Not-So-Starry Night for Copyright Applicant: Copyright Office Again Refuses To Register an AI-Generated Image
On December 11, 2023, the U.S. Copyright Office’s Review Board rejected Ankit Sahni’s request for reconsideration of the Copyright Office’s earlier refusal to register an image created using generative AI. The Office found that the image was created using generative AI software and lacked sufficient human authorship to support registration. This is at least the fourth documented rejection on these grounds by the Copyright Office, which continues to draw a line deeper and deeper into the sand as to the copyrightability of AI-generated material.
According to Sahni, he generated the artwork by inputting a photograph he had taken into the RAGHAV Artificial Intelligence Painting App. Next, as the “style” input to be applied to the photograph, he input a copy of Vincent Van Gogh’s “The Starry Night” into RAGHAV. Finally, he chose “a variable value determining the amount of style transfer.”
In his original December 2021 application to register the resulting work (called “Suryast”), Sahni named RAGHAV as co-author because, in his words, RAGHAV’s “contribution [was] distinct, disparate and independent” from Sanhi’s contribution to Suryast. On June 29, 2022, the Office refused to register Suryast because it “lack[ed] the human authorship necessary to support a copyright claim.” The Office further stated, in response to Sahni’s claims that his contributions provided requisite creative input, that “[Sahni’s] human authorship cannot be distinguished or separated from the final work produced by the computer program.”
On September 27, 2022, Sahni sought reconsideration, arguing that “the human authorship requirement does not and cannot mean a work must be created entirely by a human author.” The Office upheld its rejection, concluding that Suryast could not be registered because it was a derivative work of Sahni’s photograph “that did not contain enough original human authorship to support a registration.” In other words, because the new portions of Suryast had been created by RAGHAV, they were not “the result of human creativity or authorship.”
On July 10, 2023, Sahni sought reconsideration again on three grounds: (1) RAGHAV was merely an “assistive software tool” akin to a camera that was subject to Sahni’s creative decisions; (2) Sahni “provided the traditional elements of authorship for both the original photograph and [Suryast]” by taking the original photograph and “direct[ing] the RAGHAV tool to make changes to the colors, shapes, and style in a particular manner”; and (3) Sahni asserted that Suryast is not a derivative work because it is not “substantially similar” to the original photograph. Sahni further argued that his input resulted in Suryast containing elements such as a sunset, clouds, and the “contours of a building,” in the style of Van Gogh’s “The Starry Night.” Sahni asserted that his “total creative input in both the original photograph and [Suryast] should be considered together, and [Suryast] should be analyzed for all the traditional elements of authorship present therein.”
The Review Board was unconvinced and concluded that Suryast was not the product of human authorship, in particular because Sahni had not provided the expressive elements he identified. Sahni’s only contribution consisted of the aforementioned three inputs (base image, style image, and variable value for style transfer). However, as the Board wrote, “[t]he fact that [Suryast] contains a sunset, clouds, and a building are the result of using an AI tool that ‘generate[s] an image with the same ‘content’ as a base image, but with the ‘style’ of [a] chosen picture.” As such, RAGHAV (not Sahni) controlled “where those elements would be placed, whether they would appear in the output, and what colors would be applied to them.”
The Board also rejected Sahni’s characterization of RAGHAV as “an assistive tool” akin to a camera or photo editing software. This was in part because Sahni’s attempt to describe RAGHAV this way was at odds with his description in his original application, where he (correctly) said that RAGHAV operates by generating a new pictorial image based on features learned from user-provided images. The Board found that RAGHAV’s interpretation of Sahni’s photograph in the style of another painting was a function of how the model worked and the images on which it was trained—not of any choices made by Sahni. Moreover, while Sahni selected the numerical variable for the “strength” of the style, the Board found that this choice alone was insufficient to warrant copyright protection, and instead would be de minimis authorship not protected by copyright. See Feist Publ’ns, Inc. v. Rural Tel. Serv. Co., 499 U.S. 340, 359 (1991). The Board dispensed with the remainder of Sahni’s arguments (including a circular argument that the elements of Suryast present in the original photograph allowed for Suryast itself to be copyrighted) and affirmed the Copyright Office’s refusal to register Suryast.
With this decision, the Office and the Board continue to apply consistently the human authorship requirement. More notably, however, they also are continuing to reject arguments that human creativity is implicated in using generative AI in the same way that it is when using cameras, photo editing software, and other devices and tools. We expect that applicants will continue to lean on this latter point, and argue their own creative contributions through selection and arrangement of material, including by focusing on the degree of complexity and creativity in prompts that they enter in order to generate material using AI tools. Whether the Office will be receptive to these arguments is an open question, but to date it is hewing closely to established precedent.