Creators who generate hit products enjoy outsize success in creative industries. But too often, creators fail to learn from their initial hits, as their subsequent products lack the audience appeal of their initial hits. In this paper, I develop theory on learning transfer in creative work, focusing on how creators’ learning is shaped by the novelty (vs. typicality) of their initial hits in creative industries. I propose that creators who develop relatively novel initial hits are more likely to transfer deep learning that enhances the audience appeal of their subsequent creations, helping them generate additional hits after their initial one. I tested the theory using two studies: an archival study of 1,601 book authors and a pre-registered experiment. The archival study showed that authors with novel initial hits had better subsequent hit rates—a likely indicator of learning—than authors with more-typical initial hits. The experiment provided causal evidence that creators learn more deeply and effectively from novel than typical initial hits. Whereas prior research emphasizes that individuals learn best from multiple episodes of success, failure, or both over time, this research suggests creators who achieve novel initial hits can and do learn from one episode of extreme success.