AI for Life
Publishing AI images: who owns them, what you must label, what you can reuse
Most arguments about AI images mix up three questions that have different answers: do you own the picture, do you have to say it was made by AI, and were you allowed to feed in what you fed in. One is copyright law, one is the AI Act, one is the terms of the tool you used. Here is each one, with the official text.
1. Ownership: a prompt is not authorship
The U.S. Copyright Office published its report on the copyrightability of AI outputs in January 2025. Its conclusions are short and blunt. Copyright does not extend to purely AI-generated material, or to material where there is insufficient human control over the expressive elements. Whether a human contribution is enough has to be analysed case by case. And, based on how current widely available technology works, prompts alone do not provide sufficient control.
The Office's reasoning is worth keeping in mind, because it explains where the line falls: prompts "essentially function as instructions that convey unprotectible ideas", and even a very detailed prompt does not control how the system turns it into pixels. Ideas are not protected; expression is.
What the same report does protect is more encouraging than the headline suggests:
- your own work that is visible in the output, for example your drawing or your photograph used as an input and still perceptible in the result;
- your creative selection, coordination or arrangement of generated material, which is how a collage, a layout or an edited sequence becomes yours;
- your creative modifications of an output, meaning the work you do after the generation.
Practical consequence: if the image matters commercially, do not stop at the prompt. Your retouching, your composition and your selection are the part you can actually claim. Note that this is the position of a United States agency; rules differ between countries, and a contract can say something different from copyright law, which is the next point.
2. The tool's terms, which are not copyright
Copyright law says who is an author. The terms of the tool say what you are allowed to do with the file. The two can differ, so read the AI terms specifically, not only the general conditions.
Canva's AI Product Terms are a clear example of what to look for. You keep your rights to your input and you own your output, except for output that modifies or incorporates library content under licence: edit a Canva library photo with an AI feature and you do not own that result; generate a presentation that includes library photos and you own the presentation as a whole but not the individual photos. The same terms warn that outputs may not be unique and that other users may receive similar ones, and they state that you are solely responsible for evaluating whether an output is accurate and appropriate, with a recommendation to get independent professional advice before relying on it commercially.
Three questions to answer for any tool before you publish, in its AI terms and not in a forum thread:
- Does the licence cover commercial use, and does it change with the paid plan?
- Are there carve-outs, like Canva's licensed content, that stay outside what you own?
- What happens to your inputs: are they used to improve the service, and can you turn that off?
3. Labelling: what the AI Act actually requires
Article 50 of the AI Act splits the duty between the people who build the systems and the people who use them.
Providers of generative systems that produce synthetic audio, image, video or text must ensure the outputs are marked in a machine-readable format and detectable as artificially generated or manipulated, with solutions as effective, interoperable, robust and reliable as technically feasible. This duty does not apply where the system only performs an assistive function for standard editing or does not substantially alter the input data.
Deployers, meaning you when you publish, have a separate duty. Anyone deploying a system that generates or manipulates image, audio or video content constituting a deepfake must disclose that the content has been artificially generated or manipulated. For content that is part of a manifestly artistic, creative, satirical or fictional work, the obligation is limited to disclosing the existence of such content in an appropriate way that does not hamper the display or enjoyment of the work.
There is a second paragraph that matters to anyone running a blog or a newsletter: deployers of AI systems that generate or manipulate text published to inform the public on matters of public interest must disclose it, unless the content went through human review or editorial control and a person or organisation takes editorial responsibility for the publication.
The calendar of the AI Act has its own story, and we keep it in a separate guide on AI at work and the AI Act. The obligations above are the text of the regulation itself.
4. What you put in: faces, brands and other people's work
The input is where most real trouble starts, and it is the part no technical marking will fix. Canva's terms put it plainly: when you use the AI products you represent and warrant that you have all necessary rights, licences and permissions for your input. That sentence, in some form, is in every serious tool's terms.
So before uploading:
- A photo of a person is that person's likeness and personal data. Ask, and keep the proof you asked.
- A photo you did not take is someone's work. A screenshot from a search engine is not a licence.
- A logo or a product belongs to a company. Generating it into a scene can suggest an endorsement that does not exist.
- A child, in any form, deserves a no by default.
The checklist before you publish
- Input cleared. You have the rights to every image, face and document you fed in.
- Terms read. The AI terms of the tool, for your plan, allow the use you intend.
- Disclosure done. If it is a deepfake, say so. If it is a creative work, disclose in a way that fits the work.
- Marking kept. Do not strip metadata on purpose. If you need to know whether a given tool's marking survives your workflow, test it on one file and keep the result.
- Human part visible. If the image matters to your business, make sure your own contribution is real, not just a longer prompt.
We have not yet run our own hands-on tests on the image tools listed here, so this guide deliberately contains no price, no quality ranking and no claim about any specific tool's marking. It contains the rules, which change less often than the tools.
Tools in this guide
- Vendor information
Midjourney
Image generation with a strong visual style.
ImagesPrice: to verify
- Vendor information
Canva
Design tool with built-in AI features.
ImagesPrice: to verify
Questions
Do I own the image an AI generated from my prompt?
Under the U.S. Copyright Office's analysis, no: it concluded that copyright does not extend to purely AI-generated material, and that prompts alone do not provide sufficient human control to make you the author. What you contribute yourself, and your creative selection or modification of outputs, can still be protected.
Do I have to say an image was made by AI?
The AI Act puts a machine-readable marking obligation on providers of generative systems, and requires anyone deploying a deepfake, meaning AI-generated or manipulated image, audio or video, to disclose that the content is artificial. Platform rules and professional codes can require a visible label on top of that.
Can I use a photo of someone as the input?
Only if you have the rights. Canva's AI terms, for example, make you warrant that you have all necessary rights, licences and permissions for your input. A face belongs to a person, and image rights and data protection law apply to it.
Sources
- U.S. Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability, a Report of the Register of Copyrights
- Regulation (EU) 2024/1689 (AI Act), Article 50, transparency obligations, EUR-Lex
- Canva, AI Product Terms
One AI email a week. Only what changes something for you.