<p>This article uses the viral “Ghiblification” trend—the mass conversion of user-supplied photographs into a visual idiom associated with Studio Ghibli—to ask how law should respond when generative AI makes aesthetic patterns reproducible at scale without necessarily copying any single work. It offers a comparative doctrinal and institutional account of the United States, Japan and South Korea, using developments in the European Union and China only as contextual comparators. This article’s central claim is that a common copyright baseline—that style as such is not protected by copyright—does not yield common governance. At the training stage, US law channels disputes into fact-sensitive fair-use litigation; Japan supplies an express non-enjoyment text-and-data-mining (TDM) exception under Article 30-4 of its Copyright Act, supplemented by administrative guidance and organised industry action; and South Korea lacks an equivalent AI-specific TDM exception, relying instead on the general fair-use provision under Article 35-5 of its Copyright Act, public guidelines, and creator mobilisation directed at a platform-developer ecosystem. At the output stage, the three systems converge on the proposition that mere style evocation is unlikely to infringe copyright, but diverge in the institutional mechanisms through which creators, platforms, and developers contest borderline cases. It concludes that Ghiblification should not be addressed by creating a broad style right, but by adopting a layered framework of training-stage transparency, targeted output-side enforcement and platform procedures calibrated to avoid both over-removal and under-enforcement.</p>