Proceedings such as those against "Stability AI" show us that the similarity of an AI output to the original work is often seen by authors as a threat to their work. It is argued that AI outputs merely imitate a style - but do not infringe the rights of the artists. But where is the line between the mere use of the style and the unlawful adoption of parts of the work in the AI output?

This presentation - with a particular focus on large language models - is intended to make this boundary visible and make it easier to distinguish between reproduction, adaptation requiring consent and free re-creation in AI outputs.