Amazon Main Image Compliance Checker
Drop your main listing photo and measure the three things that get a main image pulled: whether the background is really pure white, whether the product fills at least 85% of the frame, and whether the file is big enough to zoom. Your file never leaves your device — every pixel is read locally in your browser.
What this tool actually measures
A main listing image on a large marketplace is held to a narrower standard than the other photos in the gallery, and the standard is numeric rather than aesthetic. The background is expected to be pure white — RGB 255, 255, 255, not "looks white" — because the marketplace composites your photo onto a white page and any off-white tint shows up as a visible grey rectangle. The product is expected to fill roughly 85% of the frame, so that the thumbnail actually reads at the size shoppers see it. And the file needs to be large enough on its longest side for the zoom viewer to have something to enlarge. Those thresholds belong to one marketplace, so a shot that clears them here is not automatically safe elsewhere — Etsy measures 2000 px on the shortest side and crops your first photo to 4:3 in search results, among other differences set out in the Etsy image size and crop reference.
All three are measurable, yet the advice that circulates is still "open it in an editor and poke the corners with the colour picker". That works for one photo. It does not work for a batch, and it misses the failure mode that catches people out most often: a background that is pure white in the corners but drifts to 252 or 248 halfway up an edge, because the studio lighting fell off or a JPEG save rounded it. This tool walks the whole border band rather than four sample points, so that drift is visible instead of invisible.
The fill measurement is a bounding box, not a silhouette area. Every pixel that is more than a fixed threshold darker than white is treated as product, the smallest rectangle containing all of them is found, and that rectangle's area is divided by the area of the whole image. That is the same number a human eye is estimating when it says "the product looks small in the frame", and it is deliberately generous to oddly shaped products: a thin diagonal item gets credit for the whole box it spans, exactly as the marketplace guidance intends.
Reading the tolerance setting
Strict mode requires the background to be exactly 255 in all three channels. That is the literal rule, and it is what a background cut out and filled in an editor will produce. It is also what a JPEG will almost never produce, because JPEG compression perturbs flat areas by a level or two — so a genuinely white background re-saved as JPEG can fail strict mode without anything being wrong with it. The default setting allows four levels of drift, which absorbs JPEG noise while still catching a real off-white. Use the lenient setting only to confirm a suspicion; a background that needs ten levels of slack is visibly grey.
The main-image rules, the exact value, and the reason each one exists
Most published versions of this list stop at the number. The number on its own is hard to remember and impossible to apply to an edge case, whereas the reason behind it usually tells you what to do when your situation is not covered. Values below are for the Amazon US main image and are current as of July 2026; Amazon's own image guidance sits behind a Seller Central sign-in and its details vary by marketplace and category, so confirm anything load-bearing against your own account.
| Rule | Value Amazon asks for | Why the rule exists |
|---|---|---|
| Background | RGB 255, 255, 255 | Search results and the detail page are rendered on white. Anything even slightly off-white stops disappearing into the page and starts reading as a visible grey card behind your product — and it does so unevenly across a results grid, so one seller's 248-white looks like a printing error next to everyone else's 255. |
| Minimum longest side | 1,000 px | The floor for an image to be usable at all. Below it there is not enough detail to survive being displayed on a high-density phone screen, let alone enlarged. |
| Recommended longest side | 1,600 px or more | Zoom is available from 1,000 px, but 1,600 px is the figure Amazon's own guidance names for a zoom that is worth using — and third-party guides quote anything from 1,000 to 2,000, so treat it as the safe number rather than a hard line. The reasoning holds either way: zoom only earns its place if there is detail the thumbnail did not already show, and below this the viewer magnifies pixels rather than revealing stitching, grain or print quality. This tool warns below 1,600 px for that reason. |
| Maximum longest side | 10,000 px | An upper bound on what the image pipeline will accept and process. There is no display benefit anywhere near it — a 10,000 px file is a processing cost, not a quality gain. |
| Frame fill | 85% or more | Thumbnails are rendered into a fixed box. A product occupying 60% of its frame is shown at 60% of the size of a competitor's identical product that fills 90%, in the same grid, at the same moment. The rule is not aesthetic; it is levelling the comparison. |
| File formats | JPEG, TIFF, PNG, GIF | A short list of formats the pipeline can decode reliably and reproducibly. Animated GIFs are excluded — a main image has to be a single deterministic frame, because it is what gets cached, cropped and shown in a hundred contexts. |
| Colour mode | sRGB or CMYK | Predictable colour. A file in a wide-gamut working space renders differently in every browser that ignores its profile, which turns a returns problem ("the colour is nothing like the photo") into a systematic one. |
| Not permitted in the frame | text, logos, watermarks, badges, borders, inset images | Search results are a grid of comparable things. A "Best Seller" flash or a burned-in discount badge is advertising smuggled into a slot the marketplace controls, and once one seller does it every seller must. |
| Also not permitted | props, accessories not sold, multiple products, drawings, placeholders | The main image is a contract about what arrives in the box. A styled flat-lay with a coffee cup and a plant is a lifestyle shot — it belongs at position two onward, where the buyer already knows what they are looking at. |
One row is worth flagging as softer than the others. Specialist guidance consistently reports that mannequins are not accepted on apparel main images, with a live model or a clean flat lay used instead. That is the least well-sourced item here and category practice varies, so treat it as a strong convention rather than a hard number, and check your category's own guidance before reshooting a range. If you produce apparel or wall art through a print partner, the resolution and bleed requirements a POD supplier imposes on the artwork itself are a separate constraint that applies before any of this does.
Two things this tool checks are conventions rather than published rules, and it is worth knowing which is which. Square framing is not on the requirement list — the tool reports a non-square image as a warning, never a failure, because a 1:1 file is simply the only shape guaranteed to survive every crop the marketplace applies without your product moving. File size is reported for information only. Main-image size ceilings are generous and almost nobody hits them; background purity and resolution are what actually cost people listings. And note that resolution here is pixels and nothing else. A listing image has no physical size, so it has no DPI — if your export dialogue is asking you for one, the reason that number is meaningless on screen and essential in print explains what it is actually doing.
What getting it wrong costs, in order of severity
The failures are not equally expensive, and the cheapest one to fix is the one people obsess over. An off-white background is trivially correctable — select, fill, re-export — and the damage while it is wrong is cosmetic: your tile looks slightly dirty next to the others. A background that is genuinely grey rather than compressed is worse, because it usually means the shot was taken on white paper under warm light, and no amount of levels adjustment recovers a clean edge around the product.
Fill is the one that quietly costs money. There is no penalty notice and nothing gets removed; your product simply renders smaller than everyone else's in the same results grid, forever, and the click-through gap never shows up as a policy problem because it isn't one. Resolution is similar but capped: at 1,200 px you are listable and your zoom is worthless, so the shopper who wants to inspect the weave before buying doesn't, and either leaves or returns it later. Both of these are fixed by going back to the original camera file, never by upscaling — enlarging cannot add detail that was never captured, and every AI upscaler invents texture that your product does not have.
The content rules are the ones with real teeth, because they are the ones that can get an image removed and the listing's image quality flagged rather than merely disadvantaged. Text baked into a main image, a badge, a watermark, a second product in frame, an accessory you do not actually ship — these are the ones worth a careful look before upload, and they are also, inconveniently, the ones no measuring tool can see. If the problem turns out to be that your product is the wrong shape for the frame rather than the wrong colour, the resizer will pad and re-square a batch in one pass, and the shadow generator exists because a cut-out product on pure white often looks like it is floating until you put something under it.
What a browser can measure, and what still needs your eyes
This is the honest limit of any automated image checker, including this one. A browser can read pixels. It cannot understand them. Everything in the first list below is arithmetic on an array of numbers and is as reliable as the file you fed it; everything in the second requires a judgement about meaning that no pixel test can make.
Machine-verifiable — checked here
- Background purity. Every pixel in the outer border band is compared against your chosen tolerance, and the darkest value found is reported with its coordinate in the original image.
- Frame fill. The bounding box of everything more than 12 levels darker than white, as a percentage of total image area.
- Resolution and aspect. Read from the original file dimensions, not from the reduced canvas used for analysis.
- Transparency. Whether any non-opaque pixel exists, so you know whether a background pass came from compositing or from the file itself.
- Format. Sniffed from the file's magic bytes rather than trusted from the filename extension.
Needs a human — not checked here
- Is there a prop in the shot? A saucer under a mug is pixels like any other. Nothing in a bounding box distinguishes "part of the product" from "styling".
- Is text baked into the image? Optical character recognition could find some of it; this tool does not attempt it, and it would still miss a stylised logo or a watermark set at low opacity.
- Is this the actual product? A render, a drawing or a competitor's photo measures identically to your own photograph.
- Is that accessory included? Only you know what ships in the box.
- Colour profile. The browser has already applied and discarded any embedded ICC profile by the time the pixels reach the canvas, so a wrong working space cannot be detected from here — check it in your editor before export.
One concrete gap worth knowing about: this tool accepts WebP and Amazon's accepted-format list does not include it, so a WebP file can pass every check on this page and still be refused at upload. It also does not accept TIFF, which Amazon does. Convert to PNG or JPEG before you upload, and treat a clean report here as "the measurable things are right" rather than "this image is approved".
Frequently asked questions
My background looks white but the check says it isn't. What happened?
Almost always one of three things. The photo was shot on white paper or fabric, which photographs as light grey — around 240 to 250 — no matter how bright the room was. The white was created in an editor but the image was then saved as a JPEG, which nudges flat areas by a level or two. Or there is a soft vignette or shadow falling across the edges of the frame. The report shows you the darkest pixel it found in the border band, so you can tell which one you are dealing with: 250-ish means noise, 235-ish means the paper itself.
Why does my transparent PNG pass the background check?
Because a transparent pixel is composited onto white before anything is measured, which is exactly what happens when the marketplace renders your image on its own white page. So a clean cut-out with a transparent background will pass the white test — correctly. The report tells you an alpha channel was present so you know the pass came from compositing rather than from white pixels already in the file.
How is the 85% fill percentage calculated?
It is the area of the product's bounding box divided by the area of the full image. If a 2000 × 2000 px image has a product spanning 1900 × 1700 px, the fill is (1900 × 1700) ÷ (2000 × 2000) = 80.75%, which fails. Bring it to 1900 × 1800 and you get 85.5%, which passes. The bounding box is drawn over your image in the preview so you can see exactly what the tool considered part of the product — if it is grabbing a shadow or a stray speck, that is why your number looks low.
Where does the pixel analysis actually happen?
Right here in this tab. Your file is read into an ArrayBuffer to sniff its format, decoded with createImageBitmap, then painted onto a 2,000 px analysis canvas so getImageData can walk the border band pixel by pixel. Every figure in the report is computed from that array, and no copy of the photo is transmitted anywhere — which matters when the product is not on sale yet.