Six Billion Images, One Case: What the Federal Court Ruling Means for Photographers

On September 3, Germany's Federal Court of Justice (Bundesgerichtshof) will hear a case that reaches far beyond a single dispute: a photographer is suing the organization behind one of the world's largest AI training datasets – roughly six billion image-text pairs that serve as the foundation for numerous image generators. His photos, originally licensed through image agencies, are alleged to have become part of that dataset without his consent. The lower court, the Hamburg Higher Regional Court, ruled last December that the use was permissible – based on exceptions for text and data mining in German copyright law. Now the Federal Court has to decide whether that holds up.

As someone who started out as an advertising photographer and now co-runs a studio that lives off exactly the kind of images this case is about, I read it differently than a pure legal news item.

What the lawsuit is actually about

The legal basis sounds technical at first: German copyright law permits, under certain conditions, the automated analysis of protected works for text and data mining, without every individual work needing to be licensed beforehand. The exception was originally conceived for research purposes – analyzing large bodies of text scientifically, for instance. Whether it also covers training commercial image generators at a billion-image scale is the actual question the Federal Court now has to settle. The Hamburg court said yes. If the Federal Court confirms that, the legal situation for the entire industry would be settled – in favor of AI providers. If it overturns it, a central building block of today's image generators would stand on shaky legal ground.

Why this concerns me as a photographer and studio owner

The case doesn't just concern one plaintiff. Image agencies, through which our own productions get licensed too, have for years been a preferred target for gathering training data – structured, tagged with metadata, available in bulk. Anyone who, like us, has produced advertising photography for years can't rule out that their own images are already part of some training dataset, without a single licensing fee ever having flowed. That's not an abstract fear – it's the core of what's being argued on September 3.

At the same time, I want to be honest about where I stand: we use AI tools in content production ourselves, and those tools are only as capable as they are because they were trained on enormous volumes of images. Any creative studio using AI tools doesn't sit cleanly on one side of this debate – we benefit from exactly the technology whose training basis is now being questioned in court.

What the case has to do with the value of CGI

One aspect that barely comes up in the legal debate but is central to our craft: classic, hand-modeled CGI sidesteps this entire problem from the start. A 3D model built from CAD data or original modeling isn't based on a training dataset made of other people's photographs – it's an original creation, without the copyright uncertainty currently surrounding generative image AI. That's an argument I haven't made yet in the comparison between CGI and photography: alongside cost, control, and by now energy consumption, this Federal Court case adds another dimension – the legal robustness of the image source itself.

What a ruling either way would mean

If the Federal Court confirms the Hamburg decision, much stays legally as it is for image generator providers – photographers and image agencies would have to accept that their licensed works can serve as training material without additional compensation. If the Federal Court overturns it, major AI providers would have to reassess their training datasets, possibly re-license or remove image material from their models – with consequences reaching far beyond the German market, since similar proceedings are running in parallel in several countries.

What I'd advise studios and photographers right now

Regardless of the outcome, a stocktaking is worth doing now: which of your own images sit with agencies whose licensing terms might already permit AI training? Where can future contracts be structured to explicitly exclude or compensate for that use? That's not a question to wait on until the Federal Court has ruled – contracts can be adjusted today, independent of how the September proceedings turn out.

Conclusion

September 3 won't be a footnote for photographers and studios – it may well be the day that decides whose images are actually sitting inside the tools our entire industry is using right now. Anyone who makes a living from images shouldn't just follow this case as a legal side note.