Answers are grounded in this paper and cite their sources.
This exploratory case study examines how records of Haipai Yuefenpai (Shanghai calendar poster) making were translated into text prompts. A multimodal archive included approximately 300 h of field video, historical images, and interviews. Coding 192 clips yielded 538 label assignments across seven dimensions: style, figure, process, action, material, judgement, and context. Six prompt configurations were used to generate 120 still images through Doubao’s web-based image-generation platform.Five specialists and 30 non-expert viewers assessed eight selected outputs as a set. Mean expert ratings were 4.2 for stylistic accuracy, 2.2 for craft knowledge visibility, 2.6 for material and media authenticity, and 3.0 for historical-context consistency. Representational risk averaged 4.6, with higher scores indicating greater perceived risk. Non-expert ratings averaged 4.37 for aesthetic appeal and 2.87 for willingness to use the images as references. These measures describe different constructs and are interpreted within each panel. The outputs showed recognisable poster conventions and hand-rendered effects but offered limited evidence of operational sequences or the material feedback guiding practitioners’ decisions. The proposed workflow documents links between archival observations, semantic labels, prompt wording, and output review; its application to heritage learning and institutional reuse requires further evaluation.
No takes yet. Share an insight, caveat, or question.
jiang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: