Amazon is asking sellers to disclose product imagery that contains photorealistic people generated entirely by AI. The immediate trigger is a New York rule covering synthetic performers in advertising, but the operational lesson is broader: ecommerce teams can no longer treat image generation as a file-production task with no history attached.
A convincing lifestyle image may pass a visual review and still create compliance, trust or product-accuracy risk. The team needs to know what is real, what was generated, which tool was used, who approved the output and where the asset is being served. That turns product imagery into a provenance workflow.
What the requirement actually covers
Amazon’s advertising guidance says that a photorealistic fictitious person created with generative AI is a synthetic performer. When Amazon’s own generative tools create such an asset inside the Ads Console, the platform can identify it and apply a disclosure where required. For externally produced assets, the advertiser needs a process for identifying and declaring the synthetic performer.
The rule is narrower than “all AI images are forbidden.” It focuses on realistic human depictions in advertising. A background cleanup, color correction or people-free product render is a different case. Yet ordinary image policies still apply: the visual must represent the product accurately and must not create a misleading expectation about color, scale, fit, texture or included accessories.
The minimum provenance record
Every final asset should carry a compact production record: source product photography, generation or editing tool, model and date, prompt or transformation summary, human elements used, synthetic-person status, markets and placements, approval owner and final file checksum. Keep the original product references beside the output.
This is not paperwork for its own sake. When a marketplace, regulator or customer challenges the image, the record lets the team answer quickly. It also prevents an agency handoff from turning a known synthetic asset into an unlabeled “final lifestyle image” six months later.
A pre-upload checklist for ecommerce teams
- Identify people: does the image contain a real person, a composite, a licensed stock model or a fully synthetic photorealistic person?
- Verify the product: compare color, material, dimensions, controls, packaging and included items with approved product evidence.
- Check rights: confirm ownership or permission for source photography, likenesses, locations and branded objects.
- Record the method: preserve the generation and editing trail rather than saving only the flattened final file.
- Apply disclosure: follow the platform and market requirement for each placement, including any requested metadata.
- Approve by risk: require legal or senior review for realistic people, sensitive claims and regulated categories.
Why the commercial risk is larger than compliance
Product images are evidence. If AI silently improves a fabric, changes the capacity of a bag or produces a model interaction the real product cannot support, conversion may rise before returns and complaints reveal the problem. The short-term gain then creates marketplace penalties, customer-service cost and damaged reviews.
Disclosure also affects trust. A synthetic person is not automatically deceptive, but hiding the method becomes harder to defend as platforms add labels. Brands should decide when a generated model is an acceptable production choice and when real use, fit or emotion is material to the purchase decision.
Build one workflow across marketplaces
Do not create an Amazon-only spreadsheet that lives outside the creative system. Add provenance fields to the digital asset manager, naming convention or approval tool used across Amazon, social advertising, retailer media and the brand site. The same master asset may travel through all of them.
The practical standard is simple: no final image without a known origin, an accuracy check, an accountable approver and a placement-specific disclosure decision. Generative tools can reduce production cost, but only a traceable process makes the output scalable.
This image matches the article because it separates the real product, the synthetic performer and the evidence needed to document how the image was made.
Sources
- eWeek: Amazon requires disclosure of AI-generated people
- Amazon Ads Help: synthetic performer disclosure requirements
