Is AI Art Stealing? (A Well Rounded View)
Вставка
- Опубліковано 4 жов 2024
- I wanted to educate myself on the controversy of AI Art and wrote a 12 page paper on it. Here is my 12 page paper in video form on the subject.
Original link to my paper: lightandlense....
I realize I pronounced Gaussian incorrectly, my bad. Done is better than perfect.
#stablediffusion #aiart #aiartcommunity
1:05 1. Claim: AI Generators are stealing art
14:57 2. The Dance Diffusion Problem
17:17 3. Claim: AI Generators are going to put people out of a job
22:17 4. Claim: AI Art isn’t art
25:00 5. Claim: The LAION Database has pictures of people from medical records, violent images, and non-consensual sex.
27:40 6. Claim: LAION procured every piece of art legally through Common Crawl
30:25 7. The Lensa App, Midjourney, Dalle 2 and others
31:25 8. Claim: The signatures that are generated on AI Art is proof that the art was stolen.
33:32 9. People claiming AI art as something they made by hand
33:55 10. Why is it that people are upset about AI Art but not all the other avenues that AI is used in?
34:43 11. Why is it that we have ethics about AI Art and not other things?
36:06 12. For the sake of argument let’s say that copying styles and using artists' work for reference is not OK, it is currently completely legal, how do we change that?
36:51 13. If we as humans use another artist's work for reference (much like how I understand AI does, in the overall theme) should we not credit artists that inspire us?
37:39 14. In what way can AI art (with the Stable Diffusion 2.0 data set) still be used? If I don’t reference any artist and just describe my scene, is that valid
38:20 15. How much human interaction does there need to be for AI Art to become copyrightable or “art?”
39:20 16. What does AI mean for the future?
41:03 Conclusion
1.
Simple Infographic on how Diffusion Models Work: / ive_made_an_eli5_in_ca...
VOX video on Diffusion Models: • AI art, explained
In-depth explanantion on Diffusion models: jalammar.githu...
Dave Grossman Designs v. Bortin on copyright: law.justia.com/...
Photographer vs Artist plagarism claim: petapixel.com/...
How Selling fanart and cospaly can be considered copright infringement:
lawsoup.org/le...
www.owe.com/is...
chrisoatley.co...
2.
Steven Zapata AI Video: • The End of Art: An Arg...
Harmon AI quote on Dance Diffusion: wandb.ai/wandb...
3.
ATM history: www.aei.org/ec...
The Movies that Made Us Jurassic Park: • Moments That Changed T...
Painters reaction to photography
www.andrewgatt...
UKEssays. (November 2018). Reception of photography. Retrieved from www.ukessays.c...
www.artinsocie...
4.
Quote on Duchamp: www.artsy.net/...
“Untitled” (Portrait of Ross in L.A.) by Felix Gonzalez-Torres: www.artic.edu/...
5.
Have I been Trained
Quote from VICE: www.vice.com/e...
6.
Common Crawl: commoncrawl.org/
Artstation TOS: www.artstation...
Twitter thread Gothlytical Art: / 1600240015174381569
8.
Examples of paintings in the 17th century with signatures:
artuk.org/disc...
15.
Graphic Novel made with AI pending for copyright: gizmodo.com/ai...
It's nice to see that there are still people in this world who listen to all sides of an argument and come to logical conclusions rather than just defending their ideals to the death.
Agreed: an ethical practice with AI will be a better place to continue the exploration of this technology. As an illustrator who had a large portion of my illustrations scalped by LAION 5B dataset, I was disappointed that I wasn't given the option to opt-in and now my own illustration style can be deployed against me.
Still, I am intrigued by having an AI assistant to help me tackle larger projects. I actually have a lot of confidence that we'll find an equitable solution going forward. I also hope that image generators also help to inspire its users to also make attempts to learning art and to express through it and not just rely on prompting it.
I fail to see the advantage of having an assistant whose sole existence devalues your work.
@@BC-sh7en It wouldn't? I could imagine intentionally training the AI assist program in my style and give it character models. After that, feeding it a script for a comic series.
@@generalsci3831 keep telling yourself that if it helps you feel safe.
Luddite spotted
@@danver8433 brainless fool who supports art theft spotted.
Its weird how you keep talking about "AI vs human" as if you don't get its lines of code made by a company. All your arguments should be framed as AI companies vs individual artists
Companies are made out of individual humans. 🤨
I think its clear he is talking about ai programs and how they work, the rest is suggested. If the video was about the companies behind ai he would talk about the companies instead. I hope this is clear enough.
What’s weird is you telling another person how they should do anything.
It's not lines of code in that sense, it is very literally training a machine learning model. I get that this is hard to understand but when you ask for a dog under a tree on a sunny day in the park. The model is being searched for what all those things mean, and finding them all within an image of noise. Subject to certain criteria. Nobody programs the model, this is not what is happening. The model is a repository of how to get back to a generic labelled object, though Gaussian noise distribution. It's actually a set of weights, not an image per se.
@@praxis22 No its literally lines of code. You keep describing AI ML as if its sentient. Someone designed it to recognize patterns. Not only that when they trained the dataset they scraped image data from places like art station and Pinterest. This is intentional design and it infringement
Maybe originally it was a proof of concept and these companies had no idea about how to monetize. But you bet your ass when they started seeing traction from users making a certain type of image they went back and tweaked their shit to appeal to that market
Thanks for the very Nuanced View on These AI Art Debate.
I have Grown very Tired of seeing this Discourse Online, since the Loudest are usually the Extreme Side of Both End, I have Grown Tired of Artist Arguing, and I have Grown Tired of AI Bros Arguing.
So this Video is a Breath of Fresh Air to Me.
Thanks for doing such a thorough job at laying out the issues.
There are a few points that I would like to clarify / respond to.
1. image generation vs music generation training. The problem there is largely the scale of data and individual perceptiveness. The latent information in images is much greater than in music, so more training data is necessary (on the surface, audio is 1D and images are 2D). You also have additional concepts in images such as depth and position, which don't really have an equivalent in audio. And it's harder for a human to pick up the nuances in audio, whereas our brains are really good at seeing visual anomalies.
That said, a generative image AI can be trained solely on public domain data, but collecting that data is far more challenging than for audio due to scale (you pretty much have to train a model like SD on 2B images, and a model like Parti requires 5B+ images - otherwise it will overfit).
2. I disagree with your statement that AI Art is Art. I think it's a new class of visual representation, but it's not "Art" as the model has no intentionality. In your example of modern art, the artist made a deliberate intention for those pieces. I do think that the generated images can become Art though, once it has gone through an intentional process by a human (i.e. collecting pieces, using img2img, etc.)
3. The LAION training data. The first steps of SD were trained on LAION 2B-en, which can contain inappropriate material (there's 2B images in there, it would be insane to expect humans to curate them by hand). However, that was only at the 256x256 stage. Once the model increased to training at 512x512, it was trained on the "laion-improved-aesthetics" set, which is VERY unlikely to contain any of those images. Each image in the aes dataset has an aesthetics score, and frankly medical images, images of violence, and images of graphic nudity, are not given a high aesthetics score and thus culled from the training set. It's possible that there are still some images of artistic nudity though.
4. I think a big problem with the ethics of training stable diffusion by 3rd parties comes from the influx of people who look at the technology as something to be exploited. From people in the ML community, overfitting is considered highly unethical, and in many cases can lead to serious legal repercussions (e.g. overfitting a classified for a safety critical system could be criminal negligence). The same level of ethics should apply to image generators as well, with some degree of leeway from large datasets. It's okay if stable diffusion overfits on the Mona Lisa, because there's probably 1,000 images of it in LAION, but it's not okay for it to overfit on contemporary artists who are not over-represented in the dataset. Note that this overfitting is not based on style (that's perfectly fine, as the model is doing what it's supposed to - find patterns), but it should not be able to reproduce a specific work from a contemporary artist (to the point of clear copyright infringement).
So the over-fitting done by people on reddit, especially to harass artists, is something that everyone in the ML community would agree is unethical.
I also think there is probably room to suggest that the LION captions should have artists names removed, to prevent directly targeting their styles (possibly remove any names of people altogether, including celebrities). This does not mean removing those images from the training set, but just altering the captions which is cheap to do in a pre-processing step without reducing the training size.
Then I'm on the fence when it comes to follow-up training (e.g. using textual inversion) to pull out a specific style. I think it's probably okay under the condition that the result is not shared online, is used in a piece with further human authorship, or if it's combined with other styles. I wouldn't want to take away someone's ability to create transformative generations.
5. In terms of taking ownership and control, that's not really feasible due to infrastructure. Training a model like SD from scratch on a single A6000 GPU (which most people can't afford) would take 12+ years. So you really need to have some community driven infrastructure if you want to hope to even approach the compute power of the big companies.
6. In terms of amending copyright to account for AI, I think this could also be a very slippery slope. Chances are that the law will either end up being too explicit and end up being ineffectual, or too vague and also making Photoshop, Clip Studio, and Illustrator illegal, since they all use AI under the hood.
Good write up.
For me the issue should be about data privacy. It would be pretty easy to remove people's name & trademarked things from the tags and the UI for the prompts
Imagine an AI that scrapes personal pictures from FB/ Insta/ other social medical then users making deep fake CP of kids
I also think the big trademark holders like Disney will litigate against these AI companies bc they won't be able to control the usage of their trademarks
@@kami-gun I'm not sure what you mean by data privacy. If it's publicly accessible on the internet, then there's no expectation of privacy (semantically and legally speaking).
Similarly, LAION does not include any pictures scraped from private accounts on FB and Instagram... that would be hacking since it's not publicly accessible (and be very illegal). If the pictures are not on a private account, then again there's no expectation of privacy.
Someone in the private network can always scrape images themselves and finetune the model. But they could also download the images and do photo-shop based deepfakes too. So that's not really an argument against generative AI, but an argument for how terrible and unethical humans can be.
As for trademarks... If anyone were to shut down generative AI, it would be a big company like Disney. However, I suspect that Disney may want to use these models in their workflow, which means having it trained on their IP is beneficial. Besides the model itself won't generate their IP unless prompted to do so by some end user. And in that case, they can always take legal action - just like they sometimes do against artists who infringe on their IP.
So in that sense, generative AI at best benefits Disney, and at worst is neutral.
Also, it's not currently AI companies vs Artists. It's AI companies AND individual enthusiasts AND artists against artists. However, it could easily become AI companies against artists if vague laws are passed (only big companies will be able to afford to train and run the models, meaning we would be entirely at their whim - as opposed to being able to run and fine tune the model on your own GPU with your own private data).
And to clarify, it's not lines of code - it's basically random numbers that were computed. The lines of code that describe how to interpret those weights are akin to a surface in which a drawing or painting sits upon.
Furthermore, Stable Diffusion and Midjourney are based on Latent Diffusion, which was developed by a team at a university and then the work was used by the bigger companies. That original code still works as a method to run Stable Diffusion as well, which show how little Stability AI added to the code (it's all weights and training that they put into it).
1. Very interesting points, and logically makes sense. I will need to do more research on the topic.
2. That's great! I guess we would need to define what is a "deliberate intention" like if I take my time to put in words into an AI Generator and form it just through txt2img is that deliberate enough? How do we determine how much time is needed?I don't know. For arguments sake say it isn't then I would agree on the latter portion of your statement.
3. Ooooo thank you for telling me about the previous LAION 2b dataset I think that really helps with any "potential" issues.
4. Honestly I have to agree with you here. People are always going to exploit something it's what people do. The best we can do is mitigate it how we can. I think the suggestions you gave are things to consider and what I've thought about a lot too.
5. Yeah I know that training a model by itself is crazy expensive and next to impossible. This was more in reference to I guess "editing" a model on my own work and using my own base images and my own name to create/edit or animate images.
6. Yeah copyright it feels to me only helps the big companies and doesn't really help small artists very much. And as you bring up AI in general and how it could be affected it's already so ingrained in tech.
@@hjups Data privacy means nobody opted in to have their data used to train these models.
These AI companies arent stupid. The only way they could monetize their tech is have it be used as a copyright infringement machine. Its designed to work like that. Its like how torrents are used for piracy even though they can be used for other things
My point is code is just code. You have to look at actions of the people and the business model. A business that charges money so users can easily infringe should be banned. An AI that required infringement to give it value in the first place need to be destroyed. LAION need to be destroyed. Its obvious they exploit fair use in the guise of research when in reality they use infringing materials to make profit
@@kami-gun Opt-in is not legally necessary for training. If it was, AI progress would stop in its tracks, and a bunch of features in photoshop would need to be removed (content aware fill, smart selection, etc. - those were all trained on something, and I bet Adobe wasn't ethical about the sources)
Basically, having an opt-in system would limit the training set to an extreme degree, to the point where the model either could not be directed, or would be a literal copying machine that would only be able to produce variations of public domain works like the Mona Lisa. Would you be okay with that? You certainly wouldn't be able to use it for reference in a bigger workflow in that case.
Keep in mind that photographers were ALSO need to opt-in. They hold copyright to their works as well, and that includes EVERY photo used unless it is specifically licensed in the public domain.
But luckily as I said, opt-in is not required due to ToS. Instagram, Pintrest, Artstation, etc. can do what every they want with your posted works as per the ToS. If that clauses was not in there, then they would be violating copyright by displaying your posts on their website. And since LAION scraped from these sites and not the personal computers of artists, the only ones who can argue legal violation are the big companies like Instagram. And I bet you Meta is using the public data they legally have a license to (as per the ToS) to train their own image generation models.
And no, the AI is not designed to be a copyright infringement machine. It's designed to learn patterns, and decontextualize them based on conditional input. Also, LAION is not illegal in any sense of the form, and that has been held in court decisions. If you didn't want your data used in that way, then you shouldn't have publicly posted it on the internet. This is something that's been well known since the start of the internet, once it's online, it's always online, no matter how hard you try to delete it. And unfortunately ignorance is no excuse.
As for code and AI companies, is your main issue that these companies got their hands on the technology and are selling a service? (most of the service income goes to running the computers by the way).
If so, then you should be okay with people instead using the original Latent Diffusion model, or expanding it on their own personal hardware and releasing it publicly for free and without further profit motive, right?
Latent Diffusion was created and trained for academic purposes, even though it used LAION (which in the US is 100% legal regardless of copyright due to DMCA exceptions to academic institutions - although it was a university in German, so that could get a little muddy). The reason for creating it and training it was to advance the field of generative image AI, and it very much did that. There's also tons of other image generators available that were developed for academic research.
Technically, even stable diffusion was generated for academic research purposes and then used by Stability to turn it into a product - that's often how research works in engineering.
This is probably one of my biggest annoyances with the anti-generative AI movement... there is a tendency to think that these Evil companies popped up out of nowhere thinking "hey, let's put artists out of a job", and then created a powerful AI to help them do that. That's not how the world works... The generative AI are based on decades of academic research, which have been funded by governments (US and EU) and companies for the purpose of advancing our technological understanding. It just happened that this year marked the perfect storm for technology to reach a tipping point where it can become commercially viable, computing hardware became fast enough, and the capital existed to create the AI startups (Midjourney and Stability). But even still, those startups are building upon the academic research which can already do most of the things that Stable Diffusion can, especially fine-tuning and overfitting on existing works. In fact, StyleGAN could also do that, and it was released 4 years ago.
Meanwhile, if you want to make the argument that this academic research should not exist or be progressing in this way, then that's also a very slippery slope. The networks used by Stable Diffusion is the same architecture used to detect tumors in MRI and CT scans. If we couldn't train it on the big sets like LAION, then we wouldn't be able to measure sizable improvements to justify continuing the research. And many of the improvements made for image generation are applicable to improving model size and complexity. So slowing the progress here could very well mean that you or one of your family members could end up dying from some health complication that would have otherwise been picked up by the AI. Everything in academia and engineering is connected, even in unforeseen ways. There's a bigger picture here beyond "they're stealing my art", that said, people breaking all sense of AI ethics by how they "use" the model is something that I think needs to be addressed.
As far as AI art goes vs other uses of data (and I don't think anyone can really object to taking inspiration from copyrighted works), I think it may be that the results are so visually compelling - and better than what many artists can produce.
There's also the question of how easily one expects to be able to transition into a new vocation or expand their skillset.
The education system in general doesn't teach people to be very flexible, nor optimize for learning quickly. So if you don't feel that you can learn something new easily, you're likely going to want to be able to hold on to what you have.
We're kinda at a point where a significant change is needed in terms of how we think about creating value. But also... As with the benefits of all other forms of automation, reducing the cost of expert level work is going to increase everyone's access to more of what they need for cheaper.
AI doctors would be a massive pain for doctors - but good for humanity as a whole... And that example probably shows some of what we'd want to address. Like, would we need to cancel their student loans and help with them developing new skills? How far do we need to try to maintain their standard of living as it is? ... Lots to think about.
I don't think it's a case of "Big Tech will make a ton of money and everyone else will suffer". AI access is very democratized.
Like, if we consider DALLE 2. You got charged $0.10 for an image (I forget what the actual price is). But you can incorporate it into your own work or sell it... So you make a disproportionally larger amount of money from it than OpenAI does (which also has to pay for server bills, as well as the up front research needed to create the models)
Even when you buy an iPhone, does Apple benefit more or do you benefit more given how much it enables in your life?
@Maryyy マリ Own what? There are open source models, and the tech is largely documented at this point. How you "own" something anyone can use and develop?
Artist should join together to demand a total banning of AI copy machines. This must be in the courts with a collective demand against this companies.
Hey man I'm open to having a discussion about the topic but this feels very aggressive and not open at all. Do you mean to say that you just don't want copyrighted material to be involved?
@@kylelee5966 If AI is not stopped now, tomorrow will be too late for human society. Zero tolerance against people who use technology against humanity.
For point 2: It's more than just double standard. You also need to take the nature of the medium the AI learns on into account. Music and text AI are not the same as image AI, as they have simpler mathematical model than an image, and I have yet to see any evidence of accidental copying of copyrighted works using Stable Diffusion. Under some circumstances the model would remember how things look like in detail but those generally only applies to public domain masterpieces that got photobashed a lot (like Mona Lisa) and things that have a specific look (like iPhone 5). Copyrighted artworks most of the time don't follow these 2 so it's very hard if not impossible for one to accidentally copy them by just prompting generic concepts. It has to be done deliberately by the user (through e.g. img2img or models that are badly trained on only a few persons' work) but that also shifts the responsibility to the user.
Yes I bring up this point later that it doesn't matter the medium I use to make copyright art it's the user who is at fault.
Train it in real life photos and let AI interpret it with its own style.
that would be q really cool experiment indeed
Unfortunately, that's not how the AI works. If you train it on real life photos, it can only reproduce real life photos (although undertraining it can produce "painted" works due to the gaussian noise).
Essentially what you're suggesting is like giving a blind person sight and asking them to "create art in their own style", without knowing anything about art, drawing, painting, etc. All they know is what they can now see with their eyes. Sure, they could probably produce "art" eventually, but they do so by learning about the world and interacting with it, either seeing existing art, or experimenting with various medium.... But the AI models' have no notion of interacting with their environment, they can only learn based on what they "see".
AI needs existing artwork to know what to do and what to avoid, it needs reference.
@@hjups Sure but if something exists with the right texture, with clever timing you can get AI to draw object X for texture then change the prompt mid run into subject Y, to draw subject Y using object X as texturing to generate something that is not an photo of subject Y.
@@asdf30111 I guess that's possible, yes. But it's also a hacky solution, and will likely not produce an esthetically pleasing result. You could also claim that we could take the output from a different layer within the model and call that "art" just like DeepDream.
Regardless, I don't think that's a suitable argument. In my example, it would be akin to telling someone "reproduce that apple, but give it fur like a dog". That says nothing about a "style" other than effectively performing a collage.
Here's a thought artist, A lot of us who use AI feel for you and would love to see you all make real money forever and always. I can't speak for the entire internet and it's toxic population of pirates but that is NOT everyone. What I would do and many, many others is pay you, the artist directly, for your AI models. You could then, over time keep a following of honest fans who would subscribe to your updates and other offerings. We all want peace and prosperity and this could be the road map for the AI community to harmonize with the artist community. Lets all live long and prosper. Peace to you all and please consider it.
That's actually something that I, as a guy who makes money with art independently, has thought about. Like, if an artist was used as a prompt, the pay-service that you'd use would give some compensation (even if it was a fraction of a penny) to the artist used as a prompt.
... The problem would be solved at that point. I'd be happy.
How do you propose for artists to get paid equivalent amounts, especially with countries who refuse to respect the copyright statutes of other countries? (Example: Chinese companies are notorious for directly stealing from American creators...even using their own images to try and sell their duplicates)? Truly looking to see your thoughts on this, not to argue.
You nailed it....AI is great, but it can't do 360 game design...yet.
AI requires images and not just artwork for reference, they process ridiculously fast and so can distinguish all images simultaneously and do something new or sorta similar, they use GPUs and stuff so are faster and more efficient than us but with the weaker AIs with weaker hardware, yes you will get just copies and terrible results among just not training it properly like that one Chinese AI. My pfp is a changed Earwig pic by the Artbreeder AI so you know I'm using the better kind, I don't settle for generic or basic for making digital art.
Interesting essay, if only people were as thorough as you and able to put it in layman terms.
Stealing aside, why non-artists demands AI Art can be put next to handcraft art when photoshops art and collage arts are in their own separate category of art?
Heck, I've seen people posting in pixiv containing AI image exactly similar to the original art it was sourced from.
Also people were protesting using Midjourney to generate massive amounts of Disney images to be sold as merchandise. Midjourney immediately refunds the spokeperson's subscription of the protest movement. If Midjourney refunded that, doesn't that apply to other watermark traced AI images on the internet?
AI needs artwork for reference, it has no knowledge of styles or anything if it is given nothing, it can't make anything new if you don't teach it and feed it data.
I can understand why artists are intimidated by this new technology and I can also understand that all it runs down to is the matter of ethics as well. I'm cautiously neutral about the whole AI ordeal but it can work out for both worlds if only people could atleast find a middle ground in this and stop overreacting.
Let’s hope so. I think the dust will settle when copywrite lawsuits arise from this. I heard Disney is kinda trying to get Mid-journey to stop allowing Disney owned characters from being produced. I think it’ll be a UA-cam type situation, where AI software will have to police itself to avoid lawsuit under new copywriter laws/precedent. The only question is when and how.
Hmmm I haven't heard any stuff about Disney yet other than people were trying to make a ruckus of it and at the end of the day it's the user that's the issue not the generator. At least that's how I understand copyright. What do you think?
@@RussellKlimas It's just rumored. Also, there's a lawsuit being done by the creators of Naruto (Japan has stringent copyright laws, so anime Ai art might get a severe blow) and Peta (not sure why they are involved).
Edit. Turns out the Naruto thing was also a rumor.
@@viridianacortes9642 Hmm I see only time will tell then. I'll be curious how it falls out.
Very educational, well done!
As an artist, at the beginning, I was angry, but for reasons people use ai to copy the style of artists and used their identities also it came to using the identity of a dead artist and claiming that this is his work
,
As an artist, I don't feel threatened that artificial intelligence will take over my work because doing art has to have alot of decision on every line!! that you put when you work in an industry like making movies or making games.
Artificial intelligence is like a child who can draw anything, but is ignorant of the very small details that you want
I think the artists will not be replaced but I will have the possibility to work faster
Like animation, as an artist, I suffer very much when I do in between when I do animation. If artificial intelligence helps me, this will make me focus on the key animation more.
ALSO I CAN DO 60fps ! lol
There is something I want all human beings to demand at all costs!!!!!!!!!!!!!!!!!!!!!!!
Why do we destroy and replace ourselves with artificial intelligence?
There's an amazing thing we can do with artificial intelligence
which is education!!!!O.O
Why do we use artificial intelligence to draw a picture for you
Why not use artificial intelligence to teach you drawing, programming, cooking, writing, music, physics, etc.?
It's a nice feeling to share with others something you made yourself and people love something made by human
Artificial intelligence will also know you and know how to teach you, so that it may invent a new method OF teaching that we did not know about OoO!!
I am more excited about this than a program that creates unlimited images for me without understanding what I want????
People education is the best option to use artificial intelligence. Please send this message
Don't let machines replace us, we live to WORK what we love
very well put together video, good on you for doing such good research and putting your thoughts into words in such a clear way
I don't mind AI, I would use it as a tool as long as artists work is not being fed in to the AI without consent no matter what the technical purpose is, especially when these AI companies and AI artists are profiting from them, it's unethical and pathetic. However as an artist myself with 20 years of experience I don't feel threatened, I trust myself, my skill and experience, it will not effect me financially. However I'm not okay for my art being used for any purposes I do not approve, because I put hundreds hours in a single piece and I have a say in its use.
Any person that ever handcrafted anything creative, can you see where I'm coming from ?
I really love how unbiased this is and how it's aware that if these tech gets out of hands, only the wealthy and the powerful will benefit from it and will leave us average joes to in such a state they hold all the cards and will exploit them. I like the advancement of technology it's what makes humanity powerful but when used unregulated many people will suffer. I only hope people will make the right decision
That was the goal! I just want to try and layout the problems that I can find and look and ask for solutions. Only together can we solve these issues instead of fighting against one another.
I think one obvious point to understand here is that creating new styles with AI isn't hard. So even if you removed all copyright material as sources of inspiration, getting to something as good or better isn't necessarily difficult... Once the incentive is there, someone will do it.
All you need is to pick up on a style of variations you like, and feed them back into the model, and iterate.
Beyond that it's an issue of being able to describe what you want properly, but that's also easy enough to iterate on.
Using an artist's name is really just a shortcut for setting a certain number of variables - if you generate new styles, you can similarly label those as shortcuts for the AI to recognize.
Already, many people are creating their own artistic styles using these AI tools - so that they can have something unique, that isn't like the work of existing artists.
What you're really going to be seeing in the future is a vast amount of art being generated, and also a vast number of new styles evolving.
all i've seen in this comment is a bunch of nonsensical assumptions and claims without evidence.
@@hepzibah4573 Can you elaborate?
Great video Russell! Really made me think further on the subject! Nice lightsabers :P
Ill be honest. I reached 12:26 and barely caught half of what you were saying. Not because you didnt spell it out or talk loud enough but more so because the way the video is formed is like having a conversation without being able to say something back in real time. I too have sought conversation on this in particular with artists as I am an artist myself in support of AI art as a concept and tool but not for monetary gain. However the conversations I end up having are far more enriching by the end of it. I dont think you making this video was a terrible Idea. But there is simply nothing to break up the monotony of... nothing happening. No sounds, no music, no graphics for several minutes and mostly just you talking to the camera.
It is... too structured if that makes sense. It is almost like an AI wrote it to perform and act exactly like it does. It follows certain plots you want covered, it follows certain arguments, it shows certain graphs when neccesary and thats it. But there seems to be no passion in it. Just pure unbridled logic.
That's fine! And I thank you for your honesty and even making it 12 minutes in! I was going with the structure of what other people were doing (like Steven Zapata) and I tried it with music and just felt weird. But that's what the comment section is for, conversation! I just made a video version because some people don't want to read a 12 page paper.
Such a nice person.
All fair, but like most you forget that these models improve so fast that in 1-2 years noone cares about 2D digital art anymore. Everything will just get oversaturated and human made art will just lose value because we cant distinguish it anymore from machine created. Ultimately we had a period now where alot of people could call them selves artists and now it seems that art will be for the privileged again. Art will be in gallerys etc. or analoge or digital performances and everything else will be just noise.
I am by no means an AI expert and I expect AI to get better and better because that's technology. I think with having the watermark option that will help in regards to being able to tell them apart. The future you describe is very possible and there is only one way to find out what will happen. The only thing we can do now is try and shape the future how we want it to look whatever that may be.
Our current art may well be the cave paintings of future art. Art is not drawing or a skill set, nor is it restricted by anything other than our imagination. History does not allow me to look so low towards our creative and adaptive potential.
@@cesar4729 True but its somewhat different now since were on the way to reach singularity where we build a machine who can do anything better than humans. Art too. We need to ask ourselfs if we truly want that.
@@RussellKlimas You don't mind not being able to do millions of calculations like a processor, or being slower than a car, or not being able to fly like an eagle. That's not how it works.
How capable something is should not affect us as long as the human ego is able to maintain its self-awareness.
A singularity is relevant in the point of whether it is controlled or not. I don't see any problem with super intelligence helping, as super computers already exist and we don't try to match their capabilities, but use them to our advantage.
Q: Do you think that AI art is stealing?
ChatGPT: AI art is a form of artistic expression that involves the use of artificial intelligence and machine learning techniques to create or assist in the creation of art. It is not necessarily stealing, but it is important to consider intellectual property issues when creating or using AI art.
In general, the creator of a work is considered to be the owner of the intellectual property rights in that work. However, in the case of AI art, the question of who owns the intellectual property rights can be more complex. If the AI is creating the art completely independently, it may be difficult to determine who the creator is. In other cases, the AI may be used as a tool to assist the artist in creating the work, in which case the artist would generally be considered the creator and owner of the intellectual property rights in the work.
It is important to carefully consider the specific circumstances and to obtain the appropriate licenses or permissions when creating or using AI art. This may involve seeking the permission of the AI creators or developers, or obtaining a license to use the AI technology. It may also be necessary to consider the rights of any third parties whose work is used or incorporated into the AI art.
Excellent video, thank you very much!
This is a well rounded view.
The only "art" in "AI art" is in the visions and act of coding by those writing down instructions, and how you process, remix, and work with the material it generates.
Otherwise we are just dealing with machine illustration in regards to visual it producing images etc, until a time when it becomes a self aware entity with its own visions, goals and preferens.
If we are going that art is subjective are we just talking about a base output from AI Art Generators? What if I mix in multiple elements by my own hand? Does it become "art" for you then? Or can you define when the use of AI could become art?
@@RussellKlimas
First and foremost we must remember that the word "art" refers to more than one phenomena. But if we go with one that most people can sympathize with it has to do with a person applying its skill, greater or smaller, in order to make something that represent something that the person want to express. Something reflecting that inner idea or feeling is fashioned and added to the outside world.
I would say that AI art can be art if the process of bringing it fourth is viewed as such. Coding can be your "paintbrush and canvas" as can your prompting and the context you place the generated image in, or it can be how you remix, or augment the image.
But this has very little to do with legality or if it is accepted as art by those who are willing to pay for or write of it as such. The later would require and be represented by users of the technology active within the "artworld" field (your parallel with Andy Warhol is relevant).
What this is more about is the tools of production and distribution outgrowing the rules that ultimately based on a logic from a time where human labour and scarce resources played a much more central role in society and certain roles had to be upheld and reproduced in order for society to "function well".
@@Thedeepseanomad that was beautifully written, so elegant. I'm not being sarcastic here.
I guess the only thing I have in response to that is if their is AI art in a gallery and the human doesn't know if it's AI or human and yet they still connect with it on some level does that make it art? Otherwise your argument or definition is gorgeous.
As for the tools of production I can totally see that, and I'm afraid that if we don't put in safeguards for humans of some kind things will spiral very quickly.
@@RussellKlimas How you view and evaluate objects depends on what you know or assume about the object.
The same can be said regarding what labels we put on things.
It is not just the sight, sound etc of something that earns a designation, but how we understand it´s relation to other things.
Both human made paintings, authentic and forgeries alike can end up in an art gallery. and so cam AI designs. If they all belong just in that gallery, is however another matter.
💯💯💯 they do stupid things with their painting in images
U know it doesnt steal Art right?
Then to all Artists out here from now start Using this A.i lovers Art..
Anything u draw i shall trace it n call it fair Use n dont include the law
Bro please watch the video if you haven't already and get back to me. Until then as stated in the video I won't be conversing. You don't know what my viewpoint is otherwise.
@@RussellKlimas i m nt Fighting nor Arguing with u my message in ur comments is to A.i lovers...not here to disrespect u...i shall watch it
you keep deleting my comments, I bring you 10000 more
I haven't deleted anything my man.
AI art is so much better because it's fast and cheap. It's like someone invented cars and horse carriage drivers are complaining that it's taking away their jobs.
Not living at all is the fastest way to not live I would argue. I also see that people like to pay money to mobile games so they dont have to play the game.
Horses are there to make human transport easier.
Cars were invented to make human transport better.
Art is there so people can express theirselves,
Ai art was invented to ... express it for you.
@@Mente_Fugaz Art is there so people can generate products for sale or value. The self-expression part is just an invention of hippie types. AI art was inveted to speed up the process of what art was originally intended for.
@@lorenzomizushal3980 what was originally intended for?
@@lorenzomizushal3980 wtf hippie ? 1960 ?
I dont agree with the first point. Sure, AI gives noise to the pictures and then from that noise it tries doing another picture, but with the same idea. However, THIS IS NOT REFERENCING. This is not how a human brain works in when they are doing art. There were cases that the AI had given out images extremely close to the original picture it was trained on, meaning that it had a hard time not replicating. It is incapable of referencing, given by how it works compared to the human brain. Francois Chollet explained that these softwares are NOWHERE near close to how a brain functions. Also, Fanart itself is illegal - HOWEVER, it is considered as free advertisement to a point until the fanart is making more profit than them.
You are referencing overfitting which I have also made happen on purpose to see if I could make it happen. I used a Van Gogh model and using starry night as a keyword and some others was eventually able to make it create starry night again to like a 95% match. Overfitting is the exact opposite of what AI/ML are trying to do because it doesn't make sense to do that. What would be the benefit? To try and avoid copyright because AI generated it? Wouldn't it just be easier to make AI make perfect copies? I don't know it feels like A LOT of work to make a copying machine. Would love to tackle this further with you!
As for Francois Chollet that's an awesome resource, however I think in order for that to hold any weight we need to have 100's or 1000's of AI experts way in on it for it to be good data, just like how we do it with doctors. I am more than happy to be wrong here because I love to learn and grow I just don't like taking one persons word for it.
As for the fan art free advertisement point can you give me some references that state this? I can't find anything. It's illegal all around and as I'm sure you know there is no way a fan artist could make more money than Disney, and Disney DOES NOT mess around. And Disney has the money to go after every artist while some companies just don't, so let it slide. So that argument (to me) doesn't make a lot of sense.
Love how you have no idea about how the AI process goes,yet still find ways to justify actual rampant IP theft in the art community .
@@danver8433 Feel free to enlighten me my man! I'm here to learn. Did you watch the video? Until you have done that I would prefer not to engage in conversation. I did a lot of research for this paper, and a lot of my questions are in there.
@@kylelee5966 Yeah we as people are easy to jump on the bandwagon and pit ourselves against one another. I'm guilty of it as well. There is always a way to compromise and find a way to benefit both sides, though it might not be perfect. We need it to be us vs the problem not against each other. Totally changes how we approach issues in my opinion.