You can train models on copyrighted materials. https://towardsdatascience.com/the-most-important-supreme-co...