Not really. Because it blankets all of AI by using a strawman against the entire medium by using vague and arbitrary examples as the whole. When most skilled AI artists, spend hours prompting and teaching the ai though photo learning, manipulation, edits, and masking. Which very much makes it intellectual property of the one creating it by any standard previously held for copyright. This is like saying that a DSLR photo can't be copyright protected because of the enhanced quality with things like autofocus. Or digital art when you use brushes, filters, shapers and other generated aspects which don't require the actual ability of the user outside of aesthetic and creative decisions to perform. And last I checked we don't give Nikon, or adobe the rights to our art when using them to create art. With any amount of thought this quickly becomes hypocritical to other positions and clearly needs to be more thoroughly assessed. There needs to be a set criteria of personal interpretation and action which dictates rather AI art is copyright protected or not. Just like we do for Fair use. Ai art should be copyrightable and specifically by the person creating it, rather than who created the program. But there should be a large enough delineation from the AI's initial results and your own to show fair use and thus copyright protected. Because the initial art produced is the art the coder envisioned based on the criteria. Anything past that begins to become your own interpretation. because it's changing on the back end the variables the programmer set to adjust for your aesthetic composition. So like any fair use case. Minimal changes wouldn't be enough. But at a certain point you cross the threshold and it becomes fair use.
@@zephirol4638 If you're putting in work in your piece, it can totally be copyrighted. The prevention of AI works being copyrighted is for people who are only prompting/generating images, rather than curating their datasets and taking their work further and beyond the minimum.
@@rodrigobarraza Fair enough, if they're accounting for that then it sounds fine. Just by how the video put the TLDR it was sounding to me more like ai as a whole was just a no go unless you used a completely different medium to make a complete parody of whatever you configured through manipulation of data sets. Guess that's what I get for assuming without actually reading the actual proposal. lol thanks for the clarification.
@@ehtresih9540 Of course it's going to be enforced this way, that's how these regulations work. As someone that's started a couple of tech start-ups in the past, these types of legislations are important for individuals and corporations alike, especially when it comes to technologies that begin to advance rapidly after being around for many decades. Having money means that you can afford more, better lawyers, for longer, and is a complete different issue than what you're trying to bring up, but ultimately, these cases will be treated similarly, as long as the proper channels are taken, and proper precautions are met.
Well done! I AM an attorney and you did a great job at hitting the basics. At the end of the day, case-by-case analysis will rule the foreseeable future. Use AI as your brainstorming mate and then modify with your own God-given imagination and talent.
One thing . AI art generator is just it a generator. In order for it to be a art generator. It requires Prompts it is something that is required. And prompts are what ever one can dream up its your Thoughts In Grate details that you hav to put in to the generator. That makes it my art. Now some wold say no because the generator still did the work . but with out the In put of you thar is nothing. Now if someone can sit and feed a AI Details to create a specific story line in to a script. Then had a AI Create a complete movie Based on the script now this is seen often in some Digital movies on TV. And thay are copyright. So in my Opinion if someone can mack a AI movie and copyright it then the AI art you do should also be copyright protected because it is your Is thoughts and ideas and dreams .
@@BRAVEN32m12 Sounds good in theory but not in practice. Your prompts are the same concept as commissioning an artist to visual your ideas. I've done commissioned paintings for customers that still don't own the copyrights to....they paid a lower price for the right to use it for their one project project and nothing else. I still own the copyrights to the image.
What is art to a blind man? What is copyright to a man made into machine? ... A legal paradox but one in which will inevitably arrive... Evolve or be left behind, Mankind.
I like this, Ai art should be used as stock photos for artists to use for their own creations but not sold as copyrighted material this gives way more room for Ai to be used as a tool to make new stunning pieces of animation, music, videogames, etc cutting hundreds of hours of tedious work and letting the artists spend more time perfecting their vision. I'm hopeful, I just wish big companies didn't take this as a free card to fire everyone cause it's a wasted opportunity not having all those people working on different parts of the creative process
Unfortunately, that last statement has always been standard business practice. When companies get ahold of a new technology that cuts the workload in half, they don't allow their workers to work for half as long, but rather fire half their workers...
Thank you for this video. It's very informative for someone like myself as an artist that doesn't quite know what to make of the AI landscape. The conversation tends to have a lot of emotions from both sides which makes it hard to approach. So your video is refreshing in it's succinctness and even keeled explanation and from a perspective as a working artist.
Thanks William. Ya, these topics stir plenty of emotions in me as well, but getting upset I feel won't do us any good, I feel talking about these issues rationally will help us get to the meat of the arguments, and hopefully find solutions that help propel us forward. We'll see how it all shakes out.
You'll been fine as a pro artist, depending on your field. If your a freelance artist youll lose your cheap clients not your loyal ones. I use AI art for my projects. I tried using regular artists but the field lacks quality control n can range from affordable to extremely expensive. The hiring process alone dampers all projects n leaves AI as a viable, affordable, user friendly process option. The hiring poetion became so unviable i stopped using art all together until AI art came around. Do i still want human art? Yes, but i dont want the increasing risk of quality control to harm my project.
This is an interesting approach to the whole A.I.-topic. The major question for me is, how will this copyright regulation be controlled? If an A.I. generates an images, and nobody keeps track of the process, how would someone prove, that there was an A.I. involved?
There is software out there that can detect if an image was made with AI. But it's unclear to me how reliable that software is, and will it be able to keep working as the technology evolves.
Copyrights are not simple.(edit:registering a copyrighted character is what I'm talking about here, not "natural copyright" which is easy - or automatic in theory - also I'm definitely not an expert in US copyrights, but have some experience of EU's) If you want to copyright a character, you need tons of images and details and descriptions to do it(copyright applications for animated characters can be tens or hundreds of pages). So if you want to copyright a character, AI based or not, you need tons of documentation and details about the character. If you simply AI generate a character, you won't have cohesive style guide for the character that is needed, so either you generate a ton more cohesive images and information for the character: the design history, setting, the defining elements, etc (text&images); enough info and images to get a copyright and lie about the process, or you simply tell the process involves AI and you still likely get a copyright if you did all the rest of the work and not just one image of a character. As for copyrighted AI art, there will be some ways to tell from other detector AIs to metadata to tell-tale elements of AI generation, but it's difficult. Better question might be whether anyone cares about digital art copyrights really? No one is breaking a bank from digital art, even less so with generative AI's existing, so if someone copyrights a generated image as a real digital image, it makes a little difference in real world because digital art had very little value to begin with. If you keep copyrighting AI generated stuff and some legit artist complains about you lying in your copyright, I think you might be asked to produce somekind of proof of legitimacy for your art, like the original files, though I might be wrong about this. On the other hand if your digital AI art becomes super famous and valuable, it will be found out it's AI generated pretty quick by closer inspection and it'll lose the value. Very pessimistic way to look at it, but it is what it is for digital artists... Musicians and 3d modelers and likely all other digital art professionals as well, will face these same questions and problems soon :P
@@z3dar Ya honestly seems to me personally I've always been super pro public domain. So I'm considering running two sets of projects one Pen name where I experiment with new tech and release everything I do public domain and one where I use my further developed skills later on to create serialized content that I can copyright the characters and such with. These early stages I don't care to much if anyone takes the stuff I'm making. It's mainly practice and if they like it enough to reuse then I'm personally flattered.
I was afraid that the U.S. government wasn’t going to understand how the generative A.I. works. They absolutely nailed it and are starting off with an incredible approach. Super happy with the development and I’m ecstatic they sided with the artists :)
This is a step in the right direction. AI-produced work should be part of a global commons, not copyrighted and sold. People should be able to access AI-generated art and make derivative works from it to their heart's content.
Great video, simple to follow and clearly explained! I'm an artist in the games industry, and this stuff has made me lose sleep for weeks. Your video has really put my mind at ease.
They are talking Islam and do not understand it. Who will sue if you just start selling? They wont budget copiers to compete with you. But will let anyone steal your sales with your own product even if you are there first and basically it _is your idea? Like forcing lottery winners to shed their winning ticket because, well, because they won and all others did not!
Makes perfect sense. It’s no different from how I, as a graphic artist, buy elements from places like creative market, then use those elements to create a completely original unique design that is of my own design. As an artist I’m capable of making these elements myself from scratch, but buying elements premade helps me work faster and more efficiently, and I am supporting another artist in the process. It’s artist collaboration. I see using AI exactly the same way.
for now it isn't at all sadly. Ai bros sell NFTs or make big tits asian girls. And I say it's around 95% of its use currently, while also killing new artists and current artists, and of course the artists that the Ai is based upon
As I understand it, the US Copyright Office will consider each application on a case by case basis taking into account both the AI generated portion and the human author's contribution to decide whether it gets copyright. The irony is, in future, the copyright office will probably use an AI to give the decision on each application.
The truth is that you are right but that they use the AI for each request could be more advantageous since imagine analyzing hundreds if not thousands of AI images each, with an AI that analyzes them together with a person in charge it will be faster to rule if Will it be copyrighted or not?
Its funny - the first thing I was thinking is say you work with a publisher and they find a way to use AI to reproduce even your "human" work on an image, then say "see he used ai" and void your copyright and use your images for themselves and their own profit.
Definitely needed this! Great info! I appreciate the in-depth overview of what is deemed as a standard, in the US Copyright Office concerning generating AI.
This is a decent start, I think. My filling related to the technology as creative can be best described as mixed rn. I'm not aganist it, and I can see its potential benefits but also feel it has the potential to effect humanity negatively in the long run if left completely to its own device. I am glad we're having these discussions on the ground level.
I’m not worried about it being left to its own devices. I’m worried companies like Amazon or Disney will try to monopolize the market for the max profit.
WTH it only just hit me that I'm watching Neil Blevins!!! Your 3ds Max scripts have saved countless hours of my time!! You sir are a legend in the 3D community!!
(3:08) The material (or "elements") in the image is not copyrighted but the image itself is copyrighted, and you own that copyright. Regardless of how little work you put into creating that image, or how simple the process (cut and paste), you created it. A copyright is NOT dependent on image complexity or level of completion and does not prohibit the uses of textures generated by A.I. The owners of the A.I., according to the rule, can't claim a copyright on the textures, but if someone else uses your cut and paste image in their own video or published media without your permission, you are able to claim a copyright infringement. (3:45) The middle image is NOT copyrightable by you or the A.I. owner, according to the policy. The image you provided of your own creation, according to the rule, is simply a part of the instructions you gave to the "artist", but you are correct, you own the copyright to the character and image on the left and right. In this particular case, the subject in the middle image only shares the base color of the original 3D model - it's not even a depiction of the original model, and only influenced by the original image. You are correct on the other two and according to the rule, you can freely use the subject image produced by the A.I. to create the copyrighted image on the right. (6:50) The point of the policy is to prevent someone claiming ownership of a complete, unaltered image produced by an A.I. tool, and to prevent the owners of the A.I. tool from claiming copyright infringement of all material, or bits and pieces of art ever produced by the A.I. The image you show of the alien in the video can not be copyrighted by anyone in it's unaltered state, according to the policy. Understand, it is the image itself that can't be copyrighted. If it is your concept and your description that generated the character, regardless of how vague the description, you have ownership of both the look and attributes of the character - the concept of the character belongs to you. Any image you latter produce of that character is fully copyrightable by you, both the image and the character - the same as your final image of the robot. (4:35) *Probably the most important thing to point out:* Not only can you not copyright the "fan art" of the Spiderman character, you can be hit for copyright infringement for publishing these images.- *Except in the case of this video.* You are using the images to illustrate and explain a point of the discussion, which is covered under "fair use" rules, probably the most misunderstood and abused copyright rule of them all.
"The point of the policy is to prevent someone claiming ownership of a complete, unaltered image produced by an A.I. tool, and to prevent the owners of the A.I. tool from claiming copyright infringement of all material" Is it really saying that, though? I thought the statement was that just using a text prompt was not enough control over the final image to be considered as having enough artistic input/intent to be copyrighted? It seems plausible that someone using ControlNet to control the pose of a character, shape of objects, gradient of lighting, color palette, particular style trained on a LoRA, different prompts for different areas of the image, etc could be enough intent to be copyrighted, even though the AI still produces the final image. The question on the input side seems more about just how much of the above is needed to be provided.
@@ShawnFumo I didn't design the policy. *That's what it says.* Your argument has merit, but as you point out; the problem is, who decides how much detail is required in the prompt to warrant a copyright on the AI produced image? What it also says is even if the overall image is produced by a machine, if an artist (human) develops the AI's final image further, as in the final robot image, then that refined image belongs to the artist and is copyrighted. One of the foreseen problems is aside from who was first, two different people could give two different prompts and the AI might produce very similar images. Which of the two people gets to copyright the image, and can the one person then sue the other for copyright infringement. The idea of copyrighted AI art is a kettle of worms. The best approach is to not allow the AI to claim a copyright on anything it produces, and to prevent people from claiming copyrights on any unaltered AI produced image. THAT is the core of this policy.
Really informative video. It's refreshing to hear a calm, nuanced discussion that doesn't get swept up in all the clickbait headlines and stupid internet discourse.
@SD789 Ah, yes, the most reliable source of medical advice: some guy crawling through youtube comments, looking for someone to get offended at. I think I'll probably listen to my doctor's advice over yours, lol
💐🙏✍🦊 If I could only thank you enough, for presenting this timely and meaningful video on the new AI copyright laws, as they are changing vis-a-vis AI generated art. I really appreciate this information, and have downloaded the document as you have suggested. I have not read it yet, and must replay your video for a better understanding, but something tells me. Artists will be getting the short end of the stick, (pardon the Americanism) when it comes down to the law, and what a mess for those who will be responsible for determining what is and what is not the legal case. I hope I am wrong, as I am a designer and digital artist myself, but we shall see. It will also be interesting to see how this impacts artists who mint their Crypto-art on the Blockchain, in terms of this new law and its ramifications. Thank you so much again for sharing, and for your insightful video here.
4:27 Regarding using an image prompt as being more copyrightable, it is a fuzzy area perhaps depending on whether that image was originally your own copyright. Otherwise I see that a reference image is just "additional instructions", like asking an artist doing your commision to "make it look kinda like this other reference image" which is quite common (what I see on pixiv and deviantart commision requests anyway).
Glad to see they are actually taking proactive steps. I was not sure if they would even address this stuff for another 10 years. Honestly, as someone who is not really an artist, I would assume the genie is out of the bottle. You could maybe request to have your work not included in the training data but in the end it matters little. I suspect we will see a purge of certain artists work from the existing data sets to comply with differing international copyright laws. At least in countries that pretend to care about international copyright laws. Unfortunately/Fortunately, Pandoras box is open. At this point they have enough progress in the tech to have it generate its own training data, rendering human artwork almost unnecessary in the data set. The artists of the world will have to be content with the fact that this new technology will give them more powerful creative tools in exchange for an entire reimagining of the artistic world. Music, Movies, Books, none are immune. People really need to start mentally preparing for a gigantic shift in our reality. I don't know if people are ready for what's coming.
You are making it seem deeper than it is. It won't be a "gigantic shift in our reality", artists never frowned upon the use of AI as a tool, they didn't like it trying to take over their jobs by using their art without permission nor royalties, and AI giving power to random people to literally steal their work and impersonate them (it happened way more times than needed to). You're also standing on the wrong side of the street, you want raw AI generated images to be able to be copyrighted and sold, completely ignoring how damaging and dangerous that is. I guess being lazy, uncreative and a thief is requirement enough for you to get granted money without putting minimal effort on anything. It's ok if you want to play-pretend and use AI to make beautiful pictures by just prompting, you can do that as much as you want. But expecting artists to give up their jobs just so you (or anyone just using AI) don't get mad and pretend to care about a field you never really payed attention to before AI is ridiculous. Most people using AI don't even care about art, or art related jobs, they just want to be popular on social media. And time has proven that stealing other people's work/content is the fastest way for talentless and skillless people to find "success" online. "This new technology will give them more powerful creative tools" The tool you're talking about isn't creative, the artists are creative. You can literally make a photorealistic portrait by using a regular pencil. You don't need a "more powerful creative tool" to make it. The argument about AI was that it would speed up the process, which on industry level is essential as time = money, but not for random Twitter users that want to benefit from the art of hardworking, skillfull, talented artists. Although the argument behind "why AI is good" has changed more times than a Twitter user's profile picture... Everything that an AI can do, a human did it first. The AI is just doing it faster because it doesn't require any phyisical nor mental effort to do it. Without the models trained on millions of stolen artworks, the AI is as efficient and "creative" as a 3yo child. Limiting how public models can be used (to avoid impersonation and theft) plus the copyright laws is the correct way AI should be treated. The random user from Twitter doesn't need the AI to be able to produce a high quality piece of art on prompt, neither does the artist; because the artist can work from the ground up, the AI will speed up the initual (sketches, thumbnails, etc) and final steps (post processing), whilst everything that goes in between will be made by the artist.
@@4nanaide Thats a lot of words to ignore everything I said to jump to the defense of artists and artistic integrity without even acknowledging I stated I am not an artist and have little skin in the game. I do know children, and if you think one can produce even a fraction of art that Midjourney does you have no clue what a 3yr old is capable of. Very little outside of basic coloring and stick figures. My point still stands though. They no longer need original artists to build a data set. They have enough proof of concept already done to allow the AI to work using a data set it originally creates on its own. Additionally, there will always be foreign entities that don't give a shit about any copyright law you will come up with. Artists have mainly only a few choices. Make physical art and hope people do not accuse you of just hand painting/drawing a copy of an AI generated image you are trying to pass as original. Or know that any digital art you ever create will never be assumed to be created by hand and that AI was involved in the work. Ultimately, the major benefits will be the tools within Photoshop and other creative tools giving you the ability to create even more impressive works. But if it is too good, know most people will assume it was made by an AI.
@@Infectd "[...] am not an artist and have little skin in the game. " No, I understand that and it shows, as everything you're saying is wrong. "I do know children [...]" I doubt so, and I'll prove that in a second. "[...] and if you think one can produce even a fraction of art that Midjourney does you have no clue what a 3yr old is capable of. Very little outside of basic coloring and stick figures." That's literally what I said. If you grab MidJourney and take away its dataset which was fed with stolen artwork, there's little to nothing that MidJourney can do. The "AI" only has the ability to reproduce and mix what was already done, it can't do anything new. Without a dataset that consists on millions of images, the AI is even worse than a child; but even a child can create better art, original, and without requiring being exposed to millions upon millions of images. And I'm not just saying it, you can fact check and get a piece of reality by looking at the works of Alexandre Cabannel at 13 years of age, on times in which technology was so primitive you could say it didn't exist. He didn't need "more powerful creative tools" to make amazing art, he had creativity and skills that puts MidJourney in shambles; he just needed a canvas, oil paint and brushes. The rest was not being lazy, something that in current times most people are. "[...] They have enough proof of concept already done to allow the AI to work using a data set it originally creates on its own [...]" You not only aren't an artist, but also don't know how the "AI" technology works. It isn't creating anything on its own, it is always stealing and taking from everything that has been made by humans. This "AI" doesn't create, it isn't creative; it generates - and to generate; it needs data, data that it gets from foreign sources that it doesn't understand - commonly fed on by its developers. I repeat, the AI cannot create; it only generates - because it isn't an Artificial Intelligence; it is a Machine Learning Algorithm, which is the closest humanity has gotten into creating actual Artificial Intelligence, but still isn't. It only learns and repeats patterns, and how to solve problems that deny it from reaching its predisposed goal - in this case; generating (not creating) art. The AI hype and propaganda was made up to sell the product, and because people tend to be ignorant towards how technology works (and also because is easier to explain or say "AI" to a bigger audience than to say or explain "Machine Learning Algorithm"). Your last statement about "artists having only a few choices" and "if your art is too good then people will assume is AI" is so fundamentally ignorant I can't believe you really said it. Do you know about a little something called streaming? You talk big time about technology this and technology that; but forget that people can literally stream themselves drawing, showing their creative process, and trust me; it doesn't even come close to how AI generates art. There are also apps on which people draw in real time together online, and that's how they get to meet each other and recognize thus shout-out good artists. Also, AI can only fool those that know little about art, and on top of that have poor taste on it. It doesn't matter how much time passes, AI generated art will always be recognizable and distinguised. Only those that in the past got fooled by a filter over a picture to make it pass as "hand drawn" or "cartoony" will keep falling for the lies and deception of "AI". Note: There are secret uses to AI for art that only artists have managed to figure out and implement, making it so the artwork generated doesn't look AI-ish neither "too good" as you said (uncanny is the word you were looking for), and it doesn't require a massive dataset with millions of stolen artwork, neither powerful computers nor servers. Is way simpler and technology-basic than you're making it to be. It still requires you to be an artist, not a prompter.
@@4nanaide ok so there is a lot here and I don't want this to fall into an argument. First keep in mind 90% of people are not artists. The vast majority are not the types that look at art and think the artist was trying to portray the struggle of blah blah blah. They don't go to galleries or art exhibits That's only artists and a small part of society. Second, the AI will always exist and can be trained on royalty free art, which will grow massively by artists who do like this new AI and it's capabilities. If nothing more than out of fascination of the tech. Once it has that it can generate more art that can be looped back into the data set to refine the technique. As an example we can generate people today that are photo realistic and they have already taken the photos from that AI to train other AI on how to recognize and identify human features. Now I am not familiar with the artist you mentioned, but of course there will be prodigies. But if that prodigy today does not do their art live, people will always question how a 13yr old did that when tools exist to allow them to cheat. Not everyone has the desire to be on public display on twitch streams or RUclips. I am tired of this it's machine learning not AI argument. The two are very similar and one leads to the other. It's not an AGI but AI is still a fair label. As far as how it learns and creates, we cannot make claims on it compared to how humans create art. We do not understand the human brain. Though I will say just watching my kid learn to crawl and walk and learn to speak, it's not that different than watching an AI be trained in an environment to learn to walk to crawl. It's accelerated with a human brain, but that's more a limitation of our technology. That gap will close very rapidly. See NVidias new Data center super computers designed specifically for AI. To date, all AI has been achieved on technologies designed for standard computers with tensor cores being newly introduced. These new super computers will push AI much further and faster as they are a tool specificallt designed for this task. I am not saying all Artists are done. What I am saying is their value to society will be diminished. Less will be necessary as far as independent artists are concerned. But the starving artist has always been a trope for a reason. Most of your disagreement with me relies on this tech not improving. Considering I have been watching it for 3yrs now pretty steadily, I can say for a fact the tech is improving at an exponential rate. Now that we have the hardware, it will be even faster. None of this should matter though as artists are largely the anti capitalism types who produce art out of passion and not for monetary reward.
@@Infectd The first half of your reply is just redundant non-sense that fails to understand what I said. I even gave you dates on times where Alexandre Cabannel - the artist behind many masterpieces today celebrated, as well as being a core reason on why art excelled past academic grounds; making him one of the greatests old masters, but as you said; this doesn't matter to you and most people, because you don't care about art, just about pretty pictures and how "good" they look, not their contents or portrayed skill. "I am tired of this it's machine learning not AI argument." You're tired of AI being called what it is (it's even in its name "Generative AI"), although you can literally do a simple search and learn how it works, and you'll realise that the AI dreams are nothing but dreams. Do you know why Chat-GPT doesn't give answers to topics its dataset wasn't trained on? Because it can't learn new things without a dev giving it a new dataset with said topic on it. Do you know why it can also give you wrong answers? Because it isn't precise, nor "Intelligent"; it only generates what it was given to generate. That's why its devs have to update the dataset/models constantly, because the world around it evolves, whilst the dataset doesn't. It's a static file containing scraped information. You can train it on information from the 90s, and asking it questions of today will result on the AI telling you that such thing doesn't exist, because it doesn't know it does, and it can't learn it on its own. It only learns from datasets. "[...] the starving artist has always been a trope for a reason" It's called learning to paint/draw/illustrate, and not learning to market yourself or not knowing how to sell your art. It has nothing to do with art or "artists" but with people. You knowing how to make a copy, paint a house, paint portraits, etc, doesn't mean that you're going to be a successful artist. The same way if you are good at sports but never go out of your way to try to get into a professional team, you will never make use of your sport skills. People go to art school because they like the idea of being an artist, but they don't like art or the process of making art; thus they fail. Although only bad artists become "starving artists", and the same happens on sports, medicine, and many other fields. It's not about the career, is about the person. "Most of your disagreement with me relies on this tech not improving." Not at all. I literally said that the technology is good enough as it is, artists can make use of it as a tool, as it was supposedly "intended" to be, whilst non-artists can only pretend to be artists. Making it so the generative AI makes all the art without flaws and ease of use doesn't make it a tool, but a replacement. A replacement for a problem that never existed. You're tired of your AI dreams being called Machine Learning, and I'm tired of people saying that this new AI will help artists create more beautiful and impressive artwork, when that was never an issue to begin with. As I said before, you don't even need technology to create amazing art. A canvas, brush and oil paint, or pen and paper, and you're set to create. If you want to create, even primitive tools will be powerful enough to make you create amazing things. If you want to pretend and play around, apply no effort and learn no skills, then you will need every piece of technology available to do the work for you, the worst is that you aren't even creating nor learning, you're generating and forgetting. Tired of people saying "this will help you create! but will cost your job tho." when creating was never a problem, and now that you found a solution for a problem that didn't exist, you are taking the livelihood for those that didn't have said problem. Literally not a solution nor a tool, a replacement which isn't even impressive as without already existing art, it can't create. It's like attaching knifes to your hands, now you can kill and cut, but you won't be able to do anything else. "None of this should matter though as artists are largely the anti capitalism types who produce art out of passion and not for monetary reward." And who told you that? Your communist friend from college? Art, as sports, cuisine, etc, can be done out of passion and for monetary gain. The process of creating a piece of art, step-by-step, is what artists are passionate about. The fulfilling creative journey, not the end result. You will be proud of your pieces, love some more than others; and that love comes from you remembering what you went through while creating. The same applies with the pieces you hate, the process probably wasn't the most optimal, or the end result wasn't worth the process, but you still made it, and that piece exists to remind you of your mistakes - or of what you're capable of. The "startving artists", "frustrated chefs", "failed football/basketball, etc players" comes from people that enjoys, or rather desires, the popularity that comes from doing those things. People that would love to show off pieces, titles and trophies more than they would love to draw/paint, cook or play sports. The process of drawing/painting, cooking or playing/training is nothing but a hassle to them, they just want to be recognized, and that mentality is what leads them to failing. Nothing about talent or being a prodigy. You can't run a marathon if all you do is be lazy and uncommitted
Consider the case where the "prompt" is something along the lines of: "You are a writer's assistant. Proof-read the following material and correct any grammatical errors, and make sure it follows the Style Guide. === " This might even be built into the word processor, and is seen by the operator as no different from the traditional spell checker and grammar checker.
Thanks titaniccor! Yup, the international side of copyright is something I don't have a lot of experience with, but I'm going to do a little research on the topic as I'm interested to know how all the jurisdictions interact.
That’s some great interpretation, and the copyright office approach sounds sensible - prompting an AI for images is pretty much a “search” operation in a latent space, which others can easily reproduce and doesn’t include creative elements other than knowing what to search for The HUGE gray area here is when you train a model with your own, copyrighted work, and generate images out of it. They’re variations if your original art, using your style and your content as an input. In that case, would an image you produce with a prompt be copyrightable?
Not necessarily. Google was legally allowed to scan copyrighted books for their search engine as they were not putting exact copies of the books up for search in full. The same argument could be used as to why the companies are allowed to train on copyrighted works.
I'd say yes, it's copyrightable because, with the case of standard "off the shelf" AI image generation, a human author (you, the artist) had zero involvement in the process, up to and including the prompt input. But if the model was trained with your own images, a specific human author (you) was involved from the beginning, continuously through to the final image.
The analogy would be: If you were to teach a human student with your own copyrighted work, an make them create new art out of it. They’re variations of your original art, using your style and your content as an input. In that case, would the work produced by the student be copyrightable by you, or by the student?
Excellent presentation! It also reiterates what so many of us have been saying. AI is not going to put creators out of business because owning IP is foundational to corporate capitalism. No company is going to build a business around public domain assets.There must be barriers to entry against competitors to attract investors.
I've been an artist for over 40 years. this is good news. I have some ideas for playing with AI for tiny pieces that could go into a larger image. I may have to do half a dozen images for ideas for certain parts of the image but I'm going to gesso them to a large canvas, week them as I paint over them and fill in the blank areas with my own drawing and paintwork. I know that I can copyright that and get it into one of my local galleries. This is good news. Thank you.
Glad you found this helpful, yes, I think there’s a lot of positive applications of ai which involve deeper human involvement that is also copyrightable, so enjoy experimenting!
That's just fine, I figure. I know artists who are using AI generation for their backgrounds. They do all of the foreground work, they do the characters, but the AI does the background and the fiddly bits they don't want to bother with. There's no problem with that.
Thanks Noah! This video did really explode, although since I mostly post about making art, as opposed to copyright and AI, not sure I'll be able to keep the large crowd happy. But I will be doing a short update video in a few weeks on the same subject just to answer a few more questions. Glad you enjoyed the video!
A corporation is a abstract legal entity, yet it can hold patents and copyrights. A corporation is seen as a human being in the eyes of the law. So just incorporate your AI and then they can hold copyrights legally.
Hi Neil, Img2img and controlnet would be two interesting points. Eg: Img2img I have a unique drawing of mine. I load it into stable diffusion and low diffusion and get it to colour, or shade, or edit my work? Where do I stand? Controlnet. I use controlnet to transfer poses from my own original works onto prompted or Img2img creations. Where would that stand? Thank you for your time.
Yup, if I had a ton of time, I'd love to make say 20 different images using different AI techniques, send them all to the copyright office with detailed descriptions of how I made them, and see which get copyright and which don't. There's so much variation out there to how these tools can be used even if the "text prompt" is the most common method.
@@ArtOfSoulburn Perhaps we should be thinking more about what the case ought to be, what is right and moral and what AI artists deserve, and if there are any legitimate reasons as to why AI art shouldn't be copyrightable by the person who generated it. And perhaps we should be petitioning our legislators and policy makers to do what actually makes sense, rather than letting them control the situation, and just submitting a bunch of works to see what they 'decide' or not. They are public servants who answer to the people, we get them to do what we want or they get sacked, that's how that whole dynamic is supposed to work, the people make the law. So do you want to live in a world where AI artists can copyright their work, or not? That's the questions, and then based on what you decide, do something about it, and compel the public servants to do their job accordingly.
For example using stable diffusion and the inpainting method, which allows the user to mask and area and to generate a replacement for the area in question. However the masking and painting is completely done by the which results in something that simply would not have occurred had simply been the prompt with the generative AI without them asking in place. And one little change to the masking and the whole result could change, if you bleed over into the background, you will find the background will modify what results within the actual area that one wished to modify. As such I think that using AI as a tool in this way is much different
Thanks Travis, glad you liked the video and have found my script useful.
Год назад+1
The question I have i about download, like ShutterStock have images with watermarks and those are there to show that the image can not be used for free, but Open-AI saw this as they have the right to download all example images and use them to create a commercial product, so on their server it will be at least 500.000 copyrighted images that been share around server that been processing those images. Had it been one copyrighted movie (or music) downloaded to a computer many had seen this as a crime, so why do we see it different when it come to copyrighted images? Btw, great video and good comment for the document you read, thanks.
It needs to be a little more lenient but 'case by case' applies. Someone who puts a lot of work into achieving their vision, absolutely deserves copyright. If someone uses many resources, lots of inpainting, tweaking, iterating, post-editing, etc - they should be able to copyright the final product indeed. TLDR: AI GENERATED art should not be copyrighted. AI ASSISTED art should. Otherwise, photography should not be copyrightable, as the person only framed the shot, they didn't create the thing depicted themselves.
Actually, there is a lot more work in photography than there is implied in AI art, where you just give orders to a prompt generator. It makes complete sense that you should be able to draw on top of it for it to be copyrightable.
@@artorhen “A lot more work in photography” than in AI art where “you just give orders to a prompt generator” As a photographer and someone using AI illustrator tech, you’re both very wrong and right. With a camera you aim at what you want and hit a button to take a bunch of photos. That’s the equivalent of throwing a prompt in and churning out a bunch of ai gen instances. Any yahoo can point and shoot and the camera these days does 90% of the work automagically. Focus, color correction, depth of field, stability, exposure... the human just aimed it and spammed. After that you decide what’s decent and keep those, which is part of hte creative process for both methods. Then you have ai cases where you select a model based on style and content, secondary templates, engineer the prompt with weights, and set various scales (like CFG, second phase stuff, various internal settings), adjust the prompt contents to zero in on the representation of what you’re imagining, inpaint parts to make it more cohesive and eliminate unwanted artifacts, and post-process using paint tools - and then you may end up putting that back into the AI for touch ups. That’s not even considering original sketching and manual correction, or using actual photographs as models, to tweak and refine the inputs along with the above process. Seems a lot more like the photograph process. Fiddling with camera settings, film type, lighting, positioning, choosing contents to put in it, framing, then altering the thing in photo manip software.
@@hayuseen6683 I don't think a bunch of adjustments makes for an owned piece of any kind. If I sit for an hour and adjust the settings on my phone to change the theme and everything on it, that doesn't make me an artist or a hard worker. It doesn't make me the same as someone who makes a phone or who makes a theme for a phone. Same with ChatGPT, if I just ask it to write for me, then I no longer do any actual work myself, the AI makes it. And when the AI simply takes different pieces from other places, which show very clearly in the end result, then it can't be used for commercial purposes.
@@artorhen I don't see anything addressing what I said, just strawman analogies unrelated to the real world. You ignored all the important bits. Have you ever used the AI illustrators, researched how they operate, experimented with how to use them? Serious question, not rhetorical. Back at you: Making a bunch of adjustments on a camera doesn't make for an owned piece of any kind, you're just pushing a button.
@@hayuseen6683 Yes, I have used the AI tools, I would not call them illustrators. I also used art program with implemented AI that actually are tools for making art by working based on sketches and images instead of words and prompts. To me it seems like you are completely biased towards these pieces of technology before they even prove to have a good use for humanity, which I would call almost fanatic. And I am glad you agree that making adjustments on a camera is not all that a photographer does. Much less an artist. Therefore the previous argument is rendered completely redundant.
Regarding the example at 3:50, i believe the image in the middle is still not copyrightable, even though you gave it an image prompt at the moment you sent the whole prompt (so image+text), the image given is still part of the AI neural network "training", the differences being that when you have just a text, the training comes from millions of images (each having a bit or more "weight") and when you give it image+text, the training consists of also those million images plus the image given in the prompt, that now would have a big weight when estimating what final generated image you should get (roughly said "training", of course the process is more complex, but i believe these are the basics here). So unless they will add an article or law text handling this specific exception (with all the technical details involved, which have to be full divulged probably), the generated images should have the same copyrightable status, right?
I think you're wrong about not being able to copyright the character in 6:52 . A character is more than just one-off design like we see in that midjourney image. A character is also in its iconography and basically most stuff that allow someone familiar with the work to recognize the character at a glance.
I'll be interested to see the implications of AI generation based solely on a creator's own works: an artist has a large body of work, and uses an AI to draw from their own work, exclusively, to generate new images. As I understand this, the AI generated material is not CR, but would be if further reworked/edited. What is the test, I wonder, of how much following rework/editing is required to qualify for copyrightable status?
So from my understanding, it's actually impossible to use only a single artist's data, unless they've painted several million paintings. What's more likely is you have a database of millions of copyright free work setting a base, then you train it on several dozen works of a specific artist, and basically tell the AI to favor the artist. Again though, my understanding of the tech is not perfect. As for how much reworking is necessary, that's always an open question, and why the copyright office has to take each case as it sees it. The whole change 30% rule you hear sometimes I've been told is not actually a law, it's just a vague suggestion, so I suspect you have to go with your gut and then see what happens.
The training data doesn't matter. Just because you look at van Gogh's paintings and make one in the same style doesn't give copyright to his heirs. AI does exactly the same as a human brain looking at material. It extracts an essence in a high dimensional space, call it manifold, and produces a projection of a new idea through this space.
@@franezdiy I believe lora is using stable diffusion's dataset of millions of images as a base, and using those 50 images to fine tune the model. That's the impression I get from reading about it anyways.
This is really hard to handle by modern laws ngl. How about I create an outline and then use controlnet in SD to do the rendering. How do you define what is copyrighted there? How about I "inpaint" the image until I complete a sketch but fully rendered. blows my head xD thanks for the video btw!
Yup, that's why they keep saying "case by case basis". If I had the time I'd love to make a dozen pieces of art with different AI related techniques, try and get them copyrighted, and get a more specific list from the dept about what is and is not copyrightable. But I suspect over time we'll see more cases and get better and better guidelines. Personally I think extreme inpainting probably counts as "changed enough" to get copyright.
@@ArtOfSoulburn yeah, case by case to me sounds like "bruh idk, s***t changes too fast" 😂😂 But realy love the idea of seeing what they see as copyrightable and what they dont
I think it would be a similar case of photobashing, or using sound samples in a song. How much of the original is intact vs how much of it was your own work? After a mostly subjective threshold, it no longer becomes "AI made" and would be copyrightable. If you use an outline to make a full image, I'm guessing probably not copyrightable, since none of the outline remains, but yeah that outline is copyrighted lol
@@pixelpuppy One would think that, but I've actually discovered there are some slightly different rules depending on whether your original image sources are copyrighten or whether they are public domain. I'm chatting with a few lawyers now, and will likely do a new video soon discussing some of those differences. But in the meantime, ya, hopefully we see a practical example to test the policy.
The policy is vague and ambiguous intentionally. They've purposefully made moronic blanket statements and completely disregarded any and all grey area, allowing them to use the "case by case basis" system. In other words, the policy is irrelevant, everything will be decided arbitrarily by some authority. As usual, they just want control over the situation, what's right or wrong, or what's fair or ethical has nothing to do with it.
One question which the ruling doesn't seem to address: Let's say you have a character you've created yourself. You have 2D and/or 3D representations of it made by yourself. If you use generative AI to generate images of that character via prompt-only, is the result still considered a public domain character? What if you use your own images to train a model and generate images of your own character(s)? Is it then public domain?
Pretty sure that copyright/trademark of a character would still apply. But in theory someone could edit out the character and use the rest of the image. Might also have an easier time doing something like changing the character a bit and arguing it is parody, but not sure.
If the character you prompt looks like your copyrighted character, then I would think it was still copyrightable, but it’s an interesting question. I do character animation, and I’ll set up the keframes and the computer does the inbetweens - so obviously there is lots of human input. But if you’re just giving directions and the AI is doing all the animating that could be a problem - would it depend on how explicit your directions are? Also, if your character and your script is copyrighted I would think no one could use your movie because of that no matter how it was generated - but I’m no lawyer.
From the wording, I assume that the character is still copyrightable as long as you have proof of it outside of the generated work - you just don't have rights to the generated work. The most obvious example is that if I do an AI generation of say, Olaf from Frozen, it's very obvious that the character still belongs to Disney.
That’s what I keep telling fellow artists. Instead of searching the web for hours looking for just the right reference material, you can create reference images using AI. But they still whine and reject the notion.
There’s value to both approaches. Anyways, people have always been pushed out of jobs by increasing automation. But the machines, even if they are “intelligences” don’t have any real agency, any political agency and are not legal persons. They aren’t the ones “killing” concept artists.
@@judilynn9569 imagine you have a portfolio of years of your work put in a dataset which now generates work based on you, better than you, faster than you without your consent to the existance of said dataset. Reference used by real artists don't get to be compared because: 1. ai can't unlearn 2. ai can "master study" thousands of artists in the span of several days and basicallt put those artists out of studio jobs because an ai company would rather ask for forgiveness than permission.
@@niklogus9426 ask forgiveness for what? the copy right office just legitimized this as legal. They don't need any permission. Its called societal progress. Non Ai Artists are currently just getting in the way of progress. They are obsolete, get rid of them. Go ahead and complain about how its disrespectful but keep being a hypocrite and watching movies with special effects that used AI.
Thank you for this awesome summary. I am not a US citizen and I'm based away from the US, Does this mean that every country has to make their own decisions about this? Or is a think that you think the other countries will follow?
Thanks. And as far as I know each country has its own laws, although I suspect the US making these decisions will influence many other countries. If there's other major developments in the international scene on this I'll maybe do a followup video.
I don't see this holding up to the test of time. Likely what will happen is after movie studios find they can simply replace all of the actors, lighting people, etc with AI. Then they will try to get this changed. Maybe to say you have to own the AI in order to copyright the work. OR the case by case basis will lean always where movie studios and game studios and what they want.
I'm no lawyer, but I dispute the idea that you can't copyright a character if it is "AI generated." I think the issue here is that the copyright applies to PUBLISHED works, so if you put the raw AI image out there as the initial publication, then that would be "the character," and would be public domain. However, I feel like if an AI generates a character design, and then another artist creates original art that is derivative of that design, such as making a live costume or a 3D character or drawing them into comic panels, and then that finished work is what is actually _published_ into the world, then that would become the initial artistic work, and all copyright would flow from _it._ I think that in a legal sense, it would be similar to hiring an artist to do work for hire, in which they have signed away all copyright they have to the character. What _you_ do with it is then what generates a copyright. If Spider-Man had never existed, and yet an AI spit out an image of Spider-Man, and then you drew a comic of this Spider-Man, and this is the first version of the character released _to the public,_ then you should own the copyright on Spider-Man, and nobody else should be able to draw that character
4:25 The middle image cannot be copyrighted as it is the direct product without any edition, of an AI. The manner on which the prompt is made, is irrelevant, and what the AI does is analyze the image in search of patterns, to then reproduce something different from those patterns. It is no different than inserting a text prompt. Furthermore, it might actually be copyright infringement, if the image used is copyrighted. If I commission you to do me a Mickey Mouse picture, and show you a Mickey Mouse picture whose copyrights are owned by Disney, you as an artist will also be subjected to be violating Disney's copyright, if you copied it (or it's patterns). Machine or Human, the situation is the same. Not a lawyer either btw, just offering my own perspective based on logic.
Imagine a future where a company can lobby the local government to claim that an artist used an AI to generate a famous work, in order to avoid a lawsuit because they stole the artists work and were caught.
@@rigelb9025 well it would be easy for Ai to generate hidden tags that not even a human could see, think of it like having a QR overlayed on the image unique to the Ai
As a 3D artist, I find that AI art generation is best suited as a source of inspiration, suggesting ideas, developing mood boards etc. NEVER as a substitute for original work.
Anyone who actually messed with art generators knows it's very difficult to get exactly what you want. It's a lot of trial and error, a lot of generations, a lot of fixing and tweaking. To finish a perfect project you need a little bit of coding/programming, a little bit of drawing/photoshop and whole a lot of promp crafting. I don't think role of an artist will change, just what the job entails will change. That happened before when we made the jump to the digital medium, it'll happen again.
No, lots of the AI art I have seen is superior to LOTS of paid art. I've generated thousands and thousands of images, many of which are great and can be further changed refined with almost infinite choice.
That's a legal grey area. From my understanding, technically since the art can't be copyrighted, it's public domain, so Marvel could use it, but they don't have exclusive use of it, since its public domain, and so would be unlikely to use it, since they want only stuff they have full copyright over.
Nice video. You did miss one thing and that's trademarks. Mickey Mouse's copyright expires in 2024 but its unlikely you'll be able to use Mickey because he's trademarked. Any replication of Mickey-like characters which one may confuse with Mickey would infringe upon it. I also think there are plenty of loop holes that will be tested in court. AI has evolved so fast that you can do stick drawings which get converted to art. The line where how much one must do to be the author needs clarification. My gut instinct will be along the lines of, a human must have created 51% or more of the art to be deemed the author but we'll see.
If someone uses AI with their own characters and art style, I feel like they should be able to have copyrights over that because they trained their own personal model with their own art style. However, this becomes a problem later if they release the model to the public because they could then say other people are using this style too, so how do we know you made it?" In which case, people could prove that they made it by sharing the video they made of them making the model. But this won't work either, and the reason is that you can't actually copyright a style, only your characters. But characters often look so similar that it would still be kind of hard to do that. People have been using multiple models to mix styles together similar to how real artists take inspiration from different parts of different artists' styles, mixing them together to create something new. People have been uploading models very specifically for certain poses. And these are all uploaded for free for use. Nobody is charging for them, however people can support your Patreon if you ask them to (since generating large models can be long, labor and cost intensive). If people could copyright a style, then people imitating Disney's style would be in trouble even with their fanart. Not to mention if you copy another creator's style, who's to say that creator didn't copy from someone else originally to get some major aspects of their line work? The shading? The way they draw noses, blush, highlights.... you can't prove much. Someone can say "that looks like that person's style" and you can bring up a lot of other artists that look similar. Even before AI art, you'd see lots of pretty anime art on Instagram being posted, and everyone would do the same style because they saw that is what sells. On youtube when people notice an editing style is holding people's attentions, suddenly you see that everywhere. people will follow the trends to create something that sells, and that's just facts. And people can mix art styles together until they create a model that really is their own style until someone else takes inspiration and makes a model from what they post as well, or follows tutorials showing how someone learned from their style to replicate it by hand. I don't think this will work because you can put a ton of anime girls from a wide variety of series' by each other that have the same hair and eye color and not even be able to tell they are from different series', or who they are. Most shows, even before AI, would basically copy-paste characters, barely change anything, and even use the same voice actors for characters with certain traits. Almost every tsundere character has the same voice actress, for instance, and most anime characters all have the same face with very minor changes to the bodies. If the show has natural hair colors and lots of girls, you can't tell them apart unless they have their hair styled. So if you see them with their hair down, you think it's another girl. That's honestly pretty bad. If they are looking for sameness, how would regular anime studios that are mostly copy-pasting their own work be able to get away with this? Not to mention, why would they keep drawing it when they can easily train their repetitive styles and just do very minor tweaks using real artists to fix the anatomy after? Plus, if they get a character design and pay someone who knows how to make 3d look perfectly 2d, 3d models to video is coming out soon, so people could do most of the animation with their own bodies with only some changes being adding some effects to make things look more cartoony and less perfect. I feel like this is going to harm artists everywhere, and I feel like everyone on the outside of the AI scene needs to know something. This AI stuff is only dangerous if you let big companies take it, try to monopolize it and control it, and charge for it. We are trying to make sure art stays free for everyone, sharing our models and not keeping them to ourselves. Even people who have drawn everything by hand always can and should be using it to help speed up your work to keep up with demand, and to basically have a helper. We are gaining tools that are making it so big companies have no power over us. Why should we work for someone else? Why should we work on a series that may get cancelled because of a lack of funding when we can do it all on our own at home with ease? Why should we work for someone who overworks us, under-appreciates us, will replace us the second we die, doesn't care if we are starving, will replace us if we strike, and wants to replace us with machines anyway? All companies everywhere want to get humans--beings that require breaks, food, water, compensation, clothes, air conditioning/heating--out of the picture because they want to generate the highest amount of revenue possible. But they aren't at a point where they can successfully replace us yet, but we are already at a point where we can make these things at home on our own with no team. Why should we work for them? They are losing power over us. We can take art into our own hands and have more artistic freedom than before.
Thanks for this video.. what I’m wondering about most is that next step - can the AI legally learn from copyrighted material?.. as someone who has a law degree (never practiced) and is an artist, I think the route I’d feel most comfortable with is erasing the current databases completely and starting up a voluntary contribution system with creators, copyright owners and the public domain.. kinda like Wikipedia? Taking a step back to take some steps forward.. when thinking about the idea of “derivative work”, imho, that’s the only thing that makes sense to me.. To me, inspiration and the realm of imagination are more murky and less defined/tangible than actually using bits of data from a pinpointed, copyrighted source.. I welcome the tool/technology.. but not to rehash and utilize work against the will of creators.. but if it is aligned with the will of creators, then game on! And I would imagine that new database would grow pretty rapidly..
This is very promising! It will help establish a system that will influence the laws and court cases in favor of protecting creators and limit the negative impact of those using AI and trying to pass it off as their property.
Thank you so much for making this video! 07:57 This part is so sad, indeed, only time will tell. Profit prevails in the end but I'm still hopeful there would be other avenues for concept artists but yeah... most artists would be displaced with these new tools
I had on an additional thought about your robot and using an image to create another image. If you create an image using prompts and then attempt to create additional images using the original generated art, you probably wont be able to copyright it, unless you changed it enough prior to uploading the new image as a base for the additional images. It seems the best use of AI is to generate a quick image to use as a foundation for your own creative works.
I'm a little confused with the idea that any generated character is in the public domain. It means that you can just generate a bajilion random things and say that no one can copyright them because they are generated. You can run a story generator to generate a million random characters, and then use them to block other people's copyright claims. It's a bit silly but still a weird loophole in the description. Also, there was no mention of AI using training data that is of copyrighted work, or of non copyrighted work but without consent. I think that's what artists worry more about since at that point concept artists can become used just for creating training data rather than actual original creations. The originality of the art direction can also suffer since any art would be generated from some other artwork's art direction. I think ai software made using copyrighted artwork, or artwork sampled without permission should be liable for copyright infringement or legal action. Not having that would allow for even greater disrespect for the role of the artist than already exists amongst some developers and parts of the industry.
Ya, I don't think even the copyright office fully realizes the scale of AI, being able to pump out millions of pieces of media in a matter of hours or days, and the implications that will have. Although I suppose there's an argument to be made that there are so many combinations of things that maybe even a computer can't make them all. The bit about training on copyrighted work is #6 in the video, and in the policy it's mentioned only briefly and they basically said a decision on this is pending. I suspect they won't make a policy choice until the 2 lawsuits are done. And yes, that is a big important question, but since this video was focused on their policy document, I felt a more detailed discussion of that topic is better left for another video.
I feel like copyright issues with AI should be pretty straightforward. The biggest complaint I've heard from artists is that AI is copy pasting the artwork used to train the AI which is a big misconception. Lower quality AI may do that but not any AI that's actually impressive. Training material isn't used to copy and paste an object it's used to teach the AI what an object looks like, what makes a certain style tick, etc. With that the AI can determine how to draw an object in a specific style while only knowing what the object is and what makes the desired style that style. Using an image as a prompt is more questionable to me but it doesn't feel much different than an artist using another person's art as reference material and end up drawing something similar.
Great overview! Thank you! 👏👏👏I will predict that these laws and regulations will have to evolve as quickly as the technology is. It is interesting that the "changed enough" premise is documented as it is as, IMHO, this appears to weaken existing EULAs at the majority of "For Sale" content repositories that say no derivative content can be sold as your own.
Thanks. Ya, the issue with "Changed enough" is it's a value judgement. Which it kinda has to be. But that means you can never be 100% sure as an artist how much you need to change something to be safe. So you basically have to do your thing, send it to the copyright office, explain what you did, and see what they think. And yes, I suspect we'll see a ton of people try all sorts of different things, and each will get its own ruling. So there may need to be a video 2 at some point :)
That's not what "derivative" means. A work that is colloquially "derivative" could legally qualify as "transformative" and be protected under fair use. They themselves specifically noted that works containing photobashing materials can be copyrighted, for instance. The important point here is that these policies and their nomenclature are intended to protect artists from infringment of their copyrights. Since an artificially generated image may or may not be infringing on an artist's copyright, discretion on your part is necessary. I'd suspect the threshold to be similar or identical to fair use. As I understand it End User License Agreements are at the discretion of the website, and can't be "weakened" by the policies of the US Copyright Office. A violation of the Constitution or a law that contains specific prohibitions are the only things that could do that. A website could probably make the plundering of your copyright a condition of their website's use, it just wouldn't be a popular or appreciated policy among the general populace, and the site likely wouldn't survive long enough to see divends. What you're saying suggests you couldn't sell public domain works as your own. Which legally you can do. Ultimately these repositories don't want you stealing peoples' work, setting aside the AI's inherent lack of discretion about plagiarism, you can't plagiarize the robot you instructed, therefore changing the work to suit the policies of the US Copyright Office *might* not even be necessary to stay within the spirit of your EULA, which means you *might* be able to sell it without changing it at all, you just don't have the right to claim it as your exclusive intellectual property. Which is usually desirable, if you're a professional artist. I'm not a lawyer, but from all I've ever learned about this stuff that sounds correct
Around 7:45 when you're talking about the copyrightability of the character, I have to argue somewhat with your assertion that the character is not copyrightable. Reason being, there really is no character yet. It's a depiction of a *potential* character. I'd argue that you need to assign an identity to the depiction to actually turn it into a "character" (which could include naming it, assigning character traits to it, giving it a background story, etc) at which point human creativity has entered the scenario, and thus does becomes copyrightable. But before you do those things, yes I agree it's not copyrightable, my reasoning being not because it's AI-generated, but because no real "character" exists yet.
There could be some truth there, in this example I am just talking about the look of a character, but a true character needs to be more fleshed out. One thing I've discovered talking to lawyers about this stuff is that just because something could be copyrighted doesn't mean it will be. So just the look of the character could be copyrighted, but it may be likely that the copyright office will ask for more detail (name, character traits, etc) before they're willing to give you the copyright. Which means that all of the rules they put in their policy are somewhat fluid, which is frustrating. There's just so many potential exceptions there. But anyways, great point, thanks for sharing!
This strikes me as an exact rerun of the initial question of software copyright. The initial ruling was that software could not be copyrighted. The judge held up a cassette tape with a program on it and said that because he could not read it, it could not be copyrighted. That judge got "educated" pretty soon afterwards and the decision was overturned.
I have used the "prompting an AI generator is the same as commissioning an artist" argument for a while now. It is crazy how many people think they are artists because they put in some text into a computer. Sure, "artist" is pretty subjective... but in this sense, you would be a prompt artist, not a visual artist. Creative jobs wont be hindered much by this tech. For example actual concept artists will just use the tech, instead of thinking the creative company is going to hire outside prompt artists to replace their creative talent. I could be wrong, but I doubt I am going to see "Looking for AI prompter. Must have 2+ years experience in writing AI prompts. Art experience not necessary" in any real creative job postings.
i honestly think big companies will fight to change the law when it comes to this . makes to much sense for them to go "well we made this program, and we trained it on our own art so we own what it makes"
thankfully the tech to create images is, even though highly complicated, reproducable by smaller teams. there are already tons of open source solutions for creating images and they may be not the top image creators out there but at the moment they are good enough. even if these free tools are always a year behind, in just a few years they will produce quality that is easily enough for most consumers out there.
That that's a good question, if I train my own model on images I took of a real subject and I use my model then it should definitely be considered mine right?
In accordance with the argument, the copyright office is suggesting that no photograph is copyrightable by a photographer. The artist uses an Artificial Instrument to just point at a sunset as though a prompt. To capture the essence of the moment with automatic settings without any photographic knowledge to create the image. The logic used suggests that every photographer that has used a digital camera cannot copyright his work. Going a step further, you can commission a photographer to take photographs for you. This transfers the copyright to the person commissioning the works. If a person has paid for the right to generate the art they are the copyright holder. You are correct you are not a lawyer and this Gini cannot be put back in the bottle.
Here's a scenario that I'd like to get some clarity on. If I prompt generate an element, then Photobash it to be how I want it, then I use that as an image prompt for a final iteration / polishing of my bash, is the new work now copyrightable? The process of Photobashing would be required in order to accomplish the final work to which you could not arrive to without it.
How much "the fix" part from machine in your final arr, I believe will be one of considering factors to decide that those final image can be copyrighted or not . But like mentioned in video . Not every case is equal .
Thank you for this summary Neil. You've been doing an excellent job covering the AI topic from the concept artist/art director perspective. It is true that painting over some prompts could greatly reduce the enjoyment of the work. I know I'd hate it.
Glad you've found these videos helpful. Ya, if my job was 100% tweaking prompt images I'd not enjoy it either. I might go off and brew beer for a living :) Using prompts to create photobash elements, that I'd likely be fine with, but it all depends on how much I feel I am able to personally contribute to the final image. We'll see where the industry goes.
At @4:32 I don't think you can copyright the middle image of the robot. It's equivalent to commissioning artwork from an artist and sending them a sketch instead of written instructions. You have the rights to the original sketch. The artist has the right to the work they create from the sketch, unless (as is usually the case), you have them sign a contract assigning the rights to you. But an AI cannot sign a contract (yet). Like you, IANAL.
Yup, that would be the argument as it why it's not copyrightable. I'm hoping someone tries it so we can get an official ruling, since image prompts are pretty common, not as common as text prompts, but still frequently used.
Image transformations are definitely one of most tricky aspects of this. In video example the orginal image is pretty far from ai result, but you can generate stuff that would still be recognizable as the original image and yet visibly modified by ai. And then there's turntable character sheet generators, inpainting etc etc
@@bazejkaczmarek8271 Yup, these guidelines provided by the copyright office are going to need dozens of revisions over the coming years, maybe the coming months :)
I feel like the two copyright issues being decided are a bit in opposition to each other. Like if the AI can't train on copyrighted images because it isn't a human and isn't really "inspired" and is just a tool. Then at the same time how can it be said that it can't be copyrighted because we're only giving directions to another entity that isn't just a tool?
Great video! I think what this means is that when we use AI for something we might want to copyright, we should use original images as part of the prompts. Even if you just draw a stickfigure or something as the base image prompt, it'd probably have decent chance of being copyrightable art. I don't think its a surprise that the AI characters are not copyrightable as they come out, that'd be insane(think of patent trolls, but with characters), but I do think that AI *based* characters can be copyrighted. Typically character copyright applications are not "one or couple images of the character" but rather dozens and dozens of pages showing and describing the character in high detail, from style to personality to gait to colors and much more. They'd have character sheets that show all the angles, different outfits, etc. Copyrighting a character is not a simple task and for that reason I think AI-generated characters *can* be copyrighted: the work to get an AI character to the point of having a chance at copyright would need so much work and art and decisions that the resulting character would likely be considered copyrightable at that point. (especially if the original prompts that the character was based on had an original image as part of the prompt as well.)
You can do all this in a couple of hours or faster if you use AI as substitute. I mean you basically just need to train a model that fills in the stuff depending on your description, you just need to proof read and modify it to your liking. I guess we need all to rethink "simple task" when you basically can automate anything in a process except the idea.
there is also the style used to produce the image, such as pixar style. the models are trained in producing specific styles that will always show even if you add your own picture. maybe have a look at creative commons I think this is a fair way of looking at it
It's absolutely critical that AI can be trained on copyrighted artwork. If not, AI in the western world dies and we put two different standards towards AI and the rest of the world. A human brain can be trained on copyrighted stuff, why not AI? This would just become a total litigation hell, new patent troll playground and if any AI could survive it, it would only be the big companies AI who can defend against frivolous and not do frivolous lawsuits.
"A human brain can be trained on copyrighted stuff, why not AI?" - the answer is on the question itself - because one is a HUMAN and the other is not. We can issue rights for people, not machines.
I couldn't agree more. If you think things like RUclips's broken Content ID or false DMCA take-down notices are bad, imagine that times ten thousand. I'm not optimistic about this. I fear we will make the wrong decision, and the most powerful tools in the world will be owned by the countries that we are least friendly with.
@@lordavius So ... how do you regulate one, without something that bleeds into, or impacts, the other? That IMO is the biggest problem I have with this point, it seems hard to translate that into something that won't fuck people, and artistic people, over.
@Gondor avalon the answer is pretty clear to me already. One is a human being, which the already in place laws apply. The other is a machine, a tool that is not eligible for any rights (including copyrights). The other nuances are already exemplified and talked about on the video itself - are you using the tool to create yourself some new kind of work? Like Kitbashing or reference building? Or are you (like millions of so called prompt artists who are doing waifu) just getting the end result and posting it as your "art"? This has already been widely discussed since the inception of image searches on the internet. Imagine the prompt is an "image search" that is Kitbashing to the pixel to find the image that better resembles your "search". Can I copyright a Google search result? What if Google did some fancy copy/paste on the images themselves, would I still be able to copyright them? I guess no.
@@lordavius "Imagine the prompt is an 'image search' that is Kitbashing to the pixel to find the image that better resembles your 'search'." Why? That's not even close to how these models work. Generative AI models are not doing "some fancy copy/paste" any more than you're doing some fancy copy/paste of letters to write your comment. It just doesn't work that way. In order to do an image search, the AI would need to have a database of images to search, and it simply doesn't.
What about the input material used for training different models? For example githubs copilot, it uses copyrighted (open source, but may still require additional terms as attribution to the original author etc) as training material for the AI model. Does the model / github obey the original copyright license when code generated does not have these required attributions? The same applies to generated pictures etc, the original training material was probably copyrighted and had a human author, does generated pictures violate those copyrights? If your middle mechwarrior (the one generated from a direct image prompt) is copyrighted to you, then all images generated are probably copyrighted by someone unknown.
It would be good for all concept artist to be skilled at 3D modeling and animation so they can take the concept to the next level, view it from all angles, and even give it a voice to ensure that, enough inspired from AI art, it becomes something totally new, with investment from the artist.
@4:40 I believe you are conflating trade marks with copyright. Specific works of art can be copyrighted, but characters themselves are trademarked IIUC. Please correct me if I'm wrong.
You know, as a guy who LOVES making AI art? I'm okay with this. Feels like turnabout is fair play. I mean I can still MAKE things with the AI, I just can't copyright it. That's okay. Also I realized this is necessary because a Switch game came out with AI generated art (The Internship) and the prompts they used are so basic and lazy I basically created one of their characters myself on accident two months before the game came out when I was first experimenting with NovelAI! I just hope the thing about AI trained on copyrighted images doesn't blow up in everyone's face.
There is certainly a big discussion to be had about the fact that even if you can't copyright a piece of AI art and get exclusive rights to it doesn't mean people making AI art as a hobby won't be totally changing the industry. I mean, if everyone decides to stop watching traditional films, and just watch free AI generated films, the movie industry as an industry making money and employing people goes away. Or if you make a youtube channel that shows Ai artwork, you can't monetize the channel, even if someone could exactly copy your channel and try and monetize it. We're in for some wild times.
In practice, every AI-generated piece of art is 100% copyrightable. If I copy someone else's AI piece and they sue me, I will automatically lose in court, because I cannot prove it is an AI piece. They can always claim that they made some edits. As a consequence, the US copyright policy is quite irrelevant.
Not true because most of the current AI generators imbed their images with an invisible watermark that proves that it is, in fact, AI generated. Most of them also back up every single image that they've ever produced. It's not that hard to find it, unless you're running a local Stable Diffusion system and even then, we don't know. The real question is going to be, how much "editing" is it going to require to be considered "sufficient"? That will be for the courts to decide.
@@BitchspotBlog A lot of us are running our own SD. And I'm pretty sure it won't be long before we're able to make models just as easily as images... what then? I put these sets of images into this model... that means all images should be copyrightable?
Every image coming out of AI is a OOAK, place signature on it, done. This is the classic "Are Andy Warhol Xerox copies of Soup cans someone else designed copyrightable". Well, is it transformative? One cannot enforce an unprovable position; Copyright office is not going to be able to discern or prove an AI assisted image vs human only image, and to what degree... and yes, AI art is transformative by definition.
@@BitchspotBlog You are right, instead of saying "every AI-generated piece of art is 100% copyrightable" I should have said, "every piece of AI art that I generated on my own machine is 100% copyrightable". In practice that is not important distinction, local SD is easy to install and future versions will be even easier to install and less hardware hungry. But yes, if you find an AI image on the web that has the watermark then you can probably copy it.
I am curious as to whether ai art by a paid subscription (especially if the company says you own the copyright if you "made" the art) would allow it to be seen as a work-for-hire where you are instructing an employee to create a piece of art. If that were the case any of it would be copyrighted.
With current laws it looks like the answer is no. The company can say you have the copyright because you paid them, but the US gov doesn't recognize that copyright. The government would have to pass new laws.
So I’m getting clarification on this point from a lawyer, sounds like there may be extra nuance to it, will probably post a second video with extra info in the coming week. This topic is deep :)
Maybe but the real problem is, with the flood of art being done from A.I. that's very likely going to reduce the value of human art, especially because money talks in this world, if someone can create art for a game for instant using art, why would he employ a human to do it when it's cheaper and faster to do it with A.I. If the creative industry is not careful, they could reduce their own value so much that many won't want to use their work because they don't need too when there is far more being done with A.I.
@@paul1979uk2000 AI art can be 'human art' as well. There's a lot of overlap there. Even a raw generation can and usually does require artistic vision via a prompt, someone not only has to learn things about art, like color schemes, styles, composition, camera angles etc. They have to learn how to use the tool and communicate with it, crafting a linguistic prompt - you can't just type any old garbage if you want a good result, there is an art to crafting prompts, i don't see it that different than other wordsmiths, like poets or authors. Not all prompts are created equal, and you can tell that by the results. Nevermind the fact you can also use AI the same way you would with photoshop, to edit and digitally manipulate your hand drawn scribbles or process a draft image you made in MSpaint or anything along those lines. I see no reason why AI art can't be copyrighted, it doesn't just make itself into existence, the person who generates it should own it, the tools used to create it should be irrelevant. As long as it doesn't infringe upon another's IP, then what's the problem? It's not like the AI is going to claim rights or ownership of the content, is it? So dumb. But the policy makers are just a bunch of old fools who don't like change or understand new technology, this stupid shit ought to be phased out in a generation or two at most, hopefully less.
Actually I am impressed. I was expecting worse, but their approach is pretty good.
Not really. Because it blankets all of AI by using a strawman against the entire medium by using vague and arbitrary examples as the whole. When most skilled AI artists, spend hours prompting and teaching the ai though photo learning, manipulation, edits, and masking. Which very much makes it intellectual property of the one creating it by any standard previously held for copyright. This is like saying that a DSLR photo can't be copyright protected because of the enhanced quality with things like autofocus. Or digital art when you use brushes, filters, shapers and other generated aspects which don't require the actual ability of the user outside of aesthetic and creative decisions to perform. And last I checked we don't give Nikon, or adobe the rights to our art when using them to create art.
With any amount of thought this quickly becomes hypocritical to other positions and clearly needs to be more thoroughly assessed.
There needs to be a set criteria of personal interpretation and action which dictates rather AI art is copyright protected or not. Just like we do for Fair use. Ai art should be copyrightable and specifically by the person creating it, rather than who created the program. But there should be a large enough delineation from the AI's initial results and your own to show fair use and thus copyright protected. Because the initial art produced is the art the coder envisioned based on the criteria. Anything past that begins to become your own interpretation. because it's changing on the back end the variables the programmer set to adjust for your aesthetic composition. So like any fair use case. Minimal changes wouldn't be enough. But at a certain point you cross the threshold and it becomes fair use.
@@zephirol4638 If you're putting in work in your piece, it can totally be copyrighted. The prevention of AI works being copyrighted is for people who are only prompting/generating images, rather than curating their datasets and taking their work further and beyond the minimum.
@@rodrigobarraza Fair enough, if they're accounting for that then it sounds fine. Just by how the video put the TLDR it was sounding to me more like ai as a whole was just a no go unless you used a completely different medium to make a complete parody of whatever you configured through manipulation of data sets. Guess that's what I get for assuming without actually reading the actual proposal. lol thanks for the clarification.
@@rodrigobarraza do you really think it's going to be enforced that way? No it's going to be enforced to screw everyone that isn't rich
@@ehtresih9540 Of course it's going to be enforced this way, that's how these regulations work. As someone that's started a couple of tech start-ups in the past, these types of legislations are important for individuals and corporations alike, especially when it comes to technologies that begin to advance rapidly after being around for many decades.
Having money means that you can afford more, better lawyers, for longer, and is a complete different issue than what you're trying to bring up, but ultimately, these cases will be treated similarly, as long as the proper channels are taken, and proper precautions are met.
Well done! I AM an attorney and you did a great job at hitting the basics. At the end of the day, case-by-case analysis will rule the foreseeable future. Use AI as your brainstorming mate and then modify with your own God-given imagination and talent.
One thing . AI art generator is just it a generator. In order for it to be a art generator. It requires Prompts it is something that is required. And prompts are what ever one can dream up its your Thoughts In Grate details that you hav to put in to the generator. That makes it my art. Now some wold say no because the generator still did the work . but with out the In put of you thar is nothing.
Now if someone can sit and feed a AI Details to create a specific story line in to a script. Then had a AI Create a complete movie Based on the script now this is seen often in some Digital movies on TV. And thay are copyright. So in my Opinion if someone can mack a AI movie and copyright it then the AI art you do should also be copyright protected because it is your Is thoughts and ideas and dreams .
Information must be free. Art must be free. Capitalism must die.
@@BRAVEN32m12 Sounds good in theory but not in practice. Your prompts are the same concept as commissioning an artist to visual your ideas. I've done commissioned paintings for customers that still don't own the copyrights to....they paid a lower price for the right to use it for their one project project and nothing else. I still own the copyrights to the image.
@@Jason197987 moste ov my work is blended with hand drawn and photographs
What is art to a blind man?
What is copyright to a man made into machine?
... A legal paradox but one in which will inevitably arrive...
Evolve or be left behind, Mankind.
I like this, Ai art should be used as stock photos for artists to use for their own creations but not sold as copyrighted material this gives way more room for Ai to be used as a tool to make new stunning pieces of animation, music, videogames, etc cutting hundreds of hours of tedious work and letting the artists spend more time perfecting their vision. I'm hopeful, I just wish big companies didn't take this as a free card to fire everyone cause it's a wasted opportunity not having all those people working on different parts of the creative process
Well said
Those companies will just put themselves out of business eventually, so is evolution
Human... Your species isn't alone.
But I won't explain it. Humanity will act as predestined.
Unfortunately, that last statement has always been standard business practice.
When companies get ahold of a new technology that cuts the workload in half, they don't allow their workers to work for half as long, but rather fire half their workers...
tell that to a tech bro..
Thank you for this video. It's very informative for someone like myself as an artist that doesn't quite know what to make of the AI landscape. The conversation tends to have a lot of emotions from both sides which makes it hard to approach. So your video is refreshing in it's succinctness and even keeled explanation and from a perspective as a working artist.
Thanks William. Ya, these topics stir plenty of emotions in me as well, but getting upset I feel won't do us any good, I feel talking about these issues rationally will help us get to the meat of the arguments, and hopefully find solutions that help propel us forward. We'll see how it all shakes out.
AI = copyright infrigement. Nothing more.
People with power hate creative people.
You'll been fine as a pro artist, depending on your field. If your a freelance artist youll lose your cheap clients not your loyal ones. I use AI art for my projects. I tried using regular artists but the field lacks quality control n can range from affordable to extremely expensive. The hiring process alone dampers all projects n leaves AI as a viable, affordable, user friendly process option. The hiring poetion became so unviable i stopped using art all together until AI art came around. Do i still want human art? Yes, but i dont want the increasing risk of quality control to harm my project.
It's cute the copyright office or any other government office has any sway over AI. It's up against our precious stock holders so it will loose.
This is an interesting approach to the whole A.I.-topic. The major question for me is, how will this copyright regulation be controlled? If an A.I. generates an images, and nobody keeps track of the process, how would someone prove, that there was an A.I. involved?
There is software out there that can detect if an image was made with AI. But it's unclear to me how reliable that software is, and will it be able to keep working as the technology evolves.
as with any legal dispute, evidence helps, but its all down to the argument.
Copyrights are not simple.(edit:registering a copyrighted character is what I'm talking about here, not "natural copyright" which is easy - or automatic in theory - also I'm definitely not an expert in US copyrights, but have some experience of EU's) If you want to copyright a character, you need tons of images and details and descriptions to do it(copyright applications for animated characters can be tens or hundreds of pages). So if you want to copyright a character, AI based or not, you need tons of documentation and details about the character.
If you simply AI generate a character, you won't have cohesive style guide for the character that is needed, so either you generate a ton more cohesive images and information for the character: the design history, setting, the defining elements, etc (text&images); enough info and images to get a copyright and lie about the process, or you simply tell the process involves AI and you still likely get a copyright if you did all the rest of the work and not just one image of a character.
As for copyrighted AI art, there will be some ways to tell from other detector AIs to metadata to tell-tale elements of AI generation, but it's difficult. Better question might be whether anyone cares about digital art copyrights really? No one is breaking a bank from digital art, even less so with generative AI's existing, so if someone copyrights a generated image as a real digital image, it makes a little difference in real world because digital art had very little value to begin with. If you keep copyrighting AI generated stuff and some legit artist complains about you lying in your copyright, I think you might be asked to produce somekind of proof of legitimacy for your art, like the original files, though I might be wrong about this. On the other hand if your digital AI art becomes super famous and valuable, it will be found out it's AI generated pretty quick by closer inspection and it'll lose the value. Very pessimistic way to look at it, but it is what it is for digital artists...
Musicians and 3d modelers and likely all other digital art professionals as well, will face these same questions and problems soon :P
@@z3dar Ya honestly seems to me personally I've always been super pro public domain. So I'm considering running two sets of projects one Pen name where I experiment with new tech and release everything I do public domain and one where I use my further developed skills later on to create serialized content that I can copyright the characters and such with. These early stages I don't care to much if anyone takes the stuff I'm making. It's mainly practice and if they like it enough to reuse then I'm personally flattered.
it will work as well as the RIAA did stopping napster
I was afraid that the U.S. government wasn’t going to understand how the generative A.I. works. They absolutely nailed it and are starting off with an incredible approach. Super happy with the development and I’m ecstatic they sided with the artists :)
This is a step in the right direction. AI-produced work should be part of a global commons, not copyrighted and sold. People should be able to access AI-generated art and make derivative works from it to their heart's content.
Just to note, even if it can't be copyrighted, it still can be sold. You just can't stop anyone else from selling the same image :)
@@ArtOfSoulburn
Hoping we can make that no longer profitable
Great video, simple to follow and clearly explained! I'm an artist in the games industry, and this stuff has made me lose sleep for weeks. Your video has really put my mind at ease.
A therapist would also be helpful, seeing as something as small as this causes sleep loss...
So basically we are going to need to "show our work", as my old math teacher used to demand.
haha! Yup, seems like it.
It's not going to work either.
AI models are already able to replicate the gradual process, drafts etc
Hey, my math teachers said the same thing.😂
They are talking Islam and do not understand it. Who will sue if you just start selling? They wont budget copiers to compete with you. But will let anyone steal your sales with your own product even if you are there first and basically it _is your idea? Like forcing lottery winners to shed their winning ticket because, well, because they won and all others did not!
Ultimately there is nothing that can stop AI art and soon enough AI video, games etc.
This is the perfect explanation for someone who understands things better visually!!
Makes perfect sense. It’s no different from how I, as a graphic artist, buy elements from places like creative market, then use those elements to create a completely original unique design that is of my own design. As an artist I’m capable of making these elements myself from scratch, but buying elements premade helps me work faster and more efficiently, and I am supporting another artist in the process. It’s artist collaboration. I see using AI exactly the same way.
for now it isn't at all sadly. Ai bros sell NFTs or make big tits asian girls. And I say it's around 95% of its use currently, while also killing new artists and current artists, and of course the artists that the Ai is based upon
To the point, clear, concise, and no bad jokes. Thank you!
As I understand it, the US Copyright Office will consider each application on a case by case basis taking into account both the AI generated portion and the human author's contribution to decide whether it gets copyright. The irony is, in future, the copyright office will probably use an AI to give the decision on each application.
The truth is that you are right but that they use the AI for each request could be more advantageous since imagine analyzing hundreds if not thousands of AI images each, with an AI that analyzes them together with a person in charge it will be faster to rule if Will it be copyrighted or not?
Just like google penalises bot written content from search results using AI.
Yep...Sounds about right...Errrrr...Copyright. 😁🤪
They definitely will use AI in the future
Its funny - the first thing I was thinking is say you work with a publisher and they find a way to use AI to reproduce even your "human" work on an image, then say "see he used ai" and void your copyright and use your images for themselves and their own profit.
Thank you for taking the time to research and read the boring text and then you explain it to all of us, lazy people.
Haha, no problem :)
Definitely needed this! Great info! I appreciate the in-depth overview of what is deemed as a standard, in the US Copyright Office concerning generating AI.
This video is appreciated. It will surely be ignored by many, but it needed to be posted.
This is a decent start, I think. My filling related to the technology as creative can be best described as mixed rn. I'm not aganist it, and I can see its potential benefits but also feel it has the potential to effect humanity negatively in the long run if left completely to its own device. I am glad we're having these discussions on the ground level.
I’m not worried about it being left to its own devices.
I’m worried companies like Amazon or Disney will try to monopolize the market for the max profit.
@@cara-seyun That's actually what I meant tbh. By "own device" the people behind the technology and corps.
WTH it only just hit me that I'm watching Neil Blevins!!!
Your 3ds Max scripts have saved countless hours of my time!!
You sir are a legend in the 3D community!!
Ah thanks, glad you've found my scripts to be useful!
Wow that was a surprisingly sane response from the US Copyright Office. Thanks for the detailed explanation.
Traditional Elements of Authorship
(3:08) The material (or "elements") in the image is not copyrighted but the image itself is copyrighted, and you own that copyright. Regardless of how little work you put into creating that image, or how simple the process (cut and paste), you created it. A copyright is NOT dependent on image complexity or level of completion and does not prohibit the uses of textures generated by A.I. The owners of the A.I., according to the rule, can't claim a copyright on the textures, but if someone else uses your cut and paste image in their own video or published media without your permission, you are able to claim a copyright infringement.
(3:45) The middle image is NOT copyrightable by you or the A.I. owner, according to the policy. The image you provided of your own creation, according to the rule, is simply a part of the instructions you gave to the "artist", but you are correct, you own the copyright to the character and image on the left and right. In this particular case, the subject in the middle image only shares the base color of the original 3D model - it's not even a depiction of the original model, and only influenced by the original image. You are correct on the other two and according to the rule, you can freely use the subject image produced by the A.I. to create the copyrighted image on the right.
(6:50) The point of the policy is to prevent someone claiming ownership of a complete, unaltered image produced by an A.I. tool, and to prevent the owners of the A.I. tool from claiming copyright infringement of all material, or bits and pieces of art ever produced by the A.I. The image you show of the alien in the video can not be copyrighted by anyone in it's unaltered state, according to the policy. Understand, it is the image itself that can't be copyrighted. If it is your concept and your description that generated the character, regardless of how vague the description, you have ownership of both the look and attributes of the character - the concept of the character belongs to you. Any image you latter produce of that character is fully copyrightable by you, both the image and the character - the same as your final image of the robot.
(4:35) *Probably the most important thing to point out:* Not only can you not copyright the "fan art" of the Spiderman character, you can be hit for copyright infringement for publishing these images.- *Except in the case of this video.* You are using the images to illustrate and explain a point of the discussion, which is covered under "fair use" rules, probably the most misunderstood and abused copyright rule of them all.
"The point of the policy is to prevent someone claiming ownership of a complete, unaltered image produced by an A.I. tool, and to prevent the owners of the A.I. tool from claiming copyright infringement of all material" Is it really saying that, though? I thought the statement was that just using a text prompt was not enough control over the final image to be considered as having enough artistic input/intent to be copyrighted? It seems plausible that someone using ControlNet to control the pose of a character, shape of objects, gradient of lighting, color palette, particular style trained on a LoRA, different prompts for different areas of the image, etc could be enough intent to be copyrighted, even though the AI still produces the final image. The question on the input side seems more about just how much of the above is needed to be provided.
@@ShawnFumo I didn't design the policy. *That's what it says.*
Your argument has merit, but as you point out; the problem is, who decides how much detail is required in the prompt to warrant a copyright on the AI produced image?
What it also says is even if the overall image is produced by a machine, if an artist (human) develops the AI's final image further, as in the final robot image, then that refined image belongs to the artist and is copyrighted.
One of the foreseen problems is aside from who was first, two different people could give two different prompts and the AI might produce very similar images. Which of the two people gets to copyright the image, and can the one person then sue the other for copyright infringement.
The idea of copyrighted AI art is a kettle of worms. The best approach is to not allow the AI to claim a copyright on anything it produces, and to prevent people from claiming copyrights on any unaltered AI produced image. THAT is the core of this policy.
Really informative video. It's refreshing to hear a calm, nuanced discussion that doesn't get swept up in all the clickbait headlines and stupid internet discourse.
Thanks, ya, I'm not looking for excitement or controversy, just trying to analyze and discuss facts, glad you've enjoyed it.
@SD789 Ah, yes, the most reliable source of medical advice: some guy crawling through youtube comments, looking for someone to get offended at. I think I'll probably listen to my doctor's advice over yours, lol
@@ArtOfSoulburn 🎉😢😢😢😢
@hyperix shows how much you know lmao. Stop watching fox/crowder and do some real homework
Excellent information. Thank you!
💐🙏✍🦊 If I could only thank you enough, for presenting this timely and meaningful video on the new AI copyright laws, as they are changing vis-a-vis AI generated art. I really appreciate this information, and have downloaded the document as you have suggested. I have not read it yet, and must replay your video for a better understanding, but something tells me. Artists will be getting the short end of the stick, (pardon the Americanism) when it comes down to the law, and what a mess for those who will be responsible for determining what is and what is not the legal case. I hope I am wrong, as I am a designer and digital artist myself, but we shall see. It will also be interesting to see how this impacts artists who mint their Crypto-art on the Blockchain, in terms of this new law and its ramifications. Thank you so much again for sharing, and for your insightful video here.
Thanks for doing all of that dry work and sharing it with us, Neil ♥
Great video! Thanks you very much for taking the time to make this.
No problem, glad you liked it!
4:27 Regarding using an image prompt as being more copyrightable, it is a fuzzy area perhaps depending on whether that image was originally your own copyright. Otherwise I see that a reference image is just "additional instructions", like asking an artist doing your commision to "make it look kinda like this other reference image" which is quite common (what I see on pixiv and deviantart commision requests anyway).
Glad to see they are actually taking proactive steps. I was not sure if they would even address this stuff for another 10 years. Honestly, as someone who is not really an artist, I would assume the genie is out of the bottle. You could maybe request to have your work not included in the training data but in the end it matters little. I suspect we will see a purge of certain artists work from the existing data sets to comply with differing international copyright laws. At least in countries that pretend to care about international copyright laws. Unfortunately/Fortunately, Pandoras box is open. At this point they have enough progress in the tech to have it generate its own training data, rendering human artwork almost unnecessary in the data set. The artists of the world will have to be content with the fact that this new technology will give them more powerful creative tools in exchange for an entire reimagining of the artistic world. Music, Movies, Books, none are immune. People really need to start mentally preparing for a gigantic shift in our reality. I don't know if people are ready for what's coming.
You are making it seem deeper than it is.
It won't be a "gigantic shift in our reality", artists never frowned upon the use of AI as a tool, they didn't like it trying to take over their jobs
by using their art without permission nor royalties, and AI giving power to random people to literally steal their work and impersonate them
(it happened way more times than needed to). You're also standing on the wrong side of the street, you want raw AI generated images to
be able to be copyrighted and sold, completely ignoring how damaging and dangerous that is. I guess being lazy, uncreative and a thief is
requirement enough for you to get granted money without putting minimal effort on anything.
It's ok if you want to play-pretend and use AI to make beautiful pictures by just prompting, you can do that as much as you want. But expecting
artists to give up their jobs just so you (or anyone just using AI) don't get mad and pretend to care about a field you never really payed attention
to before AI is ridiculous. Most people using AI don't even care about art, or art related jobs, they just want to be popular on social media. And
time has proven that stealing other people's work/content is the fastest way for talentless and skillless people to find "success" online.
"This new technology will give them more powerful creative tools"
The tool you're talking about isn't creative, the artists are creative. You can literally make a photorealistic portrait by using a regular pencil. You
don't need a "more powerful creative tool" to make it. The argument about AI was that it would speed up the process, which on industry level is
essential as time = money, but not for random Twitter users that want to benefit from the art of hardworking, skillfull, talented artists. Although
the argument behind "why AI is good" has changed more times than a Twitter user's profile picture...
Everything that an AI can do, a human did it first. The AI is just doing it faster because it doesn't require any phyisical nor mental effort to do it.
Without the models trained on millions of stolen artworks, the AI is as efficient and "creative" as a 3yo child. Limiting how public models can be
used (to avoid impersonation and theft) plus the copyright laws is the correct way AI should be treated. The random user from Twitter doesn't
need the AI to be able to produce a high quality piece of art on prompt, neither does the artist; because the artist can work from the ground up,
the AI will speed up the initual (sketches, thumbnails, etc) and final steps (post processing), whilst everything that goes in between will be made
by the artist.
@@4nanaide Thats a lot of words to ignore everything I said to jump to the defense of artists and artistic integrity without even acknowledging I stated I am not an artist and have little skin in the game.
I do know children, and if you think one can produce even a fraction of art that Midjourney does you have no clue what a 3yr old is capable of. Very little outside of basic coloring and stick figures. My point still stands though. They no longer need original artists to build a data set. They have enough proof of concept already done to allow the AI to work using a data set it originally creates on its own. Additionally, there will always be foreign entities that don't give a shit about any copyright law you will come up with.
Artists have mainly only a few choices. Make physical art and hope people do not accuse you of just hand painting/drawing a copy of an AI generated image you are trying to pass as original. Or know that any digital art you ever create will never be assumed to be created by hand and that AI was involved in the work. Ultimately, the major benefits will be the tools within Photoshop and other creative tools giving you the ability to create even more impressive works. But if it is too good, know most people will assume it was made by an AI.
@@Infectd "[...] am not an artist and have little skin in the game. "
No, I understand that and it shows, as everything you're saying is wrong.
"I do know children [...]"
I doubt so, and I'll prove that in a second.
"[...] and if you think one can produce even a fraction of art that Midjourney does you have no clue what a 3yr old is capable of.
Very little outside of basic coloring and stick figures."
That's literally what I said. If you grab MidJourney and take away its dataset which was fed with stolen artwork, there's little to
nothing that MidJourney can do. The "AI" only has the ability to reproduce and mix what was already done, it can't do anything
new. Without a dataset that consists on millions of images, the AI is even worse than a child; but even a child can create better
art, original, and without requiring being exposed to millions upon millions of images.
And I'm not just saying it, you can fact check and get a piece of reality by looking at the works of Alexandre Cabannel at 13 years
of age, on times in which technology was so primitive you could say it didn't exist.
He didn't need "more powerful creative tools" to make amazing art, he had creativity and skills that puts MidJourney in shambles;
he just needed a canvas, oil paint and brushes. The rest was not being lazy, something that in current times most people are.
"[...] They have enough proof of concept already done to allow the AI to work using a data set it originally creates on its own [...]"
You not only aren't an artist, but also don't know how the "AI" technology works. It isn't creating anything on its own, it is always
stealing and taking from everything that has been made by humans. This "AI" doesn't create, it isn't creative; it generates - and to
generate; it needs data, data that it gets from foreign sources that it doesn't understand - commonly fed on by its developers.
I repeat, the AI cannot create; it only generates - because it isn't an Artificial Intelligence; it is a Machine Learning Algorithm, which
is the closest humanity has gotten into creating actual Artificial Intelligence, but still isn't. It only learns and repeats patterns, and
how to solve problems that deny it from reaching its predisposed goal - in this case; generating (not creating) art. The AI hype and
propaganda was made up to sell the product, and because people tend to be ignorant towards how technology works (and also
because is easier to explain or say "AI" to a bigger audience than to say or explain "Machine Learning Algorithm").
Your last statement about "artists having only a few choices" and "if your art is too good then people will assume is AI" is so
fundamentally ignorant I can't believe you really said it. Do you know about a little something called streaming? You talk big
time about technology this and technology that; but forget that people can literally stream themselves drawing, showing their
creative process, and trust me; it doesn't even come close to how AI generates art. There are also apps on which people draw
in real time together online, and that's how they get to meet each other and recognize thus shout-out good artists.
Also, AI can only fool those that know little about art, and on top of that have poor taste on it. It doesn't matter how much time
passes, AI generated art will always be recognizable and distinguised. Only those that in the past got fooled by a filter over a
picture to make it pass as "hand drawn" or "cartoony" will keep falling for the lies and deception of "AI".
Note: There are secret uses to AI for art that only artists have managed to figure out and implement, making it so the artwork
generated doesn't look AI-ish neither "too good" as you said (uncanny is the word you were looking for), and it doesn't
require a massive dataset with millions of stolen artwork, neither powerful computers nor servers. Is way simpler and
technology-basic than you're making it to be. It still requires you to be an artist, not a prompter.
@@4nanaide ok so there is a lot here and I don't want this to fall into an argument.
First keep in mind 90% of people are not artists. The vast majority are not the types that look at art and think the artist was trying to portray the struggle of blah blah blah. They don't go to galleries or art exhibits That's only artists and a small part of society. Second, the AI will always exist and can be trained on royalty free art, which will grow massively by artists who do like this new AI and it's capabilities. If nothing more than out of fascination of the tech. Once it has that it can generate more art that can be looped back into the data set to refine the technique. As an example we can generate people today that are photo realistic and they have already taken the photos from that AI to train other AI on how to recognize and identify human features.
Now I am not familiar with the artist you mentioned, but of course there will be prodigies. But if that prodigy today does not do their art live, people will always question how a 13yr old did that when tools exist to allow them to cheat. Not everyone has the desire to be on public display on twitch streams or RUclips.
I am tired of this it's machine learning not AI argument. The two are very similar and one leads to the other. It's not an AGI but AI is still a fair label. As far as how it learns and creates, we cannot make claims on it compared to how humans create art. We do not understand the human brain. Though I will say just watching my kid learn to crawl and walk and learn to speak, it's not that different than watching an AI be trained in an environment to learn to walk to crawl. It's accelerated with a human brain, but that's more a limitation of our technology. That gap will close very rapidly. See NVidias new Data center super computers designed specifically for AI. To date, all AI has been achieved on technologies designed for standard computers with tensor cores being newly introduced. These new super computers will push AI much further and faster as they are a tool specificallt designed for this task.
I am not saying all Artists are done. What I am saying is their value to society will be diminished. Less will be necessary as far as independent artists are concerned. But the starving artist has always been a trope for a reason.
Most of your disagreement with me relies on this tech not improving. Considering I have been watching it for 3yrs now pretty steadily, I can say for a fact the tech is improving at an exponential rate. Now that we have the hardware, it will be even faster.
None of this should matter though as artists are largely the anti capitalism types who produce art out of passion and not for monetary reward.
@@Infectd The first half of your reply is just redundant non-sense that fails to understand what I said. I even gave you dates on times where
Alexandre Cabannel - the artist behind many masterpieces today celebrated, as well as being a core reason on why art excelled past academic
grounds; making him one of the greatests old masters, but as you said; this doesn't matter to you and most people, because you don't care
about art, just about pretty pictures and how "good" they look, not their contents or portrayed skill.
"I am tired of this it's machine learning not AI argument."
You're tired of AI being called what it is (it's even in its name "Generative AI"), although you can literally do a simple search and learn how it
works, and you'll realise that the AI dreams are nothing but dreams. Do you know why Chat-GPT doesn't give answers to topics its dataset
wasn't trained on? Because it can't learn new things without a dev giving it a new dataset with said topic on it.
Do you know why it can also give you wrong answers? Because it isn't precise, nor "Intelligent"; it only generates what it was given to generate.
That's why its devs have to update the dataset/models constantly, because the world around it evolves, whilst the dataset doesn't. It's a static
file containing scraped information. You can train it on information from the 90s, and asking it questions of today will result on the AI telling you
that such thing doesn't exist, because it doesn't know it does, and it can't learn it on its own. It only learns from datasets.
"[...] the starving artist has always been a trope for a reason"
It's called learning to paint/draw/illustrate, and not learning to market yourself or not knowing how to sell your art. It has nothing to do with
art or "artists" but with people. You knowing how to make a copy, paint a house, paint portraits, etc, doesn't mean that you're going to be a
successful artist.
The same way if you are good at sports but never go out of your way to try to get into a professional team, you will never make use of your
sport skills. People go to art school because they like the idea of being an artist, but they don't like art or the process of making art; thus they
fail. Although only bad artists become "starving artists", and the same happens on sports, medicine, and many other fields. It's not about the
career, is about the person.
"Most of your disagreement with me relies on this tech not improving."
Not at all. I literally said that the technology is good enough as it is, artists can make use of it as a tool, as it was supposedly "intended" to
be, whilst non-artists can only pretend to be artists. Making it so the generative AI makes all the art without flaws and ease of use doesn't
make it a tool, but a replacement. A replacement for a problem that never existed.
You're tired of your AI dreams being called Machine Learning, and I'm tired of people saying that this new AI will help artists create more
beautiful and impressive artwork, when that was never an issue to begin with. As I said before, you don't even need technology to create
amazing art. A canvas, brush and oil paint, or pen and paper, and you're set to create. If you want to create, even primitive tools will be
powerful enough to make you create amazing things. If you want to pretend and play around, apply no effort and learn no skills, then you
will need every piece of technology available to do the work for you, the worst is that you aren't even creating nor learning, you're generating
and forgetting.
Tired of people saying "this will help you create! but will cost your job tho." when creating was never a problem, and now that you found
a solution for a problem that didn't exist, you are taking the livelihood for those that didn't have said problem. Literally not a solution nor
a tool, a replacement which isn't even impressive as without already existing art, it can't create. It's like attaching knifes to your hands,
now you can kill and cut, but you won't be able to do anything else.
"None of this should matter though as artists are largely the anti capitalism types who produce art out of passion and not for monetary reward."
And who told you that? Your communist friend from college? Art, as sports, cuisine, etc, can be done out of passion and for monetary gain. The
process of creating a piece of art, step-by-step, is what artists are passionate about. The fulfilling creative journey, not the end result.
You will be proud of your pieces, love some more than others; and that love comes from you remembering what you went through while creating.
The same applies with the pieces you hate, the process probably wasn't the most optimal, or the end result wasn't worth the process, but you still
made it, and that piece exists to remind you of your mistakes - or of what you're capable of.
The "startving artists", "frustrated chefs", "failed football/basketball, etc players" comes from people that enjoys, or rather desires, the popularity
that comes from doing those things. People that would love to show off pieces, titles and trophies more than they would love to draw/paint, cook
or play sports. The process of drawing/painting, cooking or playing/training is nothing but a hassle to them, they just want to be recognized, and
that mentality is what leads them to failing. Nothing about talent or being a prodigy. You can't run a marathon if all you do is be lazy and uncommitted
Consider the case where the "prompt" is something along the lines of: "You are a writer's assistant. Proof-read the following material and correct any grammatical errors, and make sure it follows the Style Guide. === "
This might even be built into the word processor, and is seen by the operator as no different from the traditional spell checker and grammar checker.
Great info Neil. Not sure what copyright laws govern this in Europe, but it's interesting to see how it's been handled in the US Jurisdiction.
Thanks titaniccor! Yup, the international side of copyright is something I don't have a lot of experience with, but I'm going to do a little research on the topic as I'm interested to know how all the jurisdictions interact.
The difference is that US law puts great store on originality, whereas UK law does not.
Thank you for helping to bring some sense to this confusing issue. Good work.
Very interesting development in this ai art sphere, interested to see how other countries follow.
Best Ai video I have seen so far.
That’s some great interpretation, and the copyright office approach sounds sensible - prompting an AI for images is pretty much a “search” operation in a latent space, which others can easily reproduce and doesn’t include creative elements other than knowing what to search for
The HUGE gray area here is when you train a model with your own, copyrighted work, and generate images out of it. They’re variations if your original art, using your style and your content as an input. In that case, would an image you produce with a prompt be copyrightable?
yep. copyrightable.
Not necessarily. Google was legally allowed to scan copyrighted books for their search engine as they were not putting exact copies of the books up for search in full. The same argument could be used as to why the companies are allowed to train on copyrighted works.
I'd say yes, it's copyrightable because, with the case of standard "off the shelf" AI image generation, a human author (you, the artist) had zero involvement in the process, up to and including the prompt input. But if the model was trained with your own images, a specific human author (you) was involved from the beginning, continuously through to the final image.
The analogy would be:
If you were to teach a human student with your own copyrighted work, an make them create new art out of it. They’re variations of your original art, using your style and your content as an input. In that case, would the work produced by the student be copyrightable by you, or by the student?
@@正先じょえん are u comparing a machine learning model to a student?
Thank you for taking the time to make this video.
Yeah that's the legal approach that makes the most sense, to be honest.
It's what I expected.
Excellent presentation! It also reiterates what so many of us have been saying. AI is not going to put creators out of business because owning IP is foundational to corporate capitalism. No company is going to build a business around public domain assets.There must be barriers to entry against competitors to attract investors.
I've been an artist for over 40 years. this is good news. I have some ideas for playing with AI for tiny pieces that could go into a larger image. I may have to do half a dozen images for ideas for certain parts of the image but I'm going to gesso them to a large canvas, week them as I paint over them and fill in the blank areas with my own drawing and paintwork. I know that I can copyright that and get it into one of my local galleries. This is good news. Thank you.
Glad you found this helpful, yes, I think there’s a lot of positive applications of ai which involve deeper human involvement that is also copyrightable, so enjoy experimenting!
That's just fine, I figure. I know artists who are using AI generation for their backgrounds. They do all of the foreground work, they do the characters, but the AI does the background and the fiddly bits they don't want to bother with. There's no problem with that.
Thank you!!! Copyrights are always such a deep wood for me. This video is very clear
really helpful content, this lets you think of many possibilities in AI domain and I am excited for the future of creative and digital artists.
Such a well-produced informative video. Thanks for the effort. Your channel is going to 10x this year if you post consistently.
Thanks Noah! This video did really explode, although since I mostly post about making art, as opposed to copyright and AI, not sure I'll be able to keep the large crowd happy. But I will be doing a short update video in a few weeks on the same subject just to answer a few more questions. Glad you enjoyed the video!
A corporation is a abstract legal entity, yet it can hold patents and copyrights. A corporation is seen as a human being in the eyes of the law. So just incorporate your AI and then they can hold copyrights legally.
Haha! I bet someone will try that :)
This channel definitely deserves a subscription and a like!
Hi Neil, Img2img and controlnet would be two interesting points.
Eg: Img2img I have a unique drawing of mine. I load it into stable diffusion and low diffusion and get it to colour, or shade, or edit my work? Where do I stand?
Controlnet. I use controlnet to transfer poses from my own original works onto prompted or Img2img creations. Where would that stand?
Thank you for your time.
Yup, if I had a ton of time, I'd love to make say 20 different images using different AI techniques, send them all to the copyright office with detailed descriptions of how I made them, and see which get copyright and which don't. There's so much variation out there to how these tools can be used even if the "text prompt" is the most common method.
@@ArtOfSoulburn Perhaps we should be thinking more about what the case ought to be, what is right and moral and what AI artists deserve, and if there are any legitimate reasons as to why AI art shouldn't be copyrightable by the person who generated it. And perhaps we should be petitioning our legislators and policy makers to do what actually makes sense, rather than letting them control the situation, and just submitting a bunch of works to see what they 'decide' or not. They are public servants who answer to the people, we get them to do what we want or they get sacked, that's how that whole dynamic is supposed to work, the people make the law.
So do you want to live in a world where AI artists can copyright their work, or not? That's the questions, and then based on what you decide, do something about it, and compel the public servants to do their job accordingly.
For example using stable diffusion and the inpainting method, which allows the user to mask and area and to generate a replacement for the area in question. However the masking and painting is completely done by the which results in something that simply would not have occurred had simply been the prompt with the generative AI without them asking in place. And one little change to the masking and the whole result could change, if you bleed over into the background, you will find the background will modify what results within the actual area that one wished to modify. As such I think that using AI as a tool in this way is much different
Yup, that is another test I'd love to see someone send the copyright office and see what they decide.
Thanks a lot for this Neil! Been a big fan of your work since Soulburn scripts :)
Thanks Travis, glad you liked the video and have found my script useful.
The question I have i about download, like ShutterStock have images with watermarks and those are there to show that the image can not be used for free, but Open-AI saw this as they have the right to download all example images and use them to create a commercial product, so on their server it will be at least 500.000 copyrighted images that been share around server that been processing those images. Had it been one copyrighted movie (or music) downloaded to a computer many had seen this as a crime, so why do we see it different when it come to copyrighted images?
Btw, great video and good comment for the document you read, thanks.
It needs to be a little more lenient but 'case by case' applies. Someone who puts a lot of work into achieving their vision, absolutely deserves copyright. If someone uses many resources, lots of inpainting, tweaking, iterating, post-editing, etc - they should be able to copyright the final product indeed.
TLDR: AI GENERATED art should not be copyrighted. AI ASSISTED art should. Otherwise, photography should not be copyrightable, as the person only framed the shot, they didn't create the thing depicted themselves.
Actually, there is a lot more work in photography than there is implied in AI art, where you just give orders to a prompt generator. It makes complete sense that you should be able to draw on top of it for it to be copyrightable.
@@artorhen
“A lot more work in photography” than in AI art where
“you just give orders to a prompt generator”
As a photographer and someone using AI illustrator tech, you’re both very wrong and right. With a camera you aim at what you want and hit a button to take a bunch of photos. That’s the equivalent of throwing a prompt in and churning out a bunch of ai gen instances. Any yahoo can point and shoot and the camera these days does 90% of the work automagically. Focus, color correction, depth of field, stability, exposure... the human just aimed it and spammed. After that you decide what’s decent and keep those, which is part of hte creative process for both methods.
Then you have ai cases where you select a model based on style and content, secondary templates, engineer the prompt with weights, and set various scales (like CFG, second phase stuff, various internal settings), adjust the prompt contents to zero in on the representation of what you’re imagining, inpaint parts to make it more cohesive and eliminate unwanted artifacts, and post-process using paint tools - and then you may end up putting that back into the AI for touch ups. That’s not even considering original sketching and manual correction, or using actual photographs as models, to tweak and refine the inputs along with the above process.
Seems a lot more like the photograph process. Fiddling with camera settings, film type, lighting, positioning, choosing contents to put in it, framing, then altering the thing in photo manip software.
@@hayuseen6683 I don't think a bunch of adjustments makes for an owned piece of any kind. If I sit for an hour and adjust the settings on my phone to change the theme and everything on it, that doesn't make me an artist or a hard worker. It doesn't make me the same as someone who makes a phone or who makes a theme for a phone. Same with ChatGPT, if I just ask it to write for me, then I no longer do any actual work myself, the AI makes it. And when the AI simply takes different pieces from other places, which show very clearly in the end result, then it can't be used for commercial purposes.
@@artorhen
I don't see anything addressing what I said, just strawman analogies unrelated to the real world. You ignored all the important bits. Have you ever used the AI illustrators, researched how they operate, experimented with how to use them? Serious question, not rhetorical.
Back at you: Making a bunch of adjustments on a camera doesn't make for an owned piece of any kind, you're just pushing a button.
@@hayuseen6683 Yes, I have used the AI tools, I would not call them illustrators. I also used art program with implemented AI that actually are tools for making art by working based on sketches and images instead of words and prompts. To me it seems like you are completely biased towards these pieces of technology before they even prove to have a good use for humanity, which I would call almost fanatic. And I am glad you agree that making adjustments on a camera is not all that a photographer does. Much less an artist. Therefore the previous argument is rendered completely redundant.
Regarding the example at 3:50, i believe the image in the middle is still not copyrightable, even though you gave it an image prompt at the moment you sent the whole prompt (so image+text), the image given is still part of the AI neural network "training", the differences being that when you have just a text, the training comes from millions of images (each having a bit or more "weight") and when you give it image+text, the training consists of also those million images plus the image given in the prompt, that now would have a big weight when estimating what final generated image you should get (roughly said "training", of course the process is more complex, but i believe these are the basics here). So unless they will add an article or law text handling this specific exception (with all the technical details involved, which have to be full divulged probably), the generated images should have the same copyrightable status, right?
Thanks for the breakdown!
No problem!
I think you're wrong about not being able to copyright the character in 6:52 . A character is more than just one-off design like we see in that midjourney image. A character is also in its iconography and basically most stuff that allow someone familiar with the work to recognize the character at a glance.
I'll be interested to see the implications of AI generation based solely on a creator's own works: an artist has a large body of work, and uses an AI to draw from their own work, exclusively, to generate new images. As I understand this, the AI generated material is not CR, but would be if further reworked/edited. What is the test, I wonder, of how much following rework/editing is required to qualify for copyrightable status?
So from my understanding, it's actually impossible to use only a single artist's data, unless they've painted several million paintings. What's more likely is you have a database of millions of copyright free work setting a base, then you train it on several dozen works of a specific artist, and basically tell the AI to favor the artist. Again though, my understanding of the tech is not perfect. As for how much reworking is necessary, that's always an open question, and why the copyright office has to take each case as it sees it. The whole change 30% rule you hear sometimes I've been told is not actually a law, it's just a vague suggestion, so I suspect you have to go with your gut and then see what happens.
The training data doesn't matter. Just because you look at van Gogh's paintings and make one in the same style doesn't give copyright to his heirs. AI does exactly the same as a human brain looking at material. It extracts an essence in a high dimensional space, call it manifold, and produces a projection of a new idea through this space.
@@ArtOfSoulburn nah you can train small models with less than 50 images, its called lora
@@franezdiy I believe lora is using stable diffusion's dataset of millions of images as a base, and using those 50 images to fine tune the model. That's the impression I get from reading about it anyways.
@@ArtOfSoulburn sure, just saying an artist can actually train/fine tune it to his own style
Great examples and summation of the new rules.
This is really hard to handle by modern laws ngl. How about I create an outline and then use controlnet in SD to do the rendering. How do you define what is copyrighted there? How about I "inpaint" the image until I complete a sketch but fully rendered.
blows my head xD
thanks for the video btw!
Yup, that's why they keep saying "case by case basis". If I had the time I'd love to make a dozen pieces of art with different AI related techniques, try and get them copyrighted, and get a more specific list from the dept about what is and is not copyrightable. But I suspect over time we'll see more cases and get better and better guidelines. Personally I think extreme inpainting probably counts as "changed enough" to get copyright.
@@ArtOfSoulburn yeah, case by case to me sounds like "bruh idk, s***t changes too fast" 😂😂
But realy love the idea of seeing what they see as copyrightable and what they dont
I think it would be a similar case of photobashing, or using sound samples in a song. How much of the original is intact vs how much of it was your own work? After a mostly subjective threshold, it no longer becomes "AI made" and would be copyrightable. If you use an outline to make a full image, I'm guessing probably not copyrightable, since none of the outline remains, but yeah that outline is copyrighted lol
@@pixelpuppy One would think that, but I've actually discovered there are some slightly different rules depending on whether your original image sources are copyrighten or whether they are public domain. I'm chatting with a few lawyers now, and will likely do a new video soon discussing some of those differences. But in the meantime, ya, hopefully we see a practical example to test the policy.
The policy is vague and ambiguous intentionally. They've purposefully made moronic blanket statements and completely disregarded any and all grey area, allowing them to use the "case by case basis" system. In other words, the policy is irrelevant, everything will be decided arbitrarily by some authority. As usual, they just want control over the situation, what's right or wrong, or what's fair or ethical has nothing to do with it.
Great solution to this issue (extent of copyright coverage regarding AI-assisted tools for creating images).
One question which the ruling doesn't seem to address:
Let's say you have a character you've created yourself. You have 2D and/or 3D representations of it made by yourself. If you use generative AI to generate images of that character via prompt-only, is the result still considered a public domain character?
What if you use your own images to train a model and generate images of your own character(s)? Is it then public domain?
Pretty sure that copyright/trademark of a character would still apply. But in theory someone could edit out the character and use the rest of the image. Might also have an easier time doing something like changing the character a bit and arguing it is parody, but not sure.
If the character you prompt looks like your copyrighted character, then I would think it was still copyrightable, but it’s an interesting question. I do character animation, and I’ll set up the keframes and the computer does the inbetweens - so obviously there is lots of human input. But if you’re just giving directions and the AI is doing all the animating that could be a problem - would it depend on how explicit your directions are? Also, if your character and your script is copyrighted I would think no one could use your movie because of that no matter how it was generated - but I’m no lawyer.
From the wording, I assume that the character is still copyrightable as long as you have proof of it outside of the generated work - you just don't have rights to the generated work. The most obvious example is that if I do an AI generation of say, Olaf from Frozen, it's very obvious that the character still belongs to Disney.
Super useful video, seems like a very reasonable way to handle this.
I think concept artists can still survive if you generate a bunch of AI art as reference images only
That’s what I do. Im a traditional artist and it gives me prompts that are visual.
That’s what I keep telling fellow artists. Instead of searching the web for hours looking for just the right reference material, you can create reference images using AI. But they still whine and reject the notion.
There’s value to both approaches. Anyways, people have always been pushed out of jobs by increasing automation. But the machines, even if they are “intelligences” don’t have any real agency, any political agency and are not legal persons. They aren’t the ones “killing” concept artists.
@@judilynn9569 imagine you have a portfolio of years of your work put in a dataset which now generates work based on you, better than you, faster than you without your consent to the existance of said dataset. Reference used by real artists don't get to be compared because:
1. ai can't unlearn
2. ai can "master study" thousands of artists in the span of several days and basicallt put those artists out of studio jobs because an ai company would rather ask for forgiveness than permission.
@@niklogus9426 ask forgiveness for what? the copy right office just legitimized this as legal. They don't need any permission. Its called societal progress. Non Ai Artists are currently just getting in the way of progress. They are obsolete, get rid of them. Go ahead and complain about how its disrespectful but keep being a hypocrite and watching movies with special effects that used AI.
Great video. Very informative.
Thank you for this awesome summary. I am not a US citizen and I'm based away from the US, Does this mean that every country has to make their own decisions about this? Or is a think that you think the other countries will follow?
Thanks. And as far as I know each country has its own laws, although I suspect the US making these decisions will influence many other countries. If there's other major developments in the international scene on this I'll maybe do a followup video.
US is a big bully and other countries would be forced to follow its rule eventually.
I don't see this holding up to the test of time. Likely what will happen is after movie studios find they can simply replace all of the actors, lighting people, etc with AI. Then they will try to get this changed. Maybe to say you have to own the AI in order to copyright the work.
OR the case by case basis will lean always where movie studios and game studios and what they want.
Great thank you so much for breaking this down to a simple format. As a fellow Artist like really appreciate it.
I'm no lawyer, but I dispute the idea that you can't copyright a character if it is "AI generated." I think the issue here is that the copyright applies to PUBLISHED works, so if you put the raw AI image out there as the initial publication, then that would be "the character," and would be public domain. However, I feel like if an AI generates a character design, and then another artist creates original art that is derivative of that design, such as making a live costume or a 3D character or drawing them into comic panels, and then that finished work is what is actually _published_ into the world, then that would become the initial artistic work, and all copyright would flow from _it._ I think that in a legal sense, it would be similar to hiring an artist to do work for hire, in which they have signed away all copyright they have to the character. What _you_ do with it is then what generates a copyright. If Spider-Man had never existed, and yet an AI spit out an image of Spider-Man, and then you drew a comic of this Spider-Man, and this is the first version of the character released _to the public,_ then you should own the copyright on Spider-Man, and nobody else should be able to draw that character
Thanks for this video, very informative, and also interesting to hear your insights on the implications for concept art.
Thanks, glad you've enjoyed it!
4:25 The middle image cannot be copyrighted as it is the direct product without any edition, of an AI. The manner on which the prompt is made, is irrelevant, and what the AI does is analyze the image in search of patterns, to then reproduce something different from those patterns. It is no different than inserting a text prompt. Furthermore, it might actually be copyright infringement, if the image used is copyrighted. If I commission you to do me a Mickey Mouse picture, and show you a Mickey Mouse picture whose copyrights are owned by Disney, you as an artist will also be subjected to be violating Disney's copyright, if you copied it (or it's patterns). Machine or Human, the situation is the same.
Not a lawyer either btw, just offering my own perspective based on logic.
Imagine a future where a company can lobby the local government to claim that an artist used an AI to generate a famous work, in order to avoid a lawsuit because they stole the artists work and were caught.
Now imagine a future where AI can sue you for copying its work. And that future is closer than we might realize.
I definitely have no problem since I found ai art to be completely useless for me
@@rigelb9025 well it would be easy for Ai to generate hidden tags that not even a human could see, think of it like having a QR overlayed on the image unique to the Ai
The proof of burden is on the company.
Thank goodness Clip Studio Paint has the option to record the entire creation so at least if you use that you can easily show you made it without AI.
Makes sense. I would be more interested in all the artists complaining training AI on their art was against copyright. What's their take on that?
That's part 6 of the video, they haven't decided yet, and will likely not decide till the 2 lawsuits happen.
As a 3D artist, I find that AI art generation is best suited as a source of inspiration, suggesting ideas, developing mood boards etc. NEVER as a substitute for original work.
As an entrepreneur I replaced all my art needs with AI. Now my annual salary bill and costs for digital art is around $240.
Bye bye artists.
Anyone who actually messed with art generators knows it's very difficult to get exactly what you want. It's a lot of trial and error, a lot of generations, a lot of fixing and tweaking. To finish a perfect project you need a little bit of coding/programming, a little bit of drawing/photoshop and whole a lot of promp crafting. I don't think role of an artist will change, just what the job entails will change. That happened before when we made the jump to the digital medium, it'll happen again.
I'm doing photobashing but using only generated images, and I definitely feel like it stands as a final product, seems like I get copyright on it too.
No, lots of the AI art I have seen is superior to LOTS of paid art. I've generated thousands and thousands of images, many of which are great and can be further changed refined with almost infinite choice.
As a nonartist, I wonder what the difference is between an AI using an art style for inspiration and a human doing the same.
5:20
If it is not copyrighted can any one then use it (including Marvel)?
Or do you get some copyright but Marvel also get some copyright?
That's a legal grey area. From my understanding, technically since the art can't be copyrighted, it's public domain, so Marvel could use it, but they don't have exclusive use of it, since its public domain, and so would be unlikely to use it, since they want only stuff they have full copyright over.
Nice video. You did miss one thing and that's trademarks. Mickey Mouse's copyright expires in 2024 but its unlikely you'll be able to use Mickey because he's trademarked. Any replication of Mickey-like characters which one may confuse with Mickey would infringe upon it.
I also think there are plenty of loop holes that will be tested in court. AI has evolved so fast that you can do stick drawings which get converted to art. The line where how much one must do to be the author needs clarification. My gut instinct will be along the lines of, a human must have created 51% or more of the art to be deemed the author but we'll see.
Ya, didn't approach trademarks in this video, but that's certainly another meaty topic. Glad you liked it.
If someone uses AI with their own characters and art style, I feel like they should be able to have copyrights over that because they trained their own personal model with their own art style. However, this becomes a problem later if they release the model to the public because they could then say other people are using this style too, so how do we know you made it?" In which case, people could prove that they made it by sharing the video they made of them making the model.
But this won't work either, and the reason is that you can't actually copyright a style, only your characters. But characters often look so similar that it would still be kind of hard to do that. People have been using multiple models to mix styles together similar to how real artists take inspiration from different parts of different artists' styles, mixing them together to create something new. People have been uploading models very specifically for certain poses. And these are all uploaded for free for use. Nobody is charging for them, however people can support your Patreon if you ask them to (since generating large models can be long, labor and cost intensive).
If people could copyright a style, then people imitating Disney's style would be in trouble even with their fanart. Not to mention if you copy another creator's style, who's to say that creator didn't copy from someone else originally to get some major aspects of their line work? The shading? The way they draw noses, blush, highlights.... you can't prove much. Someone can say "that looks like that person's style" and you can bring up a lot of other artists that look similar.
Even before AI art, you'd see lots of pretty anime art on Instagram being posted, and everyone would do the same style because they saw that is what sells. On youtube when people notice an editing style is holding people's attentions, suddenly you see that everywhere. people will follow the trends to create something that sells, and that's just facts. And people can mix art styles together until they create a model that really is their own style until someone else takes inspiration and makes a model from what they post as well, or follows tutorials showing how someone learned from their style to replicate it by hand.
I don't think this will work because you can put a ton of anime girls from a wide variety of series' by each other that have the same hair and eye color and not even be able to tell they are from different series', or who they are. Most shows, even before AI, would basically copy-paste characters, barely change anything, and even use the same voice actors for characters with certain traits. Almost every tsundere character has the same voice actress, for instance, and most anime characters all have the same face with very minor changes to the bodies. If the show has natural hair colors and lots of girls, you can't tell them apart unless they have their hair styled. So if you see them with their hair down, you think it's another girl. That's honestly pretty bad.
If they are looking for sameness, how would regular anime studios that are mostly copy-pasting their own work be able to get away with this? Not to mention, why would they keep drawing it when they can easily train their repetitive styles and just do very minor tweaks using real artists to fix the anatomy after? Plus, if they get a character design and pay someone who knows how to make 3d look perfectly 2d, 3d models to video is coming out soon, so people could do most of the animation with their own bodies with only some changes being adding some effects to make things look more cartoony and less perfect.
I feel like this is going to harm artists everywhere, and I feel like everyone on the outside of the AI scene needs to know something. This AI stuff is only dangerous if you let big companies take it, try to monopolize it and control it, and charge for it. We are trying to make sure art stays free for everyone, sharing our models and not keeping them to ourselves. Even people who have drawn everything by hand always can and should be using it to help speed up your work to keep up with demand, and to basically have a helper. We are gaining tools that are making it so big companies have no power over us. Why should we work for someone else? Why should we work on a series that may get cancelled because of a lack of funding when we can do it all on our own at home with ease? Why should we work for someone who overworks us, under-appreciates us, will replace us the second we die, doesn't care if we are starving, will replace us if we strike, and wants to replace us with machines anyway? All companies everywhere want to get humans--beings that require breaks, food, water, compensation, clothes, air conditioning/heating--out of the picture because they want to generate the highest amount of revenue possible. But they aren't at a point where they can successfully replace us yet, but we are already at a point where we can make these things at home on our own with no team. Why should we work for them? They are losing power over us. We can take art into our own hands and have more artistic freedom than before.
Thanks for this video.. what I’m wondering about most is that next step - can the AI legally learn from copyrighted material?.. as someone who has a law degree (never practiced) and is an artist, I think the route I’d feel most comfortable with is erasing the current databases completely and starting up a voluntary contribution system with creators, copyright owners and the public domain.. kinda like Wikipedia? Taking a step back to take some steps forward.. when thinking about the idea of “derivative work”, imho, that’s the only thing that makes sense to me.. To me, inspiration and the realm of imagination are more murky and less defined/tangible than actually using bits of data from a pinpointed, copyrighted source.. I welcome the tool/technology.. but not to rehash and utilize work against the will of creators.. but if it is aligned with the will of creators, then game on! And I would imagine that new database would grow pretty rapidly..
This is very promising! It will help establish a system that will influence the laws and court cases in favor of protecting creators and limit the negative impact of those using AI and trying to pass it off as their property.
Thanks for the info🙏
Thank you so much for making this video! 07:57 This part is so sad, indeed, only time will tell. Profit prevails in the end but I'm still hopeful there would be other avenues for concept artists but yeah... most artists would be displaced with these new tools
I had on an additional thought about your robot and using an image to create another image. If you create an image using prompts and then attempt to create additional images using the original generated art, you probably wont be able to copyright it, unless you changed it enough prior to uploading the new image as a base for the additional images. It seems the best use of AI is to generate a quick image to use as a foundation for your own creative works.
I'm a little confused with the idea that any generated character is in the public domain. It means that you can just generate a bajilion random things and say that no one can copyright them because they are generated. You can run a story generator to generate a million random characters, and then use them to block other people's copyright claims.
It's a bit silly but still a weird loophole in the description.
Also, there was no mention of AI using training data that is of copyrighted work, or of non copyrighted work but without consent. I think that's what artists worry more about since at that point concept artists can become used just for creating training data rather than actual original creations. The originality of the art direction can also suffer since any art would be generated from some other artwork's art direction.
I think ai software made using copyrighted artwork, or artwork sampled without permission should be liable for copyright infringement or legal action. Not having that would allow for even greater disrespect for the role of the artist than already exists amongst some developers and parts of the industry.
Ya, I don't think even the copyright office fully realizes the scale of AI, being able to pump out millions of pieces of media in a matter of hours or days, and the implications that will have. Although I suppose there's an argument to be made that there are so many combinations of things that maybe even a computer can't make them all. The bit about training on copyrighted work is #6 in the video, and in the policy it's mentioned only briefly and they basically said a decision on this is pending. I suspect they won't make a policy choice until the 2 lawsuits are done. And yes, that is a big important question, but since this video was focused on their policy document, I felt a more detailed discussion of that topic is better left for another video.
@@ArtOfSoulburn thanks for the response. I shared the vid on my linked in, I'll do the same with the next
I feel like copyright issues with AI should be pretty straightforward. The biggest complaint I've heard from artists is that AI is copy pasting the artwork used to train the AI which is a big misconception. Lower quality AI may do that but not any AI that's actually impressive. Training material isn't used to copy and paste an object it's used to teach the AI what an object looks like, what makes a certain style tick, etc. With that the AI can determine how to draw an object in a specific style while only knowing what the object is and what makes the desired style that style. Using an image as a prompt is more questionable to me but it doesn't feel much different than an artist using another person's art as reference material and end up drawing something similar.
Great overview! Thank you! 👏👏👏I will predict that these laws and regulations will have to evolve as quickly as the technology is. It is interesting that the "changed enough" premise is documented as it is as, IMHO, this appears to weaken existing EULAs at the majority of "For Sale" content repositories that say no derivative content can be sold as your own.
Thanks. Ya, the issue with "Changed enough" is it's a value judgement. Which it kinda has to be. But that means you can never be 100% sure as an artist how much you need to change something to be safe. So you basically have to do your thing, send it to the copyright office, explain what you did, and see what they think. And yes, I suspect we'll see a ton of people try all sorts of different things, and each will get its own ruling. So there may need to be a video 2 at some point :)
That's not what "derivative" means. A work that is colloquially "derivative" could legally qualify as "transformative" and be protected under fair use. They themselves specifically noted that works containing photobashing materials can be copyrighted, for instance. The important point here is that these policies and their nomenclature are intended to protect artists from infringment of their copyrights. Since an artificially generated image may or may not be infringing on an artist's copyright, discretion on your part is necessary. I'd suspect the threshold to be similar or identical to fair use. As I understand it End User License Agreements are at the discretion of the website, and can't be "weakened" by the policies of the US Copyright Office. A violation of the Constitution or a law that contains specific prohibitions are the only things that could do that. A website could probably make the plundering of your copyright a condition of their website's use, it just wouldn't be a popular or appreciated policy among the general populace, and the site likely wouldn't survive long enough to see divends.
What you're saying suggests you couldn't sell public domain works as your own. Which legally you can do.
Ultimately these repositories don't want you stealing peoples' work, setting aside the AI's inherent lack of discretion about plagiarism, you can't plagiarize the robot you instructed, therefore changing the work to suit the policies of the US Copyright Office *might* not even be necessary to stay within the spirit of your EULA, which means you *might* be able to sell it without changing it at all, you just don't have the right to claim it as your exclusive intellectual property. Which is usually desirable, if you're a professional artist.
I'm not a lawyer, but from all I've ever learned about this stuff that sounds correct
Around 7:45 when you're talking about the copyrightability of the character, I have to argue somewhat with your assertion that the character is not copyrightable. Reason being, there really is no character yet. It's a depiction of a *potential* character. I'd argue that you need to assign an identity to the depiction to actually turn it into a "character" (which could include naming it, assigning character traits to it, giving it a background story, etc) at which point human creativity has entered the scenario, and thus does becomes copyrightable. But before you do those things, yes I agree it's not copyrightable, my reasoning being not because it's AI-generated, but because no real "character" exists yet.
There could be some truth there, in this example I am just talking about the look of a character, but a true character needs to be more fleshed out. One thing I've discovered talking to lawyers about this stuff is that just because something could be copyrighted doesn't mean it will be. So just the look of the character could be copyrighted, but it may be likely that the copyright office will ask for more detail (name, character traits, etc) before they're willing to give you the copyright. Which means that all of the rules they put in their policy are somewhat fluid, which is frustrating. There's just so many potential exceptions there. But anyways, great point, thanks for sharing!
The number of times you had to say “Copyright” to create this content tells a lot about how dedicated you seem
This strikes me as an exact rerun of the initial question of software copyright.
The initial ruling was that software could not be copyrighted. The judge held up a cassette tape with a program on it and said that because he could not read it, it could not be copyrighted.
That judge got "educated" pretty soon afterwards and the decision was overturned.
Please explain where the exact rerun is. AI images are not the work of a human, unlike the software, so the baseline is completely different.
I have used the "prompting an AI generator is the same as commissioning an artist" argument for a while now. It is crazy how many people think they are artists because they put in some text into a computer. Sure, "artist" is pretty subjective... but in this sense, you would be a prompt artist, not a visual artist. Creative jobs wont be hindered much by this tech. For example actual concept artists will just use the tech, instead of thinking the creative company is going to hire outside prompt artists to replace their creative talent. I could be wrong, but I doubt I am going to see "Looking for AI prompter. Must have 2+ years experience in writing AI prompts. Art experience not necessary" in any real creative job postings.
Good observations, good examples.
i honestly think big companies will fight to change the law when it comes to this . makes to much sense for them to go "well we made this program, and we trained it on our own art so we own what it makes"
Yup, I suspect we'll see a bunch of that too. Will be interesting to see if they succeed
I'm entirely for that. Then artists can make models of *their* art and sell them just like every other artwork.
@@miranda.cooper my god, that is fucking pure genius!
thankfully the tech to create images is, even though highly complicated, reproducable by smaller teams. there are already tons of open source solutions for creating images and they may be not the top image creators out there but at the moment they are good enough.
even if these free tools are always a year behind, in just a few years they will produce quality that is easily enough for most consumers out there.
That that's a good question, if I train my own model on images I took of a real subject and I use my model then it should definitely be considered mine right?
Very nice summary, thank you.
Glad you found it helpful
In accordance with the argument, the copyright office is suggesting that no photograph is copyrightable by a photographer. The artist uses an Artificial Instrument to just point at a sunset as though a prompt. To capture the essence of the moment with automatic settings without any photographic knowledge to create the image. The logic used suggests that every photographer that has used a digital camera cannot copyright his work. Going a step further, you can commission a photographer to take photographs for you. This transfers the copyright to the person commissioning the works. If a person has paid for the right to generate the art they are the copyright holder. You are correct you are not a lawyer and this Gini cannot be put back in the bottle.
Here's a scenario that I'd like to get some clarity on.
If I prompt generate an element, then Photobash it to be how I want it, then I use that as an image prompt for a final iteration / polishing of my bash, is the new work now copyrightable? The process of Photobashing would be required in order to accomplish the final work to which you could not arrive to without it.
How much "the fix" part from machine in your final arr, I believe will be one of considering factors to decide that those final image can be copyrighted or not . But like mentioned in video . Not every case is equal .
Thank you for this summary Neil. You've been doing an excellent job covering the AI topic from the concept artist/art director perspective.
It is true that painting over some prompts could greatly reduce the enjoyment of the work. I know I'd hate it.
Glad you've found these videos helpful. Ya, if my job was 100% tweaking prompt images I'd not enjoy it either. I might go off and brew beer for a living :) Using prompts to create photobash elements, that I'd likely be fine with, but it all depends on how much I feel I am able to personally contribute to the final image. We'll see where the industry goes.
At @4:32 I don't think you can copyright the middle image of the robot. It's equivalent to commissioning artwork from an artist and sending them a sketch instead of written instructions. You have the rights to the original sketch. The artist has the right to the work they create from the sketch, unless (as is usually the case), you have them sign a contract assigning the rights to you. But an AI cannot sign a contract (yet). Like you, IANAL.
Yup, that would be the argument as it why it's not copyrightable. I'm hoping someone tries it so we can get an official ruling, since image prompts are pretty common, not as common as text prompts, but still frequently used.
Image transformations are definitely one of most tricky aspects of this. In video example the orginal image is pretty far from ai result, but you can generate stuff that would still be recognizable as the original image and yet visibly modified by ai. And then there's turntable character sheet generators, inpainting etc etc
@@bazejkaczmarek8271 Yup, these guidelines provided by the copyright office are going to need dozens of revisions over the coming years, maybe the coming months :)
I feel like the two copyright issues being decided are a bit in opposition to each other. Like if the AI can't train on copyrighted images because it isn't a human and isn't really "inspired" and is just a tool. Then at the same time how can it be said that it can't be copyrighted because we're only giving directions to another entity that isn't just a tool?
Great video! I think what this means is that when we use AI for something we might want to copyright, we should use original images as part of the prompts. Even if you just draw a stickfigure or something as the base image prompt, it'd probably have decent chance of being copyrightable art.
I don't think its a surprise that the AI characters are not copyrightable as they come out, that'd be insane(think of patent trolls, but with characters), but I do think that AI *based* characters can be copyrighted.
Typically character copyright applications are not "one or couple images of the character" but rather dozens and dozens of pages showing and describing the character in high detail, from style to personality to gait to colors and much more. They'd have character sheets that show all the angles, different outfits, etc.
Copyrighting a character is not a simple task and for that reason I think AI-generated characters *can* be copyrighted: the work to get an AI character to the point of having a chance at copyright would need so much work and art and decisions that the resulting character would likely be considered copyrightable at that point. (especially if the original prompts that the character was based on had an original image as part of the prompt as well.)
You can do all this in a couple of hours or faster if you use AI as substitute. I mean you basically just need to train a model that fills in the stuff depending on your description, you just need to proof read and modify it to your liking. I guess we need all to rethink "simple task" when you basically can automate anything in a process except the idea.
@@smurfo-pax4423 Fair point, I suppose.
there is also the style used to produce the image, such as pixar style. the models are trained in producing specific styles that will always show even if you add your own picture. maybe have a look at creative commons I think this is a fair way of looking at it
Thanks. Clear summary of the policy.
It's absolutely critical that AI can be trained on copyrighted artwork. If not, AI in the western world dies and we put two different standards towards AI and the rest of the world. A human brain can be trained on copyrighted stuff, why not AI? This would just become a total litigation hell, new patent troll playground and if any AI could survive it, it would only be the big companies AI who can defend against frivolous and not do frivolous lawsuits.
"A human brain can be trained on copyrighted stuff, why not AI?" - the answer is on the question itself - because one is a HUMAN and the other is not. We can issue rights for people, not machines.
I couldn't agree more. If you think things like RUclips's broken Content ID or false DMCA take-down notices are bad, imagine that times ten thousand. I'm not optimistic about this. I fear we will make the wrong decision, and the most powerful tools in the world will be owned by the countries that we are least friendly with.
@@lordavius So ... how do you regulate one, without something that bleeds into, or impacts, the other? That IMO is the biggest problem I have with this point, it seems hard to translate that into something that won't fuck people, and artistic people, over.
@Gondor avalon the answer is pretty clear to me already. One is a human being, which the already in place laws apply. The other is a machine, a tool that is not eligible for any rights (including copyrights). The other nuances are already exemplified and talked about on the video itself - are you using the tool to create yourself some new kind of work? Like Kitbashing or reference building? Or are you (like millions of so called prompt artists who are doing waifu) just getting the end result and posting it as your "art"?
This has already been widely discussed since the inception of image searches on the internet. Imagine the prompt is an "image search" that is Kitbashing to the pixel to find the image that better resembles your "search". Can I copyright a Google search result? What if Google did some fancy copy/paste on the images themselves, would I still be able to copyright them?
I guess no.
@@lordavius "Imagine the prompt is an 'image search' that is Kitbashing to the pixel to find the image that better resembles your 'search'."
Why? That's not even close to how these models work. Generative AI models are not doing "some fancy copy/paste" any more than you're doing some fancy copy/paste of letters to write your comment. It just doesn't work that way.
In order to do an image search, the AI would need to have a database of images to search, and it simply doesn't.
What about the input material used for training different models? For example githubs copilot, it uses copyrighted (open source, but may still require additional terms as attribution to the original author etc) as training material for the AI model. Does the model / github obey the original copyright license when code generated does not have these required attributions? The same applies to generated pictures etc, the original training material was probably copyrighted and had a human author, does generated pictures violate those copyrights? If your middle mechwarrior (the one generated from a direct image prompt) is copyrighted to you, then all images generated are probably copyrighted by someone unknown.
It would be good for all concept artist to be skilled at 3D modeling and animation so they can take the concept to the next level, view it from all angles, and even give it a voice to ensure that, enough inspired from AI art, it becomes something totally new, with investment from the artist.
@4:40 I believe you are conflating trade marks with copyright. Specific works of art can be copyrighted, but characters themselves are trademarked IIUC. Please correct me if I'm wrong.
You know, as a guy who LOVES making AI art? I'm okay with this. Feels like turnabout is fair play. I mean I can still MAKE things with the AI, I just can't copyright it. That's okay. Also I realized this is necessary because a Switch game came out with AI generated art (The Internship) and the prompts they used are so basic and lazy I basically created one of their characters myself on accident two months before the game came out when I was first experimenting with NovelAI!
I just hope the thing about AI trained on copyrighted images doesn't blow up in everyone's face.
There is certainly a big discussion to be had about the fact that even if you can't copyright a piece of AI art and get exclusive rights to it doesn't mean people making AI art as a hobby won't be totally changing the industry. I mean, if everyone decides to stop watching traditional films, and just watch free AI generated films, the movie industry as an industry making money and employing people goes away. Or if you make a youtube channel that shows Ai artwork, you can't monetize the channel, even if someone could exactly copy your channel and try and monetize it. We're in for some wild times.
Thank you for this wonderful breakdown!!
In practice, every AI-generated piece of art is 100% copyrightable. If I copy someone else's AI piece and they sue me, I will automatically lose in court, because I cannot prove it is an AI piece. They can always claim that they made some edits. As a consequence, the US copyright policy is quite irrelevant.
Not true because most of the current AI generators imbed their images with an invisible watermark that proves that it is, in fact, AI generated. Most of them also back up every single image that they've ever produced. It's not that hard to find it, unless you're running a local Stable Diffusion system and even then, we don't know. The real question is going to be, how much "editing" is it going to require to be considered "sufficient"? That will be for the courts to decide.
@@BitchspotBlog A lot of us are running our own SD. And I'm pretty sure it won't be long before we're able to make models just as easily as images... what then? I put these sets of images into this model... that means all images should be copyrightable?
Every image coming out of AI is a OOAK, place signature on it, done. This is the classic "Are Andy Warhol Xerox copies of Soup cans someone else designed copyrightable". Well, is it transformative? One cannot enforce an unprovable position; Copyright office is not going to be able to discern or prove an AI assisted image vs human only image, and to what degree... and yes, AI art is transformative by definition.
@@thetruthserum2816 As long as you aren't trying to claim it is the original, then that's illegal.
@@BitchspotBlog You are right, instead of saying "every AI-generated piece of art is 100% copyrightable" I should have said, "every piece of AI art that I generated on my own machine is 100% copyrightable". In practice that is not important distinction, local SD is easy to install and future versions will be even easier to install and less hardware hungry. But yes, if you find an AI image on the web that has the watermark then you can probably copy it.
I am curious as to whether ai art by a paid subscription (especially if the company says you own the copyright if you "made" the art) would allow it to be seen as a work-for-hire where you are instructing an employee to create a piece of art. If that were the case any of it would be copyrighted.
With current laws it looks like the answer is no. The company can say you have the copyright because you paid them, but the US gov doesn't recognize that copyright. The government would have to pass new laws.
It is ridiculous that direct copy of a public domain art can be copyrighted, but digital art cannot be because “not enough human input”
So I’m getting clarification on this point from a lawyer, sounds like there may be extra nuance to it, will probably post a second video with extra info in the coming week. This topic is deep :)
Maybe but the real problem is, with the flood of art being done from A.I. that's very likely going to reduce the value of human art, especially because money talks in this world, if someone can create art for a game for instant using art, why would he employ a human to do it when it's cheaper and faster to do it with A.I.
If the creative industry is not careful, they could reduce their own value so much that many won't want to use their work because they don't need too when there is far more being done with A.I.
@@paul1979uk2000 oh, no! too much art! :)
@@paul1979uk2000 So Adobe Firefly is useless then. Got it.
@@paul1979uk2000 AI art can be 'human art' as well. There's a lot of overlap there. Even a raw generation can and usually does require artistic vision via a prompt, someone not only has to learn things about art, like color schemes, styles, composition, camera angles etc. They have to learn how to use the tool and communicate with it, crafting a linguistic prompt - you can't just type any old garbage if you want a good result, there is an art to crafting prompts, i don't see it that different than other wordsmiths, like poets or authors. Not all prompts are created equal, and you can tell that by the results.
Nevermind the fact you can also use AI the same way you would with photoshop, to edit and digitally manipulate your hand drawn scribbles or process a draft image you made in MSpaint or anything along those lines.
I see no reason why AI art can't be copyrighted, it doesn't just make itself into existence, the person who generates it should own it, the tools used to create it should be irrelevant. As long as it doesn't infringe upon another's IP, then what's the problem? It's not like the AI is going to claim rights or ownership of the content, is it?
So dumb. But the policy makers are just a bunch of old fools who don't like change or understand new technology, this stupid shit ought to be phased out in a generation or two at most, hopefully less.