Why AI Product Photos Look Fake

The eight things that give it away

Buyers rarely think "that is AI." They think something is slightly off and move on. These are the specific causes.

The lighting does not match. Your product was photographed under soft window light and the generated scene has warm afternoon sun coming from the right. Two light sources, one image, and the brain notices before the eye does.

There is no contact shadow. Where an object meets a surface it casts a small dark area. Without it, the product appears to hover a millimetre above the table. This is the single most common tell.

Shadows point the wrong way. The scene has light from the left, the product's own shadow falls to the left. Physically impossible, immediately wrong.

The scale is off. A mug rendered at the size of a vase, or a ring that would not fit on any hand. Generated scenes have no idea how big your product actually is.

The perspective does not match. Your photo was taken from eye level, the scene is viewed from above. The product sits in the frame like a sticker.

The edges are too sharp. A cutout pasted onto a background has hard, clean edges. Real photographs have slight softness where the subject meets the background, especially at shallow depth of field.

The reflections are wrong. A glossy product should reflect the scene it is in. If it is still reflecting your kitchen while sitting on a marble counter, it reads as pasted.

Everything is too clean. No dust, no crumb, no fingerprint, no slightly crooked thing in the background. Real photographs contain small imperfections, and their absence is uncanny.

Half of it is fixed before you upload

The photograph you feed in determines most of what comes out. Five things about your source photo make a large difference.

Light it flat and even

Soft, diffused, even light is the most flexible starting point, because it can be relit convincingly in almost any direction. A source photo with hard directional light carries that direction into every generated scene, and it will conflict with the lighting the scene wants.

Overcast daylight from a window is close to ideal. Direct sun is close to the worst.

Shoot from the angle you want

If you want your product on a table viewed from slightly above, photograph it from slightly above. Perspective cannot be invented convincingly, and a mismatch here is the hardest tell to fix afterward.

Include a little of the surface

Photographing your product on a plain surface, with a bit of that surface visible around it, gives the tool a real contact shadow to work from. Cutting the product out completely removes the information that makes it sit somewhere.

Keep the background plain and contrasting

Plain, and a different tone from your product. This makes extraction clean and reduces the hard cutout edges that give the game away.

Get it sharp

Any softness in your source becomes worse in the output. This matters most on texture, which is where buyers look.

Free toolEtsy Image CheckerCheck your source photo is sharp and large enough before you generate anything from it.

What to check in the output

Four checks, in this order, before you upload anything to a listing.

The contact shadow. Look where the product meets the surface. Is there a small dark area? Does it match the direction of the light in the scene? If the product looks like it is floating, reject it.

The light direction. Find the brightest area of the scene, then find the brightest area of your product. Are they on the same side? If not, the image is contradictory no matter how good it looks otherwise.

The scale. Compare your product to anything else in the scene. A mug beside a window, a ring on a surface, a candle next to a book. Does the size make sense?

The product itself. Zoom in. Has any part of it been altered, softened or invented? On text, fine chain, engraving or intricate detail this happens more than sellers realise, and it is the one failure that turns into a return.

Which products are hardest

Some things simply resist this, and knowing which saves you time.

Hardest: anything reflective or transparent. Glass, chrome, polished metal, resin. Reflections have to match a scene that did not exist when the photo was taken, and they usually do not.

Also difficult: fine repeating detail. Chain, lace, filigree, embroidery, engraved lettering. These get softened or invented at small scale.

Easiest: matte, solid, simple shapes. Ceramics, wood, fabric, candles, soap, leather. These relight convincingly because there is nothing complicated happening at their surface.

If your product is in the first group, plan on your own photographs doing more of the work and generated scenes doing less.

The imperfection problem

The one failure that cannot be fixed by better inputs is that generated scenes are too tidy.

Real kitchen counters have a crumb. Real shelves are slightly dusty. Real linen is not perfectly folded. Generated scenes tend toward catalogue perfection, and at some threshold that reads as artificial even when nothing specific is wrong.

There is no clean fix, but there is a practical one. Do not make every photo in your listing a generated scene. A listing with three styled images and five honest photographs of the real item on your actual workbench does not trigger the same reaction, because the honest photos anchor it.

Where this leaves you

The reasonable position, if you use these tools, is that they are good at settings and bad at products.

Photograph your product yourself, properly, in flat even light, from the angle you want. Use generated scenes for the contexts you cannot stage. Check the contact shadow and the light direction on everything before it goes into a listing.

That produces images that hold up, and it is a workflow rather than a shortcut.