Amazon vs Shopify vs Pinterest: One Product Hero, Four Ratios

We built this tool because we were that seller. One studio-shot photo of a ceramic mug, four sales channels, four different aspect ratios, and a Friday night spent re-cropping everything in Photoshop while the photographer’s invoice sat unpaid. We wanted a way to take the one good shot we already had and let the canvas grow around it, cleanly, in a few seconds, without the product itself getting touched.

We have now run the same workflow across about seventy SKUs of our own and a few dozen we tested with friends in the apparel, candle, and small-electronics niches. This post is what we found. It covers the four ratios that cover almost every e-commerce surface you publish to, the prompt hints we lean on, and the failure cases that still need a human eye.

Why one source photo is not enough

Every channel has its own crop. Amazon’s main image policy still requires a 1:1 square on a pure white background, with the product filling at least 85% of the frame. Shopify’s default home-page hero slot renders best at 1.91:1, the same ratio Facebook and LinkedIn use for link previews. Instagram’s feed favors 4:5 portrait because it eats more vertical real estate on a phone. Pinterest rewards 2:3 vertical pins because the home feed is a tall mosaic and shorter pins lose out.

If you only have one studio shot, usually 4:3 or 3:2 from the camera body, three of those four ratios mean either crushing the product or chopping its edges. Re-shooting four times is expensive. Cropping in is lossy. The cleaner answer is to leave the product alone and grow the background.

That is the whole pitch behind outpainting for e-commerce. The product is sacred. The canvas around it is generated.

The four ratios you actually need

We will walk through what each platform expects, what we put in the prompt hint, and the small audit step we run before publishing.

Amazon main image: 1:1, white seamless

Amazon’s listing engine throws up a soft warning if your main image is not square, and the policy team can suppress a listing if the background is not pure white (#FFFFFF). When we extend a 3:2 source to 1:1, we lose roughly 33% of the width or gain roughly 50% of the height, depending on which side we pad. We always pad the height, never the width, because the product was framed for a horizontal camera and we want the original composition intact.

Prompt hint we use: “pure white seamless studio background, soft contact shadow under the product, no gradient, no horizon line, no surface texture.” The phrase “soft contact shadow” matters. Without it, products start to float, and floating products look pasted. With it, the model usually grounds the object in a believable way.

Audit check: open the output, eyedrop the four corners. If any corner reads above #FAFAFA, we regenerate. Amazon’s QA bots are surprisingly strict.

You can run this directly at /aspect/1-1/.

Shopify hero: 1.91:1, lifestyle context

The home-page hero on most Shopify themes is wide and short, and a square product floating in the middle looks lonely there. We use 1.91:1 and ask the model to extend a believable surface, not a void.

Prompt hint we use for the candle SKUs: “extend warm wooden tabletop, soft afternoon window light from the left, blurred linen fabric in background, no other props, no text.” For ceramics: “extend matte concrete countertop, soft overcast daylight, slight depth of field falloff toward edges.” We avoid asking for specific props (“flowers”, “books”, “coffee cup”) because the model often miscounts or distorts them. Empty surface plus light direction is the safer recipe.

The same shot scales out cleanly to a Facebook link preview and a LinkedIn share card, both 1.91:1, so this one ratio earns its keep three times over. Try it at /aspect/1-91-1/.

Instagram feed: 4:5, vertical breathing room

Instagram lets you post 1:1 or 4:5, and 4:5 wins on engagement because it occupies more screen on a phone. From a 3:2 source, going to 4:5 means adding roughly 67% more height. Most of that height becomes background above the product.

Prompt hint we use: “extend the same studio backdrop upward, no new objects, soft falloff to slightly darker tone at the top edge for caption legibility.” That last clause matters. Instagram lets users overlay a caption, and a slightly darker top edge keeps the white text readable when someone resizes through the editor. Try it at /aspect/4-5/.

Pinterest pin: 2:3, story height

Pinterest’s algorithm visibly favors pins between 2:3 and 1:2.1. We aim for 2:3 because anything taller starts getting truncated in the home feed. From the same 3:2 source, this means adding about 125% more height. That is a lot of generated pixels, and it is the ratio where mistakes show up most.

Prompt hint we use: “extend the studio backdrop downward, allow space for a title bar at the bottom, no shadows in the lower third.” That last constraint is important because most Pinterest creators drop a title overlay near the bottom of the pin, and a generated shadow there fights the text.

A real walkthrough: one mug, four outputs

Source: a 3:2 studio shot of a matte black ceramic mug, centered, on a white sweep, soft fill from camera left, single soft shadow extending camera right. Roughly 4000 by 2667 pixels.

The three-step agent we built (vision read, prompt synthesis, outpaint generation) runs the same way every time. We upload once, pick the target ratio, and the vision step reads the source and tells the prompt step what is on screen. The prompt step writes the extension instructions. The outpaint step generates the new pixels. One credit, one image. Four ratios, four credits, about ninety seconds total.

Output 1, Amazon 1:1: square, pure white, original composition kept, contact shadow continued naturally to the right. Passed Amazon’s main image checker on the first try.

Output 2, Shopify 1.91:1: wide, mug centered slightly left of frame, white sweep extended on both sides into a clean studio environment, faint warm tone added camera left to suggest window light. Passed our team’s eyeball check.

Output 3, Instagram 4:5: vertical, mug in lower two-thirds, top third clean white with a barely-perceptible falloff for caption space. We added the caption “matte black, made to last” and it sat clean.

Output 4, Pinterest 2:3: tall, mug in upper third, lower two-thirds clean white sweep extending down, ready for a title overlay. Passed the visual scan.

Total time including the upload: under two minutes.

The failure cases we still ship by hand

This is the honest section. We do not pretend this works on every product the first time.

Glass and reflective products are the hardest. The model has to imagine what would have been reflected in the surface had the photo extended further, and it often invents a phantom light source or a phantom prop that is not in the original shot. For glass, we either shoot at the target ratio in-camera or do a smaller extension and live with the tighter crop.

Products with text labels facing the edge of the frame can lose label legibility when the model misreads the contour. We caught this on a tea-tin SKU where the brand name on the right edge ghosted slightly. The fix is to flip the source so the label is centered before extending.

Products with shadows already touching the left or right edge of the frame cause the model to misread the shadow as part of the product. The output ends up with a strange smear. The fix is to crop a few pixels of margin off the source before uploading, so the shadow has a clear endpoint inside the frame.

Cluttered or busy backgrounds in the source confuse the prompt step. If you shot the product on a wood table at a coffee shop, the model has to invent more wood and more shop, and it usually invents wrong. For e-commerce we recommend going back to a clean studio source whenever you can, and using this tool for the canvas, not for staging.

What we recommend

Shoot once, on a clean white sweep, framed for a horizontal sensor. That gives you the best raw input for every ratio.

Run the four extensions in one sitting, audit each one against the corner check, the caption-space check, and the title-overlay check before you save.

For glass, mirrors, and label-edge products, do not push the extension past 50% on any side. Crop tighter and accept the smaller canvas.

If you want the workflow start-to-finish in one place, we wrote up the e-commerce flow at /use/ecommerce/.