Amazon AI image compliance,
handled before upload
Every file is checked on import, so you only tag what is actually missing.
Nothing is uploaded — reading and writing both happen in your browser.
Detection queue
Existing tags are read the moment a file is imported
Drop files in to check them first
Existing tags detected automatically · JPG / PNG up to 50MB · video up to 500MB · nothing uploaded
Detection results for each file will appear here
Your work stays
yours.
Which files need the tag
Amazon's requirement is narrower than it first sounds: it covers photorealistic people generated entirely by AI. Not “images that used AI”, and not every AI-generated image. The test is whether the person on screen reads as a real human who does not exist.
| Content | Tag needed? |
|---|---|
| AI models and AI-generated lifestyle figures | Yes |
| AI-generated photorealistic bystanders in the background | Yes |
| Images and video inside A+ content | Yes |
| Photos of real people, even retouched or background-swapped with AI | No |
| Cartoons, illustrations and other non-photorealistic figures | No |
| Characters from film, television or games | No |
| Product-only shots with no people in frame | No |
For the keyword itself, the field it belongs in, and three free ways to add it to a single file, see the tag explained.
Why tagging is not the last step
Amazon reads the tag at the moment of upload. Between tagging and that moment a file usually passes through retouching software, compression tools, asset libraries and agency portals. We measured what those cost you: not one of eight processing steps kept the tag on both JPEG and PNG.
Which makes the order matter: tag late, verify last. Put the file you are actually about to upload through the checker — not the master it came from.
What this workspace does
- Checks on import — every file is read for existing tags before anything is written
- Skips what already complies — no rewriting, the original file is passed through untouched
- Writes in bulk — into XMP
dc:subjectfor images, the field Amazon specifies - Exports a ZIP — with a per-file report recording each file's state before processing and what was done to it
Formats and limits
JPG and PNG up to 50MB each; MP4, MOV and WebM up to 500MB each. No limit on how many. Everything happens in your own browser — no file is uploaded anywhere, and there is no account to create.
One caveat worth stating plainly: Amazon has published the metadata field for images only — the field for video has not been specified. Video handling here writes container tags and is marked experimental; it is not a compliance guarantee.
And if you also run Google, Meta or TikTok ads, those channels need the disclosure visible on the creative itself — metadata alone will not do. The label generator burns one into the image.
A workflow that survives the trip to Seller Central
The order of operations matters more than any single step, because the tag is fragile in exactly one direction: anything that re-encodes an image can drop it, and nothing puts it back.
- Finish the creative first. Retouching, resizing, compression, format conversion — get all of it done before tagging, not after
- Tag the final files. Drop the folder above; anything already carrying the tag is skipped rather than rewritten
- Verify what you are about to upload. Not the master, not the folder you tagged an hour ago — the exact files going to Seller Central
- Keep the ZIP report. It records each file's state before processing and what was written, which is the only evidence you will have if a listing is ever questioned
If your assets pass through an agency or a shared asset library between steps 2 and 3, treat step 3 as mandatory rather than optional — that hand-off is where tags most often disappear.
If you also advertise off Amazon
Amazon is the only major channel that reads a tag out of the file. Everywhere else the disclosure has to be visible on the creative or set in the ad platform, and the same image often needs both:
| Channel | What satisfies it |
|---|---|
| Amazon | The XMP tag — Amazon renders the disclosure itself |
| Google Ads, YouTube | An AI label setting in the ads product, with visible overlays for EU, India and New York campaigns |
| Meta | Meta's own label, applied when it detects AI or is told |
| TikTok | A per-ad toggle, which resets when a campaign is duplicated |
For those channels, burn a visible label into the creative — placed away from the edges that automated cropping trims first.
Common questions
Does tagging change how the image looks?
No. The tag lives in metadata, not pixels. Files come out byte-identical apart from the metadata block, and image quality is untouched — nothing is re-encoded.
Can I undo it?
Keep your originals; this writes new files into a ZIP rather than modifying what you dropped in. The files on your disk are never touched.
Why not just use exiftool?
You can, and the commands are here. This exists for the case where you would rather not install anything, and because it checks before writing so a catalogue that is already half-tagged does not get rewritten wholesale.