On February 21, 2023, the Copyright Office eclipsed its prior decisions in the area of AI authorship when it partially cancelled Kristina Kashtanova’s registration for a comic book titled Zarya of the Dawn. In doing so, the Office found that the AI program Kashtanova used—Midjourney—was primarily responsible for the visual output that the Office chose to exclude from Kashtanova’s registration. (Midjourney is an AI program that creates images from textual descriptions, much like OpenAI’s DALL-E.) The decision not only highlights tension between the human authorship requirements of copyright law and the means of expression that authors can use, but it also raises the question: Can AI-generated works ever be protected under U.S. copyright law?
On September 15, 2022, the U.S. Copyright Office issued a registration for Kashtanova’s comic book, Zarya of the Dawn, which consists of text written by Kashtanova and images created using Midjourney. Thereafter, the Office learned of Kashtanova’s public statements that she used an AI program to create Zarya of the Dawn, initiated cancellation of the registration, and provided Kashtanova with an opportunity to show cause why the registration should not be canceled. In response, Kashtanova argued that she used Midjourney the way a photographer uses a camera, or the way a graphic designer uses Adobe Photoshop, and that her “core creative input”—i.e., text “prompts” and “previously developed” images—as well as her “iterative process” of selecting Midjourney images led to her creation of Zarya of the Dawn.
While the Office found that Kashtanova’s selection and arrangement of text and images within the Zarya of the Dawn comic book was protectable, it found that Kashtanova did not author the actual images within the comic book because Midjourney generates images in an “unpredictable way,” such that Kashtanova did not “actually form” those images. The Office based its conclusion on “the significant distance between what a user may direct Midjourney to create and the visual material Midjourney actually produces” finding that “Midjourney users lack sufficient control over generated images to be treated as the ‘master mind’ behind them.” The Office also rejected Kashtanova’s argument that she authored the images through her “creative, human-authored prompts” because Midjourney prompts are mere “suggestions” not “orders.”
While consistent with the human authorship requirement reflected in previous Office and judicial decisions, this decision raises several important questions.
First, if copyright authorship is defeated by mere use of “unpredictable” tools, how does that impact experimentation in art? To create artworks—many of which are generally understood to be at the “core” of copyright protection—artists routinely take risks and experiment with different tools, mediums and modes of creation, all of which is, by definition, unpredictable. How many of the most important works of art would be denied protection if they were evaluated based on whether the means of expression used were predictable? If we follow the Office’s reasoning to its logical conclusion, would it lead to untenable results?
Second, although the Office reached the same conclusion as to all Midjourney images, reasoning that Midjourney users lack creative control, can users exercise a greater degree of control over Midjourney images? For example, Kashtanova generated some images with text prompts, and others with “previously developed” images. Did Kashtanova exert more creative control when she provided visual content for Midjourney to mimic? And could some users’ text prompts be detailed and nuanced enough to cross the line from “suggestion” to “order”?
Third, how much creative control is enough to support authorship under U.S. copyright law, and to what extent does the requisite creative control vary with the means of creative expression used? Whereas the Office is accustomed to finding photographs protectable, and thus unlikely to question whether a photograph was taken with creative control over lighting, angle, timing, etc., its decision on Zarya of the Dawn suggests that AI-generated works will receive heightened scrutiny. Such scrutiny may be unwarranted; for example, one could argue that Kashtanova exercised the same degree of control over Midjourney that a film director exercises on set, where there can often be improvisation by the actors and actresses. Assuming that Kashtanova made significant contributions to the images in Zarya of the Dawn, is she any less deserving of joint authorship than a film director? See Garcia v. Google, Inc., 766 F.3d 929, 933 (9th Cir. 2014), on reh’g en banc, 786 F.3d 733 (9th Cir. 2015) (discussing joint authorship in the context of a film.)
A New Dawn for Copyright in AI-Generated Works?
While the Office’s decision on Zarya of the Dawn may not bode well for copyright protection over AI-generated works, it may also be an outlier. Kashtanova’s statement of sole authorship was inconsistent with her public statements about Midjourney, and her use of “Zendaya” in text prompts suggests that Zarya was more celebrity copy than original creation. Further still, the Office limited its decision to Midjourney, noting that “[i]t is possible that other AI offerings that can generate expressive material operate differently than Midjourney does.”
Clearly, then, the broader question of copyrightability of AI-generated content remains open. Creators, developers, and content platforms should pay careful attention as these issues continue to percolate through the Office and the federal courts, and the copyright issues around AI-generated works further develop and are hopefully clarified.