Two years ago, the Colorado State Fair’s annual art competition sparked controversy after awarding a blue ribbon to a work generated in part by A.I. Jason Allen, the artist responsible for the work, said at the time, “I won, and I didn’t break any rules.” The judges agreed, despite outrage from some that the work lacked artistic merit. While the fairground judges saw fit to credit Allen as author of the work, it remains to be seen whether the federal judiciary will do the same. The U.S. Copyright Office refused to register Allen’s work for copyright based on lack of human authorship, and Allen has recently filed suit in Colorado federal district court to reverse the Copyright Office’s decision.
Allen’s case centers on the question of whether human authorship is present in A.I.-generated images where human input is limited to the textual prompts given to guide the A.I.’s work. Allen’s work, entitled “Théâtre D’opéra Spatial,” depicts Victorian performers looking out from a stage towards an otherworldly portal in a surreal synthesis of past and future. To create the work, Allen used Midjourney (an A.I.-driven text-to-image software) to generate an iterative series of images refined over the course of 624 prompting sessions undertaken by Allen. Allen then upscaled the resolution of the final image using Gigapixel AI and added additional elements using Adobe Photoshop.
The Copyright Office has thus far taken a dim view of A.I-generated images. Last February, it partially canceled a copyright registration for a graphic novel upon learning that the novel’s images were produced by Midjourney. Although the text and arrangement of the images in the graphic novel were human authored, the Copyright Office concluded that the individual images themselves were not, and thus they could not be encompassed by the scope of the resulting copyright registration. Further guidance issued by the Copyright Office last March has reinforced this position. Applicants are now instructed to “disclose the inclusion of AI-generated content in a work.” Further, under this guidance, “AI-generated content that is more than de minimis should be explicitly excluded from the application.”
The Copyright Office relied on its prior precedent in rejecting Allen’s application, which insisted upon copyright protection for his entire work—not just the elements he added manually to the base image via Photoshop. In Allen’s view, A.I. image generation is merely the latest technological innovation that artists can leverage to realize their creative visions. Just as human authorship exists in photographs captured by cameras and musical albums crafted in digital audio workstations, argues Allen, so too should human input be recognized as authorship in the generation of A.I.-assisted artworks using third-party tools.
Although the Copyright Office has not yet answered Allen’s complaint, its rejection letter provides a framework for its likely defense. In that letter, the Copyright Office quoted Midjourney’s website disclosure in asserting that “Midjourney does not interpret prompts as specific instructions to create a particular expressive result, because Midjourney does not understand grammar, sentence structure, or words like humans.” Relying on this understanding of Midjourney’s processing of prompts, the Copyright Office reasoned that, “when an AI technology receives solely a prompt from a human and produces complex written, visual, or musical works in response, the ‘traditional elements of authorship’ are determined and executed by the technology—not the human user.” Thus, the Copyright Office concluded, the traditional elements of authorship in Théâtre D’opéra Spatial were determined by Midjourney, not Allen.
While it remains to be seen whether the Copyright Office’s position will prevail in federal district court (Allen also posits in his complaint that the court should approach the question “with a fresh set of eyes” post-Loper Bright, the recent case where the U.S. Supreme Court overturned the judicial rule of interpretation that directs courts to defer to agency interpretations of statutes), other countries have already begun grappling with the issue. For example, last year, the Beijing Internet Court ruled that an A.I.-generated image was copyrightable, holding that a picture generated by using over 100 textual prompts met China’s copyright requirements that a work be original and demonstrate “intellectual achievement.” And in the UK, where “computer-generated works” are protected and authorship is granted to “the person by whom the arrangements necessary for the creation of the work are undertaken,” A.I.-generated images could presumably achieve copyright protection (although determination of ownership in such cases may be complicated).
Allen’s complaint raises important questions about the role of A.I. image generation and the human inputs and/or manipulation involved in the creative process. Who actually forms the image that results from the prompting process? Are there elements of authorship that A.I. solely controls, which other technologies—such as cameras—do not? And if there is a line between human- and A.I.-authored works, where should it lie? As software tools like Midjourney evolve and become more widely accessible, the practical need to delineate between the roles of humans and machines in the creative process will only increase. Yet as these tools likewise become more responsive to human input, this line-drawing may become even harder. We’ll report further as this case progresses.