Follow Slashdot blog updates by subscribing to our blog RSS feed

 



Forgot your password?
typodupeerror
×
AI The Courts United States

AI-Created Images Lose US Copyrights In Test For New Technology (reuters.com) 100

Images in a graphic novel that were created using the artificial-intelligence system Midjourney should not have been granted copyright protection, the U.S. Copyright Office said in a letter seen by Reuters. From the report: "Zarya of the Dawn" author Kristina Kashtanova is entitled to a copyright for the parts of the book she wrote and arranged, but not for images she made using Midjourney, the office said in its letter, dated Tuesday. The decision is one of the first by a U.S. court or agency on the scope of copyright protection for works created with AI, and comes amid the meteoric rise of generative AI software like Midjourney, Dall-E and ChatGPT.

The Copyright Office said in its letter that it would reissue its registration for "Zarya of the Dawn" to omit images that "are not the product of human authorship." [...] Midjourney is an AI-based system that generates images based on text prompts entered by users. Kashtanova wrote the text of "Zarya of the Dawn," and Midjourney created the book's images based on her prompts.

This discussion has been archived. No new comments can be posted.

AI-Created Images Lose US Copyrights In Test For New Technology

Comments Filter:
  • by Local ID10T ( 790134 ) <ID10T.L.USER@gmail.com> on Wednesday February 22, 2023 @08:39PM (#63316605) Homepage

    At what point is the tool doing the creative work instead of the artist?

    "The fact that Midjourney's specific output cannot be predicted by users makes Midjourney different for copyright purposes than other tools used by artists" -US Copyright Office

    • Re: (Score:3, Interesting)

      by Powercntrl ( 458442 )

      Well, it's the same situation with people posting content generated by ChatGPT in forums. Do we consider that to be a valuable contribution, or should we treat it as unwelcome noise since the person behind it didn't do anything creative beyond giving the AI a prompt?

      • The problem is that AI has no creativity. ChatGPT could never tell you about a thing that isn't composed of aspects that are all part of its training set. An artist's use of emotions and personal experience is more important to the creative process than viewing other works of art, and AI has none of those things. I think most people don't understand this because most people don't experience life passionately and intensely enough to be an artist.
        • Good art is a method of communication; there needs to be a message being stated and the challenge is in conveying complex things broadly. Such as conveying a specific emotion that is recreated in another. Being clever and making aspects interpretative and abstract allows longer and broader appeal as well. A lot of art fails to do this.

          The machine is a great seamless collage maker and will someday convey get good at literal messages but the depth and purpose will not be there - it will take some human itera

          • by micheas ( 231635 )

            AI isn't a collage maker. It creates much like a human does by looking at patterns in the world and duplicating them.

            Artists get a lot of inspiration from looking at other artists' work and reading books and plays. You can look and find elements of almost every work of art in a prior artist's works. Especially if you know which artists influenced that artist.

            The idea that a piece of art is entirely a new work that is entirely subject to copyright is a fiction that artists have liked to believe, but the real

        • by unrtst ( 777550 )

          It creates new things that never before existed based on past experiences.
          ChatGPT does the same thing.

          How passionately and intensely one experienced life is not a prerequisite for art. Hell, the blind can make visual art and the deaf can sing. There is loads of well known art by unknown authors.

          • I would argue that the blind and deaf actually experience life in a more passionate way, so interesting you used that as an example. So you say the blind can produce visual art they have not seen. Can ChatGPT introduce any element from something it has not seen? Or can every aspect of what it creates be traced back to something it has seen already. Has it ever talked about anything that is not in its training set?
            • by micheas ( 231635 )

              Generative AI is capable of talking about things, not in its training set, and introduces elements that it has not seen.

              Sometimes it gets it right and sometimes it is comically wrong. (Comically wrong because it takes things to their logical extremes and has no clue that they are extreme, it is just filling in the blanks and projecting the future based on the present.

              Ask Chat GPT about something you have in-depth knowledge about and you will be amazed at how insightful it is and how comically wrong it is a

              • How does it know about things not in it's training data? That's impossible. It's mixing its training data in incorrect ways, not going beyond its training data.
                • by micheas ( 231635 )

                  It's not impossible it is looking for patterns and sequences and extrapolating logical conclusions from those patterns. It curve fits sequences and extrapolates what should also fall on that curve that it came up with, whether it is possible in the real world or not.

                  Generative AI has come up with possible vaccines that nobody had considered

                  • The key word is logical. It can only come up with logical conclusions. Much of creativity involves the illogical.

                    AI may have come up with a new possible vaccine, but that isn't an example of creativity because it is only working within the bounds of the rules of chemistry. It cannot come up with anything beyond the rules it has been trained on. It would be like calling a dice roll creative because the result can be 1 through 6. Just because we cannot conceive all the possible results with our own min
          • The same thing if you ask your three year old to make up a story about the teddy bear. It's novel and unique, but do you really want to spam the short story web site with thousands of stories from three year olds? The AI may make up better stories but ultimately there's vast amounts of randomness in it, any jarring discontinuities are not clever uses of creativity but just randomness.

            AI isn't a good use in these areas. The purpose for the research in the first place is about natural language processing,

            • But if a 3 year old writes a good story, would you still reject it? If an AI writes a good story... then what? Should we just through it away and forget?
        • There, I fixed it for you...

          The problem is that many humans have no creativity. Many humans could never tell you about a thing that isn't composed of aspects that are all part of its experience. An artist's use of emotions and personal experience is more important to the creative process than viewing other works of art, and many humans have none of those things. I think most people don't understand this because most people don't experience life passionately and intensely enough to be an artist.

      • by AmiMoJo ( 196126 )

        At the moment people have to pay artists to generate assets for their content. There are a lot of small YouTube channels, for example, with logos and short video sequences that were made by artists who got paid for their work.

        With AI, a lot of that work will be replaced by entering prompts and selecting the output you like.

        Music will be next. YouTube offers a library of music guaranteed not to trigger copyright strikes, but of course thousands of different channels are all using the same ones. A few pay for

        • That is basically the "ditch digger replaced by backhoe" situation.

          Rather than hiring laborers to do the work, you use a tool to do the work. The only difference being that perhaps the backhoe requires more training to operate safely.

    • I would argue as soon as you modify one pixel - itâ(TM)s yours. The ai image could be considered a canvas.
      • It has to be substantial. If you restore a film that has entered the public domain you get can get a copyright on the restored version. But simply adding a watermark won't get you one.

        On the other hand, maps get by with just adding a few fake roads to nowhere and are granted "copyright" because it's not simply a collection of facts anymore.

        • by micheas ( 231635 )

          Well part of it is that 250 years ago copyright law in the US covered books and maps for 15 years with things like newspapers, paintings, signs, and posters not being eligible for copyright protection.

    • The writer does not have to be able to paint. The author had an idea and all the chapters and all the scenes and all the frames that are required. The AI draws the frames and the author puts them together to a comic book. Now, how is it different if the author would had hired an real artist to draw the same frames? The author would not had been able to predict the output of the artist, same as with the AI. So, I am not sure I can see a problem with this book.
      • by Entrope ( 68843 )

        When an artist draws pictures for a book, by default they get copyrights for those pictures -- and obviously the artist knows what they are generating. The writer only gets control of the copyrights if the contract between the two specifically says so. That's the difference.

        • Right. In the case of the AI, the pictures should not be copyrighted, but the book itself should be fine. And, as there is no contract between the writer and the AI, the writer gets full control of the book. Would this be a fair approach? Also, if I hire an artist to draw something for me, why would the copyright be artist's? It is I who came up with the idea and spent money. The copyright for the pictures should be mine then?
          • Also, if I hire an artist to draw something for me, why would the copyright be artist's? It is I who came up with the idea and spent money. The copyright for the pictures should be mine then?

            That is known as work-for-hire. It is a specific legally defined situation in which the copyright belongs to the person directing the work instead of the person creating the work. In modern times, it is usually specified in the contract who owns the copyright on the created works in order to avoid dispute.

            • by Entrope ( 68843 )

              Work for hire is even more specific than that. If there's not an employer-employee relationship, it only applies to certain kinds of works where there is a written, signed agreement about the work being a work for hire. (See the definition in the codified Copyright Act.)

              A contract can also specify that the creator of a work will transfer copyright in the work to someone else. That doesn't make it a work made for hire, but the customer still ends up in control. The existence of (at least) those two paths

          • The person who draws the art for your book with assign the copyright to you, possibly as a part of the contract. The artist could also allow you to use the images while retaining the copyright himself for use in other areas. It is up to your arrangement.

            The difference here might be in getting a logo for your website - get an AI too to generate it and then you've got a nice logo. But because it's AI generated it doesn't get copyrights, but you _can_ trademark it. The same as if you had a generic computer

    • Can you predict exactly how a brush stroke will appear if you are painting on a canvas? I assume you have a decent idea on how it will look like, but can you actually be 100% sure? Maybe you are 90% sure on how the stroke looks. I don't agree with anyone claiming 100% since painting on a canvas is analog, with muscles which are not necessarily exact and accurate to the sub mm range.

      Presumably it's only a matter of time before AI image trickery gets to that percentage or higher for someone who knows how to g

    • by gwjgwj ( 727408 )
      The result of using the brush is also not completely predictable. Does it mean, that the paintings are not copyrightable?
    • "The fact that Midjourney's specific output cannot be predicted by users makes Midjourney different for copyright purposes than other tools used by artists" -US Copyright Office

      Define "predict". Does this statement invalidate all previous works of art where the artist could not predict the final outcome? Like splashing paints randomly onto a canvas? Or being lucky enough to photograph a spontaneous event that the photographer could not have possibly predicted? No more "happy accidents"?

      • So if the author had said "that's nice, but add some shadows over the face, put the eyes closer together, add an enigmatic smile, a bit more.. to the left, and.. hold!" then it might get more of an actual authorship perhaps.

    • by Reziac ( 43301 ) *

      So how about, say, a photo of the mostly-unpredictable pattern made by a kaleidoscope??

  • specific output (Score:3, Insightful)

    by Iamthecheese ( 1264298 ) on Wednesday February 22, 2023 @08:48PM (#63316619)
    >"The fact that Midjourney's specific output cannot be predicted by users makes Midjourney different for copyright purposes than other tools used by artists"

    A painter cannot predict the precise line hes about to draw. If he's experienced he can get very, very close, though. A sculptor cannot predict the precise lines of fracture he's going to create. Only writers have the privilege of near-perfect precision. I'm not making a judgement as to the merits of AI-created art, but all visual arts involve some lack of precision. This is only different in the degree of lack of precision.
    • Re: (Score:3, Interesting)

      by CaptQuark ( 2706165 )

      A painter cannot predict the exact lines they are about to draw, but they control the brush or the pen and can decide whether to keep the brush stroke, modify it, or scrap it altogether.

      An example is a water color image. The artist cannot control the amount the water color spreads on the paper, but they have experience to know the average wicking action each type of media exhibits. To say the painter cannot predict the exact line of the paint, ink, charcoal, or pencil makes it sound like every painting is

      • Knowing what to ask for and which is good, if any, is most of the creativity...
        • So if I ask a friend who is a good artist to paint the picture of the Cocker Spaniel, is he the creator or am I? Who would you give the copyright to? The creator or the "visionary"? And since the machine cannot be granted a copyright, it does NOT automatically transfer to the "visionary".

          • Your friend would get the copyright because his personal experience and emotion went into that painting regardless of what you told him to paint. Unless you could prove you told him how thick to make every stroke or the exact shades of colors to use or how to do the shading with the colors. If you think the subject of the painting is what makes it a creative work than you don't understand art at all.
          • by guruevi ( 827432 )

            It's not about who is the creator, it's about who (if anyone) owns the copyright. And in your example it depends, did you pay the artist to do the work? Was there some form of agreement on the copyright regardless of payment? What was the jurisdiction in which the painting get created? Did said artist sublet their work without your knowledge? Is the artist considered a robot or a human?

            Copyright is a legal construct, an artificial rather than natural right, if it weren't for copyright law, you'd only have c

        • Knowing what to ask for and which is good, if any, is most of the creativity...

          Spoken like a man who's never painted.

          Heres the thing, the creativity, the REAL creativity, is IN the strokes. Any fool can come up with an idea. We all have them. But knowing how to compose the picture, colour selection, how to apply paints to get certain effects, and choosing what techniques, at any point your decision tree of choices opens to a million more choices, and only one will be the one you choose, and all the others

          • Re:specific output (Score:4, Insightful)

            by SandorZoo ( 2318398 ) on Thursday February 23, 2023 @07:27AM (#63317349)

            Knowing what to ask for and which is good, if any, is most of the creativity...

            Spoken like a man who's never painted.

            I think photography is a better comparison. Consider the amount of creativity that goes into most snapshot style photos. Often it's just the thought "I want a photo of that", point, and click. That's the lower bar on the amount creative input needed to justify a copyright on an image.

            I've spent some time with Stability Diffusion, and found I needed to continually refine the prompts, and try many different random seeds and badly drawn source images (and perhaps do a bit of infill) to get something I like. It's feels like much more of a creative process than most of my photos.

            AI is just another tool. It let's people like me, with almost no natural drawing ability, create some pretty stunning images. But it's me doing the creating.

          • "Oh no, I sneezed and it blurred the paint over there... No problem, I'll just turn it into a happy little tree!"

      • Asking Midjouney to create a image of a "Cocker Spaniel playing with a red ball" and being given four choices of what might be close to what the querent had in mind does not make him the artist.

        Which brings me to my new app, rights to stroke where you ask mid journey to create an image of a “cocker Spaniel” and then it supplies conventional drawing tools where “an artist” can trace them over thus creating the arrangement needed to secure full rights. Now you don’t need imagination, skill, or even much time at all! /s

      • by Anonymous Coward

        but they control the brush or the pen and can decide whether to keep the brush stroke, modify it, or scrap it altogether.

        You mean just like someone can do with the AI... Or are you one of the fuckwits who think everyone just puts in a few words and accepts what comes out?

        If that was true they wouldn't have outpainting extensions to expand an image, or inpainting to replace / fix literally any element of the image. Plenty of people spends hours or DAYS on a single image to get it perfect.

      • If it were any other way, the big players would generate hundreds of trillions of images, .. and own the copyright to everything adjacent to the existing dataset
        • by guruevi ( 827432 )

          Which you can technically already do. You can do so with sounds and to some extent with words and images as well. And people have done it, there is a complete set of musical four-chord progressions generated and published so nobody can claim copyright on them. You can do the same with words, although your space becomes a bit larger, if there were a mathematical representation of all "word progressions" (which is what these GPT models are approaching) you can technically represent any book ever written or th

        • by gwjgwj ( 727408 )
          So just copyright all the natural numbers. All the images are there, somewhere.
      • by kmoser ( 1469707 )
        Is a painting in which an artist splashes paint on the canvas randomly any less worthy of copyright than an image in which every brush stroke is created with specific intention? Is human-created artwork made from found items any less random than the regurgitations of an AI?
      • The artist cannot control the amount the water color spreads on the paper, but they have experience to know the average wicking action each type of media exhibits.

        So you're saying that inexperienced artists are not eligible to have their works copyrighted?

    • by Entrope ( 68843 )

      A painter cannot predict the precise line hes about to draw.

      Those fine details are not the part that copyright protects, though. If you take a painting and slightly tweak every brush stroke, a court will rule that you created a derivative work and not give you copyright. See, for example, https://www.law.cornell.edu/su... [cornell.edu] for a case where the Supreme Court is deciding whether a much more fundamental transformation still requires a copyright license.

      Another example is Shepard Fairy's "Hope" poster of Barack Obama, which was based on a picture from the Associated Pr

  • "The fact that Midjourney's specific output cannot be predicted by users makes Midjourney different for copyright purposes than other tools used by artists"

    It depends on the artistic purpose of the unpredictability. Consider other works made using indeterminency, like Cage's Music of Changes or to a lesser degree Pollock's splatter paintings. The point of these is in the randomness. Midjourney is used more to replace the creative toil, although you could argue there is some random intent to 'see what it com

    • Thank you for your post. I want to admit that I used chatGPT to write this response. It is a useful tool that I find helpful in generating ideas and organizing my thoughts.

      Regarding your statement about copyright, I can see your point of view. However, as an AI language model, I am not capable of owning copyright to any response I generate. The responsibility of copyright ownership ultimately falls on the human who inputs the prompts and selects the output. So, in this case, the copyright ownership of this

  • Using an AI to create an image is no different than using a compiler to create an executable. The creative input (and thus the copyrightable part) is the source code. If the input the AI is uncreative or duplicates previously known work than you truly have nothing to copyright. If it is creative and unique then you should be able to copyright it.

    • The creative input (and thus the copyrightable part) is the source code

      That's sort of the point. The creative input is the training data. That doesn't mean that the product of its "learning" isn't unique at all, but the relative effects your prompt has vs. the training data is relatively small. They're trying not to open a can of worms on the training data....for now.

      • Can you copyright a prompt, or the combination of a prompt and model version, anything to prevent someone else from getting exactly the same result?
      • by micheas ( 231635 )

        The creative input (and thus the copyrightable part) is the source code

        That's sort of the point. The creative input is the training data. That doesn't mean that the product of its "learning" isn't unique at all, but the relative effects your prompt has vs. the training data is relatively small. They're trying not to open a can of worms on the training data....for now.

        Unfortunately, Getty Images and Rupert Murdochs news empire have implied that they are going to try and do just that.

  • so, i'm curious... (Score:3, Insightful)

    by Jayhawk0123 ( 8440955 ) on Wednesday February 22, 2023 @10:33PM (#63316797)

    when pharmaceutical companies (or any other industry) create compounds and medicines based off of AI (fuck, i hate myself for using AI.. it's not AI.. it's fucken machine learning) ... does this then mean that the companies can't patent it?

    When you use AI to help improve designs for jet engines, wing designs, etc? same thing...

    works for me...

    If i create a painting using photoshop.. can i copyright the works? how is AI any different from a tool like photoshop? Blurring? erasing object from the background? etc?

    • by Anonymous Coward

      Yes, denying copyright and patents based on AI-generated works, that would be fair.

      But now we have an interesting point of debate: these learning models have no doubt consumed copyright materials as part of their training datasets, so have they now invalidated previous copyrights? If I ask Midjourney to "produce for me an anthropomorphic picture of a black mouse with big ears that's red wearing pants and clown boots" and it somehow comes out looking surprisingly like Mickey Mouse then who's on the hook for

    • by Tony Isaac ( 1301187 ) on Wednesday February 22, 2023 @11:53PM (#63316905) Homepage

      Copyrights and patents are two distinct branches of law. Yes, they both deal with intellectual property, but in different ways.

      Copyrights cover original works such as works of art. Copyright is automatic. If you write or draw something, it is automatically copyrighted unless you specifically transfer it to the public domain.

      Patents deal with inventions, and those inventions don't have to be original, you just have to be the first to register and commercialize your patent. Patents are not automatic, they require a legal process that is both expensive and arduous.

      • Is it copyrightable if you make some edits to a ML created image?

        Maybe you blurred some area, made some contrast changes, etc?

        The end product is a human created product after all.

        • Those questions may not yet have been resolved by the courts. But to answer, I'd suggest taking "AI" out of the equation. So,

          Is it copyrightable if you make some edits to a ML created image?

          We could instead ask, "Is it copyrightable if you make some edits to a *human* created image?"

          The answer, I think, would be the same in both cases.

          Maybe you blurred some area, made some contrast changes, etc?

          If you took a copyrighted image created by a human, and blurred an area, I'm pretty sure you would be in violation of the original copyright. So no, I don't think a small tweak like a blur or similar change, would make an otherwise non-copy

          • As I understand it currently, if a human created some image - regardless a photo or drew something etc, it's still copyright protected. So if you made some light edits, you are using a copyright protected image as your "base" to create your derived image. So you need permissions / licensing, and so on.

            According to this article, an ML created image does not have copyright. So your "base" image is already without copyright (no licensing / permissions needed) when you actually make the edits, so you are the fi

            • In your first example, you have to get permission to make a small change to the copyrighted image. This is true, you don't own it, and you also don't fully own the modified version.

              In the second example, where you take a "AI-generated" image and modify it in some small way, you are (according to the ruling) modifying a public domain image. You still don't own it, your modified image is still public domain, based on the same principle that says that the copyrighted image, even with modifications, belongs to

    • It seems to me that AI-generated art might fall under the "fair use" doctrine. The art is not truly original, but is generated from other people's works of art. If anyone should own the copyright, it's the artists who created the original art that was used as training material for the AI.

      • by erice ( 13380 )

        It seems to me that AI-generated art might fall under the "fair use" doctrine. The art is not truly original, but is generated from other people's works of art. If anyone should own the copyright, it's the artists who created the original art that was used as training material for the AI.

        Fair use would apply but it would be much trickier to navigate because the AI operator may have no idea which work was reused or how much until they get served with a lawsuit from the original artist. I don't think a blanket "It's all fair use" is ever going to fly.

        • You're right, it's a complex legal question that has not been resolved by the courts.

          The fact that you can't currently determine which original works of art were used in the composition of an AI-generated image, doesn't mean that it's not possible, only that the current implementation doesn't track sources.

      • by La Gris ( 531858 )

        The art is not truly original, but is generated from other people's works.

        No artist is creating art from thin air. Artists were trained and educated from other people's works. Every piece of art is a derivative in some way from other people work.

        No artist can produce without experiencing our World and previous artistic work.

        • That's right. And just as a human artist's "derivative" is protected under fair use of the original, I could see how AI-generated derivative works could be protected under fair use.

      • but it is original unless it's a direct copy.. teaching is not copying, it's training. if i am trained by painters, or watch how others paint, or go to an art gallery and look at paintings, then paint my version of a painting- say of a landscape, or something new.. i'm perfectly fine to say- it's mine.. and claim copyright. But if i teach an AI how to do it, and use it as my tool. I can't?

        this starts to fall apart when you look at music, art, etc... they all use machine learning to help with content creati

    • how is AI any different from a tool like photoshop?

      To me what this implies is that you can't copyright the parts of an image you've altered in photoshop which were created with content-aware fill or any of their other generative technologies.

  • Before long artists/creative people will have their nervous system linked up to the AI, thus providing input. Even something as simple as a device to measure your heartbeat, or your alpha waves, or electrical impulses from your hand as it touches a paintbrush, or a mouse.

    No judge is going to want to rule on where creativity ends and it's the AI doing the work, especially if famous artists get involved, and sell their "collaborative" work for a high price.

  • Use the AI to create the images, then pay a real artist to make a replica, with maybe a slight alteration.

    Fixed.

  • Using this precedent, it would seem Microsoft Copilot violates copyright law

    • by Entrope ( 68843 )

      How so? The obvious analogy is that code written by it cannot be copyrighted, but different logic would be needed to say it violates some copyright.

      • by guruevi ( 827432 )

        Microsoft Co-Pilot obviously copies existing code from various places and passes it off as its own. Ask it to write an SQL insert in any language and it will copy the code from Stackoverflow, bugs, injection exploits and all, you can often Google the code and find it verbatim somewhere else.

        • by Entrope ( 68843 )

          As I suggested, that is very different logic than what the Copyright Office said in this case. I was not disputing that Microsoft/Github Copilot infringes copyright, only whether this case tells us that.

  • ...preventing others from copying what the artist has created. In this case, what the author of the graphic novel has created, is a series of prompts to the Midjourney AI. So all she may be able to copyright is those prompts (but I doubt they would be copyrightable).

  • Is this book copyrightable then: https://www.rand.org/content/d... [rand.org]
  • The line between photoshop and AI is going to be *extremely* blurry.

    And these art programs are just a tool that multiplies artist ability and productivity.

    This is a really illogical artificial ruling.

  • I bet you can find bits and pieces of her "art" in other people's art. if a human did that, it would be plagiarism.

    • by micheas ( 231635 )

      You can find bits of other artists' work in all art.

      It's how humans create and communicate. The percentage of a copyrighted work that is truly unique is typically quite minimal. If something didn't contain large amounts of elements from other works it would be jarring and uninteresting.

      Copyright in the US was originally limited to books and maps. Newspapers, advertisements, magazines, and other things that were cheap to produce were not granted copyright.

      The courts have ruled that photography taken by a mon

    • by micheas ( 231635 )

      You can find bits of every artist's work in previous artists' work.

      It's how people create art, they experience art and the world around them and use that to inspire their work. You can trace schools of artwork by following the details of the works./p.

If all the world's economists were laid end to end, we wouldn't reach a conclusion. -- William Baumol

Working...