Copyright registration of AI-generated imagery

This is a thread started on our deprecated Discord server by jritzman. I’ve copied it here to keep the discussion going.

I saw this story today on LinkedIn regarding Copyright registration (or not) of AI-generated imagery. I have heard of a couple other cases (which I will try to find links to and post here as well) but this one is new to me.: Does AI art deserve a copyright? | LinkedIn

Does AI art deserve a copyright? | LinkedIn

Kris Kashtanova created a graphic novel using the text-to-image AI program Midjourney. Now the Copyright office has decided whether Kashtanova deserves the same protections as a traditional author.


The full story is at The US Copyright Office says you can’t copyright Midjourney AI-generated images - The Verge

The Verge

The US Copyright Office says you can’t copyright Midjourney AI-gene…

The images in this comic book are “not of human authorship”

A LinkedIn member posted this comment, suggesting that one could simply not inform the Copyright Office of the fact that AI was used in an image’s creation. My reply was that you could certainly withhold that information on the registration application, but per the full story linked above, the Copyright Office can rescind a registration if/when they discover the AI-generated nature of the work (such as during litigation discovery/testimony). Bill Young on LinkedIn: The US Copyright Office says you can’t copyright Midjourney AI-generated… | 53 comments

Bill Young on LinkedIn: The US Copyright Office says you can’t copy…

WOW. I’m really surprised this hasn’t had WAY more pickup - this is a pretty hot topic on LinkedIn - impacting a wide variety of industries (like, most of… | 33 comments on LinkedIn

Honestly, it seems to me to be pretty clear that using AI to generate art is NOT a work of (human) authorship simply because a human provided the “word prompts” which were used to generate that/those image(s). To illustrate why, take for example a case wherein a person asks me to create photographs of “frogs on stilts” and so I then go and create a number of images of “frogs on stilts” with varying numbers of frogs and length of stilts in various environments. Would that person be considered the author of the images because they gave me the “word prompts” from which those images were produced? I say no, and I don’t think that answer changes just because the actual creator of the imagery is an AI algorithm as opposed to a human.

Ok… so what if you create your own Dall-E/MidJourney/StableDiffusion models? For example, I could take a bunch of images that I curate, fine-tune a model, run my prompts to generate images. How is that different than Jackson Pollack using gravity and momentum to “generate” his paintings? Or Raushenburg re-organizing “found objects”?

Or Duchamp’s “readymades” for god’s sake.

GPT: “Marcel Duchamp’s Readymades are works of art that are made from pre-existing, everyday objects that are not typically considered to be art. While Duchamp created his Readymades between 1913 and 1921, his works are still protected by copyright law in many countries.”

Creating a graphic novel with prompts is a hell of a lot more creative than signing a mass-produced toilet and putting it on a pedestal.

From jritzman:

  1. Creating one’s own AI algorithm & model is an interesting issue. From what I understand, computer code can be copyrighted, and a process for making something can be patented, but my course in Copyright Law in college was quite a while back, so they obviously didn’t cover the ramifications of obtaining copyright registration for an image created with one’s own patented/copyrighted AI image generation procedure. I’d love to pose that to an Intellectual Property attorney!

  2. [4:11 PM]

As for Pollack, no AI was used in the creation of his works, so I’m not sure I see the connection. Just because his means of art generation may seem “random” that does not mean he wasn’t the artist. He still chose the size and shape of the canvas, the type(s) of paint, the colors of paint, which direction he directed the paint to go in, and when there was sufficient paint in a sufficient manner upon the canvas to stop dropping paint on it and declare it a finished piece of art. Also see someone else’s commented on a similar subject on LinkedIn: Bill Young on LinkedIn: The US Copyright Office says you can’t copyright Midjourney AI-generated… | 53 comments

Bill Young on LinkedIn: The US Copyright Office says you can’t copy…

WOW. I’m really surprised this hasn’t had WAY more pickup - this is a pretty hot topic on LinkedIn - impacting a wide variety of industries (like, most of… | 51 comments on LinkedIn

Bill Young on LinkedIn: The US Copyright Office says you can’t copy...

jritzman 02/27/2023 4:22 PM

Ditto for Raushenburg - he (a human) was responsible for determining which object(s) to present in what manner. No AI told him what item(s) to put where or how to display them. Whether a functional item(s) can be given copyright protection as a statue is, again, something I’m not qualified to answer, but as in the case of the comic book, an arrangement of previously existing works can be considered a separately copyrightable work.

jritzman 02/27/2023 9:32 PM

I just saw this on DeviantArt’s Terms of Service page (they are using a “noai” or “noimageai” directive in their headers):

March 1, 2023

jritzman 03/01/2023 3:27 AM


Microsoft and Google back Typeface, an enterprise generative AI sta…

A new generative AI startup with ties to Seattle just raised $65 million from Microsoft’s M12 and others to help enterprises produce content. Typeface emerged from stealth mode Monday and… Read More

Microsoft and Google back Typeface, an enterprise generative AI sta...

My point on Pollock and Raushenberg: of course no AI placed paint or a sink for them. You say they had to decide on their canvas size, types of paint, colors, what object to buy–and, crucially, you claim that choice is what makes them an artist (Pollock) and their work not a copyright violation (Raushenberg). I’m saying this is precisely identical to what an artist with a model does… they have to consider the correct prompt, literally pick their canvas size, specify what kind of “paint” (style) to employ, and, just as easily as it is for Pollock to let gravity guide his paint, or Raushenberg to place his sink on end and sign it, a unique never-before-seen piece of art is created for the world that would have otherwise not existed had the artist-who-uses-ai-for-their-art not made it happen.

From jritzman:

I guess the question is did the artist “make it happen” or did they simply make a detailed request of an automated system, just as a client could make a detailed request to a human artist? The client making the request is not considered the author.

That’s exactly what performance art is. Some of the work by Yoko Ono, e.g., was simply a list of instructions for an audience to follow.

From ChatGPT: “Yoko Ono is known for creating a number of works that involve instructions or “scores” for performances or installations. One of her most famous works in this style is called “Instruction Paintings,” which were first exhibited in the 1960s. These works consist of brief written instructions, often presented on small cards or other printed materials, that invite the viewer or participant to imagine or enact a particular action or situation. Some examples of instructions from this series include: “Imagine the sky dripping. Dig a hole in your garden to put it in.” “Climb a ladder and jump off it.” “Imagine the sound of a bell ringing throughout the universe.” Other works by Ono that involve instructions or participatory elements include “Grapefruit” (a book of written and conceptual instructions), “Sky Piece to Jesus Christ” (a score for a musical performance), and “Cut Piece” (a performance in which Ono invited audience members to cut off pieces of her clothing with scissors).”

So, was Yoko the artist, or the people that followed her instructions?

Looking good for AI creatives. Tweet reads as below and includes a couple highlights in original tweet:

Narrative curation or mixed process.
Released today, these are the current standards the US Copyright office has set for AI Collaborative art to be registered. Vital win: intentional, creative curation from unmodified AI output is included.

Link to the source:

Interesting point, but I’m not familiar enough with Yoko’s performance art to reply with anything but generics. The first thing which comes to mind is whether a performance art “piece” is fixed in tangible form. If not, then there is no copyright protection. See Copyright in General (FAQ) | U.S. Copyright Office.

The second question which comes to mind is whether a pure “instruction” can be covered by copyright law. An “instruction” sounds like a “process” which I think is more under the patent umbrella than copyright.

The third thing which comes to mind is whether an “instruction” aka “prompt” is simply an “idea” - there is no actual implementation of it until the “instructions” are carried out. In that case, it is very clear that such an “idea,” “concept,” “system,” or “method” is not eligible for copyright protection. See What Does Copyright Protect? (FAQ) | U.S. Copyright Office. Thus it would not even be possible to claim copyright protection in a “prompt” itself.

Maybe a controversial take, but copyright itself is already very outdated in its current implementation, and will become increasingly irrelevant very quickly if the current trajectory of AI generated content continues. Copyright effectively exists to ensure creators can profit from creation, but if anyone else can generate the same thing on their own regardless of copyright laws, then it mostly mitigates anyone’s ability to charge for their generated work.

I’d be surprised if 20 years from now I can’t sit down at a computer and ask for a full length animated movie of Tarzan but changing the setting to something from the Marvel universe. And frankly this might be the most common method of media consumption, given that it can be perfectly tailored to my preferences. At this point, unless there is some sort of establish elite control over the entirety of AI, there is really no way for Disney to profit from their IP.

I’m curious whether using AI to generate movies/TV/music/art will become the norm. Right now, my thoughts are that there could be a spike of user/AI-generated content, and I think we are seeing that what with MidJourney/DALL-E/Stable Diffusion-generated art being all the rage. Copy and paste that when AI-generated video is widely accessible.

In the long-term though, I think it will be just a “phase” that we go through as people explore what is possible. As it is, people can currently pick up painting or photography or video production (everyone’s phone shoots video - no expensive production gear needed any more) and yet people enjoy watching Netflix/HBO Max/Disney+/etc because - I think - people like to hear/see stories more than they like coming up with the ideas and producing them.

Maybe if it really becomes as easy as “Alexa, make me an album in the style of Led Zeppelin” and BAM twelve amazing rock songs appear; or “Siri, make me a couple dozen superhero movies which tie in with each other culminating in an epic two-part story” and we get an AI version of the Marvel cinematic universe (but without the trademarked characters) then ok cool, but I’m guessing that the prompt-work needed to generate an actually good album/movies will be a lot more complicated. Also, there were some bad Led Zeppelin songs and Marvel movies, and given what we saw from ChatGPT last week, I’m not holding my breath for AI to produce high-quality works any time soon.

Lastly, even if an AI system COULD generate a song/movie that I really like, I’ll never be able to share the experience of going to a concert to see a band perform that song, or to go to a movie theater and see that movie. I think there will still be a demand for studio-produced artists and movies - even if those songs/movies are AI-driven - because people still like that experience.

As for copyright law in particular - it will still protect human-generated works of art. I don’t think that just because AI is a new tool to make art that no one will use traditional tools any more. Those traditionally-generated artworks will still need copyright protection in order to protect those artists’ interests, because I think there will still be people who are curious to see what other people have created.

I think prompt engineering is likely a short term concept similar to creating punch cards in the 70s to write programs. We already have a ton of platforms that show you content based on what you’ve viewed or liked before in the past. I imagine it’d be plenty possible for an AI to generate content based on preferences too.

Agreed on the human experience - I think that is what we’re going to see a shift towards, which I’m very happy to believe.

The internet drove us all apart but we’re about to have a great reason to come back together as the internet deteriorates into a mess of information and no one is able to determine whether any of it is human in origin.

I don’t think that just because AI is a new tool to make art that no one will use traditional tools any more.

I think some people will use traditional tools, but I think they will be the vast minority. We have people today who still do hand-drawn animation projects, but this is virtually never done for any commercial application. Most of the creative products people consume are commercial in nature.

This take (people don’t like generating creative content) seems to follow my general disappointment with how expressivity on the web trended over the past 20 years. Initially, it enabled an explosion of creativity and self-expression (GeoCities, LiveJournal, MySpace) but then was systemically subverted to drive “attention” to “influencers” due to the long-term effects of capital investment… monetization of platforms, investment in brand management platforms, etc. I lament that my daughters and their generation are captured by TikTok, Instagram, and other platforms that are, essentially, ad platforms.

So, playing this out, sure we all might be able to make content easily, but it’s the advertising of this content that is locked into existing platform audiences, and we’re all still subject to the social mechanics around them.

On the other hand, maybe there is an opportunity to escape from influencers and social pressure to follow the crowd. If we lower the barriers of creativity so far that anyone can share what they want to communicate as easily by generating new content as they could by finding and linking to others’ content, maybe we eliminate the need for platforms to find expressions to communicate.

We will still follow the expressions of those who inspire us, but maybe that will be the voices of the previously disenfranchised rather than the voices of those who profit from maintaining control of the expression of others.