When using AI for product images, the part that messes up usually isn't whether the picture looks good or not.

When using AI for product images, the fail isn’t usually about whether the picture looks good—it’s about whether it’s still the same product. Generative models love to take liberties: the bottle cap’s threads get changed, the logo blurs out, the texture goes from matte to glossy, the color shifts half a shade off. The more polished the image, the more dangerous these subtle distortions are, because you won’t notice them unless you look closely.

Once it’s actually listed, customers get the real item and it doesn’t match—worst case, returns or complaints. For product images now, I always start with real photos as the base, and only use AI for swapping backgrounds, fixing lighting, or creating scenes. I never let it generate the main subject from scratch. Looking good is a bonus, but matching the actual product is the bare minimum to pass.

I got burned hard by blurry logos before. The client flipped out on the spot.

I’m using real photos as a base and then swapping the background with AI. That’s my workflow now too.

The skin tone is off by half a shade, and there’s no way to hide it—looks fine to the naked eye, but the moment you take a photo, it’s busted.

The sneakiest thing is when they change details like threads—you’d never notice during acceptance.

I usually just use it for lighting/shading, and lock the subject in place with low-strength img2img inpainting.