Remove the background from product photos: rules, hard cases and a batch workflow

Marketplace shoppers judge a product photo in about a second, and most of that judgement is the background. A grey studio sweep or a kitchen table says the seller did not finish the job; a clean white or transparent background makes the same product look like it came from a brand. That is why the big marketplaces regulate backgrounds so tightly, and why removing the background is the first step in a serious product-photo workflow. This guide covers the rules, the realistic limits of AI removal, and a workflow that ends with compliant images, using the background remover and the bulk product image resizer on this site — both run entirely in your browser.

Why the background decides more than the product

Three reasons. First, search: Amazon treats the main image as part of listing quality, and a main image that breaks the background rule can be suppressed from search results entirely — the product exists, but nobody sees it. Second, consistency: a catalogue grid where every product sits on the same white reads as one store; mixed backgrounds read as a jumble sale. Third, focus: on a small phone screen, a busy background competes with the product for the few pixels available. A clean cut-out also travels: your store's coloured sections, an email, a marketplace — no re-shooting.

The platform rules, in numbers

Amazon's main-image rules are mandatory, not advisory:

• Background: pure white, RGB 255, 255, 255. Not light grey, not off-white, not a lifestyle scene.
• Zoom: images should be 1,000 pixels or larger on the longest side or hover zoom does not work; Amazon recommends 2,000 pixels or more for the best zoom experience.
• Frame fill: the product must fill 85% or more of the image frame.
• Content: no text, logos, borders, watermarks or props that are not part of the product on the main image.

Shopify is looser: it recommends square images of 2048 × 2048 pixels, and its themes happily show transparent or styled backgrounds. The approach most sellers land on is one master set that satisfies Amazon — square, white, 85% fill — reused on Shopify. The product image sizes guide collects the full size rules for both platforms.

What AI background removal actually does

A modern remover builds a per-pixel mask — product or background — and outputs the product with the background made transparent. The remover on this site runs a segmentation model (BiRefNet) locally in your browser through Transformers.js — your photo is never uploaded. The honest constraints, stated on the tool page itself: images are JPEG, PNG or WebP up to 5 MB, processed one at a time, and the result is a transparent PNG at the original dimensions. The AI model downloads once (about 473 MB) and is cached by the browser, so later runs start immediately. One at a time sounds slow, but the mask is the judgement call of the whole workflow and deserves a human glance per product anyway.

The cases that fight back

Flyaway hair and fur. A mask is a boundary, and hair is thousands of boundaries. Strands that blend into a similar-coloured background get clipped or leave a halo. Shoot people and furry products against a plain, contrasting backdrop and the model's job gets dramatically easier.

Transparent and reflective products. Glassware, bottles, chrome and jewellery are partly made of their background: the camera sees the backdrop through them and reflected in them. Any remover — AI or human — has to guess which reflections belong to the object. Inspect these at 100%, and occasionally keep a natural-background shot as an additional image instead of forcing a cut-out.

Shadows. A product floating on white with no shadow looks pasted, because it is; a soft contact shadow anchors it. Removing a background usually removes the real shadow too, so either use a lifestyle image for that slot or accept a clean studio look and apply it consistently across the catalogue — consistency forgives what individual perfection cannot.

White products on white. A white trainer photographed on a white sweep has almost no edge for any algorithm to find. Photograph it on a mid-grey or coloured sweep, remove that background, then pad onto white: the final pixel values are identical, but the mask has something to bite on.

Choosing the output format

The remover's output is a transparent PNG, and PNG is the right master for a cut-out: lossless, alpha intact, editable later. What you upload depends on the destination. For a marketplace main image, transparency is not wanted at all — pad the cut-out onto a pure-white canvas and export JPEG at quality 0.85–0.9, the format Amazon prefers. For your own storefront, WebP keeps the transparency at a much smaller size than PNG, with PNG as the fallback. Keep the transparent PNG master either way: platform rules change, and re-cutting from the original photo is the slow path.

The batch workflow

1. Shoot with margin. Frame the product at roughly 70–80% of the shot against the plainest, most contrasting backdrop you have. Headroom lets you pad or crop to exactly 85% fill later, and a clean backdrop is the cheapest accuracy upgrade for a mask.
2. Remove backgrounds one product at a time. Run each master through the background remover and download the transparent PNG, named to match the product (sku123-main, sku123-back).
3. Inspect edges at 100%. Fringing, halos and clipped details announce themselves immediately against white. Re-shoot the failures now, not after upload.
4. Pad to canvas in one batch. In the bulk resizer, load the whole set of cut-outs, choose Pad to canvas at 2048 × 2048 with a #ffffff background, and export JPEG at quality 0.85–0.9. Padding centres every product identically and never distorts it — and because the resizer never enlarges, a too-small original stays visibly small, which is your re-shoot signal.
5. Download the ZIP and upload. Keep the transparent PNGs beside the exports. When a platform changes a rule, you re-export from the cut-outs instead of re-cutting photos or re-compressing exports.

Mistakes that undo the work

• Removing the background from shots that do not need it. Lifestyle and scale images are welcome as additional images. Only the main image carries the white-background rule.
• Trusting the mask on reflective products without checking. Glass and chrome are where AI masks guess; a 100% zoom check takes seconds and saves a suppressed listing.
• Upscaling a small cut-out to reach 1,000 pixels. Enlargement invents no detail; zoom will show mush, and the listing suffers next to competitors. Re-shoot bigger.
• Re-compressing marketplace exports for your own site. JPEG artefacts compound across generations. Export each destination from the transparent PNG master.