Framing the Formless: A Xing–Qi–Shen Framework for Culturally Aware Multimodal Human–AI Co-Creation

Abstract

Generative artificial intelligence has significantly lowered the technical threshold for visual creation. However, current systems frequently exhibit a lack of cultural context and aesthetic flattening when applied to Chinese-style scene generation. Existing models commonly offload control responsibilities to prompt engineering, making it difficult for non-expert users to consistently articulate Chinese aesthetic intentions, which require spatial order, atmospheric hierarchy, and semantic depth. This paper reframes this limitation as a human-computer interaction (HCI) problem of "cultural operability." The core objective is not merely to prompt the model to output more cultural symbols, but to translate tacit cultural knowledge into interaction dimensions that users can directly manipulate. Based on the aesthetic logic of Chinese classical gardens, we propose the Xing-Qi-Shen (Form-Atmosphere-Spirit) interaction framework. This framework is operationalized into spatial relational control, atmospheric condition regulation, and semantic guidance within a generative system. Subsequently, we developed the proof-of-concept prototype to manifest abstract cultural structures as actionable interface controls. A within-subjects experiment involving 53 participants demonstrated that the framework significantly outperformed the baseline condition across four dimensions: compositional rationality, atmospheric expressiveness, resonance and imagination, and cultural consistency. The aggregate comparison yielded a baseline mean of 2.946 versus a framework mean of 3.857 (t(52) = 6.981, p < 0.001, Cohen's dz=0.959). These findings provide initial evidence that culturally structured representations can improve the perceived quality of generated outputs.