Amazon ai generated model images fashion · 15 min read
Amazon AI-Generated Model Images in Fashion: When to Tag, How to Tag, What to Re-upload
Which AI fashion images need Amazon's contains-synthetic-performer tag, how to add it in bulk with ExifTool, what to re-upload, and how it fits Google and EU AI rules.
By Surajnarayanan SUpdated
At a glance
- Tag only images that show a photorealistic AI-generated person, never real-model shots. Amazon says not to tag those
- One keyword, one field: contains-synthetic-performer in XMP dc:subject, written before upload
- Amazon's tag isn't Google's DigitalSourceType and doesn't cover EU AI Act visible labelling. Plan all three
This guide explains platform rules as published on 8 October 2026. It is not legal advice. Check Amazon's current Seller Central help pages, and your own counsel for the laws that apply to you.
Direct answer
Amazon requires sellers to add the keyword contains-synthetic-performer to the XMP dc:subject metadata of any image or video showing a photorealistic person generated entirely by AI, before upload. In fashion, an AI model generated from mannequin, flat-lay or packshot inputs needs the tag. Real models retouched or garment-edited with AI don't, and nor do product-only shots. Amazon staff on the Seller Central forums have said older qualifying images should be tagged retroactively, though earlier staff replies framed that as best practice.
Best for
- +Amazon fashion sellers and brand teams using AI-generated models on listing images, A+ content, Stores or video
- +Catalogue teams turning mannequin, flat-lay or supplier packshots into on-model frames, who need to know which outputs need the tag
- +Ops leads planning a retroactive audit of live ASIN imagery
- +Brands selling on Amazon plus Google Shopping and/or into the EU, who want one metadata plan
Not designed for
- -Legal advice. This page explains published platform rules. Check Amazon's current help and your counsel
- -Avoiding disclosure. Nothing here helps anyone skip a required tag
- -DIY generator tutorials, prompt recipes or free-credit funnels
- -Retelling the full channel image guidelines, EU AI Act or Merchant Center guides
- -Children or minors in any imagery; celebrity likeness
The rule in one paragraph
Amazon now asks sellers to tag any product image or video that shows a photorealistic person generated entirely by AI. The tag is a single keyword, contains-synthetic-performer, written into the file's XMP dc:subject field (the standard IPTC “Keywords” field) before you upload. Amazon's help page specifies it as an rdf:li entry inside the dc:subject rdf:Bag. When the keyword is present, Amazon says it will add an indicator telling shoppers the image includes AI-generated people, “where applicable”. The seller announcement says the requirement covers “any of our worldwide stores”.
You don't tag images that only show real people (even if AI tools altered them), characters from films, TV or games, non-photorealistic people, or no people at all. That last case covers most packshots and flat-lays. The exemptions are listed in Amazon's A+ content guidelines and the announcement.
Amazon doesn't describe what the shopper indicator looks like or exactly when it appears, and CNBC reported that it is unclear what criteria Amazon will apply. This page doesn't guess.
Where the rule came from, and the dates that matter
Amazon's announcement says “recent legislation requires disclosure when images or videos in advertisements contain photorealistic AI-generated people.” CNBC reported on 23 July 2026 that Amazon told sellers about the change the previous day, after New York's synthetic-performer advertising law took effect.
New York General Business Law § 396-b defines a “synthetic performer” as a digital asset, created or modified using generative AI or a software algorithm, meant to look like a human performer “who is not recognizable as any identifiable natural performer”. It requires a conspicuous disclosure in ads where the advertiser has actual knowledge, and sets civil penalties of $1,000 for a first violation and $5,000 for later ones.
Amazon's announcement applies the requirement across its worldwide stores, not only New York. Three dates get quoted, and they come from different sources. Keep them separate rather than merging them into one “effective date”:
- 6 June 2026: Amazon staff on the Seller Central forums have said compliance has been mandatory since this date. That comes from a forum staff reply, not a help page.
- 9 June 2026: the New York Governor's office announced that the state's synthetic-performer disclosure law was in effect.
- 22–23 July 2026: Amazon told sellers about the tag requirement on 22 July, according to CNBC's 23 July report of the announcement.
Which AI fashion images need the tag (decision table)
Amazon's test, in its own words, is whether the image features people “entirely generated by AI (not based on a real person)”. Here is how that maps to common fashion production routes. Where Amazon hasn't ruled, we say so.
| Production route | Person in frame | Tag? | Why |
|---|---|---|---|
| Mannequin or ghost shot turned into an AI-generated model wearing the garment | Photoreal AI person | Yes | Entirely AI-generated person |
| Flat-lay or supplier packshot turned into on-model with an AI model | Photoreal AI person | Yes | Same as above. See supplier packshots and mannequin / flat-lay |
| Detail crop (neckline, cuff, hem) cut from an AI-model frame that still shows part of the AI person | Partial photoreal AI person | Yes (cautious default) | Amazon staff on the forums have said to apply the metadata if the content still depicts a photorealistic AI-generated human |
| AI-generated lifestyle or A+ scene with an AI model or AI bystanders | Photoreal AI people | Yes | Applies to any photoreal AI person in frame |
| AI video or try-on clip with a synthetic model | Photoreal AI person | Yes | Amazon's help examples include tagging a video file |
| Real model photo with AI retouching, relighting or background replacement | Real person | No | “Only features real people (even if they have been altered with AI tools)” |
| Real model photo with an AI garment or colourway swap | Real person | No (for this tag) | Still a real person. Garment truth is a separate QC duty, see garment fidelity |
| Packshot, flat-lay, ghost mannequin or AI-generated mannequin with no human | No person | No | “Doesn't feature any people” |
| Illustrated or clearly non-photoreal figure (sketch, cartoon) | Non-photoreal | No | “Doesn't feature photorealistic people” |
| AI-generated face on a real model's body | Mixed | Grey zone; conservative reading: tag | Amazon hasn't ruled. See the grey-zone section below |
| AI “digital twin” of a real, consenting model | Based on a real person | Grey zone | Amazon hasn't ruled. See the grey-zone section below |
Safety rule for bulk work: don't tag a whole catalogue “just in case”. Amazon's guidance says not to provide the metadata for real-people images. Tag only the files your classification step marks Yes, plus any grey-zone items you have decided to tag.
For A+ content only, Amazon also offers an “AI-generated people” checkbox in A+ Content Manager's Creative Assets, and Amazon applies the metadata for you. Amazon staff on the forums have also said that images made with Amazon's own A+ generative features, such as Creative Studio, are tagged automatically, and that for listing images, Stores and ads, tagging the file before upload is the only option. Those last two points come from forum replies, so re-check them against the help pages.
Grey zones fashion teams actually hit (read cautiously)
Amazon hasn't published a ruling on these. What follows is a cautious reading, not a rule. Document your decision for each route and re-check Amazon's help pages.
AI face on a real body
The body, pose and garment come from a real shoot, but the face is generated. Amazon exempts images that “only feature real people”, and the face in this frame isn't a real person. New York's definition covers assets “created, reproduced, or modified” so that the performer is “not recognizable as any identifiable natural performer”. A conservative reading is that the person shown is no longer a real, identifiable individual, so tag it. If you choose not to, have counsel sign off and keep the reasoning on file.
Licensed digital twin of a real model
Amazon's wording ties the tag to people “not based on a real person”. New York's definition turns on whether the performer is recognisable as an identifiable natural performer. A consented twin of a known model may fall outside both, but other likeness and consent rules can apply. This needs a legal read, not a guess. Our identity lock guide covers casting continuity, not this legal question. How a fully generated, brand-owned model differs from a real model's twin is in AI fashion models.
Partial people
Hands-only or torso-only crops from an AI-model frame: Amazon staff on the forums have said to tag if the content still depicts a photorealistic AI-generated human. Treating AI-model detail crops as Yes is the cautious default.
Composites
A real model plus AI-generated bystanders in a lifestyle scene: the AI bystanders are photoreal AI people, so the cautious reading is to tag.
How to add the tag to a single file (built-in tools)
Amazon's help page gives three routes. Use whichever fits your team, then verify.
- macOS (Preview): open the file, press Cmd+I (or Tools → Show Inspector), open the Keywords tab, click +, type contains-synthetic-performer, press Enter.
- Windows (File Explorer): right-click the file → Properties → Details. In Tags, type contains-synthetic-performer. Amazon's help says to use only the Tags field and leave Subject unchanged, because they write to different metadata locations.
- Command line (any OS): ExifTool, below.
Bulk tagging with ExifTool (tested commands)
ExifTool is a free, widely used metadata tool. Amazon's own help page uses it. The commands below were tested with ExifTool 13.25 on JPG and PNG files on 8 October 2026.
1. Work on a copy
Use a folder that holds only images classified “Tag = Yes”.
2. Check what's there now
exiftool -G1 -XMP-dc:Subject your-image.jpg3. Tag one file (Amazon's documented command)
exiftool -XMP-dc:Subject+=contains-synthetic-performer your-image.jpg+= appends to existing keywords without overwriting them. Amazon's help notes that running it twice writes the keyword twice. That doesn't break tagging, but it bloats the file. ExifTool keeps a _original backup unless you add -overwrite_original.
4. Tag a whole folder (and subfolders), safe to re-run without duplicates
exiftool -r -ext jpg -ext jpeg -ext png -overwrite_original -XMP-dc:Subject-=contains-synthetic-performer -XMP-dc:Subject+=contains-synthetic-performer ./amazon-ai-model-images/-= then += removes any existing copy of the keyword and adds one back, so the result is a single entry. Other keywords (SKU codes, colour names) stay in place. -overwrite_original skips backups, which is why step 1 says work on a copy.
5. Find files still missing the tag
exiftool -r -m -ext jpg -ext jpeg -ext png -if 'not $XMP-dc:Subject or $XMP-dc:Subject !~ /\bcontains-synthetic-performer\b/' -p '$Directory/$FileName' ./amazon-ai-model-images/On Windows Command Prompt, swap the single quotes for double quotes.
Prefer not to use the command line to check? Drop the final files into our free AI image label checker; it reads the tag in your browser.
6. Export an audit sheet for your records
exiftool -r -csv -ext jpg -ext jpeg -ext png -XMP-dc:Subject ./amazon-ai-model-images/ > amazon-tag-audit.csvTag last, verify last
Resizing, compression, some “export for web” settings and some DAM or CDN steps can strip metadata. Run the tag step on the final upload file, then run step 5 on that same file before it goes to Seller Central.
Retroactive tagging: auditing images already live on your ASINs
Amazon's help pages describe tagging “before upload” and don't mention a deadline or retroactivity. Amazon staff on the Seller Central forums have said that compliance has been mandatory since 6 June 2026, that older qualifying images should be tagged retroactively, and that non-compliance could lead to potential ASIN removal. Earlier staff replies framed updating older images as best practice for new edits. The staff answers conflict, so treat re-tagging as the safe default and check Amazon's current help pages.
A practical audit for a fashion catalogue:
- 1. Inventory. Export the image slots in use per ASIN: main, additional, A+, Brand Story, Store, video. Forum staff have described all buyer-facing media as in scope; the help pages name listing images and A+ content.
- 2. Trace each live image to its production route. Use your PIM, DAM or production records (mannequin → AI model, real shoot, packshot, and so on). A file name alone rarely tells you whether the person is real.
- 3. Classify each image with the decision table: Yes / No / Grey (with an owner's decision).
- 4. Tag the master files for every Yes (and every grey-zone item you've decided to tag), using the bulk command on a copy folder.
- 5. Re-upload. This is our practical reading, not an Amazon statement: Amazon reads the tag from the file you upload, so tagging your local copy is unlikely to change an image that's already live. Plan to upload the tagged file again. For A+ content you can use the Creative Assets checkbox on re-upload instead.
- 6. Keep the CSV audit sheet (step 6 above) with the date and who classified each route.
- 7. Gate new work. Add “Amazon AI-person tag present / correctly absent” to your pre-upload QC so the backlog doesn't grow again.
Where to start
Prioritise main and additional images on your highest-traffic ASINs first, then A+, Stores and video.
Live ASINs / Safe default
Trace, classify, tag, re-upload
Forum staff answers conflict, so plan for re-tagging rather than waiting.
Trace
Link every live image to its production route from PIM, DAM or production records.
Classify
Yes, No or Grey with a named owner, using the decision table.
Tag
Tag the master files for Yes and decided grey-zone items, on a copy folder.
Re-upload
Upload the tagged final file again and keep the CSV audit sheet.
ControlNever bulk-tag real-model images to clear the backlog faster.
Amazon's tag vs Google Merchant Center vs EU AI Act Article 50
Three rules, three different triggers. One file can carry two of them, but metadata alone doesn't cover the third.
| Rule | Amazon | Google Merchant Center | EU AI Act Article 50 |
|---|---|---|---|
| Trigger | Photoreal person entirely generated by AI | Any image created using generative AI (people or not) | Providers: machine-readable marking of AI outputs. Deployers: visible disclosure of deepfakes |
| What you add | Keyword contains-synthetic-performer in XMP dc:subject | IPTC DigitalSourceType (e.g. TrainedAlgorithmicMedia). Don't strip it | Deployers: a label perceivable by people. The Commission FAQ says deployers “cannot simply rely on the machine-readable marking” |
| When | Before upload. Forum staff have said mandatory since 6 June 2026, including older images; earlier staff replies said best practice | Ongoing Merchant Center requirement | Applies from 2 August 2026. Content generated before then need not be labelled retroactively (encouraged) |
| Source | Amazon Seller Central Help | Merchant Center Help 14743464 | Commission FAQ; Regulation (EU) 2024/1689 |
- An AI-model PDP image sold on Amazon and in Google Shopping should carry both the Amazon keyword and the IPTC DigitalSourceType. A real-model image with AI retouching gets no Amazon keyword, but may still need a DigitalSourceType if generative AI changed it. Follow Merchant Center's codes as published (CompositeSynthetic exists for composites). Details are in our Merchant Center guide.
- For EU shoppers, a photoreal synthetic model can meet the deepfake test, and a metadata keyword isn't a visible label. Whether Amazon's own indicator satisfies EU deployer duties isn't established, so don't assume it does. The same goes for New York's disclosure duty: don't assume Amazon's tag or indicator covers it for you. See our EU AI Act guide.
- Visual specs (backgrounds, crops, main-image rules) are a separate layer, covered in image guidelines by channel. Zalando takes the opposite approach to visible labels; see AI-generated product images on Zalando.
Writing both metadata fields in one pass (tested)
exiftool -r -ext jpg -ext jpeg -ext png -overwrite_original -XMP-dc:Subject-=contains-synthetic-performer -XMP-dc:Subject+=contains-synthetic-performer -XMP-iptcExt:DigitalSourceType=http://cv.iptc.org/newscodes/digitalsourcetype/trainedAlgorithmicMedia ./amazon-and-google-ai-model-images/Run this only on images that are AI-model images for both channels. Match the DigitalSourceType code to how each image was actually produced.
Disclosure doesn't replace garment truth
The tag tells shoppers a person is AI-generated. It says nothing about whether the dress in the image is the dress in the box. Print scale, neckline, closures, true colour and fabric drape still need review against the reference. That's the Accept / Rework / Hold gate.
Add the tag check to the same gate. Tag present for AI-person frames. Tag absent for real-model and product-only frames. Correct tag status is part of Accept, not an afterthought at upload.
"After generation, an agent tool checks the garment against the reference — layout, logos, QC. If those are wrong, the image goes back for rework. Frames that pass go to our retouching partners for the final clean. Nothing ships until that retouch pass is done."
DIY tagging vs managed channel delivery
Fair to DIY. Tagging itself is easy. The hard parts are classification at volume, keeping metadata intact through export, re-uploading a live backlog, and keeping Amazon, Google and EU requirements straight on the same drop calendar.
| Question | DIY / in-house | Managed AI Catalogue Production |
|---|---|---|
| Classify every frame by production route | Your team, from production records | Known at creation, because the studio made the frame |
| Tag + verify on final export | Your team runs ExifTool or editor steps | Can be added as part of channel delivery |
| Garment-truth QC | Separate review step | Accept / Rework / Hold before delivery |
| Multi-channel metadata | Operator homework per channel | Optimised to the published rules per channel |
| When it fits | Low volume, clear records, a skilled operator | Recurring volume (~250–350 jobs/day class) that needs one delivery owner |
When the question is self-serve versus studio in general, read AI fashion tools vs managed production.
Decision guide
- 1. Does the frame show a photorealistic person? If no → no Amazon tag.
- 2. Is that person entirely AI-generated (not a real model)? If yes → tag. If it's a real model, even AI-retouched or garment-swapped → don't tag.
- 3. Is it a mixed case (AI face on a real body, digital twin)? → Grey. The conservative reading is to tag AI faces. Get a legal read on twins and record the decision.
- 4. Is it a detail crop from an AI-model frame showing part of the person? → Tag (cautious default).
- 5. Is the image already live? → Tag the master and re-upload as the safe default (forum staff answers on retroactivity conflict).
- 6. Does it also go to Google Shopping? → Add IPTC DigitalSourceType too.
- 7. Do EU shoppers see it? → Plan a visible disclosure. Metadata alone isn't enough for deployers.
- 8. Is volume high and production records thin? → Route new production to AI Catalogue Production, so classification and channel delivery come with the frames.
Where 8M Studio fits
AI Catalogue Production is the soft fit when AI-model images for Amazon (and other channels) should arrive channel-ready instead of becoming your team's homework. Clients brief continuity and acceptance rules. The studio runs generation from mannequin, flat-lay or supplier inputs, then garment-truth review, retouch, SKU-mapped delivery, and channel-optimised files within an agreed scope. See managed catalogue production.
A production partner, not an app
If the prior question is whether a brief should use an AI model at all, read when not to use AI fashion images first.
"They send us mannequin or flat-lay shots; we turn those into catalogue-ready frames, handle SKUs, retouching and QC. That's a new kind of production partner, not a DIY app and not a traditional studio."
Channel rules as managed work
"Brands shouldn't have to juggle Shopify vs Amazon vs marketplace image rules themselves. We already have the tools — and we'll keep building them — so we optimise the images to those channel rules for them, instead of leaving the brand to DIY it."
The Amazon AI-person tag
We can add the tag to AI-model images as part of channel delivery.
Custom AI Production Systems fits when tagging rules, QC gates, identity continuity and multi-channel exports (Amazon keyword, Merchant Center DigitalSourceType, EU labelling workflow) must share one governed pipeline. See Custom AI Production Systems. If the brief is a launch family rather than PDP volume, that is AI Campaign Production.
What 8M Studio is not
8M Studio is not a self-serve generator and not a compliance law firm. Clients stay responsible for their own listings and legal read. There's no rate card on this page.
The ideal first call is a small or mid-size ecommerce fashion brand already selling well, roughly 250–350 product-image jobs a day, that wants managed production. Photo studios adding AI who need a partner fit the same conversation. That band describes who the work suits; it is not a capacity promise. UK brands are served remotely (for UK fashion brands); this page does not claim a UK office.
"We're an AI-native service business. We're not an app, and we're not a generic IT services shop."
Buyer checklist before your next Amazon upload
- 1. List every production route in your catalogue (real shoot, AI model, packshot, ghost, composite).
- 2. Classify each route with the decision table, and record grey-zone decisions with an owner.
- 3. Tag only “Yes” files. Never bulk-tag real-model images.
- 4. Tag the final export file, then verify with ExifTool or your editor.
- 5. Use the A+ “AI-generated people” checkbox where it applies, and tag the file itself for everything else.
- 6. Audit live ASINs: trace, classify, tag masters, re-upload, keep the CSV.
- 7. Add IPTC DigitalSourceType for Google Shopping where generative AI made or changed the image.
- 8. Plan visible disclosure for EU audiences where a synthetic model could read as real.
- 9. Keep garment-truth QC (Accept / Rework / Hold) separate from tag status. Both must pass.
- 10. Re-check Amazon's help pages. Forum answers have changed before.
Not legal advice. Check Amazon's current help pages and your counsel.
Frequently asked questions
Do AI-generated model images on Amazon need a tag?
Yes. If an image or video shows a photorealistic person generated entirely by AI, Amazon requires the keyword contains-synthetic-performer in the file's XMP dc:subject metadata before upload. Amazon says it then adds a shopper indicator where applicable.
What is the contains-synthetic-performer tag?
It's the exact metadata keyword Amazon uses to flag photorealistic AI-generated people in product media. It goes in the XMP dc:subject (IPTC Keywords) field as one entry in the keyword list. You add it with an IPTC-compatible editor, macOS Preview, Windows File Explorer's Tags field, or ExifTool.
Does a real model retouched or garment-swapped with AI need the tag?
No. Amazon says not to provide the metadata when an image only features real people, even if AI tools altered them. Garment accuracy still needs its own QC check.
Do packshots, flat-lays and ghost mannequin images need it?
No. Amazon's rule only applies when a photorealistic AI-generated person is in the frame, and product-only images feature no people. The tag becomes relevant when those inputs are turned into on-model images with an AI model.
How do I tag hundreds of images at once?
Use ExifTool on a copy folder that contains only images classified as needing the tag. exiftool -r -ext jpg -ext png -overwrite_original -XMP-dc:Subject-=contains-synthetic-performer -XMP-dc:Subject+=contains-synthetic-performer ./folder/ adds a single keyword without duplicates and keeps your other keywords. Verify the final export file before upload.
Do I have to re-tag images already live on my ASINs?
Amazon staff on the Seller Central forums have said older qualifying images should be tagged retroactively, and warned non-compliance could lead to potential ASIN removal. Earlier staff replies called it best practice. Re-tagging and re-uploading is the safe default; check Amazon's current help pages.
Does Amazon's tag also satisfy Google Merchant Center or the EU AI Act?
No, not on its own. Google Merchant Center asks for IPTC DigitalSourceType metadata on any generative-AI image. The EU AI Act expects deployers to disclose deepfakes visibly rather than rely on machine-readable marking. One file can carry both metadata fields, but EU labelling is a separate step.
Is this legal advice, and does 8M Studio handle it for clients?
This guide isn't legal advice. It explains Amazon's published rules as of 8 October 2026, so check Amazon's current policy and your counsel. 8M Studio optimises AI catalogue images to Amazon, Shopify and marketplace rules as part of managed AI Catalogue Production. It can add the contains-synthetic-performer tag to AI-model images as part of channel delivery.
Evidence and sources
- Amazon Seller Central Help — How to tag media that contains an AI-generated person - Keyword contains-synthetic-performer in XMP dc:subject (rdf:li in the rdf:Bag), Preview and File Explorer steps, ExifTool examples, duplicate behaviour and verification. Sign-in may be required. Verified 2026-10-08.
- Amazon Seller Central Help — A+ content guidelines (AI-generated people disclosure) - Tag photorealistic AI-generated people before upload, the A+ Creative Assets checkbox, when to provide the metadata and the do-not-tag cases. Verified 2026-10-08.
- Amazon seller announcement (Seller Central forum copy) - “You must now tag AI-generated people in listings and A+ content”: worldwide stores, indicator “where applicable”, exemptions. Verified 2026-10-08.
- Seller Central forum — Amazon staff reply on scope and retroactivity - Forum staff reply, not a policy page: all buyer-facing media, mandatory since 6 June 2026, retroactive tagging, potential ASIN removal, partial people, Creative Studio. Earlier staff replies elsewhere framed older images as best practice. Verified 2026-10-08.
- CNBC — Amazon makes sellers label AI-generated people in images after NY law (23 July 2026) - Reports Amazon told sellers on 22 July 2026; real people altered with AI are exempt; criteria for the indicator are unclear. Verified 2026-10-08.
- New York Governor's office — synthetic performer disclosure law in effect (9 June 2026) - Announces the law requiring disclosure when advertisements include AI-generated synthetic performers. Verified 2026-10-08.
- New York General Business Law § 396-b - Statutory definition of “synthetic performer”, the actual-knowledge disclosure duty, penalties and exemptions. Verified 2026-10-08.
- Google Merchant Center Help — AI-generated content - Generative-AI images must carry IPTC DigitalSourceType metadata (for example TrainedAlgorithmicMedia), which must not be removed. Verified 2026-10-08.
- European Commission — Transparency obligations under Article 50 AI Act (FAQ) - Applies from 2 August 2026; deployers must disclose deepfakes visibly and cannot simply rely on machine-readable marking; no retroactive labelling required. Verified 2026-10-08.
- EUR-Lex — Regulation (EU) 2024/1689 (AI Act) - Primary text of Article 50. Not legal advice.
- ExifTool by Phil Harvey - Free metadata tool used for the commands on this page (tested with ExifTool 13.25 on 2026-10-08).
Continue the decision
Classify, tag, verify, then book the call
Classify every AI-model frame, tag only the photoreal AI people, verify the final file, and re-upload what's already live. If you'd rather have AI catalogue images delivered to your channel rules than run that workflow yourself, book a production-fit call for managed AI Catalogue Production.