This is also a problem with translation apps. They have to have a dataset of real translations done by a human. But that human translator doesn’t get paid for their work being in a dataset.
7:42 This is also whats expected to happen when you freeload databases, all the morally dubious things against artists aside. A currated database, where people could opt in and be compensated, feels like a pipedream.
I think the thing about the current state of plagiarism/art theft is that it's done by individuals, as soon as a corporation starts plagiarising there's more of a public outcry. So this is a for-profit company that can "steal" art and rehash it with any person on it. I think the more niche the input, the less likely the output is transformative and therefore covered under free use
as far as the "real artist" thing goes, I defiinitely feel like the transition we're approaching is similar to recorded music. give it a decade at most and it will take hardly any skill to get the output you want, not much harder than pressing play. right now we're at the level of a self-playing piano. knowing how to play piano is not a requirement, but definitely helpful to get the output just right. but anyone can feed a paper reel in and get something passable.
Quick question: you mentioned stable diffusion was developed by hugging face. Wasn't stability ai responsible for this model? I was under the impression that hugging face just made a popular hosted instance of the first release of the model.
Hey Jordan, I actually fairly recently conducted a study within the art community in my country about how artists felt about AI art. And although we found that many carried uncertainty and fear towards the technology, many also expressed an interest in using it for references or entertainment uses. I think this video has been spot on with our findings! And I'm interested to see what AI art brings in the future and how well it could develop to tackle the issues or potentially completely ignore them. TL;DR: Great video!
As an artist, AI can create something I make in 10 hours and do it in 1 minute. TIME IS MONEY. This cheapens the value of illustrators, and portrait artists like myself to the general public. My brother just straight up downloaded lensa A.I and didn't wait for me to do his commission this week. I also watched this guy on Instagram get 22k followers in 1 month, is it envy? Ofcourse. All the hard work we put it is reduced to a mere toy. This artist on instagram says in his bio, that he's "not an artist" and even put a store up, and gets comments from people as if he were an artist. Doesn't matter if we have the same process as AI art or not. the fact is the end result is equivalent if not BETTER than us hard working artist. Will I stop making art? No, I love digital art, ut it's gonna be hard for a lot of us.
"It's very rarely 1-to-1 inspiration." Hi, artist drowning in this problem, but I can kind annecdotally claim you're sort of wrong? ALOT of AI art theft often directly name currently living artists all the time. Greg Rutowski is a common one, his name basically became a trend amongst AI art prompt writers because he does a very good high fantasy high realism style, and the fact that a lot of these AI systems could see his name, and recognize his name as part of their data collections, and then pretty much just use his art for the generations? Alot of it does become 1-to-1 because the prompt writers are pushing it to directly take from specific artists. Not 100% 1-to-1 like you're saying, but I think it's VERY relevant to that statement.
Apparently Art Station has a huge boycott against AI art at the moment. I think that may be a little extreme, but there definitely are some concerning aspects.
Lots of artists use the termed generative images to differentiate it from real art. Which by Oxford definition requires it be *made* by a human. I would say the code written by humans to generate the images has more right to be called art than the images it outputs.
Out of curiosity I asked the AI chatbot whether or not artists should be compensated for their art being used in AI image generator datasets, and the chatbot said that artists should have to have the ability to be compensated and give consent. I really don't think it is that much of an ask for artists to have the right to consent to this, especially when our art is being used to create profit. I agree that a shutterstock situation would probably be best, so that way artist can be intentional about what art they submit and be compensated when elements of their art are used in AI outputs. It seems like a pretty obvious solution and tbh I am really kinda appalled that this wasn't applied sooner and yet these AI image generators are still making profits off of our labour and skills. I feel like if AI itself can reason that it is unjust to use someone's creation without permission- it should be ubiquitous consensus.
A lot of arguments I've seen from artists against AI generated images are kind of just nonsense. A lot of them fear that this sort of thing is equivalent to "tracing," which is wild because the images being generated are in fact novel. You can say an awful lot about them, but of all things, they really aren't traced. Even though many are obviously derivative, I also reject the idea of ownership of specific styles or techniques. As a digital artist myself, I'm not very concerned with anyone trying to replicate my style, whether that's through their own hard work or by running a corpus of my art through a computer. What concerns me more is the potential for commission scams that comes with AI generated images. I haven't seen much concern about scams yet, but I expect that it's only a matter of time before people start trying to impersonate popular artists on the Internet, take commissions, and then feed the commission prompts they receive through an image generator and hope for the best. Their victims, who don't scrutinize the social media profiles of the people they're giving money to, probably won't scrutinize the image they receive either. I'm surprised that I haven't seen direct evidence of this type of scam yet. However, I mainly interact with other furry artists. Furries are everywhere, but we're not the most popular artists, by a long shot. Maybe this type of scam has happened elsewhere. I'm convinced that it will show up because that just seems to be the nature of scams. Whenever something big happens, like an election, a disease spreading around, or a cool new kind of money, there will be bastards who will weave it into a flimsy narrative that's capable of tricking vulnerable people out of a few hundred bucks. Although I'm not particularly concerned about the results of image generation, I do like the sound of legal requirements for permission to use an image in the corpus that an AI will study. I'm sure that all sorts of people will violate that requirement, but maybe it's worth a shot. Since many artists are vocally against their work being used for AI generated images, a microstock-like system of safe images may help put their minds at ease and draw the attention of AI nerds away from them. Regarding the issue of racial bias, I think the matter of the AI lightening your skin a little is a reflection of bias in art that actual people create as well. Some of it, I think, looks stylistically appropriate, as though it were created with the delicate shades of pastels. In such a case, basically any character's complexion will be lighter than it is in real life. In other cases, though, lightened skin tones definitely stand out as, well, "afraid to paint black skin accurately." Many of the examples you showed looked like they were based on art styles that employ lighter values in general, and that was probably deliberate on the part of the developers. My shaky hypothesis is that over time they chose to limit their corpus to art styles that stood the least risk of scaring people. This would mean that very few of the generated images feature dark values prominently, so you're less likely to see creepy illusions in the background. I'm sure you've seen plenty of AI generated images by now that just have freaky looking artifacts in the darkness. But then there's the collateral damage of almost none of the images replicating realistic skin tones. I've been pretty consistently impressed by AI's ability to replicate realistic lighting, so I'd bet that with a better body of work containing more PoC subjects, this app could at least replicate your complexion just fine. No guarantees that it'll stop giving you three legs, crossed eyes, insane hair, or a super thin neck, though.
I’m not a lawyer, but, from a proportion perspective (one of the factors), if you average the size of the model (gigabytes) versus the number of images (billions), you end up with less than a byte per image in the model. So, in that area, you could argue they already have a good position to work from (it also could have implications from a compensation perspective). Some of the other factors are more difficult to analyze (effect on the market value for the original work is highly contested).
It's not a style transfer model. It's a text to image diffusion model. It's not made by HuggingFace. HuggingFace is just a repository. It's made by Stability and CompVis.
Great video. I hadn't really considered the Eurocentric nature of the images that were generated, so appreciated hearing them here. I was listening to the guy from Cold Fusion's podcast and they essentially were like "this is bad for artists, but this tech is the future so what can we do?" - I think we need to be better as a society of being able to do something, but putting guard rails on it to mitigate risk or even not do it at all. Everyone is so excited about these new tech but we do have a choice to slow down.
Rather than trying to stop progress (including computers being able to do things that only humans were able to do before, such as copying other humans' painting/drawing/sculpture styles) we always do best when we use that progress to better serve our needs. So, instead of folks who had been trying to get money for their content creation, we can simply ask them what they actually need, and find others who love serving those needs to help the content creators. In other words, we can work like a collaborative species, instead of a competitive one. We can do what we love, freely and joyfully, and whomever needs what we love creating can have it, freely. Like a real global multicellular organism, instead of a planet made up of single celled mini-organisms all acting like enemies.
@@ratzelkujo7880 Trying to control nature and the laws of physics (life) is an exercise in destruction and suffering. When you try to fight against evolution - be it genetic or memetic - you always lose. Either go with the flow of life, and flourish, or you will die, basically.
I've been making use of DALL-E with varying degrees of success. Maps tend to change style randomly. As for, paint a panda drinking boba in the style of DaVinci or van Gogh or Monet, I'm okay with that as the artists are long (enough) dead so their work is in the public domain (maybe not in every country; I am not an attorney and this is not legal advice), so I was okay posting them on social media. But I also generated the prompt in the style of Salvador Dali and Andy Warhol and I decided I wasn't comfortable posting those in public.
Hello, I recently came across a program that uses "AI" to fillin missing bits of a imagine, like if you have a 2D photo and use photoshop to isolate a portion and remove it can fill in the negative space with similar images. what was this program called?! I can seem to remember any help would be appreciated
Dall-E 2, I believe that feature is called in-painting. It also does out-painting, where you can use the AI to extend a painting beyond its original canvas and fill it up believably.
The UK allows data mining for research purposes also in the UK anything that is created by a robot is actually property of the robot so if you're in the UK and you're generating a image that image is actually the property of that robots so you have to go to their terms and conditions Which is different on different applications on my application it just says do whatever you like with them as long as you don't use them for commercial use so you can't be held responsible for the copyright of those images because they belong to a generator at least here in the UK
Ursula Vernon has used generated art, showing it a few times on Twitter and Tumblr, I think. Recently she generated an image and then used digital art tools to make it into something she liked much better, with a video of how she did it. Quite a while ago, she did a surreal series over a number of tweets and months (and who knows the current status) that I think was more inspired by the generated art than painted over it.
I've always had a problem with the term "Facebook is selling your data" They don't sell your data, they sell access to you and use your data do that. It's a wholly different thing and just saying "Facebook, tiktok, IG sell your data" is doing a disservice to your audience. Some apps sell your data but the big platforms very much do not.
But what work doesn’t depend on all past works. Even today’s artist whether painting or music all owe a huge debt to past generations of progress. Even ideas and science.
Great video! AI seems to be encroaching in all aspects of our lives. Curious to hear your opinions on chat api and its impact on software engineering. Is it a real threat?
I would say no as artist aren't physical frankensteining others artwork in the way Ai does. You might say i like their sense of color, or i like their line work etc, and maybe even study them to understand. However, like handwriting, people have their own natural tendecies and subject matter that they wanna focus on. So even if you study these artist if will rarely be a one to one. AIot of Ai is taking one to one images and photobashing them together to get tne work. Honestly in tje art community its usually best to not look too similar to one or two artist as you dont wanna be a copy . You wanna be you
@@imanispence9795 the latest AI doesn't just hammer images together though. it does create its own representations without lifting directly from other individual images
We don't. We have a thing called a visual library, and even while looking at a reference, we can never replicate the same image, as it's up to reinterpretation and iteration. The badly called AI isn't capable of this.
@@quetevalgavergaaa it seems you're not familiar with how this stable diffusion AI works. it does create by interpreting from its visual library, it does not replicate the input images or just copy from its database
@@peabody3000 I didn't say that how it works, but it's not comparable to what a human brain does. It's like in the 60s the analogy of a computer being similar to our brain, even though it isn't.
Putting images and other content on the public internet is, logically, giving consent for the stuff to be seen (used) by the observers. Sure, we could try to sue those who copy/share them in other places, based on local copyright laws, but the reality of communication is that everything observable in the universe "belongs" to everyone who observes them. This is a totally separate issue from "artists" and other content creators living well. It doesn't matter what you do in life, what your passion is, what kind of activities you enjoy doing, you deserve to get what you need to do it well, so that the world can flourish. Which is why the whole competitive point scoring games that we humans have tried to force upon ourselves (money, grades, votes, "likes", etc.) as a way to organize our freely available resources make the world fail, rather than flourish. Once we start using computer technology to measure and keep track of our real needs as Earthlings, both for inputting and outputting matter and energy, we'll start being able to use all of the beautiful, creative, genius, specialized work that we humans naturally love doing to take care of ourselves, instead of trying to harm ourselves (in competition).
This is also a problem with translation apps. They have to have a dataset of real translations done by a human. But that human translator doesn’t get paid for their work being in a dataset.
7:42 This is also whats expected to happen when you freeload databases, all the morally dubious things against artists aside. A currated database, where people could opt in and be compensated, feels like a pipedream.
I think the thing about the current state of plagiarism/art theft is that it's done by individuals, as soon as a corporation starts plagiarising there's more of a public outcry. So this is a for-profit company that can "steal" art and rehash it with any person on it. I think the more niche the input, the less likely the output is transformative and therefore covered under free use
The way I run to this like the microwave before it beep😮💨
good to see you here!
as far as the "real artist" thing goes, I defiinitely feel like the transition we're approaching is similar to recorded music. give it a decade at most and it will take hardly any skill to get the output you want, not much harder than pressing play. right now we're at the level of a self-playing piano. knowing how to play piano is not a requirement, but definitely helpful to get the output just right. but anyone can feed a paper reel in and get something passable.
Quick question: you mentioned stable diffusion was developed by hugging face. Wasn't stability ai responsible for this model? I was under the impression that hugging face just made a popular hosted instance of the first release of the model.
Hey Jordan, I actually fairly recently conducted a study within the art community in my country about how artists felt about AI art. And although we found that many carried uncertainty and fear towards the technology, many also expressed an interest in using it for references or entertainment uses. I think this video has been spot on with our findings! And I'm interested to see what AI art brings in the future and how well it could develop to tackle the issues or potentially completely ignore them.
TL;DR: Great video!
As an artist, AI can create something I make in 10 hours and do it in 1 minute. TIME IS MONEY. This cheapens the value of illustrators, and portrait artists like myself to the general public. My brother just straight up downloaded lensa A.I and didn't wait for me to do his commission this week. I also watched this guy on Instagram get 22k followers in 1 month, is it envy? Ofcourse. All the hard work we put it is reduced to a mere toy. This artist on instagram says in his bio, that he's "not an artist" and even put a store up, and gets comments from people as if he were an artist. Doesn't matter if we have the same process as AI art or not. the fact is the end result is equivalent if not BETTER than us hard working artist.
Will I stop making art? No, I love digital art, ut it's gonna be hard for a lot of us.
But that is exactly what portrait painters said about photographers back in that days.
"It's very rarely 1-to-1 inspiration." Hi, artist drowning in this problem, but I can kind annecdotally claim you're sort of wrong? ALOT of AI art theft often directly name currently living artists all the time. Greg Rutowski is a common one, his name basically became a trend amongst AI art prompt writers because he does a very good high fantasy high realism style, and the fact that a lot of these AI systems could see his name, and recognize his name as part of their data collections, and then pretty much just use his art for the generations? Alot of it does become 1-to-1 because the prompt writers are pushing it to directly take from specific artists. Not 100% 1-to-1 like you're saying, but I think it's VERY relevant to that statement.
Apparently Art Station has a huge boycott against AI art at the moment. I think that may be a little extreme, but there definitely are some concerning aspects.
Lots of artists use the termed generative images to differentiate it from real art. Which by Oxford definition requires it be *made* by a human.
I would say the code written by humans to generate the images has more right to be called art than the images it outputs.
Thanks so much!
Both helpful yet dubious, an interesting sticky wicket.
Great stuff.
Have you tried Mid journey yet?
Out of curiosity I asked the AI chatbot whether or not artists should be compensated for their art being used in AI image generator datasets, and the chatbot said that artists should have to have the ability to be compensated and give consent. I really don't think it is that much of an ask for artists to have the right to consent to this, especially when our art is being used to create profit. I agree that a shutterstock situation would probably be best, so that way artist can be intentional about what art they submit and be compensated when elements of their art are used in AI outputs. It seems like a pretty obvious solution and tbh I am really kinda appalled that this wasn't applied sooner and yet these AI image generators are still making profits off of our labour and skills. I feel like if AI itself can reason that it is unjust to use someone's creation without permission- it should be ubiquitous consensus.
AI didn't reason that. It pulled the information from something someone else said on the subject and paraphrased it.
A lot of arguments I've seen from artists against AI generated images are kind of just nonsense. A lot of them fear that this sort of thing is equivalent to "tracing," which is wild because the images being generated are in fact novel. You can say an awful lot about them, but of all things, they really aren't traced. Even though many are obviously derivative, I also reject the idea of ownership of specific styles or techniques.
As a digital artist myself, I'm not very concerned with anyone trying to replicate my style, whether that's through their own hard work or by running a corpus of my art through a computer. What concerns me more is the potential for commission scams that comes with AI generated images. I haven't seen much concern about scams yet, but I expect that it's only a matter of time before people start trying to impersonate popular artists on the Internet, take commissions, and then feed the commission prompts they receive through an image generator and hope for the best. Their victims, who don't scrutinize the social media profiles of the people they're giving money to, probably won't scrutinize the image they receive either.
I'm surprised that I haven't seen direct evidence of this type of scam yet. However, I mainly interact with other furry artists. Furries are everywhere, but we're not the most popular artists, by a long shot. Maybe this type of scam has happened elsewhere. I'm convinced that it will show up because that just seems to be the nature of scams. Whenever something big happens, like an election, a disease spreading around, or a cool new kind of money, there will be bastards who will weave it into a flimsy narrative that's capable of tricking vulnerable people out of a few hundred bucks.
Although I'm not particularly concerned about the results of image generation, I do like the sound of legal requirements for permission to use an image in the corpus that an AI will study. I'm sure that all sorts of people will violate that requirement, but maybe it's worth a shot. Since many artists are vocally against their work being used for AI generated images, a microstock-like system of safe images may help put their minds at ease and draw the attention of AI nerds away from them.
Regarding the issue of racial bias, I think the matter of the AI lightening your skin a little is a reflection of bias in art that actual people create as well. Some of it, I think, looks stylistically appropriate, as though it were created with the delicate shades of pastels. In such a case, basically any character's complexion will be lighter than it is in real life. In other cases, though, lightened skin tones definitely stand out as, well, "afraid to paint black skin accurately."
Many of the examples you showed looked like they were based on art styles that employ lighter values in general, and that was probably deliberate on the part of the developers. My shaky hypothesis is that over time they chose to limit their corpus to art styles that stood the least risk of scaring people. This would mean that very few of the generated images feature dark values prominently, so you're less likely to see creepy illusions in the background. I'm sure you've seen plenty of AI generated images by now that just have freaky looking artifacts in the darkness. But then there's the collateral damage of almost none of the images replicating realistic skin tones.
I've been pretty consistently impressed by AI's ability to replicate realistic lighting, so I'd bet that with a better body of work containing more PoC subjects, this app could at least replicate your complexion just fine. No guarantees that it'll stop giving you three legs, crossed eyes, insane hair, or a super thin neck, though.
I wonder if a version of this dataset that does "properly" fit under fair use can be made.
I’m not a lawyer, but, from a proportion perspective (one of the factors), if you average the size of the model (gigabytes) versus the number of images (billions), you end up with less than a byte per image in the model. So, in that area, you could argue they already have a good position to work from (it also could have implications from a compensation perspective). Some of the other factors are more difficult to analyze (effect on the market value for the original work is highly contested).
It's not a style transfer model. It's a text to image diffusion model. It's not made by HuggingFace. HuggingFace is just a repository. It's made by Stability and CompVis.
exactly! Quite a bit of misinformation in the video :l
Great video. I hadn't really considered the Eurocentric nature of the images that were generated, so appreciated hearing them here.
I was listening to the guy from Cold Fusion's podcast and they essentially were like "this is bad for artists, but this tech is the future so what can we do?" - I think we need to be better as a society of being able to do something, but putting guard rails on it to mitigate risk or even not do it at all. Everyone is so excited about these new tech but we do have a choice to slow down.
Rather than trying to stop progress (including computers being able to do things that only humans were able to do before, such as copying other humans' painting/drawing/sculpture styles) we always do best when we use that progress to better serve our needs. So, instead of folks who had been trying to get money for their content creation, we can simply ask them what they actually need, and find others who love serving those needs to help the content creators. In other words, we can work like a collaborative species, instead of a competitive one. We can do what we love, freely and joyfully, and whomever needs what we love creating can have it, freely. Like a real global multicellular organism, instead of a planet made up of single celled mini-organisms all acting like enemies.
I think it will be better if we regulated and take the technology to another road
@@ratzelkujo7880 You can't regulate real life.
@@thewiseturtle what do you mean?
@@ratzelkujo7880 Trying to control nature and the laws of physics (life) is an exercise in destruction and suffering. When you try to fight against evolution - be it genetic or memetic - you always lose. Either go with the flow of life, and flourish, or you will die, basically.
I've been making use of DALL-E with varying degrees of success. Maps tend to change style randomly.
As for, paint a panda drinking boba in the style of DaVinci or van Gogh or Monet, I'm okay with that as the artists are long (enough) dead so their work is in the public domain (maybe not in every country; I am not an attorney and this is not legal advice), so I was okay posting them on social media. But I also generated the prompt in the style of Salvador Dali and Andy Warhol and I decided I wasn't comfortable posting those in public.
Hello, I recently came across a program that uses "AI" to fillin missing bits of a imagine, like if you have a 2D photo and use photoshop to isolate a portion and remove it can fill in the negative space with similar images. what was this program called?! I can seem to remember any help would be appreciated
Dall-E 2, I believe that feature is called in-painting. It also does out-painting, where you can use the AI to extend a painting beyond its original canvas and fill it up believably.
Thanks for the review I appreciate you bringing up this topic please keep it up. 👌🏾👌🏾😉
The UK allows data mining for research purposes also in the UK anything that is created by a robot is actually property of the robot so if you're in the UK and you're generating a image that image is actually the property of that robots so you have to go to their terms and conditions Which is different on different applications on my application it just says do whatever you like with them as long as you don't use them for commercial use so you can't be held responsible for the copyright of those images because they belong to a generator at least here in the UK
Great video!! Ty
Amazing content always! 💥✨
Ursula Vernon has used generated art, showing it a few times on Twitter and Tumblr, I think. Recently she generated an image and then used digital art tools to make it into something she liked much better, with a video of how she did it. Quite a while ago, she did a surreal series over a number of tweets and months (and who knows the current status) that I think was more inspired by the generated art than painted over it.
I've always had a problem with the term "Facebook is selling your data" They don't sell your data, they sell access to you and use your data do that. It's a wholly different thing and just saying "Facebook, tiktok, IG sell your data" is doing a disservice to your audience. Some apps sell your data but the big platforms very much do not.
But what work doesn’t depend on all past works. Even today’s artist whether painting or music all owe a huge debt to past generations of progress.
Even ideas and science.
1:11 personally, I'd rather do something way more fun. like as an AC protagonist or with wings.
10:31 good luck answering that one. it's the debate that never ends. lol
omg i love that you described givewell like a sponsor segment haha. its exciting to see folks doing stuff like that
Great video! AI seems to be encroaching in all aspects of our lives. Curious to hear your opinions on chat api and its impact on software engineering. Is it a real threat?
Maybe I mean chatgpt (?)
AI can still be used as a tool to help artists
Its not. Its really the warning that were in the end times of humanity.
The AI default is white women or white men.
i can't help wondering: don't artists themselves use a similar process to AI, where they see, learn, and adapt from artists before them?
I would say no as artist aren't physical frankensteining others artwork in the way Ai does. You might say i like their sense of color, or i like their line work etc, and maybe even study them to understand. However, like handwriting, people have their own natural tendecies and subject matter that they wanna focus on. So even if you study these artist if will rarely be a one to one. AIot of Ai is taking one to one images and photobashing them together to get tne work. Honestly in tje art community its usually best to not look too similar to one or two artist as you dont wanna be a copy . You wanna be you
@@imanispence9795 the latest AI doesn't just hammer images together though. it does create its own representations without lifting directly from other individual images
We don't. We have a thing called a visual library, and even while looking at a reference, we can never replicate the same image, as it's up to reinterpretation and iteration. The badly called AI isn't capable of this.
@@quetevalgavergaaa it seems you're not familiar with how this stable diffusion AI works. it does create by interpreting from its visual library, it does not replicate the input images or just copy from its database
@@peabody3000 I didn't say that how it works, but it's not comparable to what a human brain does. It's like in the 60s the analogy of a computer being similar to our brain, even though it isn't.
Putting images and other content on the public internet is, logically, giving consent for the stuff to be seen (used) by the observers. Sure, we could try to sue those who copy/share them in other places, based on local copyright laws, but the reality of communication is that everything observable in the universe "belongs" to everyone who observes them.
This is a totally separate issue from "artists" and other content creators living well. It doesn't matter what you do in life, what your passion is, what kind of activities you enjoy doing, you deserve to get what you need to do it well, so that the world can flourish. Which is why the whole competitive point scoring games that we humans have tried to force upon ourselves (money, grades, votes, "likes", etc.) as a way to organize our freely available resources make the world fail, rather than flourish. Once we start using computer technology to measure and keep track of our real needs as Earthlings, both for inputting and outputting matter and energy, we'll start being able to use all of the beautiful, creative, genius, specialized work that we humans naturally love doing to take care of ourselves, instead of trying to harm ourselves (in competition).
facts