E-commerce Product Retouching: Reference-Image-Driven Texture Relighting

Use a casually shot original product photo as the sole reference input, and remake it with gpt-image-2.5-flare/image-to-image into a studio-grade product image ready to be published directly on e-commerce detail pages. The approach is to break down "product photography quality" into several promptable layers: the subject's geometry and logo remain unchanged, the material's highlights and reflection relationships, the lighting direction of studio softboxes, the cleanliness of the background and floor, and the final output aspect ratio and resolution. The reference image provides the real subject shape and logo information; the prompt is only responsible for defining light, texture, and composition, thus preventing the model from redrawing the product structure. On the input side, submit only one reference image and an English editing instruction; resolution tier and aspect ratio are given as non-asset parameters. Since no background mode is specified, the model outputs an opaque background following the logic of the original image, suitable for direct use in product hero images and lifestyle images. It is suitable for e-commerce asset production that requires a batch of unified visual style for apparel, beauty, 3C accessories, food packaging, etc., and also suitable as a quick direction validation for designers before retouching.