Amazon product image requirements (2026): the rules, the specs, and AI-generated images

Short answer: the main image shows only the product you actually sell — a professional photograph of the real item on a pure white background (RGB 255/255/255), filling at least 85 percent of the frame, with no logos, watermarks, badges or text overlays. Technically you need at least 1,000 pixels on the longest side before zoom activates at all; 1,600 × 1,600 is the size worth aiming for and 10,000 pixels is the ceiling. Accepted formats are JPEG, PNG, TIFF and non-animated GIF.

The secondary images are far freer — lifestyle scenes, infographics, size charts — and they carry a large share of your conversion. This guide covers both rule sets and ends with the question almost no requirements page answers: what applies when the images are generated by AI. Current as of August 2026.

The main image: where Amazon does not negotiate

The main image is the only image that appears in search results. It decides whether your offer gets clicked at all — and it carries the strictest requirements in the whole set. Non-compliant images get removed, and in the worse case the offer is suppressed.

• Pure white background: RGB 255/255/255. A slightly grey studio backdrop is not pure white even when it looks like it on your monitor — check the value with an eyedropper, not with your eye.
• The product fills at least 85 percent of the frame. Leaving a generous margin to be safe makes your offer look smaller than the one next to it in the results list.
• Only what is actually in the box. Props that are not included — the flowers in the vase, the phone in the stand, the accessory from the larger bundle — do not belong in the main image.
• No logos, watermarks, borders, badges or text overlays. That explicitly includes Prime and Amazon marks, which sellers are not allowed to add themselves.
• A professional photograph of the real product. No drawing, no rendering placeholder, and no packaging shot standing in for the item.

Technical specs: pixels, formats, zoom

The technical limits take a minute to explain and cost more revenue than any wording decision. An image without zoom cannot be enlarged on a phone — and an image shot too small cannot be rescued later.

• At least 1,000 pixels on the longest side. Zoom only becomes available above that threshold.
• 1,600 × 1,600 is the recommended size: enough headroom for zoom without needlessly heavy files.
• Up to 10,000 pixels on the longest side is allowed. The extra resolution pays off mainly for products with fine detail — textiles, engravings, labelling, stitching.
• Accepted file formats: JPEG, PNG, TIFF and non-animated GIF. Animated GIFs are not accepted.
• Shoot large once and scale down. The reverse — upscaling an 800-pixel file — is visible in the soft edges.

Secondary images: what actually converts

The main-image rules do not apply to the rest of the set. Here you can stage, annotate and compare — and this is where it is decided whether a shopper who clicked also buys.

• Lifestyle scene: the product in use, in a realistic setting, with the audience you are writing for.
• Infographic: one feature per image, labelled large. Not five features in one graphic that nobody can read on a phone.
• Size chart or scale reference: something a shopper can measure the product against. Someone who misjudges the size returns it.
• Application steps: how the product is assembled, filled or maintained. That answers the questions which otherwise show up as criticism in reviews.
• What is in the box: exactly what ships. What the main image is not allowed to show belongs here.

Two constraints that are easy to miss

First: whether and how people may appear in product images depends on the category — apparel is governed differently from, say, products aimed at children. These rules differ between categories and can differ between marketplaces. Check the image requirements for your own category in Seller Central before you plan a shoot; there is no single blanket answer.

Second: whatever you write into an infographic is a product claim like any other. Materials, certifications, performance promises and comparisons have to be provable in an image exactly as they do in a bullet point. Graphic text gets read and checked.

On mobile, the image decides the click

More than half of all purchases happen in the mobile app. On a phone screen the main image in the results list is larger than the title and is seen before it. In practice that means the image decides the click, and the text only decides the purchase afterwards.

Two things follow. The product in the main image has to be large enough to stay recognisable at thumbnail size — the 85 percent rule is not a formality here, it is a competitive advantage. And every label on an infographic has to be readable without zooming. Review your image set once in the app, not in the desktop preview.

AI-generated product images: permitted — under the same rules

AI-generated product images are allowed. The image requirements do not distinguish by how an image was produced; they distinguish by what it shows. So an AI render is held to exactly what a studio photograph is held to: pure white for the main image, 85 percent frame fill, no overlays — and above all, the product depicted must be the product you actually sell.

That last point is where the risk sits, and it is the part almost every guide leaves out. Image models invent. They add a handle the real product does not have, give a plastic part a metal finish, put a fictional certification seal on the packaging, or build a scene that implies a property the product does not possess. A render showing a feature that does not exist is the same violation as an invented text claim — except the buyer discovers this one while unboxing.

None of which is an argument against AI images. It is an argument for quality control before upload. Five questions are enough:

• Does the render match the physical product? Put a real photograph next to it and compare shape, proportions, material, colour, controls and labelling.
• Is every claim in the graphic text backed by evidence? Each number, material and certification needs a source in the spec sheet or test report — including the ones the model dropped into the layout in passing.
• Does the main image still obey the main-image rules? Pure white rather than a near-white gradient, 85 percent frame fill, and no logos, seals or lettering the model added on its own.
• Are fantasy props removed? Anything that does not ship has to come out of the main image — including the accessory the model helpfully placed beside it.
• Does the scene imply something untrue? A product in the rain implies water resistance; a pan on an induction hob implies induction compatibility. If the property is not documented, the scene has to go.

Rejection reasons that keep coming back

• A background that is nearly white but not 255/255/255 — the most common failure in self-shot photography and in automatic background removal.
• A product that fills too little of the frame and misses the 85 percent requirement.
• Text, a logo, a watermark or a badge on the main image, Prime and Amazon marks included.
• Props in the main image that are not part of what ships.
• Fewer than 1,000 pixels on the longest side: no zoom, and visibly soft on a high-resolution phone screen.
• An animated GIF, when only non-animated GIFs are accepted.
• An AI render showing a feature, a material or a seal the real product does not have.

Keeping this up in practice

For five products this is a checklist. For five hundred products across variations and languages it is a process — and the mistakes do not happen on the first image, they happen on the two hundredth, when nobody is holding an eyedropper to the background any more.

That is where tooling earns its place. Torch Listings builds image sets from the product photos you already have, holds them to the main-image and format requirements, and checks every claim in an image against the evidence on file instead of inventing it — anything unprovable is blocked.

What no tool takes off your hands: comparing the result against the physical product. Someone who has held the item should look at the image set once before it goes live.