In February 2023, the U.S. Copyright Office told Kristina Kashtanova it would replace the registration for an eighteen-page comic book. The words in Zarya of the Dawn qualified for protection. So did the creative selection and arrangement of its words and pictures. The images generated with Midjourney would be excluded.
The book could stay exactly as it was. Its registration needed to describe something narrower. For anyone delivering a campaign, a presentation or a piece of branded artwork, that is an uncomfortable possibility: the finished file can look seamless while the copyright claim runs through it in pieces.
A headline, a photograph and a generated background may arrive together in one export. Each has its own history. When a client asks what it owns, someone needs to explain how they came together—and which contributions support a copyright claim.
The artist’s argument
Kashtanova’s lawyer, Van Lindberg, argued that Midjourney had served as an assistive tool and that Kashtanova had authored the work. He also offered an alternative: the writing and the creative arrangement of the book deserved protection even if the Office rejected authorship of the generated images. The Office accepted that narrower account.
The disagreement captures a familiar experience of making things with AI. A person can spend hours directing, rejecting and refining images, with a clear idea of what the finished work should become. From the maker’s side of the desk, the process feels full of decisions. Copyright asks which of those decisions produced protectable expression.
The Office’s 2025 report on AI and copyrightability explains its position. With generally available technology, prompts alone give users insufficient control over the expressive elements of an output. Human-written text, expressive inputs, sufficiently creative arrangements and modifications can support protection, depending on the work.
A prompt history can explain the search. The finished work must show what the person contributed.
Where the courts draw the baseline
Stephen Thaler sought registration for an artwork whose application named an AI system as its sole author. The D.C. Circuit upheld the refusal in March 2025, affirming that copyright requires human authorship. The Supreme Court declined review a year later, leaving that judgment in place.
The decision gives creative teams a firm starting point: a machine cannot be the author for copyright purposes. How much protection a person’s contribution earns in an AI-assisted production requires a closer look at the work. That broader question remained outside the court’s holding; the Supreme Court’s refusal to hear the appeal added no new ruling on it.
What the cheese picture shows
The title A Single Piece of American Cheese sounds like an unlikely entry in a creative director’s reference library. Its registration history belongs there.
Kent Keirsey, the chief executive of Invoke AI, created the work using generated material. After initially refusing registration, the Copyright Office reconsidered. It reviewed the artwork, the correspondence and a creation video running just under ten minutes.
Invoke’s lawyer described Keirsey’s process as selecting, coordinating and arranging numerous generated image fragments into a unified image—a kind of collage. The Office agreed that the selection and arrangement contained enough human authorship to support registration. The generated fragments remained outside the claim.
Here was something a reviewer could examine: separate elements, their relationship to one another, and a record showing how the composition came together. The registration covered the human-authored arrangement of those elements.
For studios, that is more useful than a headline announcing that AI art has won copyright. The decision shows what the applicant claimed, what the Office examined and where protection stopped. It also gives a production team a concrete question to ask of its own work: can we show the contribution we intend to claim?
When choosing the inputs falls short
Ankit Sahni supplied his own photograph to an AI application called RAGHAV, chose Van Gogh’s The Starry Night as a style reference and selected a value for the strength of the transformation. He argued that those choices made him the author of the resulting image, SURYAST.
The Copyright Office’s Review Board refused registration in December 2023. It found that RAGHAV determined the resulting pictorial expression and that Sahni’s choices supplied insufficient creative control. His original photograph remained a separate work, eligible for registration if it met the statutory requirements.
Read alongside American Cheese, the refusal makes the production question sharper. Keirsey’s claim succeeded on his creative selection and arrangement of generated fragments. Sahni’s choice of inputs and a transformation setting fell short. A studio needs to describe what its particular workflow allowed a person to shape—and show that contribution in the finished work.
What should travel with the file?
Consider a hypothetical campaign image: a generated landscape behind a product photograph, with a human-written headline and a foreground the designer has painted by hand. At delivery, the team flattens the composition into a JPEG.
The export is convenient. It conceals much of what someone would need to understand the authorship claim.
Studios make this harder than it needs to be when they leave authorship questions until delivery. By then, the flattened export has hidden the contributions they need to explain, and the people who made them may already be working on something else.
A useful handoff would preserve:
• The writing. Keep the human-authored copy and identify its author. Where AI helped draft it, retain meaningful versions showing what the writer contributed.
• The supplied assets. Preserve photographs, sketches and other expressive inputs, with their creators and relevant rights information.
• The working composition. Keep layers and intermediate versions that show the arrangement, painting, compositing or other changes being claimed.
• A short explanation. Describe the human contribution and identify the generated material separately.
For the hypothetical campaign, the internal handoff note could read:
Human contributions: Headline by Maya Chen; product photograph by Luis Ortega; layout and hand-painted foreground by Priya Shah. The landscape background was generated. The proposed copyright claim covers the human-authored contributions, with the generated background excluded.
The next person can understand what happened. A producer can find the original photograph. A designer can open the layers. Someone preparing a registration has a starting point for describing the work accurately.
Prompt logs belong alongside those materials when they help explain the process. They are one part of the record. The writing, inputs and working files show the contribution that a transcript of requests and rejections may leave unclear.
What the client is buying
The scope of that contribution matters when a studio talks about exclusivity. A client may receive an original composition containing generated material, with protection attached to the human-authored arrangement and other qualifying contributions. The individual generated elements can sit outside that protection.
The Copyright Office’s registration guidance requires applicants to disclose and exclude AI-generated content that is more than minimal, while describing the human authorship being claimed. A clear production record makes that explanation easier. Its value comes from the work it documents; neither an elaborate folder structure nor a particular number of edits guarantees protection.
There is also clearance to finish. Third-party photographs, trademarks, recognizable people and contractual commitments each raise questions of their own. Establishing authorship addresses one part of the delivery conversation.
The practical improvement is modest: give the handoff enough room to explain the asset. Send the working file, preserve the meaningful contributions and describe the scope before promising exclusivity.
Kashtanova’s comic survived the Office’s review with protection for its writing and arrangement. Its generated images were excluded. Both answers lived inside the same eighteen pages.
A creative team’s next delivery should be just as clear about what it made.
This feature draws on public documents. No interviews were conducted.