On September 5, 2023, the U.S. Copyright Office Review Board affirmed the Copyright Office’s refusal to register a digital artwork created with the Midjourney text-to-image GenAI tool. The Review Board held that text prompts alone did not constitute sufficient human authorship, despite the claim that the work was the result of 624 iterative prompts.
The claimant’s 600+ step process on Midjourney was described as follows:
Mr. Allen created a text prompt that began with a “big picture description” that “focuse[d] on the overall subject of the piece.” He then added a second “big picture description” to the prompt text “as a way of instructing the software that Mr. Allen is combining two ideas.” Next, he added “the overall image’s genre and category,” “certain professional artistic terms which direct the tone of the piece,” “how lifelike [Mr. Allen] wanted the piece to appear,” a description of “how colors [should be] used,” a description “to further define the composition,” “terms about what style/era the artwork should depict,” and “a writing technique that Mr. Allen has established from extensive testing” that would make the image “pop.” He then “append[ed the prompt] with various parameters which further instruct[ed] the software how to develop the image,” resulting in a final text prompt that was “executed . . . into Midjourney to complete the process” and resulted in the creation of the Midjourney Image above.
According to the Review Board, these steps were not sufficient “human authorship,” because the creative work was still being done by the Midjourney tool:
Mr. Allen’s actions as described do not make him the author of the Midjourney Image because his sole contribution to the Midjourney Image was inputting the text prompt that produced it. Although Mr. Allen describes “input[ing] numerous revisions and text prompts at least 624 times” before producing the Midjourney Image, the steps in that process were ultimately dependent on how the Midjourney system processed Mr. Allen’s prompts. According to Midjourney’s documentation, prompts “influence” what the system generates and are “interpret[ed]” by Midjourney and “compared to its training data.” As the Office has explained, “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.”
Accordingly, the Review Board agreed with the Copyright Office that the Midjourney image contained more than a de minimis amount of content created by GenAI (rather than a human), and thus, the image must be disclaimed from the copyright application. Because Allen refused to do so, registration was denied.
The Review Board left open the possibility, however, that Allen’s use of Adobe Photoshop to alter the image after its output by Midjourney could be sufficient human authorship to register copyright in the alterations. The Review Board confirmed previous Copyright Office guidance stating that “human-authored modifications of AI-generated material may [be] protected by copyright.” The Review Board clarified that this “decision does not foreclose Mr. Allen’s ability to file a new application for registration of the Work in which he disclaims the Work’s AI-generated material. In such a case, the Office could consider whether the human-authored elements of the Work can sustain a claim for copyright, an issue we have not decided here.”
Ultimately, this decision is helpful in several respects:
- It cements the principle that text prompts alone are unlikely to constitute human authorship for purposes of registering copyright in a GenAI output created using Midjourney and similar text-to-image technology.
- The decision confirms that human authorship may occur after use of a GenAI tool, in terms of alteration, selection, or arrangement of GenAI-created works. Allen may have been granted a limited registration if he disclaimed the Midjourney image and claimed authorship only of his post-output alterations.
- The Review Board acknowledged that text prompts themselves may be protectable. The Board reiterated Copyright Office guidance that “some prompts may be sufficiently creative to be protected by copyright” as literary works. So, even if a GenAI output is not subject to copyright registration for lack of human authorship, it may be possible to protect the text prompts that led to the creation of the output. (Of course, there may be other, non-substantially-similar prompts that produce similar-looking outputs.)
The current swarm of administrative challenges and court litigation addressing copyright issues with GenAI tools is just beginning to chip away at the many open questions. Stay tuned.
