🏛️ Issues and Advocacy Training
Courses › AI Image, Video and Voice Generation

Lesson 8 of 9 · about 15 minutes

Copyright of AI Output and Tool Licensing

You'll be able to

  • Explain the human authorship requirement as the Copyright Office applied it in its January 2025 report
  • Identify which parts of an AI-assisted project a person can claim
  • Know what the main tools' terms say about ownership, commercial use, public visibility and attribution
  • Understand what records support a copyright registration or settle a licensing question

You can own the rights a tool's contract gives you and still have no copyright at all. Those are two different questions, and mixing them up is the most common mistake people make with AI output.

Question one: is there a copyright?

U.S. copyright protects original expression by a human author. The U.S. Copyright Office applied that rule to generative AI in Part 2 of its Copyright and Artificial Intelligence report, published January 2025. Its conclusions, in plain terms:

  • Prompts alone are not enough. With generally available technology, a prompt is an instruction that conveys ideas. The system decides how those ideas become an image, so the person typing is not the author of the output. Revising the prompt 400 times does not change this.
  • Your own input stays yours. If you feed in your own drawing or photo (an expressive input) and it is still perceptible in the result, you are the author of that portion.
  • Selection and arrangement can be protected. Choosing and ordering AI pieces creatively, as in a comic, a collage or a video, can be protected as a whole. The AI pieces inside remain unprotected.
  • Real editing counts. Substantial creative modifications by a human can qualify, decided case by case. Tools that let you control exactly where elements go may also get there.
  • No new law needed. The Office concluded existing law handles these questions and did not recommend a special new right for AI output.

The courts agree on the core point. In Thaler v. Perlmutter, Stephen Thaler tried to register an image called "A Recent Entrance to Paradise," listing his AI system as the author. The D.C. Circuit held in March 2025 that the Copyright Act requires a human author, and the Supreme Court declined to hear the case on March 2, 2026.

A purely prompted image, voice or song has no copyright owner in the U.S. Anyone can copy it. What you can protect is what you made: your words, your photos, your edits and your arrangement.

Sorting a real project

Worked example: the 30-second explainer

ElementMade byCopyright?
Narration script (75 words)YouYes, yours
Synthetic voice reading the scriptAIThe script's words are yours; the generated performance alone is not protected
Eight images from ImageStudioAI from your promptsNo
Two photos you shot at the town hallYouYes
Music bed from MusicStudio, instrumentalAINo
On-screen titles and chartYouYes, to the extent original (facts themselves never are)
Edit: shot order, timing, cutsYouYes, as selection and arrangement

Practical result: someone could lift one of your AI images and reuse it, and you would have no copyright claim to the image. They could not copy the whole video, your script or your photos.

How registration handles AI material

The Office's March 2023 registration guidance says applicants have a duty to disclose AI-generated content that is more than minimal. It points to the Standard Application. The "Author Created" field describes what the human made, and an AI tool is never listed as an author. Under "Limitation of the Claim," the "Other" box of "Material Excluded" describes the AI content followed by "generated by artificial intelligence." For the explainer: Author Created, "script, photographs, titles, and selection and arrangement of the video"; Material Excluded, "illustrations, voice performance and music generated by artificial intelligence." Hiding the AI parts can sink the registration. The Copyright Office FAQ covers how registration works in general.

Question two: what does the contract say?

Every tool has terms of service. They decide what a user may do with output, what the company may do with your prompts and output, and whether everyone can see what you made. They cannot create a copyright that the law does not recognize, but they can restrict users or license them.

ToolWho owns output (per terms)Commercial useWatch for
OpenAI (ChatGPT, API)OpenAI assigns you whatever rights it hasAllowedOutput may not be unique; you may not present AI output as human-made; OpenAI may use content to improve services unless you opt out
MidjourneyYou own assets "to the fullest extent possible under applicable law"Allowed; companies over $1 million yearly revenue need Pro or Mega to own assetsYour content is publicly viewable and remixable by default; Stealth mode needs Pro or Mega; Midjourney gets a perpetual license to your inputs and outputs
Adobe FireflyAdobe's FAQ says outputs from non-beta features can be used commerciallyAllowedAdobe's own models trained on licensed and public domain content; the FAQ says beta features are usable commercially unless the product says otherwise; partner models offered inside Firefly (such as FLUX, Gemini or GPT Image) are not covered by Adobe's training claim and carry their own terms
ElevenLabs (Voice Studio, Dialogue Editor)Per your planPaid plans only; free plan has no commercial licenseFree-plan publications must credit elevenlabs.io or 11.ai in the title
Mureka (MusicStudio)Set by Mureka's current termsDepends on planMusicStudio's page does not state licensing terms

All terms as of October 2026. They change, sometimes quietly. For a paid or high-stakes project, the terms in force on the day of generation are the ones that matter, and an old version can vanish from the site without notice.

Traps

Trap one: output that copies someone else's work. Owning your output under a contract does not protect you if the output infringes someone else's copyright or trademark. Ask for "a cartoon mouse with round black ears and red shorts" and you may get something that a large company's lawyers recognize instantly. Prompts naming famous characters, logos or living artists' styles invite exactly this, and anything in the output that seems familiar deserves a hard look.

Trap two: your inputs. Uploading a photo you do not own as a reference image does not launder it. Upload a newspaper's photo and ask for "the same scene as a watercolor," and the result is a derivative of their photo. Your own photos, licensed photos or public domain material avoid the problem.

Common mix-ups

  • Claim: The terms say I own it, so I have the copyright. What the evidence shows: A contract can assign only the rights that exist. For purely AI material, U.S. law recognizes none.
  • Claim: Using AI anywhere means the whole work is unprotected. What the evidence shows: Human parts remain protected, and so can your selection and arrangement.
  • Claim: The copyright questions are settled worldwide. What the evidence shows: This lesson covers U.S. law. Other countries take different positions, and lawsuits over AI training data are still working through U.S. courts.

What a project log for AI-assisted work records

  • Date, tool, model and plan for every generation kept
  • Exact prompts and any reference images (and where they came from)
  • What was changed by hand after generation
  • A saved copy of each tool's terms on that date
  • Which elements are human-made, for any registration or dispute
  • The disclosure text that went out with the work

Reporting angle

Ask a public body or campaign which tool and plan made a piece of art, and request the vendor terms it relied on. Ask whether anyone registered the work and whether AI material was disclosed, as the Copyright Office guidance requires. Be careful with claims that a company "owns" AI output: a contract can assign only rights that exist, and for purely AI material U.S. law recognizes none.

Research angle

Check yourself

1. According to the Copyright Office's January 2025 report, do prompts alone make you the author of an AI image?

2. You arrange eight AI images, your own photos and your own narration into a video. What can be protected?

3. What did the Supreme Court do in Thaler v. Perlmutter in March 2026?

4. Under Midjourney's terms as of October 2026, a company earning more than $1 million a year must do what to own its outputs?

5. OpenAI's terms assign output rights to the user. What else do they say?

6. When registering a work that contains AI material, what does the Copyright Office expect?

Terms in this lesson

Human authorship requirement
The rule that U.S. copyright protects only expression created by a human being.
Selection and arrangement
The creative choice and ordering of material, which can be protected even when the individual pieces are not.
Expressive input
Your own copyrightable work, such as a drawing, given to an AI tool. It stays yours where it remains perceptible in the output.
Terms of service
The contract between you and a tool provider that sets who owns output, what you may do with it and what the provider may do with your inputs.

Full glossary

Sources