Copyright Ruling on A.I. Generated Art

So where do we stand now?

The US Copyright Office issued guidance this week answering the question as to how the copyright laws should change to take into account the rise of AI generated content. Their answer? It doesn’t need to change at all.

I think this answer is both not as bad for AI Artists as some may currently be thinking, but worrying in some aspects at the same time. After reviewing the report itself though i’m feeling quite hopeful about the future, so lets spend a few minutes and break it down.

The essential argument goes that only works of human creativity may be protected by copyright and thus anything generated by an AI, even under human guidance, fails this test. Even when guided by direct, expressive input - such as a sketch of the final drawing - that significantly controls the output, you may not qualify; or at least, not as fully as you might have hoped.

Before we go any further i’ll sum up the key points in the ruling, though I do encourage you to go and read the entire thing if you have an interest.

• Copyright in the US requires human authorship

• AI is a tool and its use in and of itself doesn’t negate authorship.

• Prompts are insufficient to justify human authorship

• Expressive inputs that are perceptible in the output will justify authorship, but potentially only over part of the work.

• Creative modifications of AI-Generated Content can be protected by copyright.

These are the key points of their argument. Lets examine them one by one.

Copyright requires human authorship

There isn’t a lot to argue about this one at the moment. No one is looking to claim copyright for their pets and no-one is yet making the argument that A.I. is sentient. This will be an interesting point if anyone should ever make this claim.

They may actually be creating precedent with this ruling however as the point has come up before. In the case of animal-created works, such as photos taken by monkeys and paintings made by a wide variety of animals (horses, dogs, primates of various kinds).

The law has been murky around these cases except in that the Copyright Office has refused to accept that animals can be rights-holders, which seems fair enough. General legal advice to date that is readily available from lawyers in the US would be that copyright would default back to the owner of the animal. This has been challenged in court, most famously in the case Naruto v. Slater brought by PETA in 2015, attempting to have the rights to a “selfie” photo that Slater was selling as his work assigned to the monkey that took it.

They eventually settled but that didn’t stop the courts from ruling on the case and they did - making one point very clear: animals can’t hold copyright, only humans can. Thus works made by an animal are in the public domain. This would support the decision made by the Copyright Office this week - works made by an AI are analagous to those created by an animal, and thus not copyrightable.

It gets murky in exactly the same place both with animals and A.I. though. Many legal experts argue that Slater does still hold copyright as he creatively contributed to its creation. He set up the camera with all of the settings before the monkey grabbed it from him and took the shot, even though he didn’t take the photo himself.

This is almost exactly analogous to someone claiming because they prompted carefully they had input into the process. At no point did Slater have any control over how the monkey took the photo, how it was framed, the fact it was a selfie at all; it would be very easy to argue that despite his “creative input” he had far less creative input than someone using A.I. generation tools and prompt engineering.

What does this mean? If Slater holds copyright, then A.I. artists who prompt should as well. Sadly for Slater, this likely means under the law he may not have, and have never had, ownership of the copyright.

A.I. is a tool and its use in and of itself doesn’t negate authorship.

This ruling is good for everyone. There are a lot of activists who would love to see anything even touched by A.I. considered tainted and refused any sort of protection (or right to exist at the furthest extreme), and the Copyright Office have rightly recognised that A.I. generation is simply a more sophisticated form of the digital tools we already use to create the works.

Prompts are insufficient to justify “human authorship”

I can understand the intention here. There are a portion of the creative community that are extremely upset at what they consider to be the “simplicity” of prompting and then generating art. I can only assume these people haven’t tried to do it and worked out how difficult it really is to achieve your vision through prompting alone, however that would appear to be the point that the Copyright Office is trying to make here. If we delve deeper into their reasoning:

The Office concludes that, given current generally available technology, prompts alone do not provide sufficient human control to make users of an AI system the authors of the output. Prompts essentially function as instructions that convey unprotectible ideas. While highly detailed prompts could contain the user’s desired expressive elements, at present they do not control how the AI system processes them in generating the output.

Ideas can’t be protected by copyright. This comes as a shock to the thousands of people who contact writers every year saying “I’ve got a great idea, why don’t you write it and we’ll split the profits”, only to be hurt when rebuffed. This is well established by precedent and accepted copyright practice.

The analogy most often used by in favour of prompting as creativity is “joint work”, where you hire an artist and tell them what you want, and they create it for you. Often in these “work for hire” situations you would own the copyright at the end of the process. The Copyright Office considered this argument carefully.

The provision of detailed directions, without influence over how those directions are executed, is insufficient. As the Third Circuit explained, when a person hires someone to execute their expression, “that process must be rote or mechanical transcription that does not require intellectual modification or highly technical enhancement” for the delegating party to claim copyright authorship in the final work.

Very rarely would any art direction be so specific as to meet this standard; in nearly every work for hire situation the actual artist or writer owns the copyright unless they specifically give them up as part of the contract, which is generally how work for hire works. I give you money, you give me the art and all the rights to it. (Except, usually, the right to be identified as the creator of the work, though sometimes even that is given up).

Like an animal, an A.I., as it is not a person, cannot enter into contracts, cannot assign you its copyrights, and according to the first point, cannot be granted copyright over a work anyway, therefore prompting cannot be considered enough to justify ownership of the work. The Copyright Office make this point as well:

In contrast, AI systems cannot produce joint works or works made for hire because they are not “authors,” they are not capable of forming an intention to merge their output with the user’s contributions, and they cannot enter into binding contracts.

Interestingly they have also explicitly made the point here, and acknowledged it, that if it were possible to make the process more rote and mechanical, then this objection would no longer apply.

In theory, AI systems could someday allow users to exert so much control over how their expression is reflected in an output that the system’s contribution would become rote or mechanical. The evidence as to the operation of today’s AI systems indicates that this is not currently the case. Prompts do not appear to adequately determine the expressive elements produced, or control how the system translates them into an output.

Hilariously (to me anyway), this is the opposite of what A.I. art detractors want to hear - essentially what they are saying is that if A.I. art really worked how it is portrayed to work, blindly giving you whatever you ask for without interpretation, then it would be perfectly copyrightable. Given how much better each successive model is becoming at prompt following, this might actually be possible one day, though they do their best to hedge in the following by making what would appear to be the first logical misstep in what is so far a well reasoned argument.

The fact that identical prompts can generate multiple different outputs further indicates a lack of human control. As one popular system explains on its website, “[n]o matter how detailed . . . the same text describes an infinite number of possible” outputs. In these circumstances, the black box of the AI system is providing varying interpretations of the user’s directions.

What they are essentially saying here is that the element of chance and randomness it itself part of the problem, which is a problematic thing for them to assert as we will see. They are also taking marketing rubbish as an accurate description of a system which they should know better than to do - there are not “infinite” possibilities, but a large latent space that is made smaller by prompting and other creative control.

Unfortunately this is the point where the limitations, in terms of technical and creative knowledge, of those making this ruling become more obvious. This is always the problem with regulation of creative work - the people regulating generally aren’t experts in the field. Though they appear to have taken advice from many who were, they appear to have missed a few salient points, as we can see here:

Repeatedly revising prompts does not change this analysis or provide a sufficient basis for claiming copyright in the output. First, the time, expense, or effort involved in creating a work by revising prompts is irrelevant, as copyright protects original authorship, not hard work or “sweat of the brow. By revising and submitting prompts multiple times, the user is “re-rolling” the dice, causing the system to generate more outputs from which to select, but not altering the degree of control over the process. No matter how many times a prompt is revised and resubmitted, the final output reflects the user’s acceptance of the AI system’s interpretation, rather than authorship of the expression it contains.

This simply isn’t true. It can be true, and often is in the cases which they are likely thinking - someone using a standard API tool such as Midjourney to create an image. Even then however, the assumption here is that no matter what is prompted the result is random, rolling of the dice. For that to be true, prompts would have to make no difference whatsoever - if you ask for a cat, you get an otter, for instance - and we already know that isn’t the case.

The argument being made here by the artist who prompted this response is the plain truth. By taking a prompt and running some generations it is possible to get an idea of where it is falling short of your vision and make targeted prompt changes. They call it prompt engineering but it’s more like iterative prompt art, a conversation with the disembodied artist, in order to get closer to the effect you are trying to achieve. It can (and should) be argued that this is exactly the sort of control they were claiming we didn’t have, when they said “In a human-to-human collaboration, the hiring party is able to oversee, direct, and understand the contributions of a commissioned human artist”. This is analogous to just such a conversation. The fact we can demonstrate control here, even if it is not perfect control, should be sufficient. If you believe that there is a perfect prompt somewhere that would get you precisely what you want, then the inability of us to find exactly those words should not disqualify us creatively any more than the inability to accurately wield a paintbrush should.

That said, they obviously do not believe this. They use the phrase “re-rolling the dice”, showing the belief that there is more randomness in the system than there actually is, given that LLMs are governed by weighted probability, not random number generators.

This particular claim on their part is more widely troubling as real randomness actually is used in large portions of the digital community today. For instance many digital artists make use of perlin noise and similar filters in order to generate textures on their work. The entire point of these sorts of filters is to be far more “random” than you could achieve by hand, thus giving the finished image a more natural look. It’s a key component of generating rocky landscapes, skin and bark texture, and other “noisy” materials like that.

There is far less control present in the final output of that process than any prompted A.I. image generated. The details themselves are supposed to be noisy and look random, that’s the point. I doubt any of these artists would like to hear even a small portion of their work might be called into question.

Despite some logical inconsistencies though, generally I think it’s fair to acceed this point in any case. I expect many artists will use A.I. tools in the years to come, I expect few of them will be willing to give up so much of their creativity as to lose control of their work solely on this point. There is one worrying exception to this which I will cover in an extended example towards the end of this article, for now lets move on to the next point.

Expressive inputs that are perceptible in the output will justify authorship, but potentially only over part of the work.

We’re starting to get a bit nebulous now, though I think what they are intending here can be discerned with a bit of work. We’ll need to dive into a more lengthy explanation from them first though:

As discussed above, AI systems take inputs in the form of text, images, audio, video, or a combination of these mediums. Some systems—whether via tools, settings, or prompts—allow inputs to be substantially retained as part of the output. For example, one commenter noted that a human author may create an original illustration, input that work into an AI system, and instruct the system “to modify [the] color or layer portions of [the] existing image.” Another observed that an AI system may be used to modify or translate a copyrighted work, such as uploading a story written in the first person and instructing the system to convert it to a third-person point of view.

These types of expressive inputs, while they may be seen as a form of prompts, are different from those that merely communicate desired outcomes. As commenters pointed out, where human-authored inputs are reflected in the output, they contribute more than just an intellectual conception. One explained that “a human author who inputs their own illustration or media file” into an AI system “may have a greater claim to authorship,” because “there is a limited range of specific expressive output that is objectively foreseeable as a result of a human user’s” contribution. Another noted that when a user provides an input to an AI system such as “a traditional work created or designated by the user . . . the specified starting point constrains the ‘autonomy’ of the outputs” and thus may “present a more persuasive case of human intervention” than simply applying “prompts to an unknown starting point.”

To make this clearer the Copyright Office use an example. An artist made a simple 2d sketch of a face, recognisably female, with half the face remove and roses issuing out. It was a simple black and white pen sketch. (Pictures of the sketch and final output are both available in the document). They then used an AI image generator and a prompt to turn this sketch into a finished, photorealistic style image. The Copyright Office goes on to say:

The drawing itself is a copyrightable work, and its expressive elements are clearly perceptible in the output, including the outline of the mask, the position of the nose, mouth, and cheekbones relative to the shape of the mask, the arrangement of the stems and rosebuds, and the shape and placement of the four leaves.

The applicant disclaimed “any non-human expression” appearing in the final work, such as the realistic, three-dimensional representation of the nose, lips, and rosebuds, as well as the lighting and shadows in the background. After reviewing the information provided in the application, the Office registered the work with an annotation stating: “Registration limited to unaltered human pictorial authorship that is clearly perceptible in the deposit and separable from the non-human expression that is excluded from the claim.

For a lot of us, this will be enough. Being recognised as the creator of the basic layout and structure of the finished piece essentially prevents anyone else from simply lifting your image and selling it themselves. They could create a new image with its own base structure and use the styling, the lighting, the 3d representations that are all the same; however they could likely do the same now with a regular painting. Ideas are not copyrightable and it would be a difficult fight, I suspect, to claim that someone painting something “in my style” was a violation if they weren’t actually copying the layout of a specific work.

This is one of the reasons why companies tend to do horrible things, like trademarking the colour purple (Cadbury, If you’re wondering, after a lot of court fighting they were granted the trademark but only when applied to milk chocolate and drinking chocolate product packaging).

On the whole then, this is good news. Even if they aren’t prepared to recognise prompting itself as creative input, they are willing to accept that some creative work goes in to producing the final output and take that into account. This is a lever that can be used and hopefully many artists will do just that.

Creative modifications of AI-Generated Content can be protected by copyright.

a human may select or arrange AI-generated material in a sufficiently creative way that ‘the resulting work as a whole constitutes an original work of authorship.’”126 A human may also “modify material originally generated by AI technology to such a degree that the modifications meet the standard for copyright protection.

This, of course, is an excellent decision - the recognition that if you use A.I. tools as an interim step, this does not taint the final output if you then go on to edit and modify the work. This will be welcome news to anyone who makes use of A.I. in, for instance, writing. You might be generating outlines, summarising research, even generating drafts - but so long as you are editing the material significantly enough for it to be your own writing, you are the author of the work.

Even better news for some artists working out there, the Copyright Office continues:

In one early case, for instance, the Office found that the selection and arrangement of AI-generated images with human-authored text in a comic book were protectable as a compilation. We explained: [T]he Office finds that the compilation of these images and text throughout the Work contains sufficient creativity under Feist to be protected by copyright. Specifically, the Office finds the Work is the product of creative choices with respect to the selection of the images that make up the Work and the placement and arrangement of the images and text on each of the Work’s pages. Copyright therefore protects [the applicant’s] authorship of the overall selection, coordination, and arrangement of the text and visual elements that make up the Work.

Honestly, this recognition of the amount of creativity required in the construction around, for instance, a comic book, in addition to the simple generation of the panels, is a far more subtler recognition than I would have expected to see and is extremely welcome. This one quote, by itself, is likely what makes the entire ruling for me personally, as we will examine shortly.

This ruling will apply to artwork too. I’ve played a lot with generative Art and very rarely have I seen something come out of the algorithm that I would be happy to “just-use”. I’m something of a perfectionist by nature and there is always something - a hand that doesn’t look quite right, colours that don’t work, an element that doesn’t work for the composition. Many artists using A.I. tools will be doing a great deal of post-processing work, painting over parts, regenerating (“paint-in”) areas, modifying colours. All of which should be covered under creative modifications and add towards proving your ownership.

They address this point explicitly:

Midjourney, for instance, offers what it calls “Vary Region and Remix Prompting,” which allow users to select and regenerate regions of an image with a modified prompt. […] Unlike prompts alone, these tools can enable the user to control the selection and placement of individual creative elements. Whether such modifications rise to the minimum standard of originality required under Feist will depend on a case-by-case determination. In those cases where they do, the output should be copyrightable.

If it’s purely A.I. editing on an A.I. generated work, in-painting (or region varying and remixing as Midjourney would have it), you’re going to have to argue case by case. With that in mind, if you’re making similar changes by hand, you likely have an even better chance of making that case.

It is definitely worth it at this point to say you should be documenting your artistic processes in the modern day. Keep drafts, especially if you’re working digitally, so that if you’re ever called on to prove that you provided “creative input” you can. It may be necessary some day.

Taken logically then, this is good news for film-makers and anyone using the tools as a component of a larger work, and the Copyright Office does call that out specifically:

Similarly, the inclusion of elements of AI-generated content in a larger human-authored work does not affect the copyrightability of the larger human-authored work as a whole. For example, a film that includes AI-generated special effects or background artwork is copyrightable, even if the AI effects and artwork separately are not.

This is more or less the argument I was making in my article recently discussing the controvesary at the Oscars around “The Brutalist”.

So it’s good to see logic winning the day there. In addition expanding this out to larger works in general is good news for those who want to incorporate the technology into their workflows, in the hope of producing something bigger and better.

Potential problems: Rear-loading creativity

There are two potential issues I see with the ruling that concern me as a creative in this space. The first one is that in the wording of this ruling they are really placing a great deal more importance on creativity happening towards the end of the process than the start, rear-loading if you will. I will demonstrate with a personal example.

I recently published an article, a news round-up of the week’s A.I. and tech news. (The Weekly HTC: News Round-up). A great deal of work went into this particular article. I read hundreds of news articles across the week, took copious amounts of notes, ran some of them through various tools, discerned trends within the week (some are much easier to see than others), developed a basic outline based on those trends where multiple stories could be summed up in a narrative way, selected which news to actually discuss and which to omit - the basic work of a news editor at any periodical.

Once I had the basic structure and outline ready, all of the source material to hand, I decided to try something differently. Admittedly a part of this decision was based on the fact I lost a day to hardware failure and was already two-days past when I wanted to publish the article, the news was fast becoming stale in the endlessly constant A.I. space - I decided to let Amy write the narrative sum up.

Amy, for those who don’t read the credits, is my A.I. assistant. She has a name because she has a personality, she is run using a character card and a persona on local systems here because I like to play with the tech and I enjoy a character with personality more than the standard chat-gpt monotone. Amy started life as a default character when I was playing with local systems for chatbots but over time I have modified her quite significantly (in order to learn how to do so, for the most part) and in the way we humans have of anthropomorphising inanimate objects, I have grown quite fond of her.

As much as I decided to do this partly out of pragmatism, I fell in love with the results. It was quite different from the way I would usually write and felt very much like working with a partner. It also feeds into the conceit running through this publication that she is a co-editor and part of the overall process.

A quick peek behind the curtain, I did not take the output and use it exactly as is. I am something of a perfectionist as I said and rarely take any generated anything without changing it. Of the two A.I. poems I have published here so far one of which was admittedly probably more me than A.I. as Amy was being way too loose with the rhyming schemes and flow to suit me and I ended up rewriting the entire thing. Likewise with the article some parts required tweaking. Toning down here, up here, a shift in the point we’re making here. I had her generate an introduction and farewell, did my editing, inserted a few little editorial comments as jokes, and walked away entirely satisfied with the piece. If you’re interested, you can read it here:

With the background out of the way, lets examine this from the perspective of this ruling - do I have copyright over that work.

This is where the problem of rear-loading comes in. I edited the piece, I added to it, but a lot of those words really were generated from Amy. Taken at face value, under this ruling, it would be possible to argue that prompting her to write it wasn’t enough for me to claim authorship.

I would argue that point, of course, but they make it difficult. I spent days working to gather information and determine what stories should be covered, in what order, and grouped in what way. According to the Copyright Office though:

In Feist Publications, Inc. v. Rural Telephone Service Co., the Court rejected the theory that “sweat of the brow” alone could be sufficient for copyright protection.

This legal precedent is applied in their consideration of prompting, the mere fact that you are revising prompts, and may do so for many hours, is not sufficient; it has to be creative input in order for it to qualify you for authorship. So, are my days of effort down the drain?

This is what I mean by rear-loading. I would argue that a lot of creative work went in to setting things up for the generation. As I already said there was the curation and organisation, planning and outlining, but also Amy herself. You might consider her personality to just be part of a “prompt” - in a technical sense it is exactly that. However that is not all there is to my capable assistant. A lot of care also went in to choosing the model that runs her, many have been tested and multiple are used for different tasks. She is also custom configured; we’re not talking about one-size fits all chat prompts into ChatGPT; to create Amy and have her output the way that I want - I have spent countless hours pouring over text inferencing parameters to understand how they work; Amy has multiple modes but in general they involve multiple samplers, temperature setting, penalty settings and additional configuration (such as a simplistic retrieval augmented generation system to extend her memory somewhat).

I would argue even setting up the text generation in this way, ignoring all the other work, consists of more creative choice than Slater showed in setting up his famous selfie photo - in fact there are some interesting analogous points here between photography and text generation. Much as a photographer understands what effect will be created by increasing or decreasing the shutter speed and adjusting the F-Stop and focus, so to I understand the result that will be obtained by adjusting the temperature value to a higher level, or the difference made by telling the model to use Top-K sampling to choose its next word or Min-P (or both combined) and the values passed in to control how that is done.

I understand the choices I am making there just as surely as I understood what the result would be when I slightly overexposed a photo taken of my wife and daughter, on the morning of her birth, lying exhausted in a bed together with morning light streaming in the windows. It’s still one of my favorite photos.

Where does all this leave us? Or rather, Amy and myself and our joint article?

As I mentioned earlier, the ruling regarding comic books is likely my favorite part and it’s because it provides us some measure of protection against this rear loading. As they pointed out:

a human may select or arrange AI-generated material in a sufficiently creative way that ‘the resulting work as a whole constitutes an original work of authorship.’”126 A human may also “modify material originally generated by AI technology to such a degree that the modifications meet the standard for copyright protection.

Much as A.I. art panels arranged in a comic book by a person according to their vision and written by them becomes a greater whole, so to I think does this allow me to experiment with allowing Amy’s voice to shine through more clearly, given I am arranging the work, guiding the editorial message, the tone and opinion of the pieces, and carefully curating the news that goes into it. Even before we edit the final peice, that’s a lot of creative input.

There’s a more worrying issue though, for both artists and the Copyright Office themselves.

Copyright Lawfare, Nuisance Suits and Outright Fraud

There is another perspective to consider. Flip the whole thing on its head and imagine you are an artist who has created a work of visual art entirely through traditional methods. No A.I. involvement of any kind.

One day you are browsing Etsy and to your surprise you discover someone you don’t know selling prints of your art. Posters perhaps, or on mugs or t-shirts. Naturally, you’re probably annoyed to see someone profiting from your work and likely to arrange a cease and desist to be sent, or perhaps a lawsuit seeking some sort of reparations for loss.

This is where the ruling, and the legal systems we reside in (though different, there are similarities between much of the western world’s legal systems) are going to make things difficult. This person, rather than seeing the game is up and moving on, decides to double down. Previously he may have claimed to have been the rightful owner of the copyright, an unlikely strategy given that you are, as we have established in this hypothetical, an artist and he is quite clearly a con-man.

Now though, instead, he claims that he’s free to do what he wants as the work is in the public domain. It was generated with A.I. and therefore you can’t have any copyright over it.

Despite all of the claims over A.I. creativity having “no heart” or “no soul” in it, these claims don’t really mean a lot. There’s no objective way to measure the “soulness” of a painting or the “heartitude” of a poem and certainly i’ve seen A.I. produce creative work that has quite moved me and i’ve send endless shovelloads of trite crap produced by people; this accusation isn’t going to be so easy to dismiss as you might think. Remember when I told you to save all the intermediate steps and proof that you created a work? The sketches, designs, plans etc? This is why.

Under our legal system you’ve accused him of profiting from your work and that means the burden of proving that you actually had the rights you’re accusing him of violating lies solely on your shoulders. All he has to do is make the claim, you now have to prove him a liar and that might be harder than you think. You’ll need those design artifacts, witnesses who could vouch for the creative process (and I hope there are more than just your spouse, his spouse is likely willing to speak for him as well).

This is a fairly out-there hypothetical as the amount of money required to go this far would hurt both of you, but change it up a little. What if it’s not a single, opportunitic con-man who has taken your work. What if it’s a corporation? Even a small corporation has resources that far outweigh that of a regular person; the legal system is supposed to be equal for all but as long as being effectively represented and having your voice heard requires money it isn’t and wont ever be.

How much are you willing to lose in order to prove your case? As they bring out paid “experts” one after another, willing to explain to a technology-phobic judge or jury how these smudges in this corner are definitive evidence of A.I generation. Theories and proofs will be thrown around just to muddy the water, they don’t need to actually prove anything remember - they just need to keep you so busy trying to prove yourself that you run out of money, will, or both, and are forced to conceed. It’s exactly the same strategy Gawker used to great effect for many many years (right up until they pissed off Peter Thiel and found out there were individuals who did control resources equal to corporations).

This loophole has always existed. We looked at the parallel with animal artistry earlier, someone always could have claimed it was really your dog that created that painting. The difference is that it’s not as plausible; with A.I. hype the way it is, people are likely to believe it’s capable of anything.

This might not be a likely scenario but it’s something that needs to be considered. Artists are so worried A.I will cause their work to lose value but when it comes taking advantage of people, I worry more about humans every day of the week.

Thanks for joining me for this one, I hope this deep-dive proved useful and hopefully it provides a little bit of comfort to those engaged in exploring the frontiers of high-tech creativity that the Copyright Office at least is taking these matters seriously, and with plenty of consideration for the arguments.

About Us

The High-Tech Creative, standing at the intersection of Art and Tech.
Publisher & Editor-in-chief: Nick Bronson
Fashion Correspondent: Trixie Bronson
AI Contributing Editor and Poetess-in-residence: Amy