Can You Copyright AI Art?

You spent an afternoon coaxing an image out of Midjourney. Forty prompts, a dozen rerolls, a careful crop, some color grading in Photoshop. It's beautiful, it's yours in every way that feels real, and you want to know one thing: can you stop someone from stealing it?

The honest answer is that part of it might be protectable and part of it almost certainly isn't, and which is which depends on what you actually did, not on how hard you worked or how good the result looks. That sounds frustrating. It is. But once you understand where the line sits, you can work on the right side of it.

Copyright protects human authorship, full stop

Start with the rule everything else hangs off of. In the United States, copyright protects works of human authorship. That phrase isn't decoration. It's the whole gate.

A machine can't be an author. The Copyright Office said so plainly in its 2025 guidance, and a federal court backed it in Thaler v. Perlmutter that same year, when Stephen Thaler tried to register an image authored entirely by his AI system with the AI named as author. The court said no. No human author, no copyright. The image fell straight into the public domain, free for anyone to use.

So the raw output of an image generator, the thing that appears when you hit generate, is not yours in the copyright sense. The model produced the specific expression: the brushstroke, the lighting, the exact arrangement of pixels. You didn't control that. You asked, it answered, and the answer is what the machine expressed, not what you authored.

"But my prompt was incredibly detailed"

This is where almost everyone pushes back, so let's deal with it head on.

Prompts alone are not enough. The Copyright Office has been blunt about this, and the length of the prompt doesn't change the answer. You can write a 500-word prompt specifying mood, palette, composition, lens, era, and a list of artists to channel. You still don't control the specific expressive choices the model makes in response. It decides where the shadow falls, how the fabric folds, which of ten thousand plausible versions of your description it actually renders. The gap between what you asked for and what came out is the gap where your authorship runs thin.

Think of it like commissioning a painter with very specific instructions. The detailed brief doesn't make you the author of the painting. The painter is. With AI, there's no human painter to hold the copyright, so the expressive part of the output just sits there, unowned.

And before you ask: no, there's no magic percentage. No "70% human" threshold, no number of edits that flips a switch. The Office decides case by case, looking at what a human actually controlled. Anyone selling you a tool that promises a human-contribution score is selling you a number that won't survive a real examiner.

Where your authorship actually lives

Here's the part that should cheer you up. The Office names three kinds of human control that can earn protection: selection, arrangement, and modification. Those are real, and they're things you probably already do.

The clearest illustration is Zarya of the Dawn, the comic Kris Kashtanova made with Midjourney images. The Office split the work. The individual AI-generated images? Not protected, the human didn't author them. But the way Kashtanova selected which images to use, arranged them into a sequence, and wrote the text? That was human authorship, and it got protection. Same work, two answers, one clean line drawn through it.

Apply that to your image. If you generated forty variations and chose one for reasons you could articulate, that's selection. If you composited three outputs into a single piece, masking and layering them into a composition the model never produced, that's arrangement. If you painted over the result, redrew the hands, changed the lighting by hand, shifted the whole mood in post, that's modification. Each of those is a place where you, the human, controlled expression.

The catch is the same one that bites everybody. Six months later, looking at a flat PNG, nobody can tell which is which. The file doesn't remember that you rerolled thirty times and hand-painted the background. It just looks like an image. Your creative control is invisible the moment you export.

When this stops being abstract

You won't think about any of this until money or theft enters the picture. Then it's all anyone thinks about.

Someone lifts your image for their product packaging. You want to send a cease and desist. Your lawyer's first question is whether you can claim copyright at all, and the second is whether you can prove the human authorship that supports it. "I made it in Midjourney" is the worst possible opening, because it invites the other side to argue the whole thing is unprotectable AI output. What you need is the record of what you did on top of the machine: the selection, the arrangement, the modification.

There's also the registration angle. If you do register a work that includes AI-generated material, the Office requires you to disclose that AI was used and to disclaim the parts you didn't author. You can't quietly register the whole thing as if a human drew every pixel. That's not a loophole to fear, it's a process to do correctly, and doing it correctly means knowing exactly which parts are yours.

Capture the process while you still have it

You can't change the law. You can change whether you have evidence when it counts.

The defensible position isn't "I worked really hard on this." It's a concrete trail: what the model produced, what you kept, what you discarded, what you reworked and why. The selection, the arrangement, the modification, recorded as they happened instead of reconstructed from memory in front of a skeptical examiner or an opposing attorney.

That's the gap Copyrightable is built to close. It captures your actual creative process and ties that trail to the finished Work as evidence, then scores it against a published Authorship Methodology, currently v0.1, that measures the kind of control you exercised rather than spitting out a meaningless percentage. It's evidence, not legal advice, and it doesn't pretend to make unprotectable output protectable. It just makes sure that when someone finally asks what you authored, you can answer with receipts.

So, can you copyright AI art? The machine's raw output, no. The expressive work you build on top of it through selection, arrangement, and modification, often yes. The work happens at the keyboard. The proof has to happen there too, while you still remember what you did.