AI art is illegal and immoral (Lecture 3/6)
Вставка
- Опубліковано 14 січ 2023
- Watch the next part early, here: • Dismantling the AI Bro...
EU crowdfunder - Help protect our art and data from AI companies (I donated to this one, it's really good)
gofund.me/1cd549ba
US crowdfunder Protecting Artists from AI Technologies
gofund.me/2df3dc07
My collection of fresh A.I. memes, you can look at, laugh at and share:
imgur.com/a/CGMSWqR
----------------------------
New to 2D animation? I will teach you to animate with my course: www.animatorguild.com/courses....
Support the production of these free videos on Patreon:
/ animatorguild
Join our community on discord: / discord
__________________
ANIMATION SOFTWARE I use: www.tvpaint.com/
The accompanying SOFTWARE I use (Adobe CC): tinyurl.com/v7fvqgo
___________________
My Website:
www.howardwimshurst.com/
I work professionally as a freelance animation producer and consultant
MY PLAYLISTS TO WATCH:
My animated films - goo.gl/8kkgqD
Animation tutorials - goo.gl/TV50zo
Discussions about animation and art - goo.gl/BwbHbI
Industry advice and freelancing discussions - goo.gl/kajjK8
My PODCAST on Spotify: open.spotify.com/show/0kKfIDK...
TUMBLR: / howardwimshurst
TWITTER: / wimsanimations - Фільми й анімація
There are SO many ways AI could have actually been used in tools to help artists. The fact is this is the laziest yet most marketable approach to AI art there could be.
This shit be hitting hard man,i drew a picture for a friend and one of them said why even bother when you can use an app or tool to do it in seconds. The general public doesn't even care.
they don't undrestand the meaning of art, everyone wants to do things fast and put no effort into things; and they want to make money with the least effort, that's why so many people get scammed.
I was halfway through painting a portrait of a friend and I just gave up because all the AI stuff hit and it ruined my enjoyment of making a nice, personal thing that people would just assume was scraped.
Did the general public ever care about art? And I would hope you gave an own style distinct from others.
I saw someone plug their kids work into AI to make it, "better!" WTF!?
@@artofjmill thats a real ashole move on the parents part.
The music industry comparison really nailed it for me. This is insane. Such double standards.
Unfortunately, music sharing was an EXACT copy of the copyrighted material. In this instance, it is a simile created by an AI that has no intent to commit a copyright infringement and only use references that are publicly available ... simply by being seen on your computer monitor. There is no need for a site file access. There are SO many differences except the potential economic impact that resulted in a radically different music industry. In this instance, the problem for the art industry is that AI can generate comparable art for much less cost and in a more convenient fashion. Until we make AI comparable in economic terms to real art, we will have a segregation in consumer markets that can't be over come by legal action alone.
@@joemoya9743 Not sure I understand your comment. I wasn't refering to music sharing. I was refering to AI generated music that does not involves usage of copyrighted music as an imput in the training of the algorithm like he said in the end of the video. From the same company, two different policies and interpretation of copyright law and ethical conduct.
@@milmundos AI-generated music uses commonly-used sequences of notes found in most music. The AI shuffles the "musical note deck" and applies the notes in a way that mimics popular and/or appealing sounds. The extent to which the AI-generated music closely resembles existing copyrighted music is a question for the courts. However, the sheer volume of music that can be generated by AI may make it difficult for artists or the music industry to seek damages. Therefore, they have to choose their legal fights and not try and stop AI-generated music. All this means is that AI-generated music is the same problem as AI-generated art with no all encompassing solution.
We were never anti-Ai, we are anti-unethical use of it.
So true, AI is beautiful, but like many other forms of technology, beautiful technology can be used to harm people.
I generally just call it image generation. We've had similar stuff for a long time with no problem because the companies using it weren't stealing art to use it. My go-to is usually Shrek because that was all the way back in 2001. They generated trees in the background :)
Yes, tech is fine, and unethical use of tech is bad.
I think part of the issue is actually adopting the term "AI" which implies that the computer is thinking in some way. It makes it sound like way more than what it actually is.
I hope artists whose rights were infringed will keep teaming up and fighting against these things - especially those who are EU citizens. The infringement is blatantly clear, while looking at the local legislation at the same time! I've come back to add that the US has that one set of copyright laws, where there's also the 'fair use' one. I do believe the industry is bigger there and they have a better chance. I can't see how international dispues (against LAION in Germany) would happen - I can imagine that's terribly difficult.
It doesnt matter, IAs are gonna success we like or not. If China create their IA they dont give a fu** about copyright ( perfect example: TIK TOK)
@@SM-gm6hr Not the subject. I didn't demonise all AI, I don't get why it's so easy for some people to just assume that. I'll accuse the brainwash that's going around. China already has an AI market. Their most advanced one is already programmed to block things related to their politics, but that's the problem of their citizens - although many don't see it is an issue already. Still, I don't live there. I didn't even know they used tik tok (because I don't use it myself) - I thought they had their own version.
@@joanabug4479 Tik Tok IS CHINESE, thats why USA wanted to ban it. And they dont care about copyright thats why in tik tok you see any song and in youtube no. And its gonna happen the same with the IAS they DONT CARE, and anyone with access to the internet are gonna be able to use them.
Moral of the story, the art industry needs to get it's copyright standards to be just as strict, if not EVEN STRICTER than the music industry.
i find it good that you made this video. The sad part is that i think we are to low because thos ai generators are "memorising" things should be deleted and build from scratch with these rules in place, named in your video. I even find it sad that so many artists that i admire begin to accept ai and how it is to this current date or are for it. I hope that this video series gets the most traction before a desaster strikes. I think that people need to get educated on it but is see the problem that because big tech companies do sell the data that we provide them so many people will just ignore it. Love your videos and I hope that for everyone we can step into a bright future where every right is being respected in any way shape or form that there is. Big love from a art apprentice from germany
Watch the next one here: ua-cam.com/video/JnWRlvpQ3xw/v-deo.html
For those who need even more convincing:
Here's an eye-opening blog entry on the irresponsible behaviour of stable diffusion: bakztfuture.substack.com/p/statement-on-stable-diffusion
Here is an official class-action lawsuit that is going down currently: stablediffusionlitigation.com/
Here is another class-action in the works: www.theverge.com/2023/1/17/23558516/ai-art-copyright-stable-diffusion-getty-images-lawsuit
---------------------------------
EU crowdfunder - Help protect our art and data from AI companies
gofund.me/1cd549ba
US crowdfunder Protecting Artists from AI Technologies
gofund.me/2df3dc07
-------------------------------
You changed the way I thought about it, thanks for the enlightnement. I wasn't a fan of AI taking over art before, but I didn't knew there was a real case that could be made agaisnt it. I was just thinking of it as an unfortunate tragedy that comes from natural technological development.
A great and concisely delivered material. But why is there only 50 views...
I haven't made it public yet :)
Oh I thought it was cause it was published 20 seconds ago..
It bull shit when they say they can't find the image owner. They can just use google image search and look at the result and find the people there and ask if the are the copyright owner.
Yep, now just do that for 5B images (or the 170M the models were trained on). If you do some simple math, you can see why that's not practical (how long would it take per image multiplied by the number of images).
@@hjups *yes* do that, please. Your argument is "But it's hard and time consuming and costly to do", welcome to the real world? Where things take time, usually costs money, and can be hard to do depending on what you want to achieve. I don't know a single artist, who didn't make sacrifices with respect to the above, but these companies and bad faith actors/users are only looking to make quick money. And then, when challenged on their unethical practices, that do in fact harm original creators, whine that 'it was hard to do'. The pinnacle of lazy, toddler entitlement.
@@pipkin5287 Clearly you didn't take time to do the simple math... 170M images at 1 minute per image, working non-stop without breaks would take 323 years to go through. As the models get bigger, more images are needed. For example, Parti was trained on most of LAION (let's say 5B images) and it wasn't enough. Assuming it was, that would have taken 9,512 years. Sure the search can be done in parallel with more people, but we're talking on the order of years to decades. It's a problem of practicality and scale (you can't beat the exponential). It's not a problem of laziness (that's as absurd as it would be to say that artists just have to work harder).
You should also consider that the researchers involved most likely didn't even consider artists or copyright when creating the models. It's common to use what ever data they can find for academic research (there are copyright exemptions in the US and EU for this). So if they're used to doing that, why would it be different for commercial purposes? I honestly didn't consider it either until artists started complaining (rightfully so), but the models had been trained and released at that point.
@@hjups Couldn't they make a registry of the exact address the image was scraped from? if they got the images sure they could get the source.
I hate AI
#2 from my understanding: Nature of the copyrighted work.
They tend to let fair use apply to non-fiction more than fiction. Because you might have many people getting the same facts and, when they publish something about a crime, for example, there will probably be a lot of similarities between what different people say because they're going off the same set of data.
Fiction tends to get a lot more protection because, while coincidences do happen, if you publish a short story about space fish with lasers, and then someone in your writing group suddenly posts an oddly similar short story about space fish with lasers, it probably is more than a coincidence.
Basically, this rule tends to favor protecting fictional works a lot more. So the art that's being stolen would also generally fall under this, too, as far as not being considered fair use. Image generation fails on all points.
Until you have a singular global legal form of enforcement, you are only going to restrict or reduce AI art in countries where they find AI actions illegal. The AI generated art will simply move to countries where the AI will continue to operate. It appears, the genie has been let out of the bottle. ---BTW... Copyright of AI art can not be verified by the US Copyright office. They can only make a rule that says they will not accept AI generated art but have no method (so far) to enforce or verify that the art is generated by AI. It never has had this capability. It is ultimately up to the courts to determine its legitimacy and whether AI art violates existing copyright laws on a case-by-case basis. This approach is costly, time-consuming, and may not result in any damages being awarded as the artist must prove financial harm. This creates a HUGE barrier for protecting the rights of artist as lawyers may be hesitant to take on such cases unless the artist covers all associated costs. The legal costs and uncertainty of newly untested AI art copyright claims make it difficult to justify fighting the legality of AI art.
What you're needing is legislative changes in law that would be implemented by the UN... and, good luck with that... since they have very little enforcement capabilities even if on the unlikely chance you get the UN to agree to a standard of legal doctrine regarding AI generated art.
It's a tough cookie for sure. Let's see how it plays out 🤔
Having trouble downloading this video even in144p, and it's not a connection issue: other videos download just fine on 1080p. I noticed a similar youtube's behaviour on some other videos/channels that try to discuss hot contreversial topics where somebody's commercial interest is involved. Adding to it stories like Facebook files makes me wonder are such concepts as "law" and "morality" even applicable to modern IT corporations at all. And yes, thank you for arguing against unethical AI on behalf of the artists!
Wow... As an artist myself, I never saw it from this point of view. I even made a video about AI, which I believe was making a valid point. Have to re-evaluate that, after watching this. Very insightful, Howard! ☝
thank you for listening!
No worries, mate!
In regards to my video, I had to take it down because it just wasn't sitting well with me.
The thing is, it's a new concept. So new laws must be made. The argument of "this doesn't apply to this law" is irrelevant.
Stealing isn't a new concept.
11:51 it would be so difficult for the ai nerds to *gasp* send an email or dm to artists and converse with actual human beings.
Thank you for spreading the information
Sad that all my friends belittle me and laugh at me when they get excited about AI art and I raise eyebrows
Wow, your comments section is *full* of AI art apologists, it's actually sad.
I aint apologising for shit haha, he's using a very cult like language.
Implying that the IDEA of AI art is something to be "apologized for," and that thinking there are bad arguments (and articulating them) is excusing the arguably bad uses.
good old capitalism
Thank you so so much for this. I’ve been but at the same time not in a moral dilemma concerning AI art. I know its bad but how do I explain it? And here the explanation is.
we can all feel that it is very wrong, but finding the right words to explain it can be challenging.
thank you again ++ for the algorhytm^^
The music industry comparison is great. Huh. I didn't know that was a thing. They are huge hypocrites.
2nd factor is determined by use of published works.If the ai image was produced with copyrighted material,and is then published,it will be less likely to support the claim of fair use.This also applies to if the work produced falls under a creative work or a factual one.Use of facts from other sources may not be copyrighted,but if the use of creative work is involved,it will not be considered fair use.
"Empowering Creators" biggest bs they say
It is empowering though.
@@HB-kl5ik yeah but they really could've done it in a respectful and moral way from the start which they didn't.It is outright disrespectful to the artists. It is empowering the need for instant gratification of some people and in return, they blindly justify the blatant misuse and stealing.
@@MrPangahas I use it as stock image alternative tbh, and I want as unique stock image as possible. Plus, I later spend hours on photoshop to get a great look possible.
Plus, I guess for that reason even OpenAI, Meta and Google are to be blamed as well. Those huge corps could've done 100x abuse for sure than a 16 year old prompter as he's saying.
@@HB-kl5ik Its a good tool no doubt and I'm actually looking forward on how it progresses I just wish they did it right from the get go.That precedence left a sour taste , I don't think anybody is against AI per se, but there needs to be some decency in how they're being used considering it feeds off artists' works in its data base.
simple answer to you
when ai generation started it was a small data set
then some asshole decided to dump all danboru images ( anime website with fanart) and train on that
it got popular and other followed him and made their own data sets....
You can steal someone else's work slap an A.I. filter then call it original. Thats still plagiarism.
> *A.I. filter*
Erm ... huh?
yes, it can - and is - being mis-used as a highly advanced plagiarism aparatus.
I am a fan of your work and have watched a lot of your content over the years. I'm fairly positive I have watched tutorials from you explicitly telling people to learn by tracing & mimicking more skilled artists work until you are good enough to do your own. Curious how you quantify that compared to a computer doing the same.
Is your argument just that computers are not people so they shouldn't be allowed the same means of learning?
I'll come back later and watch more of your series on this (it's a lot of watch time to go through). I value your perspective as you are one of my favorite animators online, but It's hard for me to justify calling AI immoral or illegal when it's functionally doing the same thing I have seen you tell your students to do for years.
They just need to change to an opt-in system for their datasets, and then it won't be illegal or immoral. Until opt-in is enforced, I stand by everything I have said.
@@HowardWimshurst Opt in doesn't change the baseline problem though. This is actually really important because if we fight the wrong battle here it could SERIOUSLY damage independent artist's ability to deal with big companies with large data libraries.
The way AI learns is fundamentally similar to how we learn. We learn by seeing something and imitating it before we make our variations based on other things we have also applied that same process to. AI just does that fast.
What seems to be argued here is that iterating on someone else's work without their permission isn't allowed. The problem is that everything is iterations. Everything uses copywritten work as a fundamental baseline to establish the style and tone of art. We are functionally arguing about copywriting a style. The downstream effect of copywriting a style is that if you can prove that someone used your work as inspiration, then you establish uour right to their work because yours acted as the precedent. Who wins in that battle?
Disney. Disney DEFINITELY wins in that battle. Whoever owns the oldest and deepest catalogues of content get to establish that they own a style and that if you used their work to build up your skills, then you don't actually own what you make. Disney will.
I have some friends who dealt with something very similar in the world of trademarking. Jukin Media decided to claim that they owned the term People Are Awesome by doing copyright takedowns on any content earlier than their first compilation videos (at the time it was a band called Hadouken). Jukin went on a tirade threatening legal action against all channels (original or not) that had that phrase or used derivatives of that phrase. Did immense financial and algorithmic damage to a ton of action sports channels across the web. Can you imagine if copyright suddenly no longer had fair use protections for iterative work? We'd suddenly have to be worried about what all the Jukin Media's of the world would do to our life's work.
By the way I feel your pain here. My main channel, Beyond Slow Motion, was one of the most freebooted channels during the 2014-2016 eras. Talking potential billions of views that I didn't get paid for. I know the pain of watching your work get used against your will to make big $ for other parties. That was literally not even transformative work either. I should own a nice house in LA for all the money Facebook didn't pay me for that stuff.
I guess I am looking at this from the lesser of two evils perspective. The lawsuits that are being brought up now and the fight that artists are fighting right now will absolutely have devastating downstream consequences if they win.
I don't have all the answers on this front, but I see the conversation that's happening now and am scared that we are about open a can of works that we REALLY don't want to be opening from a legal standpoint.
AI is making you more active on youtube huh lol
That's kinda funny
Hey there, I consider myself an artist, it's a craft I've honed and will continue to hone for the rest of my life. it brings me more peace and joy than most things in life, and i too used to be against AI.
I misunderstood what training meant and thought that it did indeed infringe on artists rights until I actually learned about how it works, how it learns and how it generates. When the facts get presented in court, it'll be found that the lawsuit doesn't actually hold any water, for a fair number of reasons. But please don't despair! That shouldn't be looked at as a bad thing!
I'm telling you, play with the tool, through something like Automatic1111 or InvokeAI, dig deep into the Machine Learning, hell even the python, and you'll realize how freaking awesome this stuff is, and how badly represented it is in that lawsuit. I've actually started dabbling with the AI myself, and its an incredible tool in all of its forms.
There are going to be bad actors, people with negative and dishonorable intentions, and we can't do anything but fight back when it arises. But, to continue to demonize this truly groundbreaking tool over what is frankly just misinformation is just going to further alienate "artists" from the larger crowd of openminded people who take the time to really dig into how these systems work.
AI is here to stay, and i'm excited by that and by the possibilities it's opened up not only for myself but everyone else. You'll see blatantly obvious AI art, and can give it the appropriate reaction, and then you will see the creatives utilize it to bring their inate creativity to a whole new level. And then you will see the purists who claim its just cheating (digital art used to be considered cheating too)
fk off soy ai bruh
Exactly ♥️ I think startups out of this, who are disregarding the very intent of Stable Diffusion are kinda in bad though. Those who are outsourcing creativity.
Though you wanna say more on how it works? I'm ready to listen.
The tool itself wasn't demonised in any way. I would gladly use it myself IF (hard if) it were built on an ethical foundation - which it is not, at the moment. Moral subjectivism is at play here, maybe you have no issue disregarding those few artists who were visibly ripped off (Mike Mignola, Greg Rutkowski, Sam Yang... none of these opted in and have voiced their disappointment). But, just like meat-eaters can't expect vegans to just support the awful dairy and meat industry, you can't honestly believe it would be their fault for "not complying". People should have the option to OPT-OUT. That's it. And those names people kept training into the models... they should be compensated for all this, if they're the reason folks decided to pay for a subscription in order to "make art like x or y". They've unknowingly given free publicity to the tool, after all. Or at the very least give credit where it is due, not pretend that "there's no names in the database that we could link to real artists"... If they want to seem trustworthy, maybe their CEO shouldn't be lying to our faces ¯\_(ツ)_/¯
@@joanabug4479 It is irrelevant what artist were used to train the AI. It is like walking through a museum and mashing up Malevich with Van Eijck when you work the canvas. It isn't some stock photo collection. It basically makes collages.
Not even collages, the way it works and creates the latent space is magical. It's a wonderful piece of code ever written. It doesn't copy but makes everything so much more unique.
Getting around the GDPR is actually pretty easy - don't make the online services or downloads available in the EU. The models then don't have to abide by the GDPR and the EU looses out on the generative AI technology.
Copyright violation is a different story, and something that the courts need to decide. Your opinion about it not being fair use or my opinion about it being fair use is irrelevant. Neither opinion is objectively true and requires a legal ruling to declare what is and is not fact.
There's strong evidence for fair use though, especially based on legal precedent, and considering the invaluable good that these models can do (which many artists ignore - things like help with tumor detection in radiology - I can go into more detail if you would like).
Also, Stable Diffusion 2.0 has removed artist names from the captions, so that's an attempt to remedy the linking of artists names to the images (there was no way to predict that such a thing would cause so much grief when training the 1.0 model).
Also I wish people would stop parroting the content of LAION 5B claiming that the models were trained on the whole thing... they were not. Please take some time to read the training methodology and filtering done to at least Stable Diffusion before repeating the claim (we don't know what Midjourney filtered though). Yes, there can be gaps in the filtering due to the sheer scale, but the filtering was pretty restrictive and the probability is quite low that harmful images made it into the training set. You should have thought more critically about the clip cited as support "true safety can only come from really careful thinking and engineering"... what hubris. No matter how careful you think about or engineer something, there will always be failure modes that you hadn't considered. So it's more of an argument of what's the best that we can reasonably be expected to do, which is exactly what I think the team at Stability did, and went above and beyond when moving to 2.0.
There's no breach of privacy or consent in regards to using publicly available photos online. That's the nature of the internet, there's no expectation of privacy when you post something publicly. Privacy concerns really only apply when you interact with something in private (e.g. clicks on facebook, or posting to a private account).
Now to answer your question:
Why couldn't the AIs been trained ethically? Cost and scale. OpenAI made a deal with Shutterstock, something that Stability couldn't afford. While Stability is valued at $1B, how much do you think shutterstock charged OpenAI? Probably more than 50% of Stability's evaluation (OpenAI, Meta, and Google can afford that though).
And for scale, these models need 100s of millions of high quality images... which would have taken decades to collect if not done through automation (web scraping). No investor in their right mind would invest in a company that won't see returns for at least 10 years. In fact, we know that the results get better as the models grow in size, and size means more images are needed. This was a problem that Google faced with Imagen where they likely let in a bunch of the LAION images that Stability filtered out (because their model is bigger) and thus couldn't release the model for public use due to the harmful nature of the training set. Basically... they ran out of good images to use! Then think about Parti being more than 5x bigger than Imagen... it definitely has issues with not enough training data.
Meanwhile, if you take the automation into consideration, you will quickly see how opt-in / opt-out doesn't really work. You would have to black-list individual images, and before doing so you would need some way to detect a list state for a given image. Meanwhile, consider that a bunch of LAION comes from Pinterest... good luck finding the artist and source of a random image posted there (I never can as a human, so I would have no luck at creating an automated tool to do so). And in some aspects, tracing these images is actually a HARDER problem than Tesla trying to make self driving cars. I don't want to be antagonistic, but unless you have a viable solution to the problem, you shouldn't criticize experts for not finding one.
Oh, and the subscription model is to pay for compute, not to make money. Stability makes most of it's money training custom models for companies like Disney... not from Dreamstudio. Think about how much money the infrastructure to run and maintain the service costs (people seem to not fathom that the GPUs cost $5k each and can only generate one user's prompt at a time).
I do agree with your criticism regarding the lies though. I know people have requested to opt-out, so that's a blatant lie. And in terms of removing names from prompts... that can't be done for the currently trained / released models. But future models can have that done going forward (which is what Stable Diffusion 2.0 did). But to the point of why it wasn't done before hand? Because AI researchers don't have crystal balls (if they did, they would spend their time selling stocks rather than making AI, after all they only care about money, right?). There was no way to predict that this would be as big of an issue. They didn't even know that these models would be so popular until AFTER they were trained. They could have just as easily been a fad just like the original DALLE.
Oh and they do not give commercial rights to what prompters make. At least in the US, courts have ruled that generated images CANNOT hold copyright (as you pointed out). So the only "right" is that they won't pursue arguable moot legal action against using generations in other works.
I also agree with your remedies though. The only caveat is that removing the ability to prompt for artists can only be done with the models going forward (based on how they are trained) and on the online services. The models already trained and released to the public cannot be retroactively updated. But I think both the first and second idea going forward are doable (an artist can have an image removed by uploading each image to the site and have it compared against the database - removing images with a high degree of similarity). The uploaded images themselves don't need to be stored either, just their embedding vectors for similarity searches with future datasets.
Opt-in, however, is not a viable option. But sites like artstation and deviantart can be removed from the dataset (i.e. black-list the domain in the URL). But other than that, opt-in would essentially require starting from 0 images which again will likely take decades to rebuild (until then all we have are the "infringing AI models"). So the best bet is a removing artists names, removing art focused websites (that does NOT include instagram and pintrest though - and if GDPR is a concern, then they could just make the models unavailable in the EU), and having a rolling opt-out, in which images can be removed from future training runs if requested.
A person's photograph in a dataset with billions of others is not "personal data". The AI doesn't even remember the picture, and has no identifying information like names, addresses, or anything similar.
They did not give their permission to have it used in data-scraping or training. Permission should be a must, regardless of your own personal opinions about it. Permission doesn't seem to matter to you, but it matters to image owners. Consent. That's all we ask for.
"The AI doesn't even remember the picture", really now? That's a new level of cognitive dissonance of I ever saw it. Of course these models store and work off of data. If they didn't have a harddrive and RAM with stored information to work from, they simply wouldn't work - just like any other computing software. I certainly hope you meant something else.
And yes, a person's picture is, in fact categorised as personal data under GDPR, as it can be used to identify you. Image data sets like LAION could just as well be used for face recognition software, as it can be used for image synthesis. And that's not even bringing into light more curated sets, focused for certain outcomes.
@@pipkin5287 They were trained on over 2 BILLION images. Stable Diffusion 1.4 is just over 4 GB in size. If it tried to store them all it would be 2 bytes per image.
All the AI has stored are the patterns(weights) that it learned. NOT the original images.
Fair use is about not asking for permission if the work is transformative enough. Which AI art generators are. In fact, you can't get more transformative than that. Also can you stop saying stolen datasets? Even if you live in the US which has no laws about datasets, there was already a case concerning datamining from authors against google in which courts decided that datamining is not copyright infringement. Not to speak about the EU, UK and Japan which have actual laws that allow for datamining and machine learning. (For the EU it's EU directive 790/2019 Articles 2, 3 and 4.).
8:20
The more data that is properly tagged you have the better the model. It doesn't even matter if the picture is bad or blurry. If appropriately tagged the AI can learn what a bad picture is and what blurry is and it can learn to find that and not do it.
8:40
Why call something legal loophole that is doing what it is intended to do? That is like saying drinking alcohol at a certain age is a legal loophole that allows me to drink. Article 2 (1) is clearly laying out what is considered a research institute. The lawmakers knew how it could be abused. And LAION is acting as a research institute according to the law since we all have access to LAION's datasets, not only Stability who financed it.
11:00
LoL no - Me and my friends have now combined properly over 250k followers on different social media websites. I know their prompts, none of us use names of artists. In fact, we've done the opposite, we compiled a list of artists to tell the AI to stay away from to get better results.
------
There is so much misinformation coming from this I can't fathom going through it point by point. I don't want to write a book. So that's it from me.
True he hasn't has his homework done at all. All the mistakes are done by prompter, not the StabilityAI. Plus, negative prompts exist. Imagine in case of Midjourney or DALL-E, you have barely any control of the style being used.
Stability has prevented a huge abuse corporations could do otherwise.
How is it not a "stolen" dataset if they argued it was for research purposes and then gave it over to AIs to make it available as a paid-for product? I'm not personally against archiving any kind of data, for research or just keeping track of it through history - but giving it over to be sold without prior consent. That's shitty. The result of the research could've shown beforehand this very need of filtering it out before having it trained. Now they're leaning on indifference, saying "it's not possible to make it forget". Mistakes were made, sure, but people were hurt. They should be held accountable until they change their products.
@@joanabug4479
What do you mean? It produces overwhelmingly unique pictures. I don't doubt in the vast amount of possibilities their will be something similar to something already existing though, so you always need to reverse image search before you publish.
But as long as the product is unique, there it's by definition transformative since something can't be called unique if it looks like something that already exists. And that makes it fair use, which allows the use of copyrighted data if you create something new with it. And by all definitions, a picture that didn't exist before and has no similarities with others is new to the world.
@@anonnymous7009 im confused, but what happens now? i mean should artist stop posting online because there is no attribution and the fear of being replaced?
@@blackwhite-zt7qt
Why the fear for replacement? Most people still can't be bothered to prompt for anything nor are they willing to sign up and learn how to get what they want from Midjourney.
Let alone learn how to use Stable Diffusion (most won't get passed the installation process).
An actual artist incorporating AI to their workflow will be 20x better and faster than any idiot typing prompts into an AI and taking a picture of a person he can't really recreate in another scene.
In fact you are able to just draw a character and color them somewhat satisfactory you can add better coloring, shading and backgrounds creating an actual novel inside the same time it took you to create one picture before.
Using from others and being used by others to express what you feel and show it to the world is what an artist is. Not someone who hoards their ideas to introduce artificial scarcity. The whole world is at your disposal to finally show what is actually in your head, what you actually feel, the story that you actually want to tell. And people are throwing that away because of some sense of ownership of style. Personally I think that's ridiculous.
Finally I can put what I imagine out to the world to see in a better approximation than ever before (the approximation to my actual fantasy is important). And the last thing I'm going to do is stop them from seeing/using or building their ideas on my fantasy. Use it, add to it, let me experience it so I can do the same.
Alright let's dismantle this
-- Copyright --
1. Not all ai art generators are for commercial use, there are plenty of free and open source ones
2. None of the actual work is used in the final product, it is only used to train it. You can have copyright over a specific piece, you can't copyright the style of your artwork lol
-- Ai is "hurting" people --
1. In the same way that photography "hurt" portrait artists when the first cameras were invented, this is inevitable just as every new technology.
Taxi drivers are not going to start lawsuits when self driving cars become dominant, and you won't be able to sue when an ai is just better at you to draw.
-- Ability to name artists in promts--
1. Dont know what eu law you are reffering to
2. By your logic, it's actually artists that encourage others to impersonate them the moment they share their art. If you don't want other people or machines to learn from you, don't share your art.
1. Yes, some Ai devs are driven by a twisted ideology rather than the usual lust for quick money.
2. The training is what we are talking about. With this NEW technology (which is not directly comparable to the invention of photography) artists and image owners did not give their permission to have their images used in training. They should have been given the chance to decline.
1. falce equivalences. So many falce equivalences. Ok, it would be as if Uber stole every taxi cab driver's car. If an AI can draw as good as a master without using laundered datasets, then so be it. But as it stands, it can't.
1. GDPR as of 2016. It's the reason that every website you visit needs to tell you about the cookies and the browsing data it collects from you, and obtain your permission.
2. This point is truly twisted. A bit like saying "if you don't want up-skirt pictures taken, don't walk outside in public wearing a dress" as if it invites that. "if you don't want your car to be keyed and its numberplate to be jacked, don't drive it out in public, keep it locked in your garrage". Showing something in public is not an invitation to steal it. And also: be careful what you wish for. Telling artists "just don't share your art" - what if we all did exactly that? Where would be your precious datasets? The world would be bleak.
@@HowardWimshurst You keep saying stealing and taking without permission, which is the main disagreement we have. Nothing is being taken, learning from a teacher is not robbing that teacher of the knowledge. If I make a UA-cam guide on subject X , I can't at the end say that you must not learn from it. Your car analogy does not work because getting keyed is not a feature of the car, its vandalism. Art is meant to be learnt from. Also yes, the open source platforms like stable diffusion are not for profit, we improve it and make models for it because we think it's a tool that should exist freely.
@@Zoykzmc From such little I know, AI isn't 'learning' to draw creatively, it's taking full portraits and illustrations and copy their style, so maybe it takes a line from style A then takes a shading method from style B and etc etc until it's finished. But the thing is, it's not being creative, because there's nothing new being created. It's just taking litte portions of other creations. It's theft. Imagine someone steals a penny from you. You'd be mildly irritated. Then they sell that penny for millions of dollars. and THEN their business de facto makes you lose your job. Now how would you feel? I'm looking forward to what you have to say.
Also, a teacher consents and is paid to give their art (but the consent is the important part). The artists who created the images the AI uses didn't consent to let it be copied.
I don't understand your youtube analogy and your car analogy.
Yes, art can be made to be copied AND IMPROVED as a base style, giving your own unique twist on it. AI isn't able to generate anything unique.
And lastly, even if stable diffusion isn't for profit, there are many other platforms that are, and what I'm assuming Howard (and probably me) wants is law regulations so that no theft is going on, not getting rid of AI completely.
@@sangmadewira4726 Well I'm sorry that you don't understand how ai works. Yes it makes unique images, if it didn't would you be able to tell from what artist they "stole" the style? No. Not to mention that not all artists have a unique style in the first place and never draw anything creative. Should that mean that they should stop drawing? And if you are really interested in learning about how ai works, watch some videos about concolutional network, the ones by 2blue 1brown explain in detail how they work.
@@Zoykzmc I understand. Unfortunately, I don't accept your apology. The reason I'm dumb and don't know Ai is because of everybody else's fault.
Anyway, You can't rely on an empirical way of measuring this. It's unrealistic to ask someone to familiarize with every single art style posted on the internet and then fact check every single AI image to see if it's stolen someone's style or not.
Besides, if you think about it, every single artist had the exact same artstyle at some point. For example, every single (digital, I'm assuming) artstyle has atleast 1 pixel involved in it. If you draw with a pixel too, does that mean you're copying everyone else's style? Well, technically yes, but it's just a pixel, who the hell cares.
But then, beginner artists will start to play around with the most basic elements such as the pixel, the line, or simple shapes. They start to experiment it, like how mixing primary colors can create tertiary colors, and end up creating a pathway leading to their own unique artstyle, and it's at this moment where their styles differentiate from others and they become truly creative.
What AI is doing is not 'experimenting' with these basic elements. At a fundamental level, they are TAKING pixels and lines and shapes from other images and mash it together into a single image. But who the hell cares, it's still pretty basic. But here's where things get problematic. At a complex, abstract level, AI starts copying other artists' own unique, complex, and abstract style and mash it together. Take Samdoesart for example. AI can imitate his style almost exactly, it can copy not his style per se, but the constituent complex elements that make up his unique style, and use (more like mash) them together to create a new image based on the prompter.
And here's the unfair part. Samdoesart took years of time and effort to create that beautiful and unique style of his, and now someone can use ai to basically create an obedient tireless 'Samdoesart' robot, and then hire (or sell) the robot instead of the actual person, effectively stealing his entire career.
It's theft. The claim that not all artists have a unique style and don't draw anything creative is a fallacy. Most beginner artists don't have a unique style or are creative and are just copying from experts themselves, but every single accomplished artist does and is and aren't (or if they are copying, it's only some minor element, not their entire style). Copying someone's constituent elements of their artstyle is theft, and, in an economic sense, is robbing them of their source of production.
ALSO will note down on your suggestion from 2b1br. Thanks for telling
"AI is the most dangerous increase in actionable knowledge that has ever befallen mankind". Here is one example: 4 AI Killer robots being developed in a lab kill with live rounds of bullets 29 scientists! Just one consequence of it's development.
I what movie?
You can opt out.
Where? How?
@@HowardWimshurst For LAION or stability, one of them, if you go to there website you can opt out of being in the future datasets.
@@waterisaneurotoxin7788 which only has any merit at all if models are disgorged and re-trained.
@@pipkin5287 That is impossible, to many people have the models downloaded.
@@waterisaneurotoxin7788 nothing is impossible, but I agree that algorithmically disgorging all privately held models is likely to be improbable. That said, if they were illegal to own, less people are likely to have them and try to profit off of them. And on top of that, is that large, corporately held models *were* disgorged, I think less individual creators are likely to be harmed by their profit models. Just saying "it's impossible" is lazy, nihilistic thinking.
I get why you're doing these and I know that putting out there the issues between the fans of the generators and actual artists is something that details how fraudulent the ai fans are.
I hate to be the downer on it but this isn't going to solve things, when these companies are taken to court they'll settle immediately because they'll sell the AI to big name corporations that can then detail the AI to create ad campaigns for various products and have got a mint for the programming.
So don’t even bother trying to educate is what you’re saying? That’s sad
@@neatdiscoveries No, if you want to DO something other than just talk about a problem you need to do something.
For instance, starting a trend on saying all art that is posted on the internet needs non-transparent watermarks right in the middle and covering 2/3rds of it to prevent AI from recognising the difference between art and watermark.
Or suggesting the complete boycotting of all sites that have AI Generation as part of their TOS
Or suggesting not posting art on the internet outside of specific formats that are far harder for AI to use as reference material (like speed paintings that have a lot of junk on screen along with the art progress and only the video of that)
there's plenty of things to do, but just talking about the problem doesn't solve anything outside of making yourself feel better for saynig how bad AI generated art and their users are. It is effectively a circle jerk as there's going to be zero people here who totally want AI art to be a thing that aren't here hate watching. Anyone interested in art is against the AI generated shit as it's not done with skill as people care about a skill used.
@@UncensoredScion Supporting the on-going lawsuit against Stable Diffusion is nothing, to you? Raising awareness and sharing or donating to the cause is also nothing? You're wrong, pal. All the programmers I know are against the shady way GitHub Copilot went about doing things and the route for artists against Stable Diffusion and their unethical data scraping is the same route programmers took to guard against what Copilot was doing. I can't wait for LAION-5B to barely reach '1B' when they take out all unethically sourced and copyright infringed data.
@@UncensoredScion talking about the issues specifically is, in fact, the first steps into defining one’s personal attitudes and actions or inactions to take, for any subject.
It’s called being logical
@@joanabug4479 This is going to be a long-winded one so bear with me, 'kay?
.
.
this video has nothing to do with those lawsuits the first one did as it featured them this is talking about the dichotomy between AI bot users and actual artists (and like I've said by the time the lawsuit is done and ready to be tried, they'll have sold the AI to a marketing corporation and settle outside of court for a limited sum and the problem will persist with others and the problem will be endless)
this video is making arguments against the people who use AI generators and pointing out their hypocrisy
Like I said I understand why he's done it, getting the thoughts out into a format that you can see helps clear your head. But it doesn't solve things and it ends up as a circle-jerking self-masturbatory ego inflation as people think talking like this is DOING something.
It's not, it has nothing to do with it.
There are ways to cripple the AI bots now, and it involves taking the initiative and making your art less appealing to see. But it WILL cripple them if a majority do things.
Ultimately the problem with the AI bot users is rationalised as a simple argument.
Show a cookie to a 5 year old, then tell that same 5 year old they don't want that cookie and see how they'll argue you to the end of the world.
The users of AI bots do not care what you, I or God himself would say about this, they want to use it and they will ignore everything you say, call you names for arguing against them and keep doing it anyway.
And as with everything when this is pointed out, people don't like to have their delusion of doing something shattered.
Liking this video doesn't solve the AI bot problem.
Donating to the lawsuits doesn't solve the AI bot problem.
Talking about the AI bot problem doesn't solve the AI bot problem.
It just makes you feel good to do these things because its the easiest and safest way to do something, no effort involved.
You want to solve the AI bot problem you need to starve the AI bot of all resources it has to use and make it useless for commercial profit.
If its about money, why is stable diffusion open source. Also they are working on music generators, there are multiple of them available now.
oh I don't know...
- a race for market dominance in a blue-ocean tech field
- 1+ BILLION dollar valuation
- Financial gain through numerous avenues and funding grants
- offers for public speaking gigs around the world
- fame, power and notoriety
- controversy hype. Shallow "front page of forbes" ambition.
- fun at the expense of generations of hard work and sacrifice in a neighboring industry.
- very little work or payouts needed because of the laundered datasets appearing to be "free for the taking" (mistaken)
- A destructive quasi-religious "ends justify the means" ideology / extreme libertarianism.
- Styling after utilitarian leaders of the past such as Hitler or Stalin.
@@HowardWimshurst what does that have to do with open source?
@@waterisaneurotoxin7788 the fact that they made it open source, without properly vetting the technology first, is even worse. It's extremely irresponsible to not even bother considering the impact it (already) has on multiple markets and fields of business *on a global scale.* Or if they did, they literally did not care and merely sought to benefit from their idiotic decision while, at the same time, hiding behind the notion that "it's open source and everyone can use it".
Don't you think some due diligence would have been the proper conduct here? If they truly had no bad faith or purely financial intention, would it not have been even more beneficial to everyone if they had at least consulted with the very fields they supposedly seek to enrich?
@@pipkin5287 This has been worked on for years, and no matter how much you vet things, with 2 billion images, some things are bound to get through.
True I’m against ai generation no creativity from true artists
if your plan is to stop the spread of the AI generated content by the lay then you are already lost. The tech is here, its gonna stay, we have to deal with it. Ps yes jobs will be lost, the market will change. These things happen all the time,
Howard I would like to respectfully disagree those generators are here to stay AI won't replace artist's jobs have you seen how it generates eyes hands the proportions /anatomy are off. The copyright office is too slow for change. Artists share a bunch of work online which is then indexed and they don't read terms of service when posting their artwork online that's on them the artists. They need to privatize their artwork and learn how to be their own business.
Instead of people being artists for someone else they need to be their own entrepreneurs making their own stories and experiences there's so much opportunity there's no excuse why you can't be successful as an artist/creator in the 21st century an artist can be an art director, publisher or content creator, director , the can learn Unreal Engine to make animated films ,etc. like there's so much stuff that the AI can't do.Artists need to adapt and use AI themselves to improve their artwork and output. I mean think of all the Renaissance Masters do you think they would have liked their artwork to be all over the world? we all copy from each other,no one creates in a vacuum. basically each artist should have their own website and limit what they share. have password protected files. like this is basic cyber security.
I've been designing characters for the past 10 years both on traditional media and in zbrush and I have never shared any artwork online but after seeing these AI generators I would love to just post all my designs because I believe that sharing is better for the collective. no one owns a artistic style. So one solution is don't post all your work online.
I think you're making an argument against a point i never made. My videos are about the data laundering, data misuse and false impersonations going on. How many artists get replaced is irrelevant. A massive crime has been committed and ignoring it sets a dangerous precedent. You have gone off on a tangent which is not our concern at present. But for what it's worth, I think your outlook is really... REALLY... sad. How sad is it that the prevailing advice is to not show your artwork out in public because it could be stolen and misused. It's like saying "don't drive your car out in public because it could get keyed. Best to keep it in the garrage under lock and key". Sad times we're living in. Your rights are being violated and your answer is to just... let them? Not put up even a hint of resistance?
Artists have emulated other artists as long as there has been art. Unless you draw something that has literally never been seen before, you are drawing inspiration from other artists. Also, AI art is much better than most artists that are currently alive today. 😂
you have poor taste, and should watch the previous lecture where I dismantle your ignorant opinion: ua-cam.com/video/bU01H9Vjnok/v-deo.html
biters have gotten their asrses kicked for generations too. just because it's digital the old rules don't stop applying.
Honest question (and I’m not trying to be clever here). As a 3D artist, if I use AI as a tool in my Blender projects to cut down the pipeline and help my workflow does that make me immoral and bad in your view?
I would advise to not use any AI text to image tools until there is a fair opt-in system established. That includes open-source software.