GenAI-Driven Image Generation Pipeline for Sustainable Garment Design and Waste Reduction in Fashion Production

The fashion industry’s linear production model generates sig- nificant pre-consumer textile waste, especially during pat- tern cutting. In response to the environmental impact of fash- ion consumption, strategies such as reuse, recycling, and re- fashioning aim to divert textiles from landfills and promote sustainable practices. However, challenges in the textile sec- tor—such as raw material variability and complex manufac- turing—require more targeted solutions. Recent studies have identified Artificial Intelligence (AI) as a promising tool to enhance sustainability, streamline production, and enable per- sonalised design. One such advancement is Generative AI (GenAI), which supports applications like virtual try-ons, fabric-to-garment transformations, and multimodal garment design via tools such as FashionGAN, StyleGAN, and Latent Diffusion Models. Despite these developments, current image generation methods struggle with preserving fabric detail and structural accuracy. This research proposes an image gener- ation pipeline that accurately reflects specific fabric textures and visual attributes, offering designers greater creative con- trol while reducing the need for physical samples—thereby minimising process waste. The system is implemented us- ing ComfyUI and LoRA-enhanced Stable Diffusion 1.5 mod- els to overcome limitations found in existing methods. To evaluate performance, quantitative metrics such as FID, KID, SSIM, LPIPS, and CLIP-S were used to assess visual qual- ity, structural similarity, and semantic alignment. A qualita- tive comparison was also conducted to evaluate fabric texture preservation and prompt consistency across models. Among the tested models, Realistic Vision v5.1 delivered the best re- sults across most metrics and is recommended for photore- alistic applications in sustainable fashion. DreamShaper v8 excelled in preserving fabric texture, while MajicMix v5 pro- duced stylised outputs more suitable for conceptual design stages. This study aims to empower fashion designers with a flexible and sustainable design model, to reduce waste, accel- erate prototyping, and explore AI-driven innovation in digital fashion.