Does Amazon Allow AI-Generated Product Images? What Passes Review

Yes, Amazon allows AI-generated product images. Its Seller Central image guide contains no ban on AI as such. The rules that matter are the ones every image must meet: the picture must accurately represent the product, the main image needs a pure white background, and any photorealistic AI-generated person needs a metadata tag before upload.

That means the real risk is not "AI" but "inaccurate". An AI tool will happily add a fifth drawer to your cabinet, smooth out a weave pattern, brighten a colour past the real thing or invent a logo. Each of those fails the accuracy rule, and a listing with no compliant main image can be taken out of search until you replace it.

Follow me on Instagram@sagnikteaches

Amazon's wording is worth reading closely, so the sections below test it against one listing. The rules below come from three Seller Central help pages: the Product image guide, the Image suppression page and the page on tagging AI-generated people. I read them in November 2026. Amazon edits these pages, so treat the structure here as reliable and check the numbers against the live page before a big upload.

Connect on LinkedInSagnik Bhattacharya

What Amazon's image guide says about AI, and what it leaves out

The guide has exactly one AI-specific section, headed "AI-generated people disclosure". It says some jurisdictions may require disclosure when product images contain photorealistic AI-generated people, and that if yours do, you must tag the image with the appropriate metadata before uploading it. Amazon then adds a disclosure to tell shoppers the image includes AI-generated people, where applicable.

Subscribe on YouTube@sagnikteaches

Notice what is missing. There is no sentence saying "AI images are prohibited", none saying "label all AI images", and none requiring you to state which tool you used. An AI-generated backdrop, an AI cleaned-up cut-out of a real product, or an AI-built infographic with no people does not trigger the tagging step. What it still has to do is pass the ordinary rules, and those are strict enough on their own.

The ordinary rules fall into three groups.

• Main image rules. A realistic, professional-quality picture that shows the true scale, quantity and colour; the product taking up 85% of the frame; the entire product visible and shown once; one unit plus any included accessories; a pure white background (RGB 255, 255, 255); no packaging unless it is an important product feature; no props that are not included.
• Rules for all images. The image must accurately represent what you sell and match the product title. No nudity, no review images or five-star graphics, no claims such as "free shipping", no seller-specific information, no Amazon logos or marks, and no look-alikes of badges such as "Amazon's Choice" or "Best seller".
• Technical rules. JPEG, TIFF, PNG or non-animated GIF; at least 500 pixels and at most 10,000 pixels on the longest side; at least 72 dpi; clear and not pixelated. Amazon recommends 1,000 pixels or more, because that is what switches on zoom.

An AI image can break any of these without anyone noticing, which is why the rest of this tutorial is a checking routine rather than a debate about whether AI is allowed. If you want the creative side, such as designing infographic slots, the sibling tutorial on AI for Amazon listing images and infographics covers it. This one is about what passes review.

The listing we will test: a cork yoga mat sold by a studio

To keep this concrete, imagine a small yoga studio that sells its own branded cork mat on Amazon. The mat is 183 cm long, 61 cm wide, 5 mm thick and comes in two colours: natural cork with a charcoal underside, and natural cork with a sage-green underside. The studio has one decent phone photograph of each, taken on a grey floor, and no budget for a photographer.

The studio's owner wants seven images: a white-background main image, two angle shots, a size-comparison infographic, a lifestyle shot of someone using the mat, a detail of the grip texture, and a colour comparison. Which of these can AI make, and which are risky? Walk it through the way Amazon's reviewers and automated checks effectively do.

The rest of the tutorial goes slot by slot, using these as the examples.

Main image: the white-background test, with numbers

The main image is where most suppressions start, because Amazon's image suppression page lists the usual causes: text, logo, graphics or watermarks; a background that is not pure white; a missing image; extra items; a cropped product; the product in packaging; props that obscure or surround it; blurriness; the product too small in the frame. A cutout made by AI is exposed to several of these at once.

Take the cork mat. The studio uploads its phone photo to an AI background-removal tool and asks for a white background. The result looks white on screen. Test it properly before upload, using any image editor with an eyedropper:

1. Sample four corners and the middle of each edge. Every sample must read 255, 255, 255. A reading of 252, 252, 252 looks white to the eye but is not "pure white" in Amazon's definition.
2. Measure the product's bounding box. The guide wants the product to fill 85% of the image. For a 2,000 by 2,000 pixel square, a rolled mat lying diagonally might measure 1,880 by 1,150 pixels; the longer dimension is 94% of the frame, so the product comfortably passes the "fills the frame" requirement. If the same mat measured 1,350 pixels on its longest side, it would be at 67% and likely flagged as too small.
3. Zoom to 300% on the edges. Look for a pale grey fringe, a "halo", where the cutout tool kept background pixels. Halos make the product edge look fuzzy and are a typical cause of the blurriness flag. The tutorial on removing backgrounds from product images deals with the hardest edges, such as glass and hair.
4. Check the shadow. A soft contact shadow is common in studio photography. A shadow that turns the background below 255 across a wide area fails the pure-white test on the strictest reading; keep shadows tight and faint, or remove them from the main image.
5. Check the longest side. The file must be at least 500 pixels; Amazon recommends 1,000 pixels or more so zoom works. A 2,000-pixel square gives plenty of margin.

Here is how a failed first attempt can look in practice, written up as an illustrative log, not real data.

Illustrative pre-upload log, cork mat MAIN image

Corner samples:       255,255,255 / 255,255,255 / 254,254,254 / 255,255,255
Edge midpoints:       255,255,255 x4
Bottom edge under mat: 241,241,241  (shadow too wide)
Product longest side:  1,720 px of 2,000 px  = 86%
Halo at 300% zoom:     pale grey on left end of roll
Verdict: FAIL on shadow and halo. Re-run cutout with "no shadow", re-check.

After a second pass with the shadow switched off, the bottom edge reads 255 and the halo is gone. The point of the log is not the figures; it is that someone looked at actual pixels rather than trusting the preview.

Where AI cutouts quietly change the product

Accuracy is the harder test, because no eyedropper catches it. AI retouching can alter the product itself. Here are failures to look for on the mat photo:

• The roll's end shows slightly different thickness. The listing says 5 mm; the picture implies 8 mm.
• The cork speckle is smoothed into a uniform surface, so the natural variation customers expect is gone.
• A strap or carrying band that the studio does not include appears from nowhere. The main image may only show accessories that are included.
• The sage-green underside looks more saturated than the printed colour. That is a colour-accuracy problem and a returns problem.

The only reliable check is putting the physical mat next to the screen. Hold the real item up beside the edited image and ask whether a buyer would feel misled when it arrives. If you would hesitate, so will the customer, and a "not as described" return costs more than the photograph.

Does AI-generated people in an image need a tag?

Yes, if the people are photorealistic and were generated entirely by AI. Amazon's rule, as the guide states it, is to add the keyword contains-synthetic-performer to the dc:subject field in the image's XMP metadata, using an IPTC-compatible metadata editor. The separate help page adds that the tag applies to video as well, and that the metadata must contain an rdf:li of that keyword inside the XMP dc:subject rdf:Bag.

The guide is just as specific about when not to tag. Skip it if:

• the image shows only real people, even if you altered them with AI-powered tools;
• it shows only characters from films, video games or other expressive works;
• it shows no people at all; or
• the people in it are not photorealistic.

So the studio's real photograph of an instructor, retouched with an AI tool, needs no tag. A fully generated, photorealistic woman in a yoga pose does. A cartoon illustration of a person in a pose does not.

How to add the tag in two minutes

Amazon's help page gives three routes. On a Mac, open the file in Preview, show the inspector, use the Keywords tab and add contains-synthetic-performer. On Windows, right-click the file, choose Properties, open the Details tab and type the keyword in the Tags field. Amazon warns that Subject and Tags write to different locations, so use Tags only and leave Subject unchanged. For a batch, the command-line tool exiftool can do it in one line.

exiftool -XMP-dc:Subject+=contains-synthetic-performer lifestyle-01.jpg lifestyle-02.jpg

One detail from the page saves headaches. The += operator appends rather than overwrites, so running it twice writes the keyword twice; duplicates do not break tagging but they bloat the file. Separately, and this is not on Amazon's page, many editors' "export for web" or "save for social" options can drop metadata, so tag the final exported file, then verify that the keyword shows in the XMP dc:subject field.

There is a trap here for studios who edit in several tools. A file tagged in step one, then opened and re-saved in a social-media design tool at step three, may have lost the tag. Make tagging the last step before upload, not the first.

The lifestyle shot: the slot most likely to cause trouble

The studio's idea is to ask an image generator for "a woman doing a stretch on a cork yoga mat in a bright studio". The output is attractive. It is also, in Amazon's terms, a picture of a product that is not the studio's product: the generated mat has the wrong proportions, no charcoal underside visible, and a texture that is closer to foam than cork. Three problems follow.

1. Accuracy. The image must represent the product you sell and match the title. A different-looking mat fails that, whether or not any automated check catches it.
2. The synthetic-person tag. The woman is a photorealistic person generated by AI, so the file needs the tag.
3. Customer expectation. If the generated mat looks thicker than 5 mm, buyers notice on arrival.

A safer method keeps the real mat in the picture. Photograph the actual mat on a floor, then use an AI tool to extend the background, add a window and warm the light around it, with no generated person. Or take a real photo of a real instructor and use AI only to tidy the room. In both cases the product pixels are real, and no tag is needed because no AI-generated person appears.

Below is a prompt that keeps the product fixed. It is an example of the wording; adapt it to your tool, and note that no prompt guarantees the product stays unchanged, which is why the visual check remains.

Keep the cork yoga mat exactly as it appears in the uploaded photo: same
length, thickness, colour, speckle pattern and charcoal underside. Do not
add straps, logos or text. Change only the surroundings: replace the grey
floor with a pale wooden studio floor, soft window light from the left,
no people. Output 2000 x 2000 px.

A typical output (illustrative) will look right at thumbnail size and slightly wrong at full size: perhaps a faint green cast on the cork, or a roll end that looks cleaner than the original. Put the original and the result at the same zoom level side by side. If anything on the product itself differs, reject the result and rerun or fall back to the plain photo.

Infographics: allowed, but the numbers must be yours

Infographic slots are usually the safest place for AI help, because AI is laying out text, not inventing a product. Amazon's rule that images must accurately represent the product still governs them, and the all-images list bans review images, five-star graphics, claims such as "free shipping" and seller-specific information. Those exclusions catch a lot of infographics.

For the mat, a clean infographic shows the dimensions (183 cm by 61 cm by 5 mm), the weight if you have measured it, the material and the two underside colours. A risky one adds "Best seller", "#1 yoga mat" or "Free carry strap" because the AI suggested badges to fill space. Remove all three: "Best seller" is on Amazon's list of badges you may not copy, the ranking claim is unverifiable, and the strap is not included.

A short review of an AI-made infographic is worth doing line by line:

1. Check every number against the product itself, with a tape measure if needed.
2. Delete any claim you cannot prove, such as "non-toxic", "eco-friendly" or "hypoallergenic" unless you hold paperwork for it.
3. Confirm no Amazon-lookalike badge appears.
4. Confirm the infographic is not your main image. Text on a main image is a listed suppression reason.

For the full design routine, including slot order and text sizes, the infographics tutorial linked earlier goes further than there is room for here.

The rejection checklist: run it on every file before upload

This is the checklist the studio uses. It works for any product. Print it or paste it into your own notes, and tick each line per file.

Items 1 to 5 are the ones most often tied to suppression in the guide's own list. Items 6 to 8 are the ones AI most often breaks without any automated check noticing, and they surface later as returns and bad reviews.

Four other listings, and how each would fare

The checklist changes in emphasis by product. Here are four quick walk-throughs on different businesses, all illustrative.

A removals firm selling packing kits

The firm sells a "starter kit" of 10 boxes, tape and a marker. The danger is quantity. An AI image that shows twelve boxes, or a tape gun that is not in the kit, breaks the "one unit and included accessories" and accuracy rules. The fix is a photo of the actual kit contents, with AI used only for the background. Add an infographic listing "10 boxes, 2 tape rolls, 1 marker" typed from the packing sheet.

A landscaping company selling raised-bed kits

Garden products sell on lifestyle, so the temptation is a main image full of generated vegetables. Amazon's image guide allows a lifestyle main image only for a limited number of product types, and then with no text or extra logos, so the plain white-background version is the safer main. The AI-generated tomatoes belong in a secondary lifestyle slot, with a note in your own file that the vegetables are not included, and a caption-free image so no claim is made. If a person appears, tag it.

A tutoring centre selling a revision workbook

Printed books and workbooks use the main image to show the cover. AI is the wrong tool here: a regenerated cover with slightly altered lettering misrepresents the product. Use the real cover file from the printer, and use AI only to build an infographic of the contents page.

A yoga studio selling a branded water bottle

Reflective steel is hard for AI cutouts. The reflection may carry a bit of the original room, so the bottle looks like it sits on a coloured floor. Check the white-background reading under the bottle and along its curve, not just in the corners.

The tidy-up that changed the product and drew returns

A pattern worth knowing is the "helpful" retouch. Imagine the studio ran its charcoal-underside photo through a sharpening and "enhance" tool. The tool read the cork speckle as noise and smoothed it. The mat in the picture now looks like a plain tan rubber sheet. Nothing about it triggers a pixel check: the background is white, the size is right, the file is sharp. The failure shows up as a run of "not as described" returns and a handful of reviews saying the real mat looks different.

Nothing in Amazon's guide is needed to explain why that matters; the all-images rule is simply that the image must accurately represent the product. The lesson is to compare the final export with the unedited original and to treat any smoothing, sharpening or "enhance" button as a change to the product, not a polish. If a before-and-after pair looks different on the product surface, use the before.

What to do when an image is suppressed

If Amazon suppresses a listing, the Image suppression page says to identify the reason, then upload a compliant replacement. It recommends replacing a non-compliant main image rather than deleting it: upload the new image with the MAIN variant code and allow up to 24 hours for the change to show. Selection is also not guaranteed. The guide says uploading an image does not guarantee it will display, because Amazon selects and arranges images from multiple selling partners.

1. Open the suppression notice and read the stated reason, such as non-white background or cropped product.
2. Reopen your original, not the previous edit, and redo the job with that one reason in mind.
3. Run all thirteen lines of the checklist again, because fixing one fault often reveals another.
4. Upload with the MAIN code and wait up to 24 hours.
5. If it fails again with the same reason, send the image to someone who has not seen it and ask them to say what is wrong with it. Fresh eyes catch halos and shadows quickly.

Keep a one-line note of what was wrong and what fixed it; it is the cheapest training you can give whoever makes the next batch of images. If your catalogue is larger than a few products, assign one person to run the checklist so the responsibility is not shared by nobody.

A light process that keeps AI images inside the rules

For a small seller the process can be short:

• Start from a real photo wherever the product is in shot. AI changes the surroundings, not the item.
• Keep every original. Name edits with a suffix so you can always compare back.
• Write the facts first. Dimensions, colours and what is in the box live in one sheet, and every infographic is built from it. The tutorial on bullets and search terms with AI uses the same fact sheet for the text side.
• Tag last. If any image contains a photorealistic AI-generated person, add the tag after the final export and verify it.
• Check against the category rules. The main-image rules above are general; clothing and multipacks have their own pages, which Amazon links from the guide, and shoes have extra rules inside it. Adult clothing, for example, goes on a standing model.

If the cost side is what drives you towards AI, compare it honestly with hiring someone: the tutorial on AI product photos against a photographer prices both, and a real photo plus light AI cleaning is often the cheapest compliant route. If you sell items made by other suppliers, replacing supplier images safely covers where the images came from.

Where the rules may change next

The synthetic-performer tag is the guide's only AI-specific rule so far, and the guide itself says disclosure may be required in some jurisdictions, which hints at more movement. Check the Product image guide each time you start a new batch, and look at the tagging help page if you use video, since it says the tag applies to media generally, including video. Do not rely on a forum post or a third-party site for the rule; Amazon's own help page is the thing a reviewer will use.

Keep the question you ask of each image simple: would a customer opening the parcel recognise the product they saw? If yes, and the checklist is ticked, you have done the job Amazon's guide sets out. If you would rather have the process mapped around your own catalogue, that is part of what my AI implementation consultation covers, and you can start from the contact page.

Questions sellers ask before uploading AI images

Do I have to label every AI image I upload to Amazon?

No. Amazon's product image guide asks for a metadata tag only on images or videos that contain photorealistic people generated entirely by AI. An AI-made background, a retouched product or an infographic with no people needs no tag. The image rules about accuracy and a white main-image background still apply to it.

Can an AI image be my main image?

Yes, if it passes the same main-image rules as a photograph: pure white background, the real product filling about 85% of the frame, shown once and uncropped, with no text, logos or props. The image must also show the real scale, quantity and colour of what the customer receives.

What does Amazon do with an image that fails?

Amazon's guide says non-compliant images may be removed, and a listing with no compliant main image may be temporarily taken out of search until you supply one. Amazon may also modify submitted images to meet its requirements. You fix it by uploading a new image with the MAIN variant code.

Is an AI-generated model wearing my clothing acceptable?

It can be, but it carries the most risk. The garment must match the real one in cut, colour and detail, the model must stand for a main image, and the file needs the contains-synthetic-performer tag if the person is photorealistic and fully AI-generated. A real photo of the garment is safer.

How do I add the tag without special software?

Amazon's help page describes built-in routes: on a Mac, open the image in Preview and add the keyword in the inspector; on Windows, use the Tags field on the file's Details tab. The command-line tool exiftool also works. Afterwards, confirm the keyword appears in the XMP dc:subject field.

Further reads

• AI Tools for Ecommerce Photography: Backgrounds, Shadows, Consistent Sets — Backgrounds and shadows that stay consistent across a catalogue.
• Amazon Seller Assistant Explained: Check Its Advice First — Amazon's own assistant, and why its advice needs checking.
• Amazon Buyer-Seller Messaging: AI Replies Within the Rules — Keep AI replies to buyers inside Amazon's messaging rules.
• Amazon Seller Virtual Assistant or AI: Which Tasks Go to Whom — Decide which listing tasks go to a person or to AI.
• How to Create Product Photos With AI on a Small Budget — A low-cost route to product photos you can then edit.
• How Small Amazon Sellers Use AI for Listings and Reviews — See where else small sellers use AI on a listing.
• How to Use AI to Stage Furniture for Product Page Room Sets — Build room sets around your real product photos, check scale, colour and construction against the piece, and label the result honestly.
• Jewellery Photography With AI: Clean Backgrounds, True to Life — Honest-retouch rules for jewellers: what AI may change in a product photo, the true-colour check to run on every batch and the wording to disclose edits.
• Amazon Plan of Action: Draft an Appeal With AI, Truthfully — Build a fact file first, then let AI draft the root cause, corrective actions and preventive actions strictly from your own records.
• Amazon's 75-Character Title Rule: Review and Override AI Rewrites — Find the titles Amazon's AI has rewritten, check them against a short rubric, and put your own 75-character version live.
• Google Merchant Center Image Requirements for AI-Generated Product Photos — A pre-upload checklist for Merchant Center product images: size and format limits, the banned overlays, and how to test that AI metadata survived.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Amazon Seller Central help pages 'Product image guide' (G1881), 'Image suppression' (GT8RSE9S9NK4LP6Z) and 'How to tag media that contains an AI-generated person' (GFXHCHYZRGJRBZA5), read 5 November 2026.