Theatre D'opera Spatial: 624 Prompts Weren't Enough

Jason Allen did the work. Nobody disputes that. To make Theatre D'opera Spatial he ran at least 624 prompt revisions through Midjourney, then took the result into Photoshop, then upscaled it with Gigapixel AI. The finished piece won first place in the digital art category at the 2022 Colorado State Fair. Then he tried to register it, and the Copyright Office said no. He appealed. On September 5, 2023, the Review Board affirmed the refusal.

If effort decided copyright, Allen would have won easily. It doesn't, and that's the point of his case.

What the Board actually said about prompts

Allen's strongest argument was the 624 number. Surely that volume of iteration, that much steering, made him the author. The Board disagreed, and its reasoning is worth quoting because it's the cleanest statement we have.

Prompts, the Board found, "influence" what Midjourney produces but don't dictate it. The artist proposes; the machine interprets and executes. Allen entered text, Midjourney decided how to render it, and the distance between his instruction and the image on the screen was filled by the tool, not by him. Running the loop 624 times doesn't close that gap. It just repeats it 624 times.

This is the same logic that narrowed Zarya of the Dawn six months earlier. A prompt is a request. The expressive output is the response. Copyright attaches to the person who authored the expression, and when a machine generates the expression in response to a request, the requester isn't the author of what came back.

The disclaimer he wouldn't sign

Here's the part that decided the case, and it's procedural, not philosophical.

The Office told Allen the Midjourney-generated material in his image was "more than de minimis," meaning more than a trivial amount, and so it had to be disclaimed. Disclaiming doesn't mean throwing the work away. It means telling the Office which parts are AI-generated so the registration covers only the human-authored remainder. Allen could have done that. His Photoshop edits and other human contributions might well have supported a thin registration over what he personally added.

He refused. Allen wanted the whole image registered as his, AI material included, or nothing. So the Office gave him nothing, because it can't register material it has been told to treat as authored by a machine. The refusal wasn't the Office declaring his entire image worthless. It was the Office unable to register a claim that included unprotectable matter the applicant wouldn't carve out.

That distinction trips people up. Allen's case is read as "AI art can't be copyrighted." What it actually holds is narrower and more practical. You must disclaim the AI-generated portions, and if you won't, your application fails on that ground alone.

Volume is not the test

The deeper teaching is about what copyright measures. It does not measure labor. It does not measure iteration count. It does not reward the artist who tried hardest. The 1976 Copyright Act protects original works of authorship fixed in a tangible medium, and authorship means a human creative contribution to the expression itself.

Allen kept pointing at his process: 624 revisions, hours of Photoshop, careful upscaling. The Board kept asking a different question. Not how much did you do, but what part of the final expressive image did you, the human, determine? For the core composition, the answer was that Midjourney determined it. The number 624 measures persistence. It doesn't convert machine-generated expression into human authorship.

Compare this to A Single Piece of American Cheese, which the Office registered in January 2025. That creator also used AI heavily. The difference wasn't effort. It was that the human selection, arrangement, and editing were identifiable and disclaimed cleanly, so the Office had a protectable human claim to register. Allen had a plausible human contribution too. He just wouldn't separate it from the rest.

What to take from a refusal

Two things, and they're both actionable.

First, your AI-generated material needs to be disclaimed, and you need to know which parts those are. You can't disclaim what you didn't track. If you don't know where the machine's output ends and your editing begins, you can't file a clean application, and a muddy one invites refusal.

Second, the human contribution that survives, your selection, your arrangement, your specific modifications, is the only thing carrying your claim. Make it real and make it provable. Allen's mistake wasn't using Midjourney. It was treating the prompt count as proof of authorship and then refusing to draw the line the Office needed.

Copyrightable draws that line as you work. It records which elements you selected, how you arranged them, and exactly what you modified, separating your human contribution from the generated material in real time. When it's time to file, you already have the disclaimer and the proof of authorship, not a memory of 624 prompts and an argument the Board has already rejected. Read the Review Board's full decision, then capture your next project the way the Office actually asks you to.