How we test

The rule behind every page

Nothing on PicBarn describes a feature we have not run ourselves, and no marketplace number appears without a link to the platform page it came from and the date we read it. When a test fails, the failure goes on the page: the recolour tool's page says it cannot change one stripe colour inside a two-colour pattern because we tried three phrasings and all three recoloured the whole garment.

Engines we run and what we learned

• Scenes: Bria product-shot. The product is cut out and locked; only the surroundings are generated. With Bria's prompt optimiser on it painted a second blender into the frame, so the optimiser is off and the product is pinned to a fixed placement.
• Virtual try-on: FASHN try-on v1.6. Stripes and prints transfer well; a flat-lay hood is read as a collar, so hoodies should be shot on a hanger or ghost mannequin.
• Personas: FLUX Pro 1.1 Ultra in raw mode, twelve synthetic people, each reviewed by eye; one was regenerated because the first render was not a photograph.
• Edit text and recolour: FLUX Kontext, one image per call. A looser prompt recoloured black stencil lettering; the current prompt keeps letters, logos and labels.
• Infographics: MOSS-Image. Four of four test images (poster, menu, social card, A+ callouts) rendered every word correctly; the product in an infographic is drawn from a description, not from your pixels, and the tab says so.
• Main image: Bria background removal, then our own composition: product trimmed, scaled to 87% of a 2,000 × 2,000 canvas, placed on RGB 255, 255, 255. Measured result on a hot-sauce bottle: corner value 255, fill 85%, zero Amazon main-image rule failures.
• Reading photos: Florence-2 for text detection (boxes in input pixels) and for the product description that seeds an infographic.

Marketplace rules

The compliance checker and the requirements pages quote Amazon Seller Central (product image requirements, image technical file requirements, A+ guidelines, the AI-people tagging page), TikTok Shop Seller University (listing policy, listing quality, AI-generated content pages), Etsy Help (image requirements, listing image rules, create a listing), Shopify Help, eBay's picture policy and Walmart Marketplace Learn. The source list with quotes is kept in the repository and each page states when it was last checked. When a figure is ours rather than the platform's, such as the 70 px callout cap height for a 2,000 px infographic, the page says so.

What the checker measures and what it cannot

It measures pixel size, file size, format, the colour of the four corners, the bounding box of the product against the frame, whether an image is square, and whether a JPEG carries the contains-synthetic-performer keyword. It cannot judge accuracy, props, claims or poses. A green row means the measurable rule passes, not that a marketplace has approved the image.

Costs we publish

One credit is about US$0.25. The provider cost behind it is below that for every tool we run; the margin pays for the four candidates you do not keep, which are free, and for failed runs, which are refunded automatically.

Corrections

If a rule on this site is out of date, the platform's page wins. Tell us through the contact page and we will change the page and the date.