Extension-semantic prompting for AI-generated traditional Chinese patterns: behavioral and semantic-aesthetic evaluation with exploratory fNIRS analysis

Background This study developed the Extension-Semantic Prompting Framework (ESPF) to organize cultural-semantic information in prompts for AI-generated traditional Chinese patterns. The framework decomposes traditional patterns into seven semantic-aesthetic dimensions and maps them onto structured prompt slots and corresponding evaluation items. Methods The experiment included 18 experts and 18 laypeople as participants. Stimuli comprised images of five traditional pattern themes generated using Stable Diffusion and Low-Rank Adaptation (LoRA) models under the ESPF and Free conditions. Liking ratings, response times (RTs), and semantic-aesthetic questionnaire responses were analyzed using cross-classified mixed-effects models, while prefrontal functional near-infrared spectroscopy (fNIRS) data were examined using region-of-interest (ROI) and general linear model analyses. Results Mixed-effects analyses showed that the ESPF condition was associated with shorter aesthetic-judgment RTs and higher framework-derived semantic-aesthetic ratings, although the latter varied significantly across pattern themes. Interpretation of the RT difference is limited by the fixed Free–ESPF presentation order. The overall condition effect on liking was not significant after accounting for participant and image variability. No stable condition-related effect was observed in the prefrontal ROIs, and exploratory between-group neural differences were not robust across adjusted and GLM analyses. Conclusion ESPF provides a structured approach to organizing cultural semantics in AI-generated traditional patterns. The selective differences observed across evaluation outcomes and pattern themes highlight the need to balance semantic control with creative openness in prompt-supported human–AI collaborative design.