Abstract: Generative AI technology empowers creators, yet simultaneously displaces labor in the market for creative works. Grounded in the constitutional protection of economic rights in copyright, rightholders do bear social obligations, but should not be made to shoulder alone the sacrifice of their economic interests for the public good; a compulsory license coupled with compensation presents a more fitting solution. Establishing a compensation mechanism also serves to incentivize human creation, sustain diversity in the marketplace of expression, and secure a sustainable supply of training data. The upstream compensation model centered on statutory licensing, however, confronts prohibitive transaction costs and distributional difficulties, and its mode of value transfer struggles to reach the creators who are genuinely affected. Drawing on three downstream compensation mechanisms found in copyright law—revenue adjustment through taxation, hardware-based levies, and royalty-sharing arrangements akin to "work-relief" programs—a progressive downstream compensation framework for data training may be envisioned: in the near term, collective management organizations could be guided to negotiate profit-sharing agreements with major AI service providers, channeling support to affected creators through project grants and skills training; in the medium to long term, a dedicated tax could be introduced, with regulators setting differentiated rates according to service type so as to furnish a stable funding source for the creative ecosystem; in the longer run, a supplementary hardware levy could be adopted to cover scenarios that profit-based exaction schemes find difficult to reach, such as on-premises enterprise deployment.