The　rise　of　artificial　intelligence-generated　content　(AIGC)　has　fueled　a　growing　demand　for　data　uploads.　Massive　data　are　transferred　from　clients　and　aggregated　on　the　cloud　for　the　AIGC　model　update.　However,　traffic　fluctuation　makes　it　difficult　to　carry　AIGC　uploads　over　conventional　end-to-end　(E2E)　connections.　In　this　paper,　we　present　a　storage-assisted　uploading　method　for　hierarchical　federated　learning　(SU-HFL)　over　optical　AIGC　networks.　SU-HFL　not　only　reduces　uploading　traffic　via　edge　aggregation　enabled　by　HFL,　but　also　relaxes　the　E2E　constraint　via　temporary　storage　on　intermediate　nodes.　Simulations　show　that　SU-HFL　outperforms　conventional　methods　in　terms　of　network　performance　and　training　accuracy.