly since then, there is quality of life and access to services which have changed greatly since then also.    Introduction of the printing press literally took decades to become anything sizable enough to cause job losses.  AI has taken less than a decade and is occurring in a time where social welfare is orders of magnitude better. 

 

Put simple the poorest of advanced nations are living much better lives than even the richest people of the 15th century and mainly it's due to technology that reduces the need for workers.   We simply cannot make a direct comparison of Device A in 1440 versus device B in 2022.    

That's patently nonsense, given the direct similarities between the reactions from the people affected by the economic realities. By your logic, you can never make any comparison between any event in history and to take this into the extreme, you can actually never make any comparison whatsoever, regardless of context. Because if your core argument is that "not all parameters of two events line up perfectly, therefore you cannot make comparative statements" then you can only compare like for like. And comparing like for like is pointless, because they are by definition identical. Congratulations, you've gotten rid of the entire concept of comparisons. 

 

And I know that you think this would be a ludicrous end point to reach from your stated stance. But that's the reality of it. And it seems to me that this is motivated by a misguided attempt to silence anyone critical or supportive of anything that has parallels across history by pretending that their insights don't matter. Because this is in my experience a typical vector for dismissing arguments, attack their form, not their content.

 

But that's broad terms. Let's get back to the actual discussion. So let's put your arguments to the test, starting with the economic realities. You claim even the poorest live better than the richest in the 15th century and that this aspect is meaningful to the comparison. That hasn't stopped artists and programmers from crying foul against these new tools for fear of losing their jobs. So the economic reality is still that the people condemning AI tools do for the same reason monks condemned the printing press in the 15th century or the luddites condemned textile machines in the 19th century.

 

This is also regardless of the timespan necessary to cause significant job loss. Because again, the motivation behind the criticism is still the same. And this is why you're wrong about focusing on these parameters which are different. In a comparison, you weigh up similarities and differences and zero in on if one side outweighs the other. And since I've demonstrated now that neither the economic conditions nor the timespan had any impact on the outcome of "new disruptive technology; outrage from people being made redundant by it", it still stands as a valid comparison.

 

To get back to that point of attacking form and not content, you've been busy coming up with reasons to dismiss comparisons outright and this has been the only time you've made any semblance of a point to actually refute the comparison based on its own merits. And I think I've demonstrated that these objections are not worthy of consideration. Were these your best examples of why the comparison shouldn't stand? Because if you end up responding, let me urge you to back up this notion with your strongest cases. Otherwise I consider this point belabored enough, the comparisons are valid.

[Holmes108]

Seems obvious to me that the very act of a computer creating art shouldn't be an issue in and of itself. But we obviously need to refine it so that it's not just spitting out images of Mickey Mouse, or paintings that are 90% similar to an existing painting. At least not in a public space. If you wanted to have your PC do such a thing for personal use, I don't see the issue.

 

The idea of copywriting a 'style' is silly to me in general (how narrow or broad we go with the term 'style' will greatly complicate that issue though), but we obviously need to enforce some copyright protections on images the PC spits out. But it's no different than having to do the same with human artists. It'll be more difficult, due to the volume of art that can be produced with that process. But the principle is the same.

 

In fact, that's all we can do. What else is there to discuss really. Banning PC art? That's obviously not realistic, so all you can do is police the output as you would a human. 

[mr moose]

22 hours ago, Avocado Diaboli said:

That's patently nonsense, given the direct similarities between the reactions from the people affected by the economic realities.

The reactions of the people can be the same, but that doesn't mean that the problem is the same nor does it mean the problem is to the same degree of scale.

 

22 hours ago, Avocado Diaboli said:

By your logic, you can never make any comparison between any event in history and to take this into the extreme, you can actually never make any comparison whatsoever, regardless of context. Because if your core argument is that "not all parameters of two events line up perfectly, therefore you cannot make comparative statements" then you can only compare like for like. And comparing like for like is pointless, because they are by definition identical. Congratulations, you've gotten rid of the entire concept of comparisons. 

I don't argue they have to line up perfectly, but they do need to be at least the same scale of effect over time.     Again you are ignoring a plethora of differences between the effects of the development of the printing press and the effects of the development of AI.   

 

22 hours ago, Avocado Diaboli said:

 

And I know that you think this would be a ludicrous end point to reach from your stated stance. But that's the reality of it. And it seems to me that this is motivated by a misguided attempt to silence anyone critical or supportive of anything that has parallels across history by pretending that their insights don't matter. Because this is in my experience a typical vector for dismissing arguments, attack their form, not their content.

I don;t even know why you would think half of that.  I'm not trying to silence any criticism, I am merely pointing out that you are being rather superficial by arguing that AI today will have the same effect that printing or power looms had 300 years ago.  You are completely ignoring the fact that our economy and society is so different now that it simply cannot have the same effect no matter how much you think it will.

 

22 hours ago, Avocado Diaboli said:

But that's broad terms. Let's get back to the actual discussion. So let's put your arguments to the test, starting with the economic realities. You claim even the poorest live better than the richest in the 15th century and that this aspect is meaningful to the comparison. That hasn't stopped artists and programmers from crying foul against these new tools for fear of losing their jobs. So the economic reality is still that the people condemning AI tools do for the same reason monks condemned the printing press in the 15th century or the luddites condemned textile machines in the 19th century.

People cry foul over everything from corporate regulation to how much they should be paid.   Just because people complain doesn't mean the issue is just as bad as some other issue that people also complain about.  Hell people in the Australia complain about the price of power while Ethiopians complain about not having food, does that mean they are effected by the cost of power in exactly the same way that an Ethiopian is effect by the lack of food? Of course not,  We live in a completely different society where we can afford a price hike in power and still have a house and food.  Ethiopians cannot.     All comparisons have to be qualified,  otherwise you are just pulling events from history with different levels of effect and trying to argue they are the same.

 

 

22 hours ago, Avocado Diaboli said:

This is also regardless of the timespan necessary to cause significant job loss. Because again, the motivation behind the criticism is still the same. And this is why you're wrong about focusing on these parameters which are different. In a comparison, you weigh up similarities and differences and zero in on if one side outweighs the other. And since I've demonstrated now that neither the economic conditions nor the timespan had any impact on the outcome of "new disruptive technology; outrage from people being made redundant by it", it still stands as a valid comparison.

You haven't actually demonstrated anything.  the concept that the length of time something takes to develop (how long people have to see it coming and make adjustments) and the economic conditions under which they live are moot in comparing these things is horribly flawed.    If I lost my job tomorrow, I would not lose my house, my car or my ability to eat, I can find another job within weeks if not days.  That is just not how it was in the 15th to 18th century.    Hell, you only have to compare workhouses (18th century version of social welfare) to today's social welfare to see that things are not the same nor are they directly comparable. 

 

22 hours ago, Avocado Diaboli said:

To get back to that point of attacking form and not content, you've been busy coming up with reasons to dismiss comparisons outright and this has been the only time you've made any semblance of a point to actually refute the comparison based on its own merits. And I think I've demonstrated that these objections are not worthy of consideration. Were these your best examples of why the comparison shouldn't stand? Because if you end up responding, let me urge you to back up this notion with your strongest cases. Otherwise I consider this point belabored enough, the comparisons are valid.

Believe whatever you want.  You are not comparing apples to apples.

 

You are literally trying to argue that economic conditions and technological development speed are the same today as they were 400 years ago.  If you need a white paper to explain why that is wrong then I don't know what to to tell.

[jagdtigger]

19 hours ago, Holmes108 said:

But it's no different than having to do the same with human artists.

Seems like someone didnt read the previous posts. Its not the same, not even remotely.

[Holmes108]

55 minutes ago, jagdtigger said:

Seems like someone didnt read the previous posts. Its not the same, not even remotely.

Sure i did, and it's rude to suggest otherwise because you happen to disagree with my opinion. I'm simply saying that I think it has to be the same, because there's not much choice. All we can do is police the copyright aspect. 

 

But I feel as long as it isn't putting out blatant ripoffs into the public as it's own,  it's a non issue. When it is, it gets dealt with accordingly. How blatant is too blatant? It'll be tricky sometimes ...

 

The same with human artists. 

[Kisai]

1 hour ago, Holmes108 said:

Sure i did, and it's rude to suggest otherwise because you happen to disagree with my opinion. I'm simply saying that I think it has to be the same, because there's not much choice. All we can do is police the copyright aspect. 

 

But I feel as long as it isn't putting out blatant ripoffs into the public as it's own,  it's a non issue. When it is, it gets dealt with accordingly. How blatant is too blatant? It'll be tricky sometimes ...

 

The same with human artists. 

 

There are two angles on both sides of the problem

 

On the input side "this has stolen artwork in it", no, it does not. The model contains weights, not samples. It can not re-create any of the training data exactly, it can only create "noise" that resembles the training data in a coarse manner. Even then you'll have a hard time finding what input is needed to recreate something resembling any specific training sample.

 

On the output side "this has stolen my style", well you can't copyright a style. You also can't copyright anything generated by AI.

 

On the input side, the "research" aspect of fair use can defend doing it, but on the output side, it can not, because if it's used commercially,  it's in violation of the spirit of the research use exception. 

 

So unless your output image is going to be purely of "fair use" exception itself (eg satire, parody, criticism, research) you pretty much CAN NOT use anything generated by AI, commercially. Because ethically it was not trained in a way that would be a cleanroom. It's trained directly off the data rather than a clean-room interpretation. 

 

That can however be done, just nobody has "done that", because AI technology isn't smart enough to do so. The English language contains at least 171,146 words not including loanwords from other languages in the Oxford English dictionary. with about 41,000 words we do not use anymore. Chinese has likely 500,000 words. So what if we had one research AI "train" off every word in the english and chinese dictionary with a common image. And then generate a new image, text and keywords for the "cleanroom" AI? That fits everything necessary to "reverse engineer" a copyrighted work as per IBM vs Compaq.

 

But....

 

Someone would have to check each and every image that it was unique, and not simply a resurfacing of a training image. That's a not insignificant amount of human time, and relies too heavily on people knowing ALL the words, in both English and Chinese.

 

But what about the other 7100 languages?

 

As you can see, there is no making a simple AI that could understand "any input query" The existing models only understand english, and even the LAION data collection only deals with data tagged in english. What needs to happen, instead of worrying about copyright, is instead creating "clean" training data that is correct and is correctly mapped in every language so that one model can can understand the input regardless of the language. 

 

We just are not there, and likely nobody will deem it financially viable to do so. Models will likely be generated in English for Western Europe and in Chinese for East Asia and generate two completely different results given the same query translated to the other language.

 

[Shreyas1]

AI generated art right now seems to have a lot of parallels to early photography, at least to me. At first you have people who dislike it because it seemingly takes away "skill" from art, but will probably evolve into an art form on its own.

Quote

Upon seeing the first daguerreotype around 1840, the French painter Paul Delaroche (1797-1856), declared: “From today, painting is dead.”

https://www.artandobject.com/news/today-painting-dead-photographys-revolutionary-effect

[Thaldor]

I don't think we need to go any further than coding, if your coding is that you bash together different github codes and call it a day, are you really a good coder? If al lyou can do is use search and patchwork stuff to work, are you really anything special? If AI was made that will do just the exact same, search code from github and slap them together and see if it works and if it doesn't change certain parameters that it does, why not should we use it rather than the coder who also consumes coffee?

 

What AI at least for now cannot do is come up with creativity and create something actually new. AI based code optimization is a thing but it also just merely goes around and tries known solutions and chooses which one is the fastest and that is closest to creativity AI can get, at least for now. AI cannot find the n:th solution that is unique and could wipe everything else off the table, AI won't be creating Wolfenstein 3D and coming up with the quirks to make real-time 3D-like graphics in the time when 3D was sci-fi concept.

 

Will we loose jobs because someone made an AI that can replace the row coders who are important but who have low chance to make anything really that important and, well to be frank, who cannot even compete with a machine in creativity? Yes and it's sad but that is life in the modern world but that has been pretty much known future for the "bad coders". Should we protect those works by setting artificial limits and restricting the development of the AIs? I don't think so, we ain't doing that anywhere else so why is "bad coders" any different? I didn't see anyone complaining when welder robots were made or CNC machines, those probably replaced way more jobs than coder AI.

 

The hard fact is, you don't need to be Linus Torvalds or Dennis Ritchie or John Carmack but it's also not good enough to be able to just slap together github codes and assets from Unity/Unreal Store, you ain't a good coder either. You don't need to be Picasso or Dali but if your original creativity is on the level of a machine copying images, you ain't that good artist either. It's harsh and painful but that's life and when it gives you lemons, it isn't enough to just make lemonade from them, you also need sugar, a lot of sugar.

[LAwLz]

On 12/27/2022 at 6:29 PM, NumLock21 said:

Using a mouse still requires hand movement. Take drawing a landscape as a example, with a mouse your hands have to actually draw the land, trees, sky, etc. With the keyboard you're just typing the word landscape and the program does it for you.

And using a keyboard doesn't require hand movements? Come on...

If the argument is "using a mouse requires hand movements but using a keyboard doesn't" then you are laughably wrong. You need hand movements to input data into a program regardless of if you use a mouse or a keyboard. 

And yes, with Photoshop you are just clicking a button and the program does thousands upon thousands of calculations for you and suddenly you got what you asked it to produce. Sounds like those people are also "lazy" and "cheating" by your definition, right? I don't think the people using Photoshop are doing any of the calculations necessary in their head. They just leave that part to the computer, those lazy wannabe-artists, right?

 

Also, you keep flip flopping. Sometimes "effort" is requires for something to be art, and sometimes it isn't. Which one is it?

 

 

 

On 12/27/2022 at 6:29 PM, NumLock21 said:

As long as you're doing most of the work manually, it's okay to use some help with the tools the program provides.

What do you define as "most of the work manually"?

When I take a picture with a camera, the computer is doing 99% (if not more) of the work automatically for me.

When I use Photoshop or some DAW program, the computer is doing 99% (if not more) of the work automatically for me.

 

We mathematically knows this to be true because if we start counting all the mathematics that goes on in the background as "work", which is what you are doing as soon as AIs gets brought up, then I as the user of the program is not doing any of that in my head. 

This is code similar to what Photoshop does when I press the export button. It's 2000 lines of code, not including the external libraries. I do not do any of this work when I click the export button. I just click a button on my computer and it does thousands upon thousands of calculations and then spits out the image I want. No math necessary. How is that any different from the AI image generator? In both cases we humans gives some vague inputs in some form (mouse movements, text prompts, config settings, audio inputs, etc), ask a program to run a bunch of math on our behalf (which is over 99% of the actual work) and then it spits out the result we want (hopefully). If it doesn't match what we want then we try and tweak the input.

 

 

Also, you didn't answer my questions:

If I work in Photoshop using macros and not a mouse, is that no longer art?

Is it "lazy" to use Photoshop and let a computer do all the hard math that goes into the functions, rather than do the math myself?

Is it lazy to let a camera decide which settings to adjust to get a good exposure?

Is it "lazy" to press a button to signal to the ISP to capture an image using complex neural networks executing in the background in the imaging pipeline?

 

I want you to give a yes or no answer to those examples. 

By the way, we can actually mathematically quantify how much work a human does by examining the software in some of these cases, because we can look at how big of a portion the input (the only thing a the human contributes) is compared to the rest of the work that is being done by the computer. If a human changes 2 variables in two lines of code, and the overall program is 10,000 lines of code with 1000 variables, then it is very hard to argue that the human contributed more than a fraction of a percent of the work made.

 

 

On 12/27/2022 at 6:29 PM, NumLock21 said:

The problem isn't artist not liking AI generated art. The problem is art-styles done by actual artist that gets stolen, where it's feed into AI, and random people can just generate that style of art, and then go and sell it for profit. The other problem is people using AI to generate an image from a specific style of art, and then paying the creator of the AI program, while the actual artist gets nothing.

How is this any different from a random human person copying someone's style of art (which everyone does, all the time, this is not up for debate)?

Are you arguing that art styles should be copyright protected? Because if that is your argument (which it seems to be) then we will have a really big issue with for example everyone creating images in the style of manga. They 100% copied that, and everyone who draws manga-style images for profit are profiting from someone else's style. Why do you not have any issue with humans copying and profiting from someone's style, but when an AI does it then it's suddenly bad?

[Arika]

On 12/28/2022 at 4:29 AM, NumLock21 said:

The problem isn't artist not liking AI generated art. The problem is art-styles done by actual artist that gets stolen, where it's feed into AI, and random people can just generate that style of art, and then go and sell it for profit.

Depends on the pretense that it's being sold as. If it's being sold as "this is <artists> work", then that is actually art theft, nothing to do with copyright or AI. People who spend their livelihoods studying and making recreations of famous works, but they don't sell it as "this is an original work by this artist".

if it's sold as "this is an AI art piece that is mimicking <artist>" then what's the problem with that? it's not being misattributed or being touted as something it's not.

 

I have trained several of my own SD models on my own art and style, which in turn has been very heavily inspired by 1 or 2 artists i follow who would have copied the style of someone else. So who do i need to get permission from to make pictures with my models?

 

Me, because i tried the model with my art?

Artist's i copied the style of?

Artists that the artists i copied the style of copied the styles of?

[NumLock21]

25 minutes ago, LAwLz said:

snip

I've already said stated what I've said. Can't really help with your over thinking.

1 minute ago, Arika S said:

snip

Customer who buys them probably won't even know who the original artist is, unless they have been following that person's work and see something they recognize immediately.

 

[Arika]

12 minutes ago, NumLock21 said:

Customer who buys them probably won't even know who the original artist is, unless they have been following that person's work and see something they recognize immediately.

 

then why does it matter? if it's being sold as AI artwork, and the customer doesn't know who the original artist is so it's not like that would influence the buyer.

[divito]

26 minutes ago, NumLock21 said:

I've already said stated what I've said.

Well, there is more than one person outlining that you keep saying contradictory things. It's wonderful that you've "stated what you've stated," but that doesn't help anything.

Humans have copied or assimilated styles of artwork for centuries, and while there have been issues of forgery or copyright issues, the art industry still advances forward with millions of artists drawing similar things, and those that are more unique with little attribution to how they learned or who inspired them. It's a very wide spectrum of talent, creativity, vision and entertainment.

An AI copying/assimilating examples of artwork is faster and more efficient than a human, but it doesn't change the core mechanics at play. If you have a problem with one, logically, you must have a problem with the other. And under that logic, new artists basically cannot happen under your worldview; there are only so many geometric shapes, shading options, or ways to draw an eye. Do you know the name of the first person to draw an eye? How should they be attributed or compensated? Can you trace the lineage of all the cave art that exists in the world to ensure no artist is infringing?

[Kisai]

5 hours ago, Arika S said:

Depends on the pretense that it's being sold as. If it's being sold as "this is <artists> work", then that is actually art theft, nothing to do with copyright or AI. People who spend their livelihoods studying and making recreations of famous works, but they don't sell it as "this is an original work by this artist".

if it's sold as "this is an AI art piece that is mimicking <artist>" then what's the problem with that? it's not being misattributed or being touted as something it's not.

 

I have trained several of my own SD models on my own art and style, which in turn has been very heavily inspired by 1 or 2 artists i follow who would have copied the style of someone else. So who do i need to get permission from to make pictures with my models?

 

Me, because i tried the model with my art?

Artist's i copied the style of?

Artists that the artists i copied the style of copied the styles of?

Depends what you intend to do with it.

 

Let's say... a popular artist just drops dead in 5 minutes. And because you have a model that was trained on their works, but also your work, you could "create new" works in their style or your style.

 

From an ethical point of view, selling or mis-attributing your AI-generated works as theirs would be unethical. However if you simply attribute it to yourself with no mention of the other artist, because the model does not actually contain their artwork, there's really nothing you have to worry about on the surface.

 

However, you will get called out for it, if you don't mention it was AI generated, because you will be unable to replicate "your style" or "their style" consistently.

 

This is a problem with all all ML presently. Giving the model more data doesn't always make it better It's still going to be a 4GB model regardless if it was trained on a single 1MB image or 1 billion 1MB images. The more input data it has, the less likely it's going to overfit an image in a way that will result in it resurfacing, unless that image is unique and shows up a lot (such as public domain paintings.)

 

"Dogs playing poker"

 

Dalle E, produces some ... bad results.

 

Google Images shows us at least 5 of the same image.

 

 

"Dogs playing Poker" in LAION

 

Most people know of "dogs playing poker" is this image:

"A Friend in Need (1903)"

 

None of the images known as "dogs playing poker" are actually called "dogs playing poker" But at least 5 images in google images are this image. Those online stores? Between 8$ and 600$

 

Would an AI devalue the original painting? No. The fact that there are multiple stores selling the exact same image shows you how little people value "unique" and "collectable" stuff. If you really want that image you can just rip it off Wikipedia.

 

However if you trained the AI, see above, those are not "paintings" of dogs, those are photos.

 

In fact, take a closer look:

Wow those eyes suck. And what is that on the right? Avacado dip? vomit?

That mashup of a dog on the right is hilarious, it's as if the AI glued three different dogs together. Also, those aren't poker cards.

 

The eyes on this one are just straight up "wtf" Also they're supposed to be playing poker, but this looks more like one dog shuffling cards. What's that dog in the middle doing? Making drinks telepathically?

 

Again, more dogs that look like they're three dogs glued together, the two in the front look like they don't even belong in the same scene, they look "chroma-key'd"  But at least those foreground dogs look like dogs.

 

Letting an AI only learn from public domain materials means it does not learn "new" things, because new public domain sources don't exist or are incorrectly licensed. You can't rely on how things are tagged.

 

And because of how people have been trained to abuse tagging and keyword systems to raise their discovery in search engines, has now come back to bite them with AI using "their" spammed results.

 

Assuming I wanted the AI to generate a billion images of "dogs playing poker" I could overwhelm Google images with it, and then an AI, not knowing it's also AI generated, might learn these "new" dogs playing poker and making discovery of the PD source disappear entirely.

 

That is the real risk from not mandating that AI art be tagged as AI generated artwork, because it may directly result in burying discovery of the sources used.