In image-to-image work the model receives a picture as well as a prompt. Depending on the method, the source fixes the composition, the subject or only the style, and a strength setting decides how far the result may move away from it. For sellers the useful form is the one where the product photo is the anchor: the garment, bag or shoe stays the subject while the setting, the model or the light change.
The source carries the information the result depends on. A blurry phone photo, a crumpled garment or a colour cast is either passed on or guessed around; a sharp, evenly lit photo of the whole product, taken straight on, leaves the model the least to invent. Parts the source does not show, such as the back of a jacket or the inside of a bag, have to be made up, and that is where errors appear.
Even from a good source, the output is a new drawing of the product, not the original pixels. The higher the strength setting, the more freely the model redraws: the scene goes first, then colour and texture drift, then construction such as stitching lines, pocket shapes and print placement. A low strength stays close to the source but keeps its flaws too, so the setting is found by comparing a few results with the real item. Marketplaces that accept AI images ask that they show the product accurately, and marking rules differ between them.