Can someone explain to me how this achieves the stated goal of "protecting from style mimicry"? Because: What this does is, it makes it harder to train models

Can someone explain to me how this achieves the stated goal of "protecting from style mimicry"? Because: What this does is, it makes it harder to train models on specific works. Okay. But if the model is sufficiently generalized, an arbitrary style could simply be described to the model (not by the artists name, but by technically describing the style), and then reproduced, even if the model never saw any training data in that style. Come to think of it: If a model is sufficiently generalized, it could reproduce any conceivable style described to it, not just those it didn't have in its training set, but even styles that didn't exist before. So yeah, mid-to-long-term, how does this "protect from style mimicry"?