AI Product Photos: Marketplace Rules and Disclosure

No marketplace requires you to label an ordinary AI-generated product photo. The rules turn on two things instead: whether a photorealistic synthetic person appears in the image, and whether the picture misrepresents the product. Amazon began requiring disclosure of AI-generated people in July 2026. Etsy requires disclosure when AI made the item. Walmart and eBay require accuracy, not labels.

That distinction is the whole subject, and almost nothing written about it gets it right. The question sellers ask is “is AI allowed?” Every rule on the books answers a different question.

The stakes are not theoretical. The EU AI Act’s transparency article applies from 2 August 2026, with fines up to 15 million euros or 3% of worldwide turnover. New York’s synthetic performer law has been in force since 9 June 2026. Both landed on the same asset type, from opposite directions.

This is what each platform and each regulator actually published, quoted from the source, with the date it was checked. It also flags the claims circulating in marketing content that are not true.

Key Takeaways

• There is no general legal duty anywhere in the US, UK or EU to label an accurate AI-generated product photo. Every real obligation is narrower than the headlines suggest.
• Trigger one is a synthetic person. Amazon now asks sellers to disclose photorealistic AI-generated people in listing imagery (Amazon staff reply, Amazon Seller Forums, July 2026). New York fines $1,000 then $5,000 per violation for undisclosed synthetic performers (nysenate.gov, July 2026).
• Trigger two is product misrepresentation, and it predates AI by decades. Walmart, eBay and Etsy all enforce accuracy rules that never mention AI and catch it anyway.
• The EU’s rule is narrower than reported. A real product shot against an AI-generated background is explicitly not a deep fake and needs no label, per the European Commission’s own worked example (digital-strategy.ec.europa.eu, July 2026).
• Google Merchant Center has the one hard technical requirement. AI-generated images must carry the IPTC DigitalSourceType metadata tag, and you must not strip it (support.google.com, July 2026).
• Meta does not require commercial advertisers to declare AI. That duty applies to social issue, election and political ads only. For everything else Meta applies the label itself.
• eBay and TikTok Shop both ship generative AI into their own seller tools. Framing marketplaces as hostile to AI imagery gets them exactly backwards.
• Promo content and the listing photo are separate surfaces with separate rules. TikTok Shop permits AI backgrounds in promo video and bans “digital renderings” on the listing image, in two different policies.

Do you have to disclose AI-generated product images?

In most cases, no. No US federal rule, no UK rule, and no EU rule requires a brand to label a product photo simply because a model generated it. The obligations that exist attach to a synthetic human appearing in the shot, or to an image that misleads about the product. An accurate AI-generated photo of a real product, with no people in it, carries no disclosure duty on any platform or in any jurisdiction covered here as of July 2026.

The confusion comes from collapsing four separate things into one word. Platform policy, advertising law, ad-platform labelling, and product-feed metadata are different systems with different triggers. A rule from one gets quoted as though it governs all four.

Sorting them out is the practical work, because the answer changes what you can ship. It also changes what you need to keep records of.

This page covers marketplaces and retail listings. The social platforms run a separate set of rules with different triggers and different mechanisms, set out in AI for social media content.

What each marketplace actually says

These are the published positions, taken from each platform’s own pages. Where a platform is silent, that silence is stated as a finding rather than filled in.

Shopify is the one row with no marketplace policy to clear, because it is your own storefront. The constraints there are platform mechanics rather than content rules, and what Shopify’s variant and media caps mean for AI imagery is the version of this question that applies. The permission question has its own answer too, and the four rulebooks that bind a Shopify store are Google Shopping, Meta, the EU AI Act and your own terms of sale.

Amazon publishes image standards that are about output, not method: pixel minimums, white background, the product filling 85% or more of the frame (sell.amazon.com, July 2026). Apparel carries extra composition rules on top, and which garment presentations Amazon accepts in the main image slot is a separate question from whether AI made the picture. Category style guides are worth reading with a date in mind, because some are badly out of step: Amazon’s published eyewear guide is stamped 2010 to 2012 and still asks for 80% frame fill against the current 85%. Those pages say nothing about generative AI either way. Separately, and recently, an Amazon staff member posted in the Amazon Seller Forums that sellers must now disclose photorealistic AI-generated people, with checkboxes for “AI-generated” and “AI-generated people” in the A+ Content upload flow. The same reply states the requirement is not retroactive for already-published content, and that simple AI retouching such as background removal and colour correction is treated like traditional photo editing (Amazon Seller Forums, July 2026).

Be precise about what that is and is not. It is an Amazon staff statement on an Amazon-owned domain, which makes it citable. It is not a policy page, and Amazon has published no enforcement deadline. Press coverage has attached a specific metadata keyword to the requirement, but that keyword appears nowhere on an Amazon-owned page, so treat the mechanism as unsettled and the requirement as real.

Walmart is the clearest of the four, and it took the opposite approach. Its Product details policy states that “Content generated by artificial intelligence must be truthful, accurate and not misleading. All claims, images and descriptions must match the actual product delivered, including size, color, quantity, materials/ingredients, features, benefits and limitations.” It adds that “You’re responsible for reviewing and substantiating AI-generated content before publication” (marketplacelearn.walmart.com, updated May 2026). There is no disclosure requirement anywhere in it. Walmart also ships a generative listing editor in Seller Center and attaches no labelling duty to its output.

Etsy splits the question in a way that catches people out. Etsy permits AI-generated products as a named category, and its Seller Policy states that “If an item is created through the use of artificial intelligence, you must disclose this in your relevant listings” (etsy.com/legal/sellers, last updated 9 June 2026). That rule is about AI making the item. Separately, Etsy requires listing photos to be “Your own photographs or video content, not stock photos, artistic renderings, or photos used by other sellers or sites.” Etsy never says “AI” in that sentence, and it does not have to. If you sell a physical object, a generated picture of it is a rendering, and the safest reading is that it does not satisfy the image rule. What that leaves an Etsy seller to actually do, from a photo taken at home, is worked through in Etsy product photos: what to shoot and what AI can fix.

eBay answered by allocating liability rather than restricting the tool. Its User Agreement says a seller assumes “full responsibility for the item offered and the accuracy and content of the listing, including listing content created using tools offered by eBay or third parties such as translation, image editing, and generative artificial intelligence tools” (ebay.com, July 2026). The binding constraints are the pre-AI ones: photos must accurately represent the item, and stock photos are banned for used, damaged or defective goods. eBay also ships an AI backdrop generator into its own listing flow, which it describes as “powered by open source deep learning, inpainting models (Stable Diffusion)” (innovation.ebayinc.com, July 2026).

TikTok Shop publishes the most detailed AI policy of the five, and it governs two surfaces with two different rules. Its AI-Generated Content page states that “AI-generated content is allowed, but you must be transparent and accurate” and, more directly, that “Content will not be restricted or penalized solely for using AI.” It names permitted uses including AI editing to “adjust lighting and colors” and “change backgrounds,” plus AI-generated virtual scenes for product demonstrations. It requires creators “to proactively disclose content that is fully generated or significantly edited using AI,” via a text note in the video or the platform’s AI-generated content toggle (seller-us.tiktok.com, updated July 2026).

The listing image is stricter and sits under a different policy. TikTok Shop’s Product Listing Policy states that “Placeholders and digital renderings of a product are not allowed” and that images “must include only what the customer will receive” (seller-us.tiktok.com, updated June 2026). So a generated background on a promo video is permitted, and a generated rendering as the listing photo is not. TikTok does not reconcile those two pages anywhere, and writing that TikTok either allows or bans AI product photos gets it wrong in both directions.

One practical constraint outranks all of that. TikTok Shop allows one video per product listing at a maximum of 5 MB, confirmed on two of its own pages. That is roughly a hundredth of what Amazon accepts, and it decides your encoding long before any policy question does. The rest of TikTok’s listing spec, including the two different pixel minimums it publishes, is in TikTok Shop product images.

The two triggers that decide every case

Strip the policies down and the same two conditions appear everywhere, in platform rules and in law. Neither of them is “you used AI.”

Trigger one: a photorealistic synthetic person. Amazon’s new disclosure requirement targets AI-generated people specifically, not AI-generated products. New York’s law, effective 9 June 2026, requires conspicuous disclosure when an ad contains a “synthetic performer,” defined as a digitally created asset “intended to create the impression that the asset is engaging in an audiovisual and/or visual performance of a human performer who is not recognizable as any identifiable natural performer.” Penalties are $1,000 for a first violation and $5,000 for subsequent ones, and it applies to any advertiser whose ads reach a New York audience regardless of where the business sits (nysenate.gov, July 2026).

The EU lands in the same place by a different route, which is covered below. The pattern is consistent: regulators went after synthetic humans, not synthetic product shots.

For anyone generating on-model imagery this is the line that matters. A packshot, a flat lay, a relit hero shot, a product composited into a styled scene: no synthetic person, no trigger. An AI model wearing the garment: trigger, in at least two jurisdictions.

Trigger two: the image misrepresents the product. This is the older rule and the one with real enforcement history. Walmart, eBay and Etsy all state it plainly, and none of them needs the word “AI” to apply it. A generated image that flatters the product, shows a colour it does not come in, or implies a size it is not, is a violation whether a camera or a model produced it. Accuracy is the constraint that survives every policy rewrite, which is why product accuracy in AI images is worth solving before compliance is even a question.

What the EU AI Act requires from 2 August 2026

Article 50 of the EU AI Act applies from 2 August 2026, and the European Commission adopted its final guidelines on 20 July 2026 (digital-strategy.ec.europa.eu, July 2026). The reporting on this has been consistently wrong in one direction: it is far narrower for brands than the coverage suggests.

Two separate obligations sit in Article 50, and only one touches a brand. The machine-readable marking duty under Article 50(2) binds providers, meaning the company that builds the AI system, not the company using it. The visible disclosure duty under Article 50(4) binds deployers, which is you when you run the tool yourself. That deployer duty triggers only when the output is a deep fake.

The Commission’s guidelines give worked examples, and they are unusually direct. Not a deep fake:

“A real product (e.g., a car) shown in an advertisement against an AI-generated background and surrounding environment as long as the ad is not likely to mislead the audience about the product’s actual representation and its characteristics and use.”

Is a deep fake, and must be visibly labelled:

“An AI-generated image of a product in advertisement or packaging that can affect the audience’s perception and mislead as to the actual product appearance, characteristics or use (e.g. making the product appear not identical to the real product, more appealing or with improved quality than in real life).”

The guidelines also treat minor edits as out of scope, naming “AI-powered colour correction, background extensions of existing content, adjustments or replacements of backgrounds for clearly aesthetic purposes, compositions and arrangements of existing products, or re-scaling of images.”

So relight, reangle, restage, swap the background, composite your real product into a scene: no label. Generate a product that looks better than the real one: label. It is the same accuracy test the marketplaces apply, written into law with a 15 million euro ceiling.

One point fashion brands should not miss. The definition of a deep fake covers invented people, and reads “persons” to include “realistic AI-generated human avatars or personas.” A photorealistic AI model wearing your garment is very likely in scope. If you generate on-model shots from a flat garment photo, assume the EU disclosure applies to those and not to your packshots. The asset-by-asset version of that call, and what a label costs when you apply one, is in labeling AI-generated fashion images.

Three more details that cut against common advice. Content generated before 2 August 2026 does not need retroactive labelling. AI-written product copy is outside Article 50(4) unless it makes health, consumer-safety or sustainability claims. And embedded metadata does not discharge the duty: the guidelines state that deployers “cannot rely on the machine-readable marking embedded in the content by the provider,” because it is not “immediately clear and distinguishable” to a person. The label has to be visible.

Where the US and UK stand

The FTC regulates deceptiveness, not AI. There is no FTC rule requiring you to disclose that an image is AI-generated. The provision that does bite is the Endorsement Guides at 16 CFR 255.2(c), which requires advertisements presenting endorsements by apparent “actual consumers” to use actual consumers “or clearly and conspicuously disclose that the persons in such advertisements are not actual consumers of the advertised product” (ecfr.gov, July 2026). Note the trigger is “not an actual consumer,” not “AI-generated.” A hired human actor has always triggered the same requirement.

The FTC has also said the opposite of what is often claimed about virtual influencers. Its own published Q&A on the Fake Reviews Rule states the rule “has no blanket prohibition on the use of AI-generated avatars in marketing” and is “drafted specifically so as to not prohibit companies from using virtual influencers” (ftc.gov, July 2026).

The UK has no statutory AI disclosure requirement. The Advertising Standards Authority put it directly: “There is no blanket legal requirement in the UK to disclose the use of AI in ads,” adding that “our existing rules apply regardless of how content is generated, edited, or targeted” (asa.org.uk, July 2026). The test is whether the audience would be misled without the disclosure.

Meta’s rule is scoped to political advertising. Meta requires advertisers to disclose AI use in social issue, elections and political ads, and states it will reject ads where the disclosure is missing. For ordinary commercial ads there is no advertiser declaration. Meta’s policy says that from 1 June 2026 it applies automated detection and adds an “AI Info” label itself, and that “No advertiser action is required” (transparency.meta.com, July 2026). The widely repeated claim that Meta now requires commercial advertisers to declare AI or have the ad rejected does not appear in Meta’s own policy text.

The technical requirement most brands miss

Google Merchant Center has the only hard, in-force, unambiguous requirement that touches an ordinary AI product image, and it is not a consumer-facing label. It is metadata.

Google states: “All images created using generative AI must contain meta data indicating that the image was AI-generated by using the IPTC DigitalSourceType TrainedAlgorithmicMedia metadata tag.” It adds the instruction most workflows break: “Don’t remove embedded metadata tags such as the IPTC DigitalSourceType property from images created using generative AI tools” (support.google.com, July 2026).

That second sentence is the practical risk, and the standards body agrees. C2PA’s own security document states that the standard “does not offer any protection against the complete removal of C2PA manifests from assets,” and its implementation guidance adds that metadata “may be routinely removed or corrupted” and that “this is common, for example, on social media platforms” (spec.c2pa.org, July 2026). IPTC puts it more bluntly still: metadata is “removable by design” (iptc.org, July 2026).

So resizing, format conversion and many CDN pipelines strip it by default. An image that left the model compliant can arrive at the feed non-compliant without anyone touching the policy question. Check the file at the end of the pipeline, not at the start.

Metadata also travels further than intended. TikTok reads C2PA Content Credentials and automatically labels content generated on other platforms, and its policy states that a platform-applied label “cannot be reversed” (seller-us.tiktok.com and tiktok.com, July 2026). TikTok says the label alone carries no penalty. It is still a label you did not choose and cannot remove.

A separate Merchant Center rule catches heavily styled imagery regardless of how it was made. Google requires the primary image to “Accurately display the entire product, and include minimal or no product staging.” A dramatic generated scene belongs in an additional image, not the main one. This is the same reason PDP image choices move conversion independently of anything about AI.

The feed’s text attributes carry their own restrictions, and they catch far more brands than the image rules do. Google’s description attribute disallows promotional wording, links and your company name, which most on-page copy contains. What the Google feed accepts against what the page accepts sets the two side by side.

On Google Ads, the July 2026 update permits AI labels rather than requiring them. Google carved an exception to its own text-overlay ban so advertisers can comply with EU, New York and India rules, and states plainly that “Use of the AI label setting in Google’s advertising products doesn’t guarantee compliance with specific regulations” (support.google.com, July 2026). Election ads keep their own separate mandatory checkbox.

What you are not required to do

Marketing content on this topic has propagated a set of claims that do not survive contact with the primary sources. Each of these is false as of July 2026.

Before the table, one example of how these claims get manufactured, because it is unusually instructive. A widely repeated line holds that TikTok Shop bans AI voiceover and bans static or animated content covering more than half the screen in product videos. Both phrases do appear on a real TikTok page. Both sit under a heading that reads “During a LIVE.” The rules govern livestreams, where the entire point is real-time interaction, and a pre-rendered video is not a livestream.

The wording of the false claim traces to one place. TikTok ships an auto-generated summary of that page as its meta description, and that summary reads “In LIVE, still or animated content over 50% of the screen, AI/pre-recorded audio, and PDP screenshots are prohibited.” The circulating claim is that sentence with the words “In LIVE” removed. The page’s own data labels the field an AI summary. An AI-written summary of an AI policy became the source of a false claim about AI policy, and it propagated because the citation checked out.

That is the failure mode to guard against generally. The link is real, the page is real, the phrase is on it, and the rule still does not say what the article says it says.

Being accurate about the ceiling matters as much as being accurate about the floor. Over-disclosing is not free. It puts a label on creative that does not need one, and it trains a team to treat a real obligation as boilerplate.

A checklist before you publish AI product content

Run this per asset, not per campaign. The triggers are asset-level.

1. Does a photorealistic person appear? If yes, expect disclosure duties in the EU and New York, plus Amazon’s checkbox on listing imagery. If the person is presented as a customer or reviewer, the FTC’s actual-consumer rule applies on top.
2. Does the image depict the product accurately? Colour, size, materials, quantity, condition, and what is in the box. This is the rule every marketplace enforces, and the one with the longest enforcement history.
3. Is the item itself AI-made? On Etsy that requires disclosure in the listing description. On a marketplace selling physical goods it usually means the listing photo needs to be a real photo.
4. Did the metadata survive? Check the IPTC DigitalSourceType tag is still on the file after every resize and export, before it reaches a Google product feed.
5. Is the asset used commercially, and does your plan permit that? On DesignerBox the commercial license starts at Pro, which is $35 a month. Confirm current terms on the pricing page before you run generated imagery on a live listing.
6. Keep the source photo. Every accuracy dispute is resolved by comparing the published image against the real product. Generating everything from your actual product shot rather than from a text prompt is what makes that comparison easy to win.

That last point is the structural answer rather than a compliance one. When every asset derives from a photograph of the real item, the accuracy trigger mostly stops being a risk, because the output is your product rather than a lookalike. That is how DesignerBox is built: one product photo in, packshots, flat lays, on-model shots and video out, across 13 image and video models on one subscription. An image costs 5 credits, and you can start free with 112.

For the platform specs those assets have to meet once they clear policy, see how many listing image slots each marketplace actually shows and what affordable AI video for product listings costs, plus the per-platform detail on Amazon’s four video slots and Etsy’s listing video rules. To compare what each model holds accurately, browse the model catalog.

The bulk catalog processor workflow applies one compliant treatment across a whole catalogue rather than per listing.

FAQ

Do I need to disclose AI-generated images on Amazon?

Only when the image contains photorealistic AI-generated people. An Amazon staff member posted in the Amazon Seller Forums in July 2026 that sellers must disclose these, with “AI-generated” and “AI-generated people” checkboxes in the A+ Content upload flow, and that the requirement is not retroactive. Amazon treats simple AI retouching such as background removal and colour correction like traditional photo editing.

Does Etsy allow AI-generated product photos?

Etsy allows AI-generated items and requires you to disclose in the listing description when AI created the item. Listing photographs are governed separately: Etsy requires “Your own photographs or video content, not stock photos, artistic renderings, or photos used by other sellers or sites.” For a physical product, a generated image is a rendering, so use a real photo of the finished item.

Is AI product photography allowed on eBay and Walmart?

Yes on both, and neither requires disclosure. eBay’s User Agreement explicitly names “generative artificial intelligence tools” as a normal way to make listing content and makes the seller responsible for accuracy. Walmart requires AI content to be “truthful, accurate and not misleading.” eBay also ships its own AI backdrop generator inside its listing flow.

Does the EU AI Act require labelling AI product images?

Only if the image is a deep fake. Article 50 applies from 2 August 2026, and the European Commission’s guidelines state a real product shown against an AI-generated background is not a deep fake as long as it does not mislead. An AI-generated image that makes the product look better than it is does require a visible label, as does a photorealistic AI-generated person.

Do I have to label AI images in Facebook or Instagram ads?

Not for ordinary commercial ads. Meta requires advertiser disclosure for social issue, elections and political ads only. For everything else Meta says it detects AI-generated media automatically and applies an “AI Info” label itself, with no advertiser action required. Claims that Meta rejects commercial ads for undeclared AI do not appear in Meta’s published policy.

What metadata do AI product images need for Google Shopping?

Google Merchant Center requires images created with generative AI to carry the IPTC DigitalSourceType TrainedAlgorithmicMedia tag, and instructs sellers not to remove it. The practical failure is silent: resizing, format conversion and CDN pipelines often strip metadata, so verify the tag survives your export before the image reaches the feed.

Can I use AI-generated models wearing my clothes?

Technically yes on every platform reviewed, but this is the highest-disclosure use case. A photorealistic AI person triggers Amazon’s listing disclosure, New York’s synthetic performer law, and the EU’s deep fake rule. If the model is presented as a customer or reviewer, the FTC’s actual-consumer disclosure applies as well. Non-photorealistic or clearly stylised figures sit outside most of these.

Does TikTok Shop allow AI-generated product videos?

Yes. TikTok Shop’s AI content policy states that “Content will not be restricted or penalized solely for using AI,” and names AI background changes and AI-generated virtual scenes as permitted uses. You must disclose content that is fully generated or significantly edited by AI. The listing image is stricter and sits under a separate policy that bans “digital renderings” of the product.

Does AI-generated product copy need disclosure?

Generally no. The European Commission’s Article 50 guidelines place company advertisements and product descriptions outside the deployer disclosure duty, with an exception for claims about health, consumer safety or sustainability. Marketplace accuracy rules still apply to the words exactly as they do to the images.

Sources

• Article 50 deployer duty, the deep fake definition, the worked product-advertising examples and the minor-edit exclusions: (digital-strategy.ec.europa.eu, guidelines adopted 20 July 2026)
• Amazon’s disclosure of photorealistic AI-generated people, the A+ Content checkboxes and the non-retroactive scope: (Amazon Seller Forums, July 2026)
• Amazon listing image standards, pixel minimums and frame coverage: (sell.amazon.com, July 2026)
• Walmart’s requirement that AI-generated content be truthful, accurate and not misleading: (marketplacelearn.walmart.com, updated May 2026)
• Etsy’s AI-item disclosure rule and the own-photographs listing image rule: (etsy.com/legal/sellers, last updated 9 June 2026)
• eBay’s User Agreement clause on seller responsibility for generative AI listing content: (ebay.com, July 2026)
• eBay’s AI backdrop generator and the models behind it: (innovation.ebayinc.com, July 2026)
• TikTok Shop’s AI-generated content policy, permitted uses and creator disclosure: (seller-us.tiktok.com, updated July 2026)
• TikTok Shop’s Product Listing Policy ban on placeholders and digital renderings: (seller-us.tiktok.com, updated June 2026)
• TikTok’s reading of C2PA Content Credentials and the irreversible platform label: (seller-us.tiktok.com and tiktok.com, July 2026)
• Google Merchant Center’s IPTC DigitalSourceType requirement, the do-not-strip instruction, the primary-image staging rule and the Google Ads AI label setting: (support.google.com, July 2026)
• C2PA on manifest removal and metadata durability: (spec.c2pa.org, July 2026)
• IPTC on metadata being removable by design: (iptc.org, July 2026)
• New York’s synthetic performer disclosure law, definition and penalties: (nysenate.gov, July 2026)
• FTC Endorsement Guides at 16 CFR 255.2(c) on actual-consumer disclosure: (ecfr.gov, July 2026)
• FTC Q&A stating the Fake Reviews Rule does not prohibit virtual influencers: (ftc.gov, July 2026)
• ASA position that there is no blanket UK legal requirement to disclose AI in ads: (asa.org.uk, July 2026)
• Meta’s AI disclosure scope for social issue, election and political ads, and its automatic “AI Info” labelling: (transparency.meta.com, July 2026)
• DesignerBox pricing, credit costs, plan allocations and feature gating verified against live product configuration, July 2026

Marketplace policies verified from Amazon, Walmart, Etsy, eBay and TikTok Shop seller documentation as of July 2026. TikTok Shop policies are US Seller Center pages and are not generalised to other regions. EU obligations verified from the European Commission’s Article 50 guidelines adopted 20 July 2026. US and UK positions verified from ftc.gov, ecfr.gov, nysenate.gov and asa.org.uk as of July 2026. Platform policies in this area are changing quickly, and this is not legal advice. Individual results vary.