How the AI works

This is the most important sentence on the page. We do not edit your photograph. An AI model looks at it and paints a new one.

That is what makes the result look like a studio shot rather than a phone photo with the background cut out. It is also why the result is a picture of your item rather than a record of it, and why you have to check it.

The same instruction goes with every shot, with only the backdrop changing. In plain words, it says:

• keep the item exactly as it is — same shape, same colour, same logo, same label text, same wear and marks, same quantity;
• remove everything that is not the item: hands, bedsheets, tiles, cables, clutter;
• relight it as though shot with a big soft key light, and take out the yellow-green cast of a ceiling bulb;
• give it one believable contact shadow, centred, square to the camera, in a 1:1 frame;
• add no text, no watermark, no border, and no props that were not already there.

Pinning the backdrop and the lighting to the surface you picked is what makes shot one and shot forty match. It is an instruction, not a mechanism — the model follows it well, and not perfectly.

In our experience the failures cluster:

• Small text. Ingredient lists, serial numbers and fine print are the first thing to go. Treat any writing on the product as unreliable.
• Logos. A well-known mark usually survives. An unusual one can come back subtly redrawn.
• Flaws. A scratch, a scuff or a faded patch is exactly the kind of detail a model will tidy away, because it is trying to make a good photograph.
• Colour. Relighting can shift a shade. A navy can come back black, a cream can come back white.
• Count and shape. If several items overlap, the model may merge them or invent an edge where it could not see one.

Look at the shot next to the real item before you post it. Ask one question: would a buyer who received this item feel the picture told the truth?

If the answer is no — the colour moved, the damage vanished, the label reads differently — shoot it again, or use your original photo. A listing photo is a representation you are making to a buyer, and you are the one making it.

Where the condition of a specific unit matters, say so in the listing, or include one honest phone photo of the actual item alongside the clean one.

Some marketplaces are beginning to ask sellers to label AI-assisted imagery, and some require it. That obligation sits with you on the platform you are listing on, so check their rules.

Our view is simpler than any rule: a photograph that honestly shows the item you are sending is fine; one that flatters it into something else is not, however it was made.

Shots are routed through OpenRouter, which lets us choose among image models rather than being tied to one. We run whichever currently gives the most faithful result for this job.

That means the model behind your shots can change. We will not change it quietly in a way that makes results noticeably worse — but if you notice a difference between a shot taken today and one from last month, this is why.

• A photo too dark or too blurred to read. The model cannot recover detail that was never captured; it will invent it instead.
• Several unrelated products in one frame. Shoot them one at a time.
• Items where exact texture is the product — fabric weave, gemstones, grain of leather. Expect to verify those closely, every time.
• Anything worn by a person. This tool is built for objects on a surface.