AI-assisted literary-to-song adaptation: a practice-based study of literary interpretation in a text-to-audio workflow

This qualitative practice-based digital humanities article examines AI-assisted literary-to-song adaptation. GenAI research has addressed writing support, music generation, literary analysis and creative work, with less attention to literary judgement in generated audio and post-production. The corpus comprises 32 songs developed by a single creator, largely based on H. P. Lovecraft and produced in a post-punk/indie-rock idiom through close reading, researcher written lyrics, ChatGPT-supported revision, Udio text-to-audio generation and human-directed assembly. Two cases are analysed: ‘The Viol Screams’ adapts The Music of Erich Zann and ‘Whispers from Yuggoth’ brings together selected sonnets from Fungi from Yuggoth. Benchmarking, listener testing and human-only comparison fall outside scope. The analysis traces source selection, lyric development, generation, evaluation, repair and post production, focusing on textual transformation, interpretive emphasis and production intervention. Across the cases, generated audio was selected and assembled using literary, musical and technical criteria, while source structure affected human work across stages. Researchers in digital humanities and human-AI creative practice can use the source-to-song workflow heuristic and lyric-strategy typology to document how literary meaning moves into generated sound. Distributed interpretive control gives adaptation and popular music scholars a case-derived concept for examining how interpretive priorities remain active in lyric, vocal, sectional and production decisions.