← All guides Buying and selling

AI product images on Amazon and Etsy

Generated product photography sells items that do not exist as pictured. What to look for as a buyer, and what sellers need to know before their listing gets removed.

· 10 min read · Best AI Image Detector

Check the physics rather than the product. Generated product images get the object convincingly right and the shadow, reflection and contact point with the surface subtly wrong, and those are where the region map concentrates.

Why generated product photography spread so fast

Product photography is expensive and slow. A studio session, a photographer, lighting, a stylist and a retoucher can cost more than a small seller's entire stock, and it has to be repeated for every variant and every seasonal refresh.

Generation removes all of that. A single reference photograph of an item can produce dozens of lifestyle shots, in different rooms, on different models, in seasons the seller has never photographed. The economics are not close.

Most of this is legitimate marketing rather than fraud. A real mug photographed on a generated kitchen counter is closer to a studio backdrop than to a lie, and platforms have been slow to draw a line because the line is genuinely hard to draw.

The problem arrives when generation moves from the setting to the product. A garment whose drape was never photographed, a colour that does not exist in the dye, a texture rendered rather than captured: these produce a buyer who returns the item and a seller who wonders why.

What to look at as a buyer

The object itself is usually convincing, because that is where generation effort concentrates. The physics of the scene is where it falls apart, and physics is cheap for a person to check.

Five checks on a product photo
  • Where the object meets the surface Contact shadows are the hardest thing to render. Look for an item floating a millimetre above its own shadow.
  • Reflections in and around it Glass, chrome and glossy packaging must reflect a room that could exist.
  • Text on the product Labels, care instructions and small print are still where generation struggles.
  • Repeated pattern Fabric weave, wood grain and knitting that drift or repeat unnaturally across the item.
  • Consistency across the listing Do the eight photographs describe one object, or eight renderings of an idea?
Where to look, in order of how often it catches something.

The last row is the strongest and the one buyers skip. Run three or four images from the same listing rather than one. A real product shoot produces variation in lighting and angle around a constant object; generation produces variation in the object itself.

Reading the map on a product shot

Studio photography scores oddly to begin with. A clean white background, even lighting and heavy retouching produce a smooth, low-noise image that already sits higher on the scale than a casual snapshot of the same item.

So the absolute number matters less here than where the heat is. A cool product on a hot background is a real item in a generated setting. A hot product on a cool background is the case that should stop a purchase.

Four patterns, and what each means for a buyer
Pattern on the mapWhat it meansRisk to you
Cool item, hot backgroundReal product, generated settingLow, this is marketing
Hot item, cool backgroundProduct rendered into a real photoHigh, the item may differ
Hot everywhereWholly generated imageHigh, nothing was photographed
Cool everywhere, one hot patchSomething removed or correctedAsk what was there
Warm and flatStudio retouchingNormal for the category

The second row is the one to act on. It describes a listing where the setting was photographed and the product was not, which is the exact shape of a picture that will not match what arrives in the box.

If you are the seller

Platform policy is moving faster here than most sellers realise, and the direction is consistent: generated imagery of the setting is broadly tolerated, generated imagery of the product is increasingly not, and undisclosed use of either is where suspensions come from.

The practical protection is the same one photographers use. Keep the original captures with their metadata, keep the layered editing files, and be able to show which parts of a published image came from a camera.

It is also worth running your own listings through a check before a competitor or a platform does. A high score on your own photography tells you which images would be hard to defend, and it is much cheaper to find that out yourself.

What actually protects a purchase

As everywhere else on this site, the decisive step is not detection. It is a specific request that a legitimate seller can satisfy in a minute and a fabricated listing cannot satisfy at all.

Ask for a photograph of the actual item, taken today, on a plain surface, next to something for scale. For handmade goods, ask for a photo of the piece in progress. Sellers with stock find this trivial and often welcome it.

The categories where it matters most

Some products survive a generated photograph and some do not. The difference is whether the picture is doing the work of a specification or the work of an advertisement.

Clothing is the worst case. Drape, weave, sheen and how a colour behaves in daylight are the whole purchase decision, and all four are exactly what generation approximates. A garment rendered rather than photographed produces a return, and often a dispute about whether the listing was accurate.

Anything sold on colour has the same problem. Paint, fabric, cosmetics, flooring and dye lots all depend on a faithful capture, and a generated image has no relationship to a physical sample. Where a colour matters, ask for a photograph in daylight against something familiar.

At the other end, a generated setting around an accurately depicted object costs a buyer nothing. Nobody expects the kitchen behind a kettle to be a real kitchen, and a rendered backdrop is closer to a studio sweep than to a misleading claim.

Questions people ask

Is it against the rules to use AI product photos?
Policy varies by platform and is changing quickly. The consistent direction is that a generated setting around a real, accurately depicted product is broadly tolerated, while generating the product itself, or using generated imagery without disclosure where disclosure is required, is what gets listings removed.
Why does a normal studio photo score high?
Because studio work is smooth, evenly lit, retouched and low in sensor noise, and all of that resembles what the model was trained to find. A clean white-background product shot sitting in the forties is ordinary rather than suspicious, which is why the region map matters more than the number.
How do I tell a generated background from a generated product?
By where the heat sits on the map. Heat around the object with the object itself cool means a real product in an artificial setting, which is marketing. Heat on the object with a cool surround means the item was rendered into a real photograph, and that is the case worth acting on.
What should I ask an Etsy seller for?
A photograph of the actual piece taken today on a plain surface, with something for scale, or a shot of the work in progress. Makers usually enjoy sending these. A seller who cannot produce one is either dropshipping or working from images rather than stock.
I am a seller and my own photos score high. Is that a problem?
Not on its own, but it is worth knowing. Keep the raw captures with metadata and the layered editing files so you can show which parts came from a camera. Running your own listings through a check before somebody else does is much cheaper than answering a suspension.
Do reviews with photos help?
Considerably, and they are underused. Customer photographs are taken in ordinary rooms on ordinary phones, which makes them both hard to fabricate at scale and a far more honest depiction of the item than any listing image. Read the picture reviews before the written ones.