AI images in copyright law: When prompting protects and why copying is allowed

The legal assessment of images generated with the help of Artificial Intelligence is becoming more precise. In a landmark ruling, the Higher Regional Court (OLG) Düsseldorf has affirmed that purely machine-generated output generally does not enjoy copyright protection. Anyone who wants to legally secure an AI image as their own personal creation must be able to prove a significant, human-creative influence. Merely feeding the software with general, open-ended text instructions (prompting) is by no means sufficient for this.

At the same time, the now published decision of April 2nd opens up greater scope for the targeted, AI-assisted imitation of existing image ideas (File No.: 20 W 2/26). According to the Düsseldorf judges, the mere motif or the underlying idea of a photograph is in the public domain and can be legally reproduced by generative systems. The condition is that the craftsmanship and design peculiarities of the original remain untouched.

The proceedings were based on a dispute in a legal grey area that has so far been little illuminated. An underwater photographer had specialized in photographing dogs diving for toys. A former cooperation partner uploaded one of her elaborately post-processed image files into an AI system and generated a new graphic using the so-called image-to-image process. This also showed a diving dog but had a significantly altered, almost comical style.

The photographer saw this as an infringement of her rights and wanted to prohibit its distribution via a warning letter using interim legal protection. After the Regional Court of Düsseldorf rejected the urgent application on the grounds that it was a permissible free adaptation, the OLG confirmed this result in the second instance. However, it significantly corrected the legal reasoning and thus set guidelines for the industry.

Human Creation vs. Consumption of Results

The responsible 20th Civil Senate makes it clear: A free adaptation requires that the newly created AI image itself possesses the quality of a protected work. Whether this is the case depends crucially on the degree of human control. While copyright protection for AI creations is not categorically excluded, it requires the personality of the human user to be expressed in the final image.

According to the ruling, this can happen through extremely detailed pre-settings, continuous, highly specific corrections during prompting, or a conscious, creative selection from a variety of intermediate results. However, according to the OLG, those who feed the AI with vague instructions and leave the actual design decision to the software are merely consuming results without their own creative contribution.

Procedurally, this line leads to a significant hurdle for AI users. Anyone who claims in court to have created an AI image as an independent work bears the full burden of presentation and proof. Since the respondent in the specific case, despite court instructions, could not or would not explain which specific creative decisions and prompts he had used, the court denied the AI graphic recognition as a work. This blocked the legal path via a free adaptation.

Farewell to the Classic Overall Impression

The fact that the photographer still lost the case is due to a fundamental reorientation in the examination of copyright infringements. Here, the OLG follows European law. Previously, in German jurisprudence, the visual overall impression of the images was decisive. The Senate is now explicitly moving away from this and applying a precise, element-oriented procedure.

According to this, it must be examined whether the creative elements that constitute the photographer's personal achievement have been adopted exactly. These can include camera settings, specific lighting, depth of field, or the chosen framing. The abstract motif, on the other hand, i.e., a dog reaching for a toy underwater, is a mere idea and not protectable. Since the AI graphic, while picking up on the approach of the model, deviated completely from the original in terms of perspective, anatomy, and dynamics, the OLG could not identify any legal infringement.

The ruling shows that photographers are protected against the identical reproduction of their technical work. However, they must accept AI-assisted reconstruction of visual ideas. IT lawyer Jens Ferner welcomes the decision. It provides "the most precise higher court doctrine to date on the interface between classic photographic protection and the use of human-created images as input for generative AI systems." It also systematically builds on the most recent guidelines from the European Court of Justice.

(nen)