What’s the difference between artificial intelligence (AI) and a monkey (or, more accurately, a crested macaque named Naruto)? Well, while the U.S. Copyright Office has made clear that nonhuman-authored works—like Naruto’s famous “monkey selfie”—are ineligible for copyright registration, according to a new letter from the Copyright Office, a graphic comic book that contains nonhuman, AI-generated images nevertheless may receive certain copyright protections.1 But the devil is in the details, and creators using AI and other stakeholders would do well to understand the Copyright Office’s nuanced, mixed decision.
As we recently reported, generative AI raises a number of novel legal issues, including whether and to what extent works created by or with the assistance of AI are copyrightable. Following its refusal last year to register the AI-authored work “A Recent Entry to Paradise,” the Copyright Office revisited its decision to register the comic book “Zarya of the Dawn” (the Work or “Zarya”) after it identified post-registration statements made on social media by claimant Kristina Kashtanova that parts of the Work had been generated using Midjourney’s eponymous AI program which creates images from text-based prompts.2 In her original copyright registration application—which the Copyright Office reviewed and registered on the same day it was submitted, September 15, 2022—Ms. Kashtanova had identified herself as the author of the Work and did not disclaim any part of the Work or indicate that any portions of it had been created by AI.
After viewing online press accounts detailing Ms. Kashtanova’s use of Midjourney to create the Work, on October 28, 2022, the Copyright Office notified her that it intended to cancel the registration for “Zarya” unless she provided sufficient additional information to support maintaining the registration—i.e., illustrating how the Work was the product of human authorship. Ms. Kashtanova’s counsel responded, principally arguing that Midjourney was merely a tool that assisted her in creating the images contained in the Work or, alternatively, that the Work is registrable as a compilation because she selected, coordinated and arranged the text and images of the Work.
The Copyright Office Cancels ‘Zarya’ and Reissues a More Limited Registration
In its February 21, 2023 letter, the Copyright Office determined that it would cancel and reissue the Work’s registration on a more limited basis, covering the original authorship that Ms. Kashtanova contributed to the Work. It stated:
We conclude that Ms. Kashtanova is the author of the Work’s text as well as the selection, coordination, and arrangement of the Work’s written and visual elements. That authorship is protected by copyright. However, … the images in the Work that were generated by the Midjourney technology are not the product of human authorship.3
In light of the Copyright Office’s determination, Ms. Kashtanova’s reissued registered copyright in the Work extends to the text she wrote and to the selection, coordination and arrangement of the text and AI-created images as a compilation, but not to the AI-generated images themselves.
The Copyright Office explained its decision by noting that Ms. Kashtanova had sufficiently clarified that “the text of the Work was written entirely by Ms. Kashtanova without the help of any other source or tool, including any generative AI program” and that she “selected, refined, cropped, positioned, framed, and arranged” the images in the Work to create the story therein.4 However, as explained below, for the AI-created images, the Copyright Office held that they were not the product of human authorship and could not be registered.
What Makes Midjourney-Generated Images Noncopyrightable Here?
It is unequivocal that only original works of authorship are protected by copyright. The Copyright Office will not register works produced by a machine or mere mechanical process that operates randomly or automatically without any creative input or intervention from a human author. As the Supreme Court has explained, the “author” of a copyrighted work is the one “who has actually formed the picture,” the one who acts as “the inventive or master mind.”5 In evaluating whether the AI-generated images in the Work were copyrightable authorship, the Copyright Office compared the process of creating images using Midjourney to other tools traditionally controlled and guided by human authors.
Notably, the Copyright Office detailed its understanding of how Midjourney works, which may well serve as a precedent for other applicants and litigants when explaining to regulatory bodies or courts text-to-image generative AI and latent diffusion processes. Specifically, the Copyright Office explained that, while the user may input text into Midjourney that the program then reads to generate images, the user’s text functions more as suggestions than instructions because Midjourney does not understand words or grammar in the way humans do; instead, it starts with a “field of visual noise, like television static” and then converts the words into “tokens” to relate to its training data in order to refine that static into human-recognizable images.6 As such, via text-based inputs, the user influences how the image is generated, but it does not dictate a specific result, limiting Ms. Kashtanova’s control and allowing Midjourney to generate the images “in an unpredictable way.”7
Ultimately, according to the Copyright Office, “[b]ecause of the significant distance between what a user may direct Midjourney to create and the visual material Midjourney actually produces, Midjourney users lack sufficient control over generated images to be treated as the ‘master mind’ behind them.”8 Therefore, the Copyright Office concluded 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, writer, or photographer”9 and found the Midjourney-created images in the Work could not receive copyright protection.10
The Copyright Office also addressed whether Ms. Kashtanova’s personal edits to the Midjourney-generated images would qualify them for copyright protection. Ms. Kashtanova explained that she modified the lips and mouth of one image of Zarya, the comic book’s protagonist. The Copyright Office concluded that these edited images could not be protected because the edits were too minor and imperceptible to achieve the modicum of creativity for copyright protection. In general, therefore, it appears that whether an artist’s sufficient revisions to an AI-generated image could qualify for copyright protection is a factual determination that remains open for debate.
The Office Holds Open the Door to Potentially Register Other AI-Assisted Works
While the Copyright Office determined that Ms. Kashtanova did not have or wield sufficient control over the Midjourney images to qualify them for copyright protection, the Copyright Office kept its determination narrow “based on the specific facts provided about Ms. Kashtanova’s use of Midjourney” and left open the possibility that other AI-generated works might qualify for copyright protection in the future, stating that “[i]t is possible that other AI offerings that can generate expressive material operate differently than Midjourney does.”11
While other AI-based programs would need to be evaluated on a case-by-case basis, at a minimum it seems that users must both exert a high level of control and be able to predict the specific output in order for AI-assisted works to be protectable by copyright. The Copyright Office distinguished Midjourney from other programs used to generate expressive works (e.g., Adobe Photoshop), stating that “when artists use editing or other assistive tools, they select what visual material to modify, choose which tools to use and what changes to make, and take specific steps to control the final image such as it amounts to the artist’s ‘own original mental conception, to which [they] gave visible form.’”12 In this case, however, the Copyright Office found that although Ms. Kashtanova provided “hundreds or thousands of descriptive prompts” to Midjourney until the “hundreds of iterations [created] as perfect a rendition of her vision as possible,” the fact that Midjourney’s specific output cannot be predicted by users nevertheless rendered the final images unprotectable by copyright.13
Why It Matters, Plus Lessons for Creators and Other IP Stakeholders
The Copyright Office’s decision regarding “Zarya” indicates that purely AI-generated or AI-assisted expressive works are not protectable by copyright if the human user lacks sufficient control over their formation and the ability to readily predict the specific output of the program. However, the Copyright Office only looked at one generative AI program, so it is possible that other AI-based programs (or a different version of Midjourney) used to create expressive works may offer more control and predictability sufficient for generated works to qualify as original works of authorship and receive copyright protection. Beyond the specific facts in the Zarya decision, what the requisite level of human intervention is and where the Copyright Office draws the line in determining human authorship versus nonhuman authorship in the context of AI remain fact-dependent and almost certainly will continue to evolve alongside new technologies in this space.
A take-home lesson here is that creators should take caution when using AI to generate images and other expressive works. If an AI-generated and/or AI-assisted expressive work is not protectable per se, that likely would make it difficult for creators to stop others from making copies, creating derivative works and/or otherwise using the work in a manner normally protected by the copyright law’s bundle of rights. This also could make it more difficult for creators to exploit AI-based works via licenses and/or other channels and impact the value of the works. (It is worth noting that generative AI platforms may also have their own limitations and/or restrictions on ownership of works created via their programs.) Equally, purchasers or licensees of intellectual property based on AI-generated expressive works should do their diligence to ensure they understand the authorship of the works and the scope of rights they will obtain under the transaction. This is especially important because as the technology advances, the line between human-generated and AI-generated artwork and other works likely will become increasingly blurred. As always, caveat emptor.
Further, when applying to register works containing or incorporating AI-based expressive works, owners should clearly specify the authorship and the scope of their rights in the application. This may help prevent the resulting registration from being vulnerable to attack or challenge by third parties and/or canceled by the Copyright Office. Applicants should consider alternative strategies for protection such as a compilation or other collective work.
As AI evolves and becomes more widespread in generating and assisting with creative works, so too will the copyright implications. Creators and other stakeholders will do well to monitor these developments—we know the Copyright Office is watching.
1 Letter from Robert J. Kasunic, Copyright Office, to Van Lindberg, Taylor English Duma LLP, re: “Zarya of the Dawn” (Registration # VAu001480196) (Feb. 21, 2023) (“Feb. 21, 2023 Office Letter”), https://www.copyright.gov/docs/zarya-of-the-dawn.pdf.
2 Feb. 21, 2023 Office Letter.
3 Id. at 1.
4 Id. at 4-5.
5 Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53, 61 (1884).
6 Feb. 21, 2023 Office Letter at 7 (quoting Prompts, MIDJOURNEY, https://docs.midjourney.com/docs/prompts and Seeds, MIDJOURNEY, https://docs.midjourney.com/docs/seeds).
7 Id. at 9.
8 Id. at 9.
9 Id. at 8.
10 Id.
11 Id. at 10.
12 Id. at 9 (alteration in original) (quoting Burrow-Giles Lithographic Co., 111 U.S. at 60).
13 Id. at 8-9.