Who owns an image or graphic created with AI? And what if someone else uses a similar output? In this article we'll use concrete cases to show why prompting alone usually isn't enough for copyright protection — and what that means for creatives and for client projects.
AI-generated content and authorship
A typical day at work for a designer today often looks like this: I need to quickly get an image for a project, and the first thought is obvious — I'll have it generated. So I open ChatGPT and, in fairly primitive language, write a prompt. Let's say that, for some reason, I need an image of a lobster next to its underwater house while figs fall around it.
Within moments I have a result I'm happy with. I drop the image into the project, publish everything, and that's it (for today). But after a while I'm surfing the internet and come across a blog where the author used the exact same image for an article about lobsters. A fair question: can I sue them for infringing my copyright?
AI vs. copyright
While looking for an answer to this question, I came across an interesting precedent from the United States in 2023. AI artist Kris Kashtanova created the graphic novel Zarya of the Dawn and tried to register it for copyright protection with the U.S. Copyright Office. (It's a supplementary tool for protecting authors in the American legal system — entering a work into the official register of protected copyrighted works.)
The Office initially did grant her the protection, but when it emerged that she had created the illustrations using the Midjourney tool, it retroactively revoked the copyright protection of the illustrations.
Kashtanova was ultimately recognized as the author of the comic's overall concept, the texts, the prompts for generating the images, and the arrangement of images and text — but not as the author of the images. The Office also rejected the argument that Kashtanova had spent hundreds of hours tuning the prompts and made hundreds of attempts to bring the generated images as close as possible to her authorial vision.
The reason for this decision is above all that current legislation is classically not prepared for the arrival of technologies like AI. When copyright laws were being formed, no one anticipated that neural networks would be able to "create" at a human level.
For an output to be protected by copyright, it must fulfill the conceptual features of an authorial work:
**1. a unique result
- the creative activity of an author
- having a literary, artistic, or scientific character
- being expressed in an objectively perceptible form.**
Typical examples are a book, a song, a painting, a photograph, or a sculpture.
In the context of AI, the first two conditions are the most problematic: the creative activity of the author and the uniqueness of the work. The law doesn't see a sufficient connection between the generation itself and the personality of the author — their creative and intellectual effort. It also questions the originality of such outputs. Even though we get a different image each time for the same prompt, the results share the same characteristic features (composition, characters, lighting, etc.) and are therefore not considered truly unrepeatable. Moreover, under the legal interpretation, creativity is tied to the free creative choices of a human that determine the specific form of the work. In other words: the result must not be random — the author has to steer it consciously and deliberately.
The conclusion? Content created solely with AI is not a protected work under copyright law.
At the same time, no one is the author: the AI user isn't, because under the current interpretation they lack sufficient creative and intellectual effort. The AI itself can't be an author at all, because under most legislative approaches worldwide only a natural person can be an author.
In practice this means the user has no copyright to their generations — neither moral nor economic rights. They can't claim authorship, demand that their name appear on the result, prevent others from ordinary use of the image, or forbid its modification. If we return to the lobster I generated with ChatGPT, then just like Kashtanova I'm simply out of luck.
AI and training data
For AI to be able to generate anything at all, it first has to learn how. And for that it needs training data — essentially all the content available on the internet, including creatives' copyrighted works that were published there.
As we know, tools like Midjourney and Stable Diffusion were trained, among other things, on the LAION database from a German non-profit organization. This database is a list of URLs of original images from the internet together with their ALT texts — i.e., the descriptions found with the images. The database's authors and the AI operators have long claimed that this doesn't infringe copyright. They rely on the fair use doctrine in the USA and on the text and data mining exception in the EU — that is, the ability to freely use data for research purposes, which includes training AI models.
AI vs. artists
In recent years, more and more precedents have appeared in which artists and creatives themselves sue AI companies for copyright infringement. One of the best-known cases is the 2023 class action — Andersen et al. v. Stability AI, Midjourney, DeviantArt. In it, the artists claim that the resulting images clearly imitate their specific style, and that they are therefore losing income from licenses and commissions.
The verdict? While the court dismissed most of the claims — for example the assertion of unjust enrichment on the part of the AI operators — it also accepted the artists' argument that providing the Stable Diffusion model to other AI operators could have infringed copyright, and that with certain prompts it is possible, to a certain extent, to reconstruct the original training data — i.e., the works of specific artists.
Another court hearing is therefore scheduled for 2026, in which experts will examine in detail how AI companies work with copyrighted works and whether they are in fact infringing copyright. If the plaintiffs manage to push their arguments through, this very case could become one of those that fundamentally reshape the regulation of AI art in the United States.
The AI Act
Concerns about copyright infringement are thus very real, and lawmakers are gradually starting to respond to them. At the European Union level, the so-called AI Act (the Artificial Intelligence Regulation) was adopted in 2023; it is gradually coming into force and represents the first comprehensive attempt to regulate AI technologies.
From 2025 it already imposes a specific obligation on operators of generative AI models: to keep technical documentation that, among other things, contains a detailed description of the training data sets used. In other words — at least on paper it should be clearer which sources the model "learned" from.
Another important obligation for operators is to label AI-generated content. The aim is to prevent users from being misled — and indirectly this also raises the question of the value of such content, especially compared with human creation.
Practical tips for generating with AI
So what to do when we've spent a lot of time generating visuals and want our outputs to be at least somewhat protected against misuse?
Above all, it's wise not to rely on the assumption that mere prompting automatically produces an output protected by copyright. In practice it makes sense to think of the result in layers: (1) AI as the foundation and (2) a personal "touch" by a human hand as the essential part.
The AI-generated lobster image isn't mine on its own, but I'm the author of all the further edits.
The key is to make sure my contribution to the final result is more significant than the AI foundation itself. But it also works the other way around: if I completely redraw the AI output, fundamentally transform it, or draw only limited, clearly delineated inspiration from it, then we're already talking about my own work.
And last but not least: always read the license terms of the specific AI tool. What can I actually do with the result? Is commercial use allowed? Does it depend on the subscription tier? With some free plans, AI operators reserve the right to keep using the generated content themselves.
That said, there are already AI artists who build their entire brand on this principle: their main value isn't the output itself, but the know-how — i.e., the recipe for achieving a certain result through prompts. In terms of the authorship of the AI generations themselves, though, their position remains problematic.
How (not) to use AI on client projects
For client projects it again makes sense to think in layers: (1) AI foundation + (2) your own processing + (3) brand elements. I consider it fair practice to inform the client in advance that AI outputs will be used in the project, and that these outputs on their own may not be protected by copyright.
The ideal scenario is one where significant further edits and creative processing build on the AI outputs. In that case the resulting work — assessed as a whole — does enjoy copyright protection.
Moreover, the moment any of the client's brand elements are reflected in the result, another protective mechanism comes into play: the trademark. That can act as a further instrument preventing misuse of the work by third parties.
An important note: if, when generating content, we copy or substantially imitate other brands, this can be considered an unfair commercial practice and a breach of competition rules. So even when working with AI, the basic rule still applies — inspiration yes, parasitism no.
AI and generated content
AI is a great tool, but you need to treat it more as a smart foundation than a finished work — real value only emerges the moment we bring our own thinking, originality, and experience into the result. So that AI relieves you of the routine and leaves more room for what's most fun about creating — being creative and unique — here's a practical cheat sheet to finish with:
Checklist: What to watch out for when generating visuals with AI
- Did the output come from a prompt alone, without any further creative intervention?
- Did I add my own creative input to the final visual?
- Does the result imitate someone else's style, brand, or a specific author too conspicuously?
- Have I verified that I can use the output commercially?
- Does the AI platform reserve the right to keep using my visual?
- Does the final output also contain my own processing or brand elements?
- Did I inform the client that we use AI when working on the project?
- Do I have the prompts and other materials that led to the final result saved?
