Presentation Information
[4Yin-A-24]A Study on Optimizing Fragrance Product Images Using AI-generated Backgrounds
〇Masaya Kirishima1, Ryosuke Araki1, Satoshi Shiibashi1 (1. ZOZO, Inc.)
Keywords:
Image Generation AI,E-commerce,Background image generation
Product images displayed on product list pages are important for attracting attention in e-commerce, especially for fragrance items that need to communicate non-visual attributes such as scent. Diffusion-based models can generate background images from product text, but their impact on user behavior in e-commerce remains unclear. This work investigates how AI-generated backgrounds influence click-through rate (CTR) on a Japanese fashion e-commerce platform. Through an online A/B test, we observed that images with AI-generated backgrounds achieved a higher CTR compared to original product images. Among these, images with visually restrained backgrounds tended to achieve higher CTR than those with complex backgrounds. To further investigate the effect of background complexity on CTR, we conducted an offline analysis by categorizing images into three levels of background complexity using an LLM. The results show that images with visually restrained backgrounds achieve the highest CTR, highlighting the impact of background complexity on user engagement.