AI-generated Content and Copyright Registration - Creative Law Center

AI-generated Content and Copyright Registration

Things are moving fast in the world of AI-generated content. The Copyright Office is doing its best to keep up. Even though is it part of the U.S. federal government bureaucracy, it is a more nimble institution than the federal judicial system or Congress. Simply put, the Copyright Office can change its policies and procedures faster than a court or Congress can act. And it has.

TL;DR: The Copyright Office’s Current Position on AI-generated Content.

If you use AI technology to create work, you can claim copyright protection for your contribution to that work. Your contribution to the work must be appreciable. If your contribution is de minimis, the work is not protectible.

Conversely, if the contribution to the creative work made by the AI technology is de minimis, the Copyright Office is not interested in even knowing about it. But if the contribution by AI to the work is appreciable, then that part of the work is excluded from copyright protection.

In copyright law, "de minimis" is a Latin expression meaning "about small things." This concept typically arises in situations where the amount of copyrighted material used is so small or insignificant that it does not constitute a copyright infringement. In the context of AI-generated content, de minimis means, essentially, so little as to be not worthy of notice.

de minimis is one of those fuzzy, legal no-bright-line concepts. Think of it as a spectrum with no AI-generated content in the work on one end of the spectrum and the entire work is created by AI on the other end with de minimis and appreciable sliding around in the middle. Where de minimis ends, appreciable begins.

AI Protectibility spectrum

Roy G. Biv spectrum diagram of AI-generated content protectibility. Click to enlarge.

Copyright Office Cancels, then Reissues, Registration for AI-generated Comic Book

You may recall that in September 2022, Kristina Kashtanova, a graphic novelist, applied for copyright registration for her comic book Zarya of the Dawn. In her application, she listed herself as the author. The Copyright Office registered the work.

Zarya of the Dawn was the first known instance of AI-generated work being registered and it garnered a lot of publicity on social media and in the press. The problem was that the Copyright Office did not know that the work was AI-generated. Kashtanova had not disclosed the use of artificial intelligence in her application.

Excerpted images from comic book Zarya of the Dawn.

When the Copyright Office learned that Kashtanova had used the AI image generator Midjourney to create the images in her comic book, Kastanova was told that the registration would be canceled unless she provided additional information. Which she did.

Kashtanova explained that she wrote the text, that she prompted Midjourney in order to generate the images, that she edited some of the images in photoshop, and that she decided which images to include in her work, how the images were to be arranged, and in what order.

Ultimately, the original registration was canceled “for failure to exclude non-human authorship.” After receiving Kastanova’s explanation, a new registration was issued (with a retroactive date) and the basis for the copyright protection was changed to “text; selection, coordination, and arrangement of text created by the author and artwork generated by artificial intelligence.”

The AI image generator’s contribution to the creation of the images was considered to be appreciable by the Copyright Office and Kastanova’s contribution to the creation of the images was de minimis. Therefore, the images themselves are not protected by copyright, but her selection, coordination, and arrangement of those images in her comic book is protected by copyright because it is the product of her creative choices.

Human Authorship Requirement for Copyright Protection

Copyright protects “an original work of authorship fixed in any tangible medium of expression.” Works of authorship are limited to the creations of human authors. The term “author,” which is used in both the Constitution and the Copyright Act, excludes non-humans. The courts are consistent on this point.

The case law goes back to 1884 in Burrow-Giles Lithographic Co. v. Sarony, in which the Supreme Court extended copyright protection to photography. More recently, in 1997, the Ninth Circuit held that a book containing words “‘authored’ by non-human spiritual beings” is only protected by copyright to the extent there is “human selection and arrangement of the revelations.”

The policy of the Copyright Office reflects these decisions. If there is no creative contribution from a human, a work cannot be registered. The human contribution to a creative work is protectible and registrable, the non-human contribution is not. 

Which leads to the question of how to fill out the copyright application form when AI technology is used in the creation of the work.

How to Complete the Copyright Application when Using AI-generated Content

There are eleven sections, or screens, in the Standard Application in the online registration portal at copyright.gov. Information about AI-generated content needs to be identified in only two of those screens: the Author screen and the Limitation screen.

The Author Screen

After you identify the author (who has to be a human), the application asks you to describe the authorship that was created by the author. If the author used AI-generated text and incorporated it into a larger textual work, “Text” is properly selected. If the author created the artwork, but used AI to write the text, “Artwork” is properly selected.

If the author contributed an appreciable amount of the text and the images, both can be checked.

If the author arranges the human and non-human content in a work, then in the “Other” field, the Copyright Offices advises use of the following language: “Selection, coordination, and arrangement of [describe human-authored content, i.e. text or artwork] created by the author and [describe AI content] generated by Artificial Intelligence.”

You do not need to go into any more detail on the Author screen. There’s no need to identify the AI technology that was used. 

screen shot of author screen from online copyright application

Screenshot of Author screen from online copyright application.

The Limitation Screen

The Limitation screen is where you exclude all AI-generated content from your application. When you exclude something, that means you are not claiming copyright ownership or protection for it. Work in the public domain has to be excluded, as does previously registered work.

Excluding content is easy to do. Remember that if the AI-generated content is de minimis, you do not have to mention your use of AI at all.

If you have incorporated text and/or images into your larger work, the Limitation of Claim screen is where you tell the Copyright Office about it. Simply provide a brief description of the AI-generated content, like “some images were generated by artificial intelligence” or “some text was generated by artificial intelligence.”

You do not need to specify which text or images were created by AI, only that there is AI-generated content in the work.

Remember to add a checkmark for the material you created, that you want protected.

Screen shot of the Limitation screen from the online copyright application.

Screenshot of the Limitation screen from the online copyright application.

Example of a de minimis contribution to a work

Because the acceptance of AI-generated content in a work registered with the Copyright Office is new, there are few concrete examples of what would constitute a de minimus contribution. In the Zarya of the Dawn application, the Copyright Office rejected the argument that Kashtanova's editing of the image created by Midjourney was sufficiently creative for copyright protection.

de minimis editing in photoshop

de minimis editing in photoshop

Kashtanova argued that she made changes to Zarya's mouth, particularly the upper lip. The Copyright Office determined that the changes were "too minor and imperceptible to supply the necessary creativity for copyright protection."

Short phrases and brief quotes have always been considered de minimis by the Copyright Office with no need to exclude them in the registration process. It strikes me that the kind of uses that can be made of AI tools that would be de minimis and not worthy of mentioning would included the use of AI to generate ideas that the writer or artist uses to create their own expression. After all, that is what copyright protects -- the expression, not the idea. AI assistance to help outline a book, to suggest character traits, to list elements when building a world, to structure a visual composition -- these would most likely be considered de minimis with no need to mention the role of AI in the copyright application.

Conclusion

What I particularly like about the approach the Copyright Office has chosen is that it takes existing law and applies it to the new AI tools. The registration system does not need to be overhauled; a new application form is not needed. The framework that is already in place can be used to protect the work of creatives from a registration perspective. We can all move forward, use the tools, and continue to protect our creativity.

But the fact that work created with AI assistance can be registered does not ease the burden of proving the extent of human authorship versus AI-generated authorship when it comes to an infringement action. If someone infringes work that you created with the help of AI, you will need to show that it is an infringement of your contribution to the work, not the AI machine’s contribution to the work.

The creative choices made by a human to produce a work are protectible by copyright. In order to prevail in an infringement action, you will need to explain what your creative choices were and how and when in the process you made them that resulted in the final work.

The more thought you put into crafting, designing, refining, and directing your prompts and the more you reject, edit, revise, rewrite, or change the output from the AI content generator, the more likely you will be viewed by a court to have created a protectible work. You need to think about the coming challenge of sorting out the difference between de minimis and appreciable AI-generated content as you create your work.

About the Author

Kathryn Goldman helps small business people, writers, artists, and creative professionals make a living from their creative work by teaching them how to protect and enforce their rights. She is an attorney who writes these posts to help you be more thoughtful about intellectual property and the law as you build your business, write your stories, and create your art.

  • Thank you for the interesting article, Kathryn.

    So far, Ai seems to generate content that has as much human emotion, with as much depth and authenticity, as what is in a poor 3rd generation photocopy.

    I’ve also noticed that when Ai is more convincing, it still lacks the depths of genuine psychological connections, lacking elements which do not seem to bother consumers who are less discerning. The problem with that is those consumers devolve to believe that contrived psychological connections are genuine.

    Well, if it’s on Google, it’s real. Right? No need to look further for any creator.

    At that point, ideal copyright becomes tainted.

    Looking forward to see how humanity can protect its original creative copyrights.

    As I’ve been saying on Instagram…

    What kind of genuine intelligence would replace itself with artificial intelligence?

    Have a good evening —

    LorenzO

    Writer Adapter Producer
    http://www.LorenzoOrzari.com

  • […] AI-generated Content and Copyright Registration. […]

  • Shelley Marshall says:

    I’ve translated books from Japanese and copyrighted them in the USA and was considering using AI for my next translation, a book that would be longer than 1,000 pages in English. Mostly to save time on input, my plan was for the AI to translate and then do the heavy editing and re-translations that would be needed. Since this is all up the air, I guess I’ll stick to search engine look-ups for now.

  • bob keck says:

    Thanks for your input. I use AI to create pieces and then assemble the pieces into an image. Also, on other AI pieces, after correcting them in Photoshop, I use colored pencils to enhance the images. So, where do these two applications of AI fall?

  • Sam says:

    Thank you for this info. As a seasoned illustrator I continue to watch with interest the evolution of the law around this.

    • It is important to stay up to date. Things are moving quickly and it’s good to know what your options are.

  • Thank you for blogging about this!
    It’s hard to know exactly what constitutes de minimus and what will be considered appreciative contribution. For example, if someone uses their own photo as input and the generated image retains more than half of the appearance of the photo, I wonder which side of the spectrum it falls on. Is it the composition that must be retained? If it’s slightly different, how much different before the original image is considered de minimus contribution?
    There’s new AI tools that let people change the placement and even poses of people in the image, I wonder in those cases will the AI element be considered de minimus?
    There are plenty of artists and illustrators right now claiming it’s impossible to copyright any AI generated art, however, according to the USCO I don’t think it’s fair to say that since there’s a lot more involvement one can have with generative tools.

    • If you own the copyright to the photo(s) the AI is trained on, it strikes me that the creator’s contribution to the resulting image will be appreciable and the AI’s de minimis. If we think about this logically, everything the AI creates is derivative of the work it is trained on. Derivative rights belong to the copyright owner of the photos. The human choses the training photos, crafts the prompts, refines the prompts, and perhaps edits the output.

      We’ll have to wait and see if I’ve chewed through the leash here.

      • EN Heim says:

        It may end up being the property of the AI machine. Ha-ha! The originators of the people that created the program.

  • A question came in by email about the fact that both grammar check and spellcheck are powered by AI. Both would be considered de minimis uses and need not be disclosed or excluded. Selecting alternative wording or phrasing is like using a thesaurus. Using spellcheck is like using a dictionary, but who still does that?

  • >
    Share
    Tweet
    Share
    Pin