When someone says a product photo tool is "not working", the tool has usually done exactly what it was asked. The complaint almost always means one of six specific things, and five of them are caused by the source photograph rather than by the software. Knowing which one you are looking at turns a frustrating afternoon into a two-minute fix.
The cutout has a light or dark halo
This is the most common complaint and the most misunderstood. Along the edge of the subject there is a thin fringe — light grey against a white background, or a dark outline against a coloured one.
The cause is that the original photograph has soft edges, not hard ones. Every boundary between subject and background in a real photograph is a gradient a few pixels wide, where the recorded colour is a blend of both. A segmentation model has to decide where along that gradient the subject ends, and if it guesses slightly wide or slightly narrow, the leftover blend becomes a visible fringe.
Three fixes, in order of effectiveness. Photograph the product against a background that contrasts strongly with the subject — a dark object on a white sweep, a white object on a mid-grey one. Then remove the background from the highest-resolution version you have, because a fringe two pixels wide at 4000 pixels becomes eight pixels wide at 1000. Finally, if a fringe survives, it is usually cheaper to nudge the edge manually than to keep re-running the automatic pass.
The shadow disappeared
Automatic removal does what it says: it identifies the background and removes it. If the product was resting on a surface, the contact shadow and the soft cast shadow behind it are part of that surface, so they go too.
The result looks wrong in a way that is hard to name — the object appears to float, and the eye registers it immediately even if the viewer cannot explain why.
The fix is deliberate re-lighting rather than more removal. Either keep the shadow by choosing a tool that supports separating it, or add a soft synthetic shadow underneath after the cutout. A product with no shadow reads as a sticker; a product with a shadow reads as an object.
The edges are jagged
Stair-stepped edges along curves and diagonals are a resolution problem, not a model problem. A segmentation model needs pixels to work with. Give it a 500-pixel-wide image and the best possible edge it can produce is one pixel wide on a 500-pixel image — which is a hard, visible step when the result is placed on a page.
Upscaling the output does not help, because it enlarges the steps along with everything else. The only real fix is to start from a larger original. If the original is all you have, the cutout will look acceptable at small display sizes and poor at large ones, and no setting changes that.
The colours shifted
A product photograph that looked correct in the tool but looks slightly dull, or noticeably more saturated, once published is a colour profile problem.
Photographs from a modern phone are often recorded in a wide-gamut profile such as Display P3, which holds more saturated colours than the sRGB profile that browsers and web pages assume. If the profile is carried through or converted inconsistently, colours either clip or shift.
Two things to check. Convert to sRGB before publishing rather than leaving a wide-gamut file for the browser to interpret. And be aware that converting to JPEG from a transparent PNG can alter the appearance of partially transparent edge pixels against whatever colour is flattened behind them — which is another reason a cutout that will sit on a white page should be exported on white rather than as a transparent PNG.
The file was refused, or came out enormous
Two separate problems that look like one.
A transparent PNG containing a photograph is genuinely large. Transparency defeats most of the compression a PNG relies on, so a cutout that was 400 KB as a JPEG can become 4 MB as a PNG. If the destination does not need transparency — it places the product on a white card anyway — export as JPEG on a white background and the size problem disappears.
If the upload was refused outright, the cause is usually one of three limits, and they are worth checking before anything else: a maximum file size, a minimum or maximum pixel dimension, and a restricted list of accepted formats. Marketplace requirements are specific and they are enforced automatically.
Fixing the source instead of the output
Nearly every complaint above is cheaper to prevent than to repair, and prevention happens before the photograph is taken.
- Contrast with the background. The single largest factor in cutout quality. A subject that differs clearly in tone from what is behind it is almost trivial to separate.
- Even, diffuse light. Hard light creates sharp shadows, which the model reads as part of the object or as background — the former leaves stray blobs, the latter eats real edges.
- Fill the frame. A product occupying a third of the image wastes two-thirds of your resolution on background that is about to be deleted.
- Shoot at the full resolution available. Every downstream step has more to work with, and pixels cannot be recovered later.
- Keep a copy of the original. Not the cutout, not the exported JPEG — the untouched file, so a better tool or a better method can be applied in six months without re-shooting.
The product photo maker handles cutting out and re-backgrounding a shot in the browser, and the background remover is the single-purpose version if all you need is the cutout. Both are worth testing on one image before committing a whole catalogue to a workflow — the halo question in particular is answered in about ten seconds, and it determines whether the rest of the set will be usable.