The short answer: Since late July 2026, Amazon has asked sellers to tag any listing image or video that shows a photorealistic person generated by AI, so that it can tell shoppers. AI backgrounds, clean-up and retouched photos of real people are not covered. Our rule is older and simpler than Amazon's: photograph the product for real, once, properly, and let AI work around the product, never on it. A generated product is a promise your parcel may not keep.
This guide covers the rule, a table of what to shoot and what to generate, a shot list for one product and the image mistakes that send orders back.
What Amazon's rule says
Amazon's notice sits behind the Seller Central login, so this summary relies on reports from seller agencies and trade press that quote it. Read the original in your own account before you act.
- What triggers it: a photorealistic person generated by AI in a product image, a lifestyle image, a video or A+ content.
- What you do: before uploading, add the keyword contains-synthetic-performer to the file's keyword metadata (the dc:subject field in XMP) using a metadata editor. Photoshop, Lightroom and similar tools can do this.
- What shoppers see: Amazon may show an indicator on the listing that the media contains AI-generated people.
- Where it applies: reports describe it as covering Amazon's stores worldwide, not only the US.
| In the image | Tag needed? |
|---|---|
| A realistic AI-generated model wearing or holding the product | Yes |
| A real model, retouched or edited with AI tools | No, as reported |
| A real photo of the product on an AI-generated background, no people | No |
| An illustration or cartoon figure that is clearly not a photograph | No |
| Product only | No |
The rule followed a New York law on "synthetic performers" in advertising, which trade press reports took effect in June 2026. That law is American. The tagging requirement, as reported, is Amazon's own and global.
A practical trap: many export and compression tools strip metadata from a file. Add the tag to the final file, then open it again and check the tag is still there before you upload.
Real or AI, image by image
The tag is the small issue. The bigger one is what AI does to accuracy. This is how we decide.
| Image | Shoot for real | AI is safe for |
|---|---|---|
| Main image | Always. The actual product, the actual pack | Removing dust, cleaning the white background |
| Angles and back of pack | Always. Shoppers read labels and ingredient lists here | Straightening, colour matching across a range |
| Scale shot | Always. A real hand or a real object beside the product | Nothing. Generated hands get proportions wrong |
| Texture and detail | Always. Fabric weave, the serum on skin, the finish of the metal | Nothing |
| Lifestyle image | The product itself, in real light | The room, the shelf, the scenery behind it |
| Infographic | The product photo at its centre | Layout ideas, icon drafts, resizing for other marketplaces |
| A person using it | A real person where you can. It is one shoot day | If you do use an AI model, tag it and expect the label |
| Planning | Mock-ups to agree the look before paying for a shoot |
Why so strict about the product? Generated products drift. A cap changes shape, a label loses a line of text, a 30 ml bottle starts to look like 100 ml. Each of those is a buyer who receives something different from the picture.
Main image specifications
These come from Amazon's own guidance for sellers. Category style guides can add to them, so check yours.
| Item | Requirement |
|---|---|
| Background | White, RGB 255, 255, 255 |
| Product size in frame | 85% or more of the frame |
| Size | 500 to 10,000 pixels on the longest side |
| For zoom | Amazon prefers more than 1,000 pixels on each side. We shoot at 2,000 or more |
| File type | JPEG (preferred), TIFF, PNG or non-animated GIF |
| Quality | Clear, not pixellated, no jagged edges |
| How many | At least one. Amazon recommends at least six |
Amazon's listing rules also keep the main image free of added text, logos, watermarks and props that are not in the box. We could not open that page without a seller login, so treat this line as widely reported and confirm it in Seller Central.
A shot list for one product
Here is a full set for Neem Lab Face Serum, 30 ml. The brand and product are made up. One product, one half-day, seven images and a video.
| # | Shot | What it must answer | Real or AI |
|---|---|---|---|
| 1 | Main: bottle and carton, front on, pure white | "What exactly am I buying?" | Real |
| 2 | Back of pack, label fully readable | "What is in it? How do I use it?" | Real |
| 3 | In a real hand | "How big is 30 ml?" | Real |
| 4 | Dropper with a drop, and a swatch on skin | "What is the texture and colour?" | Real |
| 5 | What is in the box, laid flat | "Do I get the carton, the dropper, anything else?" | Real |
| 6 | On a bathroom shelf in morning light | "Does this belong in my routine?" | Real product, background may be AI |
| 7 | Infographic: three benefits, how to use, who it is for | "Why this one?" | Real product photo, designed layout |
| 8 | Video, 20 to 30 seconds: open, dispense, apply | "Does it look like the photos?" | Real |
Shoot the product once on white with consistent light and you have the raw material for every later image: the Amazon set, the Flipkart and Nykaa crops, your own website and your ads.
Photo mistakes that cause returns
A return usually starts with a gap between the picture and the parcel. These are the gaps we see most.
Colour that is not true. A heavily warmed or brightened image makes a maroon kurta look red. Shoot with a grey card and correct to it. Do not edit until it "looks nicer".
No sense of size. Without a hand, a ruler or a familiar object, shoppers guess, and the guess is often wrong.
Props that look included. A serum photographed beside a jade roller will get a complaint from someone who expected the roller. If it is not in the box, keep it out of the main images or say so plainly.
Pack quantity left unclear. A picture of three bottles on a listing for one. Match the image to the unit being sold, and add a "what is in the box" shot.
The flattering angle only. Hiding the seam, the plain back or the thin strap delays the disappointment until delivery. Show it.
AI drift. A generated or heavily AI-edited product whose label, cap or proportions no longer match the real thing.
Old packaging. The design changed six months ago and the listing did not.
Every one of these is cheaper to fix in a photograph than in a refund.
A two-week plan
- Days 1 to 2: Audit your top ten listings against the shot list above. Mark missing shots and any image where a person or the product itself was generated.
- Days 3 to 5: Read Amazon's notice in Seller Central. Tag any image that needs it, or replace it with a real photograph.
- Week 2: Reshoot the gaps in one batch: scale, texture, back of pack and what is in the box. Check metadata on final files, then upload.
After that, compare the return reasons on those listings over the following month with the month before.
Where this fits
Images are the first thing we fix in marketplace management, because they affect clicks, conversion and returns at once. The shoot itself is planned by our product photography team so one day covers marketplaces, your own site and your ads. If you are still deciding where to sell, read marketplace or your own website.
Sources
- Amazon: Tips for taking product photos (image requirements)
- My Amazon Guy: Amazon AI image policy brings mandatory labels for AI-generated people
- GeekSeller: Amazon introduces new rules for AI-generated images
- Goat Consulting: Amazon's new rule for AI-generated images in listings
- Novadata: Amazon sellers must label AI-generated people
