AI image generators continue to raise new questions regarding copyright law. The Higher Regional Court of Düsseldorf (judgment of April 2, 2026, Case No.: I-20 W 2/26) addressed the question of what applies when a copyright-protected photo is uploaded as a template into AI software and the resulting output is subsequently published.

The Senate concluded that the publication of the AI-generated image of a dog did not infringe on the photographer’s copyright. The decisive factor is whether the creative elements of the original were recognizably reproduced.

The decision is noteworthy because it distinguishes between two points: On the one hand, AI-generated output generally cannot be considered a “work” as long as no human creative input is discernible. On the other hand, this does not automatically mean that the output is an infringing reproduction of the original photograph.

What was it about?

Among other things, the applicant offers underwater photos of dogs. She had taken and edited a photo showing a dog underwater reaching for a red toy.

The defendant had cooperated with the plaintiff in the past. Later, he uploaded the image file to AI software and had it generate a new image. He then published this image on his website.

The photographer viewed this as a copyright infringement. She filed a motion for a preliminary injunction and sought to prevent the defendant from reproducing or making the AI-generated image publicly available without her consent.

No “free adaptation,” since there was no new work

The lower court had assumed that the AI-generated image constituted a free adaptation within the meaning of Section 23(1), second sentence, of the German Copyright Act (UrhG). The Higher Regional Court of Düsseldorf corrected this interpretation.

According to Section 23(1), second sentence, of the German Copyright Act (UrhG), a work is considered a free adaptation only if a new work is created that maintains a sufficient distance from the earlier work. However, a work must be the result of personal intellectual creation.

This is not necessarily the case with AI-generated works—we explored the fundamentals of this issue in our article “Can AI Be an Author?” In the Senate’s view, a copyright-protected work can only come into existence if, despite the software-driven process, human creative decisions shape the specific output. This could be achieved, for example, through sufficiently individualized specifications, creative guidance during the prompting process, or subsequent human editing.

However, it is not sufficient for the user to provide the AI with only general, open-ended instructions or to simply select one of several AI-generated results.

In this specific case, the defendant had not explained which prompts or creative decisions he had used to generate the image. Therefore, the AI-generated image could not be considered his own copyrighted work.

Nevertheless, no copyright infringement

That didn’t help the photographer, however. After all, just because the AI-generated output is not an original work in its own right does not automatically mean that it infringes on the source photo. The decisive factor remains whether the new image incorporates protected elements from the original photo.

In this regard, the Higher Regional Court of Düsseldorf relied on the recent case law of the European Court of Justice (judgment of December 4, 2025, C-580/23 and C-795/23 – “Mio and konektra”). According to this case law, a mere comparison of overall impressions is not sufficient. Rather, the court must examine whether it is specifically the creative elements of the protected work that have been copied—that is, those elements in which the author’s personality is expressed.

In the case of photographs, such elements may include, in particular:

• Image detail,
• Perspective,
• Lighting,
• Sharpness and blur,
• Composition,
• specific photographic composition.

However, the mere idea, subject matter, or motif are not protected.

The “Dog Underwater with a Toy” image is free to use

The petitioner pointed out that both images showed a dog underwater reaching for a red toy. However, the Higher Regional Court of Düsseldorf did not consider this to be a protected element.

The subject matter itself is not protected by copyright. Only the specific photographic composition is eligible for protection.

And, in the Senate’s view, there was a clear difference: The original photo essentially showed the dog’s head and the toy. Through perspective, blurring, and photographic composition, it appeared realistic and dynamic. The dog’s body receded into the background.

In contrast, the AI-generated image had a comic-book-like quality. You could see the dog’s entire body. The dog appeared to be reaching for the toy not only with its mouth but also with its front paws, which were stretched far forward. The unique dynamism of the original photo—created by the lighting, aperture, and focus—was notably absent from the AI-generated image.

In the opinion of the Higher Regional Court of Düsseldorf, only elements in the public domain had been incorporated: the theme and the motif. That is not sufficient to constitute copyright infringement.

Nor is there a violation of the right to one’s own image

Nor was the petitioner able to successfully invoke the photographer’s ancillary copyright under § 72 of the German Copyright Act (UrhG).

While this also protects photographs that are not photographic works, only the specific photographic creation itself is protected. Therefore, even in the case of an infringement of the photographer’s right, it is specifically the photographic creation that must be reproduced.

Since the AI-generated image did not incorporate the distinctive photographic elements of the original photo, a violation of the photographer’s rights was also ruled out.

What does this mean for AI image generators?

The decision does not constitute a blanket authorization to upload third-party photos into AI tools. However, it does show that copyright reviews must be conducted on a case-by-case basis.

Anyone who uses someone else’s image as a template for AI is still operating in a legally risky area. The more closely the AI output replicates the specific design of the source image, the more likely it is that an infringement has occurred.

It may be less problematic if only a general idea or a motif that is not eligible for protection is adopted and the resulting work has a significantly different design.

At the same time, the Higher Regional Court of Düsseldorf makes it clear: Anyone who claims that AI output constitutes an original work or a free adaptation must be able to demonstrate what constitutes the human creative contribution. Without a clear explanation of the prompting, control, and selection process, this will be difficult.

Practical Tips for Creative Professionals and Businesses

For photographers, designers, and agencies, this ruling serves as a reminder to present precise arguments in cases involving AI. It is not enough to simply point out similarities in the subject matter. What matters most is which specific creative design elements were copied.

Conversely, for companies that use AI tools, the development process should be documented. Anyone working with third-party source material should keep a clear record of prompts, intermediate steps, selection decisions, and any post-processing. This can be crucial in the event of a dispute.

You should be especially careful when source images come from partners, photographers, or agencies. A previous collaboration does not automatically mean that such images may be fed into AI systems or that outputs generated from them may be published.

We discuss how to use AI-generated images in a legally compliant manner in our podcast episode, “Using AI-Generated Images Legally—But How?”

Background: AI Copyright Laws Are Becoming More Concrete

The ruling by the Higher Regional Court of Düsseldorf is one of the key early decisions regarding the copyright assessment of AI-generated images.

It sets forth two guidelines:

First: AI output is considered a work only if it reflects human creative decisions. Mere machine performance is not sufficient.

Second: An AI-generated image does not infringe on someone else’s photograph simply because it depicts the same subject. What is required is the incorporation of protected, personally distinctive design elements.

On the one hand, this decision reinforces the public domain status of ideas, themes, and motifs. On the other hand, the protection of specific photographic compositions remains unaffected.