AI Images and Copyright: Who Owns the Image?

An image generated in seconds, uploaded directly to the DAM, and used in the very next campaign shoot. What sounds like a seamless workflow actually harbors legal pitfalls that many companies still underestimate. This raises questions regarding the copyright of AI-generated images, usage rights, and labeling requirements - issues for which German law has yet to provide comprehensive answers.

Tools like Midjourney, DALL-E or Adobe Firefly generate images based on text inputs, known as prompts. The user describes what they want to see - the AI delivers the result. It sounds simple. But it is precisely at this intersection between human input and machine output that the legal crux lies: Who actually created this image?

For traditional photographs or illustrations, the answer is clear: the photographer or illustrator is the copyright holder. With AI-generated images, the answer is more complex and has not yet been definitively settled. This even applies to images from one’s own collection that have been manipulated by AI.

The German Copyright Act (UrhG) protects only personal intellectual creations by human beings (Section 2(2) UrhG). An AI is not a legal entity—neither natural nor legal—and therefore cannot be a copyright holder. But what about the user who entered the prompt?

Under current law (May 2026), simply entering text is not sufficient to be considered the author of the generated image. The user does not make any creative decisions regarding image composition, color, or form - they formulate an idea, but the AI decides how to implement it. The result is therefore considered to be in the public domain: it belongs to no one and can, in principle, be used by anyone - including for commercial purposes.

At first glance, this sounds practical. But for companies that use AI images professionally, it means one thing above all: no exclusivity. Anyone working with the same prompts will get similar or identical results. A competitor can legally use the same image without any issues. In the worst-case scenario, two companies could appear on the market with the same AI-generated product image or video. Imagine you’re watching a commercial on Prime or Disney+ and suddenly the same alpacas or sheep show up in your commercial.

For DAM systems like TESSA, where companies centrally manage all their digital assets, this has a direct implication: AI-generated images should be clearly labeled as such within the system and tagged with appropriate metadata - not only for organizational reasons, but increasingly for legal reasons as well.

There are certain scenarios in which copyright may arise:

1. Subsequent creative editing: Anyone who substantially alters an AI-generated image during post-processing—for example, through individual color corrections, compositing, or artistic interventions—may, under certain circumstances, become the author of the edited work. The key factor is that the creator’s own creative contribution must be clearly distinguishable from the AI output.
2. AI as a pure tool: Anyone who controls the AI so precisely that it serves merely as a technical tool for implementing a clearly predefined image concept of their own has a better chance of obtaining copyright protection. The more detailed and specific the prompt, the stronger the argument. Perhaps you also use your own images as source material.

Important for practical application: Anyone wishing to substantiate these protection scenarios should document all work steps. In a DAM system, it makes sense to store original versions, edit histories, and prompt texts as metadata. TESSA enables exactly that: versioning, comment fields, and structured metadata help make the creative process traceable—an important aspect should legal disputes ever arise. You can do this too if you work with the TESSA Photoshop plugin and use Photoshop AI features.

Copyright issues arise not only because AI-generated images are not protected - they can also infringe on the rights of third parties.

Training-related risks: AI models are trained on large image datasets that may contain copyrighted works. If the AI generates an output that closely resembles a protected original work, the user of the image may be liable for copyright infringement—even if that was not their intention.

Similarity to well-known works: The more an AI-generated image resembles an existing brand logo, a well-known photograph, or a protected work of art, the higher the risk of receiving a cease-and-desist letter or a claim for damages. Companies should therefore perform a reverse image search on AI-generated images before using them. A little “common sense” is also required here. You should definitely check if a similarity is likely.

Right to one’s own image: AI images depicting people who resemble real individuals may infringe on personality rights or the right to one’s own image - even if no real person is depicted. Particular caution is warranted with AI images in sensitive contexts such as advertising, politics, or satire.

For DAM users, this means that images should not be added to the asset pool without careful review. Legal reviews should be part of the upload workflow - ideally through clearly defined approval workflows, such as those supported by TESSA with its workflow and rights management features. You should definitely subject the results of AI-generated content to a visual inspection (or have them inspected). If legal violations are possible, a more detailed review should be conducted.

A scenario that is becoming increasingly common in practice: A company has an extensive library of professional product photographs—elaborately produced, licensed, and carefully tagged in the DAM. Now, AI is supposed to help make this material usable more quickly, more affordably, and in more diverse ways: swapping out backgrounds, placing products in new scenes, generating variations for different markets, and modernizing outdated images. What works technically without a hitch raises fundamental legal questions.

Who owns the rights to the source material?

Before AI is even used, a crucial preliminary question arises: Who owns the images used as input? In the case of photography commissioned from external sources, copyright generally belongs to the photographer. Companies typically acquire only licenses for use, and these often expressly prohibit editing, modification, or integration into AI systems. Anyone who uploads a licensed image to an AI image-processing platform without permission risks copyright infringement, even if the resulting image bears little visual resemblance to the original. You should therefore review the terms of your contracts with photographers and, if necessary, have them amended for future shoots.

Even with your own employee photos or internally produced images, complications can arise: If recognizable people are visible in the images, personality rights and the right to one’s own image apply—and with them, potentially additional consent requirements that preclude the use of the image material with AI.

For DAM users, this means: The rights status of every asset must be known before AI is used. In TESSA, license information, expiration dates, and usage restrictions can be stored as structured metadata—and it is precisely this information that becomes mandatory reading before an image is used as AI input.

What are the legal implications of AI output?

Let’s assume that the source material poses no legal issues—it consists of the company’s own photos, which are entirely royalty-free and produced in-house. Now, a company uploads these photos to an AI tool and uses it to generate new image variations. What are the legal implications?

This depends largely on the extent to which the AI alters the original and how much creative control the human retains. Broadly speaking, three scenarios can be distinguished:

Scenario 1 – Minor modification: The AI simply replaces the background or changes the lighting, while the product itself remains essentially unchanged. In this case, the edited image is likely to be considered a modification of the original. The copyright to the original image remains intact; the edited result generally does not qualify for separate protection, as the human creative contribution to the AI process is too minimal.

Scenario 2 – Substantial Transformation: The AI reinterprets the product, fundamentally altering its style, composition, and context. The source material is barely recognizable in the result. This raises the question of whether an independent new work has been created. If so—and if the human user has demonstrably made creative decisions in the process—the result may theoretically be eligible for its own copyright protection.

Scenario 3 – Iterative Development with Human Control: The user carries out multiple rounds of editing, actively intervenes in composition and design, combines AI results with their own elements, and makes conscious creative decisions. Under current law, this is the most promising approach for establishing copyright protection for the final result.

The Higher Regional Court of Düsseldorf Sets Initial Standards (April 2026)

A recent ruling by the Higher Regional Court (OLG) of Düsseldorf from April 2026 (Case No. I-20 W 2/26) provides important insights into future legal developments. In the case, a former business partner of an animal photographer uploaded her professional underwater photograph of a dog into AI software and generated a new image. The photographer sued for an injunction - without success.

The court first examined whether the AI-generated work could be classified as a free adaptation under German copyright law. A free adaptation differs so significantly from the original work that it is considered an independent work. For this, it is crucial that the result itself is a copyright-protected work—that is, a personal intellectual creation. This was lacking because the user had not explained which specific creative decisions he had made during the prompting process.

The court clarified: In the case of AI images generated from photos, it is not the similarity of the subject matter that matters, but rather the incorporation of protected design features. In photography, these are typically decisions such as framing, perspective, lighting, and the sharpness or blur achieved through aperture and exposure time. Since the AI image had not adopted these specific photographic decisions, the court found no copyright infringement.

The ruling sends an important message in both directions: Using someone else’s images as input for AI does not automatically infringe copyright—but having one’s own images processed by AI does not automatically grant new protection for the resulting work.

Specific risks associated with the use of employee and model photos

A common real-world scenario: A company owns licensed photos of models for product campaigns. Now, AI is to be used to depict the same products on different people, in different outfits, or in altered contexts—without producing new photos. Even if the original photos are completely royalty-free, new questions arise here:

• Has the model consented to the AI-based further processing of their image?
• Does the original license agreement permit this type of use?
• Does the AI manipulation generate new, deceptively realistic depictions of a real person that would be subject to labeling requirements as “deepfakes” under the AI Act?

These questions have not yet been conclusively resolved in law. However, they make it clear that AI manipulation of images of people requires particularly careful legal review and that the relevant consents and license documents must be readily available in the DAM.

The legal uncertainties do not argue against the use of AI to enhance your own image library—they argue for a structured approach to its use. The following principles can help:

Clarify legal status before using AI: Every image intended to be used as AI input must be checked for its license in advance. Does the license permit editing by AI? Have all persons depicted given their consent? In the DAM, this information should be stored as required fields and, if necessary, secured through approval workflows.

Document creative control: Anyone wishing to claim AI-edited images as their own creative works must be able to prove that they made creative decisions. Prompt texts, iteration steps, and selection decisions should be stored as metadata in the DAM—not as a formality, but as potential evidence.

Maintain a transparent chain of origin: For every AI-edited image, there should be a clear link to the source material in the DAM. TESSA enables assets to be linked and relationships between originals and derivatives to be mapped. This chain of origin (asset lineage) is not only important for internal organization—it can be decisive in the event of a dispute.

Implement a two-tier system: Many companies find it helpful to clearly separate AI-processed assets from original images within their DAM structure—for example, by using dedicated categories, folders, or status fields. This makes it immediately clear which images can be used without restriction and which require further review.

The legal landscape is evolving. With the EU AI Act (Regulation (EU) 2024/1689), companies will face binding regulations:

Starting in August 2025, a labeling requirement will apply to certain AI-generated content under Article 50 of the AI Act - in particular for hyper-realistic representations and so-called deepfakes.

Starting in August 2026, a broader labeling requirement for AI-generated content will take effect: Anyone publishing images, videos, or text that were significantly created by AI must indicate this - regardless of the publication channel, whether a website, social media, or retailer portal.

This labeling requirement is particularly relevant for companies that distribute AI-generated product images to retailers, journalists, or influencers via platforms such as the TESSA BrandHub. A well-designed metadata structure in the DAM is key here: If an asset’s AI origin is stored in the system, labeling can be automated during export or publication and compliance requirements can be met much more easily.

Note: This article is intended for general informational purposes only and does not constitute legal advice. For specific legal questions, we recommend consulting a lawyer specializing in copyright or IT law.