Copyright Office rules AI-generated images in Zarya of the Dawn aren't copyrightable
The office kept protection for Kashtanova's text and her arrangement of images, but stripped it from each individual image Midjourney had generated.
- Courts & copyright
- Major
The US Copyright Office notified artist Kris Kashtanova that it was reissuing the copyright registration for her graphic novel Zarya of the Dawn with narrower scope than originally granted. Kashtanova had registered the work, which combines her own text with illustrations generated using Midjourney, without disclosing that the images were AI-generated; once the office learned this, it opened a review of whether the registration should stand at all.
The office’s letter concluded that the images themselves could not be copyrighted, because Kashtanova had not authored them in the sense the law requires — she supplied text prompts to Midjourney but could not control the specific output the system produced, and copyright law protects only works of human authorship. It rejected registration of the individual images while preserving protection for the elements Kashtanova had unambiguously authored: her written text, and her selection, coordination and arrangement of the images and text into the finished work as a compilation.
The decision did not ban the use of AI tools in creative work, and it stopped short of a categorical rule that no AI-assisted image could ever be copyrighted — the office left open that sufficient human creative control over an AI system’s output might change the analysis in a future case. But applied to Midjourney as it worked at the time, the letter set the office’s operating position: a prompt is an instruction, not an act of authorship, and the resulting image is not eligible for its own copyright.
The ruling became the reference point for the Copyright Office’s subsequent guidance on AI and authorship and for later litigation, including Thaler v. Perlmutter, over whether copyright law requires a human author at all. It left artists and publishers using generative image tools with a practical asymmetry: the text and curation around an AI image could be protected, but the image itself, however distinctive, generally could not.