Amazon AI Model Images: What Sellers Need to Disclose

If you use an AI-generated person to model a necklace, hold a skincare product, wear clothing or appear in another Amazon product image, there is now an extra publishing step to add to your workflow. Amazon says product images and videos containing photorealistic AI-generated people must include specific metadata before they are added to listings and A+ content in its worldwide stores.

This is not a ban on AI model imagery. It is a disclosure requirement. The useful distinction for sellers is also narrower than “AI was used somewhere in the image.” Amazon says the requirement applies to photorealistic AI-generated people, while media that only features real people can be exempt even if those real people were altered with AI tools.

What Amazon now requires

Amazon's current seller announcement tells sellers to use an IPTC-compatible metadata editor and add the keyword “contains-synthetic-performer” to the dc:subject field in the file's XMP metadata. This needs to be done before the image or video is added to a listing or A+ content.

Amazon says that when this metadata is present, it will add an indicator to inform customers that the media includes AI-generated people, where applicable. The requirement applies across Amazon's worldwide stores.

For sellers, that means disclosure should become part of the asset-preparation process rather than something remembered at the final upload screen. If your team creates ten AI lifestyle images for a jewellery range, the relevant files should be tagged before they enter the listing or A+ workflow.

Which images need the tag?

The clearest case is a photorealistic person who was generated by AI. An AI-generated model wearing your bracelet, a synthetic person holding a bottle, or a generated fashion model wearing a T-shirt all fit the requirement Amazon describes.

Amazon also gives several explicit exemptions. You do not need this metadata when the media only features real people, even if those people have been altered with AI tools. The same exemption applies when there are no people, when people are not photorealistic, and for specified characters from expressive works such as films, TV programmes, streaming content or video games.

That creates an important practical distinction. Replacing the background behind a real jewellery model is not the same case as generating the model herself. Retouching the lighting on a photograph of a real person is also different from creating a photorealistic synthetic person from scratch.

Some edge cases can be less obvious, especially cropped body parts such as generated hands, ears or necks. Amazon's public announcement focuses on photorealistic AI-generated people rather than spelling out every possible crop. Sellers should avoid inventing their own exemption for ambiguous media. If the human element is synthetic and photorealistic, treat it conservatively and check the current Amazon help guidance before upload.

A simple workflow for ecommerce teams

The easiest way to avoid mistakes is to classify images when they are created. Keep three practical buckets: product-only AI imagery, imagery containing a real photographed person, and imagery containing a photorealistic AI-generated person. The third bucket is the one that needs the new Amazon metadata step.

Next, preserve a final master file before marketplace export. Add “contains-synthetic-performer” to the dc:subject XMP field using an IPTC-compatible metadata editor, then export the file in a way that preserves the metadata. If another optimisation tool strips metadata during resizing or compression, the tag can disappear, so metadata should be checked on the actual file that will be uploaded.

Finally, record which listing assets are synthetic-person images. This matters if the same creative is later reused in A+ content, another marketplace or a campaign. Treating provenance as part of the asset library is much easier than trying to remember how every image was made months later.

Why this matters particularly for jewellery sellers

Jewellery is one of the categories where AI model imagery is especially attractive. Photographing every necklace on a model, every ring on a hand and every earring on an ear is expensive and repetitive. Modern image models can reduce that workload, which makes Amazon's distinction between a synthetic model and an AI-edited photograph of a real model commercially important.

Imagine a seller has a real product photograph of a pendant. In one workflow, the pendant is composited onto a photograph of a real person and AI is used only to adjust the scene. In another, the complete model and environment are generated. Those workflows may look similar to a shopper, but Amazon's disclosure rule treats the presence of a photorealistic AI-generated person as the key trigger.

This is a useful reason to keep source information alongside generated assets. A folder full of finished JPEGs may not tell a marketplace manager whether the person was photographed or generated.

Disclosure does not make an inaccurate image acceptable

The metadata requirement solves a transparency problem. It does not solve a product-accuracy problem. A disclosed AI model image can still be a poor listing image if the model changes the product.

For jewellery, inspect the clasp, chain length, engraving, stone count, setting, metal colour and apparent scale. For packaged goods, inspect logos, labels, quantities and pack dimensions. For clothing, check construction details, colour and fit. The generated person should not be allowed to become an excuse for a generated SKU.

This is also where tools such as Lustra Studio fit naturally into a seller workflow: generation can reduce the number of shoots needed for model and lifestyle imagery, but the final marketplace asset still needs a human fidelity check and, where Amazon requires it, the correct disclosure metadata.

Do not rely on visual appearance alone

One mistake would be to decide whether an image needs disclosure by asking whether the person “looks AI-generated.” The rule is about how the media was created, not whether a reviewer can spot artifacts. Better models make provenance harder to infer visually, which is exactly why the production workflow needs to retain that information.

Another mistake is to assume that because Amazon adds a customer-facing indicator, the seller can add ordinary visible text such as “AI generated” instead of the specified metadata. Amazon's current instruction is explicit about the required XMP field and keyword, so sellers should follow that technical requirement rather than substitute their own label.

The practical rule

If an Amazon listing image, A+ asset or relevant video contains a photorealistic person generated by AI, add Amazon's specified metadata before uploading it. If it only contains a real person who was edited with AI, Amazon currently lists that as exempt from this particular requirement.

For teams producing AI imagery at scale, make this a repeatable asset-management step: record whether the human subject is real or synthetic, add the required metadata to synthetic-person files, verify that export has not stripped it, then complete the usual SKU-fidelity review. That is safer than trying to reconstruct an image's origin at publishing time.

Sources: Amazon Seller Central UK, “You must now tag AI-generated people in listings and A+ content”, reviewed 2 September 2026; Amazon Seller Central US, corresponding worldwide-store announcement, reviewed 2 September 2026.