Zarya of the Dawn: Is the Sun Setting on the Copyrightability of AI-Generated Works?
At the end of February, the U.S. Copyright Office made waves again when it announced it was revoking a comic book author’s copyright registration in the AI-generated images used within their book Zarya of the Dawn. This decision arrives almost exactly a year after the Office denied registration of another artwork that was claimed to be authored by an AI.
It’s clear the Copyright Office is putting up major barriers to those trying to claim ownership of AI-generated artwork. But if AI technology itself cannot author a copyrighted work, and a human who directs the AI to generate images cannot claim ownership of those images, then where does that leave us? And what does this all mean for creatives who want to use AI-generated content in their work?
Naruto, a macaque, took this self-portrait with a camera owned by photographer David Slater.
Monkey Selfies & The Copyright Act’s Human Authorship Requirement
Naruto, a macaque, took this self-portrait with a camera owned by photographer David Slater.
The Copyright Office clarified in December 2014 that it would only register an original work of authorship “provided that the work was created by a human being,” and would not register works created by animals or by machines without human intervention (we’ll come back to that later). See U.S. Copyright Office, Compendium of U.S. Copyright Office Practices § 306 (3d ed. 2021). The Office further explained that “[b]ecause copyright law is limited to ‘original intellectual conceptions of the author’, the Office will refuse to register a claim if it determines that a human being did not create the work.” Id.
This clarification was made largely in response to the saga of British nature photographer David Slater and the now-famous “monkey selfies.” In 2008 Slater traveled to Indonesia to photograph endangered Celebes crested macaques, and after noticing that the monkeys were curious about his camera, set up his equipment in such a way to encourage them to play with it and snap several photographs themselves. In 2011, an editor on Wikimedia Commons uploaded one of the resulting photographs, asserting that it was “the work of a non-human animal” and thus within the public domain. Slater opposed this interpretation, claiming that his contributions to the setup of the “selfies” should be sufficient creative contributions to make him the author of the works. Adding to the confusion, the People for the Ethical Treatment of Animals (PETA) intervened, filing suit against Slater’s publisher in the Northern District of California and arguing that the macaque itself (which PETA named “Naruto”) should be assigned the copyright.
The District Court ruled, and the 9th Circuit affirmed, that copyright law does not extend its protection to animals, so Slater was free to continue publishing the images. See Naruto v. Slater, 888 F.3d 418 (9th Cir. 2018). The courts did not clarify, however, whether Slater himself held any ownership rights to the “selfies.”
A Recent Entrance to Paradise,” an image generated by David Thaler’s program “Creativity Machine”.
A Recent Entrance to Paradise
A Recent Entrance to Paradise,” an image generated by David Thaler’s program “Creativity Machine”.
Fast-forward to February 2022, when the Copyright Office denied a request to grant an AI program copyright as author of a work of art. The algorithm “Creativity Machine,” which was created by programmer Steven Thaler and is designed to create images with minimal human interaction (again, we’ll come back to this later), created a surreal landscape image titled “A Recent Entrance to Paradise.” When filing an application to register a copyright in the work in 2019, Thaler listed Creativity Machine as the author. The Office took a dim view of this, finding that the work “lacks the human authorship necessary to support a copyright claim,” and so denied the application.
The Office explained that Thaler had “provided no evidence of sufficient creative input or intervention by a human author in the Work,” and that it would not “abandon its longstanding interpretation…that a work meets the legal and formal requirements of copyright protection only if it is created by a human author.”
The Office’s decisions against Thaler did leave open the possibility that, with “sufficient creative input or intervention by a human author,” a work created by an AI could still be protected by copyright. However…
Zarya of the Dawn
A panel from Zarya of the Dawn, generated by Midjourney from a prompt by Kristina Kashtanova.
On February 21, 2023, the Copyright Office cancelled the registration for the comic book Zarya of the Dawn because the application did not disclaim that the images contained within the comic were generated using the Midjourney AI technology. The applicant, Kristina Kashtanova (they/she), was informed via a letter to their attorney that Kashtanova would still be considered “the author of the Work’s text as well as the selection, coordination, and arrangement of the Work’s written and visual elements,” but “the images in the Work that were generated by the Midjourney technology are not the product of human authorship” and so were not “original works of authorship protected by copyright.”
A panel from Zarya of the Dawn, generated by Midjourney from a prompt by Kristina Kashtanova.
Kashtanova’s process began with a general prompt—e.g., “dark skin hands holding an old photograph”—which would result in four generated images. Kashtanova would then choose one image and through a series of “tweak[s] or change[s]” to the prompt the AI would generate a series of further iterations, and eventually arrive at the image selected for use in the comic. It was a “trial-and-error” process involving “hundreds or thousands of descriptive prompts.”
Kashtanova argued that Midjourney served “merely as an assistive tool” to their creative process, but in reviewing Kashtanova’s explanation and the Midjourney documentation itself, the Office determined that “[t]he process by which a Midjourney user obtains an ultimate satisfactory image through the tool is not the same as that of a human artist.” While the user can modify the prompt to influence subsequent iterations, “the process is not controlled by the user because it is not possible to predict what Midjourney will create ahead of time.”
In its decision, the Office stated unequivocally that “Midjourney users are not the ‘authors’ for copyright purposes of the images the technology generates.” The Office disagreed that use of textual prompts permit copyright protection of the resulting images because “Midjourney starts with randomly generated noise that evolves into a final image,” and that human-submitted prompts “function closer to suggestions than orders.” The Office also drew an analogy to the relation between a director and a commissioned artist; if Kashtanova had given the same prompts to a graphic artist, the artist, not Kashtanova, would be the author of the resulting work, and that artist would need to assign their rights in the work to Kashtanova for Kashtanova to own the work. But since AI cannot own the copyright they’ve created, then there is nothing for directors like Kashtanova to receive.
What Does This Mean for Ownership of AI-Generated Works?
The Office’s decision regarding Zarya of the Dawn leaves a massive question unanswered: who, if anyone, can claim ownership over the images created by generative AI technology? The Office appears adamant in their stance that the AI itself, as a non-human actor, cannot be considered the work’s author. Because the AI cannot be the author, it cannot own any copyright in the resultant work, and so cannot then assign those rights to another (human) party.
The Office previously left open a door for human ownership of AI work by saying that some degree of human interaction may permit the prompter to claim ownership over the resulting works; however, the Zarya of the Dawn case now almost entirely closes that door, by holding that feeding an AI thousands of descriptive prompts, adjusting them in ways to “tweak or change” the outcomes, and whittling down the results several dozen final images, does not meet that threshold. While Kashtanova can still claim protection in the selection and arrangement of these images into their comic book, the images themselves remain unprotected, which leaves them in a very vulnerable position.
So where does this leave publishers that wish to utilize AI technology in their creation of games, comic books, or other media? It means that they must be extremely careful.
If the publisher wants to protect their work, the safest approach is to use AI technology only as a source of inspiration, reference, or ideation for human artists who then create the actual art assets to be used from scratch. Relying on human authors to fix a work in a tangible medium of expression is still the most reliable way to ensure the work will be copyrightable (and that said copyright can be validly transferred to the publisher).
But beyond this, publishers need to be extremely careful to ensure that they themselves do not feed their own IP into AI generation services. As explained here, the view of the Copyright Office is that the output of AI generation services is unprotectable, which means anyone can freely copy the images as if they were in the public domain. Companies must therefore be cautious about inadvertently diluting the value of their own intellectual property by using it to create hundreds of freely copyable AI-generated images. Strong internal guidelines and policies around AI are essential to mitigate these risks – as will proactive legal counsel.