Eight AI product photo mistakes a customer notices before you do

There is a soft shadow under the product, but no dark line where it touches the surface. Without that contact point the object hovers just above the table. This usually means the model never worked out where the surface was, and another attempt is faster than a patch.

The product's shadow falls left and the shadow of the prop beside it falls right: two light sources, one of which does not exist. It happens most when several things in the frame were put together separately, so look twice at any scene with more than one object.

A bottle stands on a matte wooden table with a clean mirror image under it. Or a glass jar reflects a window that is not in the frame and could not be, given where the light comes from. A reflection has to belong to the surface and to the light.

This is the one customers catch first, because they read it. Your label says your brand name, and the generated version says almost your brand name, with one letter wrong or a character from no alphabet.

Zoom to full size on every piece of text in the image: the label, the box, the ingredient list. Text on packaging is the most common reason to throw a generated image away, so check it before you fall for the composition.

The cap is a slightly different shape. The handle joins the body in the wrong place. A seam moved. The bottle is a little taller than the real one.

This matters more than it looks. A customer who receives something that does not match the photo asks for a refund and writes a review, and "the picture was generated" is no defence. Compare with the real product, not with your memory of it.

Water splashing up from nowhere. A cloth draped as if a hand held it. Fruit balanced on a curve it would roll off. Steam rising from something cold. These read as advertising rather than as wrong, so they pass review, but they cost the plain credibility that makes a small shop worth buying from.

Scene lighting warms or cools the product: a white mug turns cream under a warm lamp, navy fabric goes black in shadow. Real photos do this too, so the test is whether a customer holding the delivered item would feel misled. If the answer is maybe, change it.

Fabric weave, brushed metal, leather grain, the small irregularities of handmade ceramics. They justify a higher price, and they are exactly what gets averaged into a smooth plastic surface. If you sell on texture, keep at least one real close-up photo in every listing.

1. Zoom to full size on every piece of text.
2. Find where the product meets the surface. Is there a dark contact line?
3. Follow every shadow in the frame. Do they all agree?
4. Put the generated image beside a real photo of the product. What changed?
5. Look at the material. Is the texture still there?

It takes under a minute and catches nearly all of it. The four signals behind points 1 to 3 are explained in why product photos look flat.

Generated scenes are good at setting, light and mood. They are still unreliable at small printed text and at the fine texture of a material.

So split the work: generate the lifestyle and context shots, and keep real photos for close-up detail and for anything whose label must be read. ProduktShot's product photos make up to 4 images in one go, so you can pick the one that passes the check instead of patching one that does not.