Amazon Main Image Requirements: the official numbers, and four that are wrong

The specification, as published

Four numbers that are widely repeated and wrong

• "Minimum 1,600 px" — the minimum is 500 px. Zoom starts at 1,000 px. The 1,600 figure comes from an old apparel guide where it applied to height. Believing it makes you reject photos that would have passed.
• "Must be square 1:1" — Amazon publishes no aspect ratio rule and pads non-square images itself. Square is still the practical choice because padding effectively reduces your fill percentage, but it is not a rule.
• "10 MB per image maximum" — this does not appear on any current Amazon help page. Keep files sane, but do not cite it as a requirement.
• "Product fills 80%" — the figure attached to the current suppression check is 85%. Shipping at 80% is how listings get flagged by someone who thought they had cleared the bar.

The 2026 addition most guides have not caught up with

If your product image contains photorealistic AI-generated people, Amazon requires you to tag the file before uploading. The tag is the keyword contains-synthetic-performer, added to the dc:subject field (XMP) using an IPTC-compatible metadata editor. Amazon then shows a disclosure to shoppers where the law requires one.

The distinction matters for anyone using AI to put a product on a model. If the person in the image was generated entirely by AI, tag it. If the image only contains real people — even ones altered with AI tools — do not. The same applies to fictional characters, to images with no people, and to people who are not photorealistic.

Practical consequence: an AI-generated model shot is not simply an image you upload. It is an image plus a metadata field. Most product-photo tools strip or ignore IPTC fields on export, so you have to re-add the tag after the tool has finished.

What gets a listing suppressed that no tool will warn you about

• An off-white background — including a clean studio backdrop that measures 250,250,250 instead of 255,255,255.
• A border on any edge, even one pixel. A white image with a thin frame still fails.
• A soft shadow that runs off the bottom of the frame. The background can be perfect and the image still fails on this alone.
• A halo left where the background was removed — a pale fringe around the product edge.
• The product touching or being cropped by the frame edge.
• Two views of the product in one image, or more than one unit shown.
• Any part of a mannequin, of any material or finish, including a hanger.

Where AI actually helps, and where it does not

AI helps with the mechanical part: separating a product from a messy background and placing it on a clean one at a predictable size. That is genuinely hours of work saved per hundred SKUs.

AI does not help with the judgement part: whether the label in the generated image still matches the label on the unit you ship, whether a prop is part of the product, whether a reflection is physically plausible. Amazon's rules are written about the product, and a model does not know what your product is.

The rule of thumb that survives contact with these marketplaces: use AI to produce a draft, then check the draft against the spec by hand — corners for colour value, fill for percentage, edges for halo, and the product itself for accuracy. Tools that skip the check are not compliant tools, they are faster ways to produce non-compliant images.

Sources

Every figure above comes from the pages below, read on the date shown. Marketplace policies change without announcement, so if you are about to rebuild a catalogue on one of these numbers, open the source and confirm it still says what it said.

• Amazon Seller Central — Product image guide — read Oct 26, 2026
• Amazon Seller Central — Image suppression — read Oct 26, 2026