Contains-synthetic-performer checker · 9 min read
AI Image Label Checker: Amazon contains-synthetic-performer and Google DigitalSourceType
Drop in your final product images. See which ones carry Amazon's AI-person keyword, Google's AI source tag and Zalando-style invisible marking, and get the exact command to fix the rest. Runs in your browser; nothing is uploaded.
By Surajnarayanan SUpdated
Channel rules as published when we read them: Amazon on 8 October 2026; Google, IPTC and Zalando on 9 October 2026. This is a metadata reader, not legal advice or a compliance check.
Direct answer
This free checker reads the metadata inside your product images in your browser. It shows whether each file carries Amazon's contains-synthetic-performer keyword, Google Merchant Center's IPTC DigitalSourceType and C2PA Content Credentials, then gives the ExifTool command for what's missing. It reads files, not pixels, so it can't tell whether an image is AI-generated.
Best for
- +Amazon, Google Shopping and Zalando sellers checking AI-made product images before upload
- +Catalogue teams auditing a mixed folder of AI-model, real-model and product-only images
- +Ops leads who want a CSV record of which files carry which label
Not designed for
- -Detecting whether an image is AI-generated. It reads metadata, not pixels
- -Legal advice or a compliance certificate
- -Checking live product URLs or video files
Check your product images
Drop the final files you plan to upload, pick the channels you sell on and say how the images were made. The checker reads each file's metadata and gives a verdict per channel: Pass, Missing, Fix, Check or Not needed. Where something is missing, it builds the ExifTool command to add only that.
Drop final product images or a folder here
JPG, PNG, WebP, TIFF, GIF, AVIF or HEIC. Up to 500 files, 100 MB each.
Runs entirely in your browser. Your images are never uploaded or stored, and the checker still works if you go offline after the page loads.
Not sure whether a garment swap on a real model counts? Our guide to Amazon's AI-generated model image rule walks through each fashion production route. Route A usually applies when supplier packshots are turned into on-model PDP images with an AI-generated model.
Check the exact file you upload. Resizing, "export for web", DAM renditions and CDN recompression can strip metadata after tagging, so run the tag step last. For Google Shopping, save the live product image and drop it in. The checker doesn't fetch URLs, so nothing leaves your browser.
AI label rules by channel (as of October 2026)
Amazon, Google and Zalando each ask for a different label in a different place, and EU rules add a visible-label question that metadata can't answer. The table summarises what each channel's published help asked for when we read it in October 2026. Re-check the live pages before a large upload.
| Channel | What it asks for | Where it goes | Applies to | Source |
|---|---|---|---|---|
| Amazon | The keyword contains-synthetic-performer | XMP dc:subject, as one rdf:li in the rdf:Bag, written before upload | Images and videos showing a photorealistic person generated entirely by AI. Not real people altered with AI, and not product-only images | Amazon Seller Central help, read 8 Oct 2026 |
| Google Shopping | IPTC DigitalSourceType, for example TrainedAlgorithmicMedia, kept and not removed | XMP Iptc4xmpExt:DigitalSourceType | Any image created using generative AI, person or not | Merchant Center Help: AI-generated content, read 9 Oct 2026 |
| Zalando | Invisible AI marking, required by December 2026. No visible on-image labels | Metadata or an invisible watermark | AI-generated images. Zalando also rejects low-quality or unnatural AI images, and content that legally needs a visible label, including deepfakes | Zalando Partner image guidelines, page dated 1 Oct 2026, read 9 Oct 2026 |
| EU shoppers | A visible label for deepfakes, decided by the deployer (usually the brand) | On or next to the image, not in metadata | AI images that could falsely appear authentic, depending on context. Not legal advice | European Commission Article 50 FAQ |
Google's rule covers every generative-AI image, not just people. See how Merchant Center handles AI product images for the field and the codes. Metadata isn't a visible label. Whether EU shoppers need one is covered in our EU AI Act guide for AI fashion product images.
Labels are one layer. Size, ratio and background rules differ too; see the fashion image guidelines by channel. Selling into the UK and EU from one catalogue? See AI catalogue production for UK fashion brands.
The five production routes and what each channel expects
The same file can pass on one channel and fail on another, because each rule depends on how the image was made. That's why the checker asks for a route. Pick one for the whole batch, then change single files where a folder mixes AI-generated models, real models and product-only shots.
| Route | Amazon keyword | Google DigitalSourceType | Zalando marking | EU visible label |
|---|---|---|---|---|
| A. AI-generated person | Needed | Needed (trainedAlgorithmicMedia) | Needed (metadata or watermark); deepfake caution | Decision needed |
| B. Real model, AI-edited | Don't add it | Asked for if generative AI changed the image (compositeWithTrainedAlgorithmicMedia) | Needed; deepfake caution if the face or body changed | May apply, depending on the edit |
| C. No person, AI-made | Not needed | Needed | Needed | No person shown; check with counsel if unsure |
| D. No generative AI | Don't add it | Not needed | Not needed | Not needed |
| E. Not sure | Check | Check | Check | Check |
Verdicts mean: Pass, the label the channel's help names is in the file. Missing, it isn't. Fix, something is there but wrong, such as a case variant, the wrong field, or Amazon's keyword on a real-model photo. Check, a person has to decide. Not needed, the channel's rule doesn't apply to this route. A pass is never a compliance certificate.
Try-on images made in ChatGPT are generative-AI images too; here's how they compare to catalogue output.
How the checker reads a file
The checker opens each file with JavaScript in your browser and reads the metadata blocks, not the pixels. It parses the standard places each format stores XMP, legacy IPTC keywords and C2PA Content Credentials, then falls back to a byte scan for formats it can't parse fully. No image bytes are sent anywhere.
| Format | XMP | Legacy IPTC keywords | C2PA presence |
|---|---|---|---|
| JPEG | APP1 XMP packet, with extended XMP reassembled | APP13 Photoshop IPTC record (Keywords) | APP11 JUMBF box labelled c2pa |
| PNG | iTXt chunk XML:com.adobe.xmp, inflated if compressed | Not used | caBX chunk |
| WebP | RIFF XMP chunk | Not used | RIFF C2PA chunk |
| TIFF, GIF, AVIF, HEIC | Byte scan for an XMP packet (best effort) | Not read | Byte scan (best effort) |
- Amazon keyword, Present: exactly one dc:subject entry equal to contains-synthetic-performer. Duplicates still pass, with a note.
- Near miss: a dc:subject entry such as Contains-Synthetic-Performer or contains synthetic performer. Amazon specifies the exact lowercase keyword.
- Legacy IPTC only: the keyword sits in old IPTC Keywords, not XMP dc:subject. Wrong field: the text is in the file, but in a title or description.
- DigitalSourceType: the element or attribute form of Iptc4xmpExt:DigitalSourceType, with or without the IPTC URI prefix. A bare code passes with a note, because Google's Help uses the short name and IPTC specifies the full URI.
- C2PA: presence only. The checker doesn't validate signatures, so it never calls a manifest valid.
Fix commands: add only what's missing
When a label is missing or wrong, the checker writes the ExifTool command for that route group only, so you never bulk-tag a mixed folder. The same tested commands are below. Run them on a copy folder of the final upload files, then drop the files into the checker again to confirm.
The commands were tested with ExifTool 13.59 on JPG and PNG files on 9 October 2026. ExifTool is free. -overwrite_original skips its _original backups, which is why you work on a copy.
Amazon keyword only (route A)
exiftool -r -ext jpg -ext jpeg -ext png -overwrite_original -XMP-dc:Subject-=contains-synthetic-performer -XMP-dc:Subject+=contains-synthetic-performer ./folder/-= then += removes any existing copy of the keyword and adds one back, so re-running it leaves exactly one keyword. Other keywords stay in place.
DigitalSourceType for fully generated images (routes A and C)
exiftool -r -ext jpg -ext jpeg -ext png -overwrite_original -XMP-iptcExt:DigitalSourceType=http://cv.iptc.org/newscodes/digitalsourcetype/trainedAlgorithmicMedia ./folder/Both, for route A images sold on Amazon and Google
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 ./folder/Edited with generative AI (route B, or route C edits)
exiftool -r -ext jpg -ext jpeg -ext png -overwrite_original -XMP-iptcExt:DigitalSourceType=http://cv.iptc.org/newscodes/digitalsourcetype/compositeWithTrainedAlgorithmicMedia ./folder/Match the code to how the image was made. compositeWithTrainedAlgorithmicMedia is IPTC's code for an image edited using generative AI.
Remove a wrongly added Amazon keyword (routes B and D)
exiftool -overwrite_original -XMP-dc:Subject-=contains-synthetic-performer file.jpgAmazon says not to tag images that only show real people, even if AI tools altered them. For a near miss, remove the variant the same way, then add the exact keyword.
What the checker can't tell you
It answers one narrow question: is the label your channel asks for actually in this file? It can't say whether an image was made with AI, whether a visible label is legally needed, or whether a listing will be accepted. Treat every Check verdict as a decision for a person.
- Pixel watermarks such as SynthID are invisible to a metadata reader. An absent label never means an image isn't AI.
- C2PA is reported as present, not verified. Signature checks aren't part of this version.
- Videos are out of scope. Amazon's rule covers video too, so tag video files with ExifTool or your editor.
- A+ content: Amazon's A+ Content Manager has an AI-generated people checkbox, and Amazon applies the metadata when you use it.
- Amazon Creative Studio images are tagged automatically, according to Amazon staff on the Seller Central forums.
- Older images: Amazon staff on the Seller Central forums have said older images should be tagged too; an earlier staff reply called it best practice.
- Channels also judge image quality and product accuracy. Garment truth needs its own review on the AI fashion imagery QC checklist.
Where 8M Studio fits
8M Studio is a managed AI production studio, not a self-serve tagging app. In managed AI Catalogue Production, 8M Studio can add the Amazon keyword and IPTC DigitalSourceType as part of channel delivery, within an agreed scope. This checker is free for anyone, client or not.
"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."
Frequently asked questions
What does this checker look for?
It reads three things inside each image file: Amazon's contains-synthetic-performer keyword in XMP dc:subject, the IPTC DigitalSourceType that Google Merchant Center asks for on generative-AI images, and whether C2PA Content Credentials are present. It then tells you, for each channel you pick, whether the file matches that channel's published rule for how the image was made.
Are my images uploaded anywhere?
No. The checker reads files with JavaScript in your browser, and no image bytes are sent to 8M Studio or anyone else. You can load the page, go offline, and it still works.
Which images need Amazon's contains-synthetic-performer tag?
Images or videos that show a photorealistic person generated entirely by AI, such as an AI model created from mannequin, flat-lay or packshot inputs. Amazon says not to provide it for images that only show real people, even if AI tools altered them, so don't bulk-tag real-model shots. Our Amazon AI-generated model images guide covers the grey areas.
Is Amazon's tag the same as Google's DigitalSourceType?
No. Amazon's keyword sits in XMP dc:subject and only covers photoreal AI-generated people. Google Merchant Center asks for IPTC DigitalSourceType on any image created with generative AI, person or not. An AI-model image sold on both channels usually needs both.
Why does my tagged image show the tag as missing?
Usually because a later step stripped the metadata: resizing, "export for web", a DAM rendition or a CDN. Check the exact file you upload, and for Google Shopping, save the live product image and check that. Run the tag step last, after every export.
Can this tool tell whether an image was made with AI?
No. It reads metadata, not pixels, so a missing tag doesn't prove an image is a real photo, and invisible watermarks such as SynthID can't be read here. It answers a narrower question: is the label your channel asks for actually in this file?
Does a pass mean my listing is compliant?
No. A pass means the metadata the channel's help page names is in the file. Channels also judge image quality and accuracy, and EU rules can require a visible label that metadata can't provide. This page isn't legal advice.
Evidence and sources
- Google Merchant Center Help says images created with generative AI must include IPTC DigitalSourceType metadata indicating AI origin, and that this metadata must not be removed. Verified 2026-09-26.
- IPTC defines trainedAlgorithmicMedia as digital media created algorithmically using a model trained on captured content. Verified 2026-09-26.
- Amazon Seller Central Help — How to tag media that contains an AI-generated person - Keyword contains-synthetic-performer in XMP dc:subject, written before upload; don't tag real people altered with AI; duplicates don't break tagging. Sign-in may be required. Verified 2026-10-08.
- Amazon Seller Central Help — A+ content guidelines (AI-generated people disclosure) - A+ AI-generated people checkbox and the do-not-tag cases. Verified 2026-10-08.
- Seller Central forum — Amazon staff reply on scope and retroactivity - Forum staff reply, not a policy page: older images, Creative Studio. Earlier staff replies 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 that real people altered with AI are exempt. Verified 2026-10-08.
- Google Merchant Center Help — AI-generated content - Generative-AI images must carry IPTC DigitalSourceType metadata, which must not be removed. Read 2026-10-09.
- Google Merchant Center Help — Image link [image_link] - Restates the requirement not to remove DigitalSourceType on product images. Read 2026-10-09.
- IPTC — Digital Source Type NewsCodes - Code list and labels, including trainedAlgorithmicMedia and compositeWithTrainedAlgorithmicMedia. Read 2026-10-09.
- IPTC Photo Metadata Standard - Digital Source Type property: XMP Iptc4xmpExt:DigitalSourceType, value is a NewsCodes URI. Read 2026-10-09.
- Zalando Partner — Zalando image guidelines - Invisible AI marking in metadata or an invisible watermark, no visible on-image labels, deepfake and low-quality AI lines. Page dated 1 October 2026; read 2026-10-09.
- European Commission — Transparency obligations under Article 50 AI Act (FAQ) - Deployers cannot simply rely on machine-readable marking to fulfil their disclosure obligation. Not legal advice.
- C2PA specifications - Where Content Credentials sit in JPEG, PNG and WebP files. Used for presence detection only.
- ExifTool by Phil Harvey - Free metadata tool used for the fix commands (tested with ExifTool 13.59 on 2026-10-09).
Continue the decision
Labels handled at delivery
Want channel-ready AI catalogue images labelled at delivery instead of fixing files yourself? Talk through your channels and production routes with 8M Studio.
Need this done for your catalogue?
Send your product photos and we return reviewed, ready-to-publish images.
Start a brief