The Product Image QA Pass to Run Before You Publish

Catalogue images almost never fail while someone is working on them. They fail after publication: a pale halo along a bag strap, one background two shades warmer than the other eleven, a shadow falling left in a row where everything else falls right, the charcoal variant sitting on the black listing.

None of that was hidden. It was simply never looked at by anyone whose job was to look. Whoever ran the cutouts assumed the uploader would catch it; the uploader assumed it was approved upstream. So the first reviewer is a customer or a marketplace's automated image check, and fixing it then means pulling a live listing instead of spending two minutes.

What follows is a pre-publish pass short enough to actually get run: five checks, one named owner, on the whole batch before anything goes live.

Zoom to 100% and follow the thinnest edge

Cutouts do not fail evenly around an outline. They fail in one place — a wire handle, the gap between a strap and the body of a bag, a lock of hair against a wall of a similar tone. Fit-to-screen, all of that looks fine. Open the file at actual pixels and trace the outline, concentrating on the thinnest, lowest-contrast sections.

Two faults matter. Fringe is leftover background clinging to the edge: pale pixels invisible on white and glaring once the image lands on a dark card. Bites are the opposite — a chunk of product that left with the background, usually where product and backdrop were nearly the same tone. Inspect against the colour the image will actually be published on, never against the transparency checkerboard, whose grey squares hide pale fringing. If you hand transparent files onward, our guide to transparent PNGs covers where the alpha channel survives.

Compare backgrounds across the whole range

Opening images one at a time is why mismatched backgrounds get published: your eye recalibrates to whatever is on screen, so every file looks correct. Put the batch in one folder view at thumbnail size and the almost-white, the warm white and the actual white separate instantly.

Mixed methods are the usual culprit: some items shot on a paper sweep and left as-is, the rest cut out and placed on a colour. They will not match unless one decision covers both groups. In remover.bg you can drop the original background out and put the product on a solid or custom colour, then use that same value on every image in the range, so consistency comes from a decision rather than luck. Write the value down where the next person will find it. If the destination demands pure white, check that surface's own published rules first — Amazon's image policy shows how specific they get, and it changes, so read the current version.

Make the shadows agree

Two things must match: whether a shadow exists at all, and which way it goes. A grid where half the products float and half sit on a soft contact shadow reads as two catalogues stitched together, and so does one lit from the left on one row and the right on the next. Treat presence and direction as a range-wide rule, not a per-image choice.

Watch too for the double shadow, where the original photograph's shadow survived the cutout and a new one was added underneath. Easy to miss, impossible to unsee once someone points at it. Our notes on adding realistic shadows cover what a believable one looks like.

Prove the image belongs to the SKU

The most expensive failure here is a technically perfect image attached to the wrong product. Variants are where it happens: charcoal published as black, the small bottle on the large listing.

Check against the listing title and the spec sheet, not your memory of the shoot. Read the label in the image. Count the buttons. Confirm the connector. Review whole variant sets in one sitting rather than approving files as they arrive — variants are only wrong relative to each other.

Look at the grid on a phone

Desktop preview flatters everything. Much of your audience meets the catalogue as a grid of small squares on a handset — check your own analytics for the split, then judge the images at that size.

At that size the edge detail you agonised over disappears and the faults that survive are the ones above: tone mismatch, products at wildly different scales — a framing decision to fix at the shoot or in your own photo editor — and shadows pointing different ways. Open the collection page on a phone, scroll at normal speed, and pull anything that jumps out.

Sign-off, and where rejects go

Name one person who owns the publish gate. Without that, the checklist belongs to everyone and therefore to nobody.

When an image is rejected, send it back to its cause rather than patching the file. If the outline failed because the item was touching the backdrop, a repaired export fixes one listing while the next shoot repeats the mistake. Log the reason in a single line — "reject: strap edge, item too close to wall, reshoot with a gap" — and let those lines rewrite the shooting brief. If you push a large catalogue through the background removal API, remember that automation makes output consistent, mistakes included — sample every batch, not only the first file.

The short version

• Edge at 100%, on the colour it will be published on
• Backgrounds compared side by side
• Shadow presence and direction identical across siblings
• Image matched to the SKU and variant
• Last look at the grid on a phone