Automatic Generation of Fashion Images using Prompting in Generative Machine Learning Models

Abstract

In the contemporary landscape of fashion, the convergence of technology and creativity has catalyzed a transformative shift, ushering in new opportunities and redefining industry standards. At the forefront of this evolution lies the integration of computer vision and artificial intelligence, revolutionizing fashion through innovation, efficiency, and refined aesthetic precision. This thesis investigates methodologies for generating tailored fashion descriptions using two distinct Large Language Models (LLMs) and a Stable Diffusion model for image creation. Emphasizing efficiency and adaptability in AI-driven fashion creativity, we depart from traditional approaches and focus on prompting techniques, such as zero-shot, one-shot and few-shot learning as well as Chain-of-Thought. Central to our methodology is Retrieval-Augmented Generation (RAG), enriching models with insights from fashion magazines, blogs, and other sources to ensure accurate and contemporary fashion representations. Evaluation combines quantitative metrics like CLIPscore with qualitative human judgment, highlighting strengths in creativity, coherence, and aesthetic appeal across diverse styles. Comparative analysis demonstrates the efficacy of techniques such as Few-shot learning and RAG with PDFs in producing descriptions and images tailored to specific fashion variables. Qualitative assessment reveals advancements in realism and visual diversity, supported by the Chain-of-Thought methodology