As text-to-image (TTI) generation becomes increasingly integrated into fashion image creation, understanding prompt language has become increasingly important. This study examines how fashion design vocabulary is reorganized as prompt language in TTI, analyzing user-generated Midjourney prompts to identify frequent fashion-related terms, their co-occurrence patterns, and the semantic network through which fashion images are linguistically constructed. A large-scale prompt corpus from the Midjourney Discord dataset was used, and fashion-related prompts were selected through two-stage filtering based on fashion items or style/aesthetic terms and visual representation language. After preprocessing, 119,409 prompts and 60,437 unique vocabulary items were analyzed using term frequency, term frequency-inverse document frequency, keyword co-occurrence network analysis, and VOSviewer cluster analysis. “Photography” was the most frequent and distinctive term, followed by visual representation terms such as “depth of field,” “volumetric,” “natural light,” and “cinematic lighting.” Co-occurrence analysis showed that “photography” functioned as a central hub, strongly connected with lighting, focus, mood, and image-quality terms. In contrast, fashion item terms such as “dress” and “suit” appeared frequently but exhibited lower co-occurrence centrality, indicating that item vocabulary was dispersed across varied prompt combinations. Cluster analysis revealed semantic groups related to atmosphere and glow, color and clothing items, lighting effects, image quality and texture, body features, editorial styling, portrait framing, and aesthetic mood. The findings suggest that in TTI environments fashion prompt language extends beyond garment description and operates as a photographic–editorial image syntax and that fashion design vocabulary is reorganized through visual representation language into AI-readable commands for generating fashion imagery.

Abstract
I. Introduction
Ⅱ. Review of Literature
    1. Text-to-Image generation and prompt engineering
    2. Attribute-based representation of fashionimages
    3. Design language and visual syntax in fashionprompts
Ⅲ. Research Method
    1. Data collection and corpus construction
    2. Data preprocessing
    3. Data analysis
Ⅳ. Result
    1. TF and TF-IDF analysis
    2. Co-occurrence network analysis
    3. Clustering analysis
Ⅴ. Discussion
    1. From garment description to photographicimage syntax
    2. Academic and practical implications
Ⅵ. Conclusion
References

• Dawool Jung(Lecturer, Dept. of Fashion Design & Merchandising, Gachon University, Korea) | 정다울 (가천대학교 패션산업학과 강사) Corresponding author