Insights

Does AI bring out the artist in us?

image

Copyright and AI

The latest update of ChatGPT introduced a feature allowing the AI chatbot to accurately mimic images in a requested style. Think of the drawing styles of animation studio Studio Ghibli, The Muppet Show, South Park…

Although AI tools capable of this have existed for a while, the recent trend of ChatGPT-generated Ghibli-style images raises many questions about whether such AI creations could infringe on intellectual property rights. When do we cross the line? Read all about it in the blog below.

Protecting creativity and interacting with AI

At the heart of protecting creative output like images lies copyright. For copyright protection to apply, the following conditions must be met:

  1. A work, expressed in a tangible form (so not mere ideas or concepts);
  2. That is the result of a creative (human) activity;
  3. And is original (it must have a unique character shaped by the author’s personal creative stamp).

The rules that specifically apply copyright law to AI models are still relatively limited: on the one hand, the EU DSM Directive provides two exceptions specifically applicable to text and data mining. On the other hand, the EU AI Act outlines obligations in Article 53 for providers of general-purpose AIs to implement a policy in line with copyright and related rights, and to disclose what has been used to train the model.

We are still awaiting case law that will clarify how these fairly new rules will be applied in practice. For now, the obligations lie with the provider during the training phase — not with the user for the output that is generated.
When it comes to a specific or conceptual style, such style (for now) cannot be protected under copyright. It is not considered a “work” and, more importantly, it cannot be defined with enough precision (although the parameters that AI tools seem able to determine may still change this in the future).

Repainting or re-AI-brushing surely can’t be infringement

The fact that the style of the output isn’t protected doesn’t mean the photo or image being “converted” by the AI doesn’t have an author. The author’s exclusive reproduction right implies that you cannot simply recreate and distribute a work — even if it’s “in a different style” or via a different medium.

A well-known example in the Belgian IP landscape is the plagiarism case involving artist Luc Tuymans, who recreated a photo taken by photographer Katrijn Van Giel. The ruling stated that Tuymans’ painting had clearly taken over elements that contributed to the originality of Van Giel’s photo. Although criticised, the reasoning suggested there were not enough clear differences with the photo to consider it a parody either.

However, this case was settled out of court and was not assessed by a higher court.

The real question is whether the copied or imitated work has reproduced the author’s creative choices — such as framing, lighting, staging, etc. — which gave the original photo its unique character, and whether there are enough differences from the original.

Remarkable Europe Luc Tuymans
Does this apply to the Ghibli photos as well?

That depends on the case: it’s possible only the composition or subject of the original photo is used, while other creative choices of the original author are not reproduced in the AI-generated work.
However, if enough of the original photo’s unique character is recognisable in the new image, there may be grounds for infringement — especially if such images are created and (commercially) distributed.

Our advice?

Use as much of your own creative input as possible, so the work is considered your own and not someone else’s.