Amazon Main Image Compliance: White Background, Frame Fill, and the Rules That Changed

The core rule, stable and unchanged

Amazon's main image requirement has stayed consistent: a pure white background at exactly RGB 255, 255, 255, the actual product rather than a stand-in or icon, filling 85% or more of the frame, with no text, watermark, logo, prop or badge graphic added on top. The file needs to fall between 500 and 10,000 pixels on the longest side, in JPEG, TIFF, PNG or non-animated GIF, with anything above 1,000 pixels preferred so the on-page zoom feature functions. None of that is new for 2026 — it is the same baseline Amazon has enforced for years, and most "new main image rules" content circulating is really this same baseline restated with a fresh publish date.

It's worth naming why that confusion happens so often: a genuinely new rule (the July 2026 title-length change, the AI-disclosure requirement below) gets published, and a wave of listicle content restates the entire image and title spec around it "for 2026," folding old, unchanged rules and new ones into one undifferentiated list. A seller skimming that content has no way to tell which line item is actually new without checking each one against Amazon's own current statement — which is the entire reason this page separates the two.

What actually changed: AI-generated people need disclosure

The real 2026 change is narrower than most coverage of it suggests. Amazon now requires sellers to disclose images containing photorealistic AI-generated people — full bodies, faces, hands — by checking an "AI-generated" or "AI-generated people" box during upload through A+ Content. Simple AI-assisted editing — background removal, colour correction, lighting adjustment on a real product photo — is treated the same as traditional photo editing and needs no disclosure, as long as the final image still accurately represents the physical product. Compliance is mandatory for new uploads going forward; Amazon has not required sellers to retroactively update images already live, though updating older images at the same time as other planned edits is treated as good practice rather than a requirement.

The distinction Amazon is drawing tracks a real difference in risk. A photorealistic invented person carries a different kind of trust question than a colour-corrected photo of a real product — a shopper reading a lifestyle image with a person in it is implicitly trusting that the scene shown is at least plausible, and a disclosure label lets Amazon (and eventually the shopper) know when that person doesn't exist. It is a narrower, more specific rule than "AI images need approval," and treating it as the latter leads sellers to over-disclose or avoid useful editing tools that were never actually in scope.

The method-doesn't-matter principle

This is a genuinely useful reframe for a catalogue team deciding how to source images at scale: the compliance question was never "is this AI," and treating it that way leads to either wasted caution (avoiding a legitimate AI tool for fear of a rule that isn't there) or a false sense of safety (assuming a photograph is automatically fine when it happens to be inaccurate in some other way). The accuracy standard is the same regardless of tooling, which simplifies the actual review checklist even as it doesn't simplify the sourcing decision.

What matters to Amazon's compliance check is not whether an image came from a camera, a 3D render, or an AI generation tool — it is whether the final result accurately represents the actual product: correct colour, correct dimensions, correct included contents, correct materials. An AI-generated white-background main image that gets the product genuinely right is compliant in the same way a studio photograph is. An AI-generated image that shows a feature the product does not have, a colourway that isn't sold, or contents that aren't included is non-compliant for the same reason a misleading photograph would be — the rule was never really about the camera, it's about accuracy.

A worked example: two AI-assisted images, two outcomes

Lay both images side by side against the same two-question checklist before uploading either: does this contain a photorealistic AI-generated person, and does the product shown match what actually ships. Answering those two questions independently, rather than treating "it used AI" as one combined verdict, is what separates a five-second disclosure checkbox from an actual compliance problem.

Image one: a real product photograph with the background swapped to pure white using an AI tool, colour-corrected, no people in the shot. That is standard editing under the current rule — no disclosure box needed, and it needs to pass the same white-background and frame-fill checks any photograph would. Image two: a fully AI-generated lifestyle shot showing a photorealistic person holding the product. That image needs the AI-generated-people disclosure checkbox at upload, separate from and in addition to meeting the accuracy standard — the product itself still has to be represented correctly regardless of the disclosure. Treating both images as "AI, so the same rule applies" misses that only one of them triggers the new disclosure requirement.

What to do when an image gets rejected under the new rule

If an upload is rejected for a missing AI-people disclosure, the fix is checking the correct box at re-upload rather than assuming the image itself is the problem — this is a metadata requirement, not a ban. If an AI-generated image is rejected on accuracy grounds instead, the underlying issue is the same one that would sink a photograph: the product shown does not match what ships. Amazon's own guidance draws this line clearly, and conflating a disclosure failure with an accuracy failure leads to fixing the wrong thing.

A practical habit for any catalogue using AI tools at all: keep a simple internal log of which images used AI-generated people versus which used AI-assisted editing on a real photograph. When a rejection notice arrives, that log answers the disclosure question in seconds instead of requiring someone to reconstruct how an image was made months after the fact — which, for a catalogue with several photographers and tools in rotation, is not always obvious just from looking at the file.

Why this distinction matters for a working catalogue

These come from Full Circle, a full-service Amazon management company with $500M+ in managed revenue across 100+ brands. Dr. CRO generates listing images from product-truth reference images of the real product rather than inventing features from a text prompt alone, which is the practical answer to the accuracy standard above — an image grounded in what the product actually looks like passes the same check a photograph has to pass, disclosure box or not.