How do I make AI product photos with a white background for Amazon?
Updated July 2026
Ask an AI about this guide
Open this article in your assistant with a ready-made question about doing the same job on your own API keys.
Amazon main images have hard technical requirements, and AI can meet them - if you drive it correctly and respect one big rule about accuracy. Here is the compliant workflow, from a single hero image to a full catalog.
Know the spec you are generating against
Amazon’s main-image requirements are enforced by automated checks, so "looks white" is not enough.
- Background must be pure white, RGB 255/255/255 - off-white or light gray triggers suppression.
- The product should fill at least 85% of the frame, fully visible and uncropped.
- Minimum 1,000 px on the longest side; 1,600 px or more recommended to enable zoom.
- No text, logos, watermarks, props, or inset images on the main image.
The critical rule: start from your real product
The main image must accurately represent the actual product, so the correct AI workflow is image-to-image: photograph your real product (a phone shot in decent light works), then use an image model to place it on a clean studio white background with professional lighting - not text-to-image inventing a product that does not exist. Models with strong reference fidelity (GPT Image, Nano Banana) preserve labels and proportions well; always compare the output against the real item before publishing, because misrepresentation is an account-level risk, not just an image rejection.
Prompt for the spec explicitly: "place this exact product on a pure white RGB 255,255,255 studio background, soft even lighting, product fills most of the frame, no props, no text". Then verify the corners with a color picker - models produce near-white more often than true white, and a levels adjustment is often needed as a final step.
Chat tools versus a catalog
For one hero image, ChatGPT or Gemini handles this fine. A catalog is different: 80 SKUs times main-plus-lifestyle shots means hundreds of generations, and chat interfaces give you sequential output, rate caps (users report roughly 50 per 3 hours on ChatGPT Plus, around 20 per day on Gemini free), visible watermarks on some tiers - disqualifying for Amazon - and no way to map images back to SKUs. The failure mode is not image quality; it is losing track of which output belongs to which product at row 30.