What Is AI Image Copyright? What Creators Need to Know in 2026

What Is AI Image Copyright? What Creators Need to Know in 2026

On 2 March 2026, the US Supreme Court declined to hear Thaler v. Perlmutter — the case that could have changed everything about who owns AI-generated art. By refusing to take it up, the Court left one rule firmly in place: under US law, a purely AI-generated image cannot be copyrighted.

That single ruling has consequences for every founder using Midjourney for their logo, every indie author using DALL-E for their book cover, and every creator building a brand on AI-generated visuals. But the story is more nuanced than the headlines suggest — and understanding the nuance is what protects you.

This guide explains the current legal position clearly, covers what you can protect, what you cannot, and what practical steps to take if AI-generated images are part of your brand or business.

Can You Copyright an AI-Generated Image?

The short answer: No — if the image was created purely by entering a text prompt.

Under US copyright law, a work must have a human author to qualify for protection. The US Copyright Office (USCO) has maintained this position since at least 2023, and the Supreme Court's refusal to revisit it in March 2026 means it is now settled law — at least for the foreseeable future.

This means:

• A Midjourney image generated from your text prompt is not copyrightable
• A DALL-E image created entirely from a description is not copyrightable
• A Stable Diffusion output where you only typed what you wanted is not copyrightable

These images effectively fall into the public domain. Anyone — including a direct competitor — can legally take the exact same output and use it without consequence, because neither you nor anyone else holds a copyright over it.

This is different from what Midjourney's Terms of Service say. Their ToS grants you commercial usage rights. But usage rights and copyright protection are not the same thing. The platform can permit you to use an image commercially. It cannot manufacture a copyright that the law refuses to recognise.

What Did the January 2025 Copyright Office Report Actually Say?

In January 2025, the USCO published Part 2 of its Copyright and Artificial Intelligence report — the most detailed official guidance to date on how copyright applies to AI-assisted creative works.

The key shift was this: the Office moved away from asking whether AI was involved, and toward asking how much human control was exercised over the expressive elements of the output.

This matters because it opens a door. The USCO has already registered hundreds of works that incorporate AI-generated material — but only the human-authored portions are protected. The AI-generated sections are explicitly disclaimed and excluded.

The practical takeaway from the January 2025 report is that copyright protection follows human decision-making over the final appearance of the work — not the effort invested in prompting, and not the sophistication of your instructions.

When Can AI-Assisted Art Be Protected?

This is the part most articles get wrong. "No copyright for AI images" is not the whole picture. Copyright can apply to elements of your work if human creative control is clearly present and clearly separable from what the AI generated.

Workflows that can produce protectable copyright:

Starting from your own artwork. If you draw a sketch, scan it, and use an AI tool to render it — your original sketch elements can be copyrighted. The AI output layer may not be, but the underlying human art is yours.

Heavy overpainting or compositing. If you take a raw AI image and manually repaint significant portions — new geometry, hand-drawn textures, altered characters — the human-authored additions can be protected. The AI-generated base typically cannot.

Selection and arrangement. If you curate, sequence, or arrange multiple AI images into a comic, lookbook, or editorial layout alongside original text and design, the arrangement as a whole — the editorial choices you made — can be copyrightable. This is called compilation copyright. The individual AI images within it are still not protected, but the human creative structure around them can be.

Inpainting with original elements. Using tools like Midjourney's Vary Region to insert your own hand-drawn elements into specific areas of a generation may qualify the replaced portions as human-authored.

The threshold: "degree of human control"

The USCO's 2025 standard is that copyright protection requires the human to have exercised control over the expressive elements of the final work — not just the idea or the instruction. Prompting describes what you want; it doesn't determine how the AI renders it. That gap — between your instruction and the AI's unpredictable execution — is precisely why prompts alone don't meet the threshold.

The Dual Risk Most Creators Miss

There are actually two separate risks at play, and most articles only explain one of them.

Risk 1: You can't protect your AI image.
Because you have no copyright in a pure AI output, a competitor can legally copy it, sell it, or build their brand on it. You have no legal recourse.

Risk 2: Your AI image might infringe someone else's copyright.
AI models are trained on billions of images — many of them copyrighted. In rare cases, an AI can reproduce elements of training images closely enough to create infringement exposure. If your AI image happens to reproduce a distinctive element from a copyrighted work, you could be liable — even though you have no copyright of your own in the output.

This dual exposure is the least-understood aspect of AI image copyright. The image simultaneously has no legal protection for you and carries potential legal risk from others. Both are live concerns in 2026.

How Do Different AI Platforms Compare?

Not all AI image tools carry the same risk profile.

Adobe Firefly is trained exclusively on licensed Adobe Stock imagery, public domain works, and openly licensed content. This significantly reduces the risk that your output reproduces copyrighted training material. However, the copyright status of the output itself is identical to every other AI tool — without sufficient human input, it is not protected.

Midjourney (paid tier) grants you commercial usage rights via ToS and says you "own" the assets. As above, this is contractual permission to use the image — it does not create copyright protection under law.

Midjourney (free tier) operates under a Creative Commons Non-Commercial licence. You cannot sell or monetise free-tier outputs.

DALL-E / OpenAI grants users rights to generated images for commercial use, subject to usage policies. Same copyright position as above — no legal copyright in the output.

Stable Diffusion (open source) gives you maximum flexibility but zero indemnity. You are entirely responsible for any training data concerns.

For commercial use, Firefly has the cleanest training data story. But no platform can give you copyright protection that the law won't recognise.

What About Trademarking an AI-Generated Logo?

This is a critical pivot that most creators don't know to make. Copyright and trademark are entirely separate systems.

Even if your AI-generated logo cannot be copyrighted, you may still be able to trademark it — provided it is sufficiently distinctive and is used in commerce to identify your brand. The USPTO has granted trademark registrations for AI-assisted logo designs when the mark is distinctive enough.

Trademark protection is arguably more useful for a logo than copyright anyway: it prevents competitors from using similar marks in related industries, regardless of how the original was made. If your business depends on its visual identity, trademark registration is the protection you should be pursuing — not copyright.

What to Avoid

Don't assume Midjourney's ToS protects you legally. Platform terms grant usage permissions. They do not substitute for federal copyright law.

Don't claim copyright on pure AI outputs when registering. The USCO requires disclosure of AI-generated content in registration applications. Failing to disclose — or knowingly claiming copyright over purely AI-generated material — can result in registration being invalidated.

Don't assume that because your image is uncopyrighted, it's also legally risk-free. As explained above, AI outputs can still infringe existing works even without carrying copyright of their own.

Don't treat minimal edits as sufficient. Cropping, resizing, colour-grading, or upscaling a pure AI output does not create sufficient human authorship. The USCO looks at whether the human authored the expressive character of the image — not whether they touched it.

How to Register AI-Assisted Work Correctly

If your workflow does involve substantial human authorship — overpainting, compositing, original source art — and you want to register the result:

1. File with the US Copyright Office at copyright.gov
2. Disclose AI involvement in the application. There is a field for this. Be specific about which elements were AI-generated.
3. Limit your claim to the human-authored portions only. The USCO will register those portions and exclude the AI-generated content.
4. Document your process before you file — screenshots of your source sketches, layers, edit history. If your registration is ever challenged, your process documentation is your evidence.
5. Describe what you created in the application, not how the AI helped. Focus on your original contributions.

Next Steps

Here is what to take away and act on from this guide:

• Pure AI image outputs — text prompt only — are not copyrightable under current US law as confirmed by the Supreme Court's March 2026 position.
• Copyright can apply to human-authored elements in AI-assisted workflows — overpainting, original source art, editorial arrangements — but only those elements.
• Platform ToS is not copyright. Commercial usage rights and legal copyright protection are different things.
• Your AI output can still infringe others' copyright even though you hold no copyright in it yourself — these are independent risks.
• Trademark is the better protection for AI-generated logos. Pursue that route if your visual brand depends on it.
• If registering a hybrid work, disclose AI involvement, limit your claim to human-authored elements, and document your process before filing.

The law in this area is not finished evolving. Courts, Congress, and the Copyright Office are still building the framework. The guidance above reflects the position as of March 2026 — check back as new rulings emerge.

This article is for educational purposes only and does not constitute legal advice. For specific copyright or trademark questions, consult a qualified IP attorney.