There is a comfortable assumption among people who generate locally: my GPU, my electricity, my model file, my image. Ownership of the hardware feels like it should settle ownership of the output. It does not. Copyright does not care whose machine did the work. It asks who did the authoring, and in the United States that question now has fairly clear answers.
This is not legal advice, and the law differs by country. It is a map of where the lines currently sit, with links to the documents that drew them.
Can You Copyright an AI-Generated Image?
Not the purely generated parts. In January 2025 the US Copyright Office published Copyright and Artificial Intelligence, Part 2: Copyrightability. Its conclusions are short and direct:
- "Copyright does not extend to purely AI-generated material, or material where there is insufficient human control over the expressive elements."
- "Based on the functioning of current generally available technology, prompts do not alone provide sufficient control."
- "Whether human contributions to AI-generated outputs are sufficient to constitute authorship must be analyzed on a case-by-case basis."
The courts agree. In Thaler v. Perlmutter, decided on 18 March 2025, the D.C. Circuit held that "the Copyright Act requires all work to be authored in the first instance by a human being." The Supreme Court denied review on 2 March 2026, so that ruling stands.
What Is Protected, Then?
A great deal, as long as a human made it. The same Copyright Office report says "the use of AI tools to assist rather than stand in for human creativity does not affect the availability of copyright protection for the output." It names three kinds of protectable human contribution:
| Human contribution | Example | Protectable? |
|---|---|---|
| Your own expressive work visible in the output | Your drawing, photo or sketch used as the starting image | Yes, your part |
| Creative selection, coordination or arrangement | Choosing, ordering and composing generated images into a comic or layout | Yes, the arrangement |
| Creative modifications of the output | Substantial repainting, compositing or editing by hand | Yes, the modifications |
| A prompt, however detailed | A 200-word description of the scene | No, not on its own |
The court was careful about this too. The human authorship rule "does not prohibit copyrighting work that was made by or with the assistance of artificial intelligence"; it only requires that the author be "the person who created, operated, or used artificial intelligence", not the machine itself.
The Case That Shows the Line
The clearest worked example is the Copyright Office's Zarya of the Dawn decision (February 2023), about a comic book illustrated with Midjourney. The Office found the author owned "the Work's text as well as the selection, coordination, and arrangement of the Work's written and visual elements." The individual Midjourney images were "not the product of human authorship" and were not protected.
So the comic was protected and the pictures inside it were not. That is the shape of the rule: the human layer is yours, the generated layer underneath it is not.
If You Register Work, You Must Say What the AI Made
The Office's registration guidance of March 2023 says applicants "have a duty to disclose the inclusion of AI-generated content in a work submitted for registration", and that AI-generated content "that is more than de minimis should be explicitly excluded from the application." Leaving it out is not a shortcut. It is a defect in the registration.
Other Countries Draw It Differently
- United Kingdom. Section 9(3) of the Copyright, Designs and Patents Act gives computer-generated works an author: "the person by whom the arrangements necessary for the creation of the work are undertaken." The government's March 2026 report on copyright and AI proposes removing that provision. It is a proposal, not yet law.
- China. In Li v Liu (November 2023), the Beijing Internet Court protected an image made with Stable Diffusion. As the Kluwer Copyright Blog summarises, the plaintiff "selected over 150 prompts, arranged their order and set specific parameters", adjusting them "until the final image aligned with his conception."
The direction differs, but every one of these systems rewards the same thing: evidence of the choices a person made.
What Running Locally Does and Does Not Change
Running a model locally does not make the output more yours under copyright. The Copyright Office test is about human control over expression, not where the computation happened.
What it does change is the contract layer. A hosted service attaches its terms of service to every image, and those terms vary in who may use the output and on what conditions. A local model has no service terms, only the model licence. That licence still matters: the FLUX.1 [dev] licence, for example, restricts commercial use of the model while saying you "may use Output for any purpose (including for commercial purposes), except as expressly prohibited herein."
How to Build a Record of Your Authorship
- Start from your own material where you can: a sketch, a photo, a composition guide.
- Edit meaningfully. Inpaint, composite and repaint with intent, not only re-roll.
- Arrange. Selection and arrangement of many outputs into a finished work is protectable.
- Keep the history. Prompts, seeds, settings and edits are the evidence of your choices if anyone ever asks.
Vision Studio keeps that history for you. The iteration tree turns every render into a node; forking one records the prompt, model, sampler, steps and seed, and each child stores a settings diff against its parent. The tree survives restarts, and the asset library is searchable by prompt and model. Canvas layers hold your masks, ControlNet guides and reference images, so the parts you contributed are separate from the parts the model generated.
None of that makes a generated image copyrightable by itself. It makes your human contribution visible, which is what every one of these legal tests asks for.
The Short Version
Your GPU makes images. You make authorship. Prompting alone does not count; your own input, your arrangement and your real edits do. Keep the record, disclose the AI parts when you register, and read the model licence before you sell.
Vision Studio is free, open-source (MIT), and runs locally on your GPU. The iteration tree, canvas layers and asset library are documented with their source files at vision-studio-x.com/features. Download it at vision-studio-x.com/download.
-Rocky
#Copyright #AIArt #CreatorRights #GenerativeAI #LocalAI #OpenSource #VisionStudio