Fashion fusion: generating trendy attire with Indian ethnic art using text-to-image generative model

Designing fashion apparel with local trends and styles, holds significant potential for retailers and consumers. Fashion designers invest considerable time and effort in creating designs based on print/pattern and attribute selection. To address the challenges posed by manual efforts and achieve automation, this article introduces a novel framework for designers, incorporating specific styles (prints/patterns) and attribute information via generative models. The proposed method fine- tunes the Stable Diffusion XL, a text-to-image (T2I) generative model, using a dataset of images featuring Kolam or Gond tribal arts from South and Central India, respectively. These images help to infuse region-specific print/pattern styles into T-shirts/Kurtas. Additionally, input attributes such as neck type and, sleeve type are integrated with text prompts during inference to produce the final fashion attire. A comprehensive evaluation was carried out and compared with state-of-the-art T2I models. The results indicate 75.78% accuracy in attribute and 76% in styles.