Filling the frame properly

Amazon states the rule — the product must fill at least 85% of the image — and does not state the geometry. Since "85% of the frame" could reasonably mean area, long edge, or bounding box, the interpretation matters, and getting it wrong in the cautious direction is what leaves a listing looking small and distant next to its competitors.

What 85% means here

We apply it as: the product's **bounding box** reaches at least 85% of the frame along its own long edge, with the other side keeping the aspect ratio and not overflowing. That is the strictest reasonable reading — reading it as 85% *by area* would let a product take only about 92% of the long edge in one direction and much less in the other, and it is the reading that leaves you closest to what the marketplace's own examples look like. Because it is an interpretation rather than policy, the fill is a slider, not a fixed number: if you read the guideline differently, move it. The presets set 85% for Amazon, 90% for Etsy, Shopify, Trendyol and eBay, and 86% for the social shapes.

Margins quietly break the rule

This is worth stating on its own, because it is a trap we walked into ourselves. If you add a comfortable margin around the frame — say 2% on each side — and *then* fill 85% of what is left, the product ends up filling about 81.6% of the actual image and the marketplace rule is missed by a tool that was trying to be tidy. The default margin here is therefore **zero**, and the audit compares the fill against the *whole* frame rather than against the padded area inside it. If you want breathing room, open the margin deliberately and raise the fill to compensate.

Anchoring: centre or base

Centring the bounding box is right when the frame has no shadow — the product sits in the middle and nothing else is in the picture. As soon as you add a contact shadow, centring looks wrong, because the shadow adds visual weight below the product and the composition reads as if the object is sliding off the top. Anchoring to the **base** instead puts the product's contact point where the eye expects the ground to be. The default here is base anchoring with the shadow on, and that combination is what makes an isolated product on white look like a photograph rather than a sticker.

One planner, two outputs

The frame maths is deliberately separated from the drawing: a pure function computes the plan — scale, position, shadow geometry — and a second step draws that plan onto a canvas. The preview, the downloaded file and the verification tests all consume the *same* plan. That is why what you see on screen is what lands in the file: there is no second code path that could drift from the first. It is a small architectural point with a very practical consequence, and it is the reason a 168-combination test sweep can meaningfully claim anything about what you will download.