Guide
AI content label
Every try-on image the API returns is marked, inside the file, as made with AI. Here's what that label is, why it's there, and the one thing you still need to do yourself.
What's in the file
Each result image carries one XMP metadata block, the standard place photo software looks for this kind of information. It holds four fields:
- Digital Source Type (IPTC):
compositeWithTrainedAlgorithmicMedia. That's the standard IPTC term for an image that combines real photographs with content made by a trained AI model, which is exactly what a try-on is: a real photo of the shopper, with the garment rendered on by AI. It's the field platforms such as Google Images read to label an image as AI-generated. - Description: “AI-generated virtual try-on preview made with Clothsy AI. Fit, colour and drape may differ from the real garment.” This is the text most photo viewers show.
- Creator tool and credit:
Clothsy AI.
Nothing about the shopper is written into the file: no names, ids, dates, locations or device details. Any metadata the image had before we served it is removed first, so these fields are the only ones it carries. The label is added to JPEG, PNG and WebP results alike, and it doesn't change a single pixel.
Because it uses open, widely supported vocabularies, software that looks for AI provenance (photo tools, social platforms, content checkers) can read it without knowing anything about Clothsy AI. You can inspect it yourself with a metadata tool such as ExifTool:
exiftool -XMP-iptcExt:DigitalSourceType -XMP-dc:Description -XMP-xmp:CreatorTool -XMP-photoshop:Credit result.jpgThe Digital Source Type it prints ends in compositeWithTrainedAlgorithmicMedia.
Why it's there
Laws are starting to require that AI-generated images can be recognised as such. Article 50 of the EU AI Act, for example, expects synthetic images to be marked in a machine-readable way. Labelling every result at the source helps you meet transparency duties like these without extra work on your side.
The label is always added and can't be switched off. It lives in the file's metadata, so it doesn't change how the image looks. Every surface gets the same label: the API and SDK, the website widget, the Shopify and WooCommerce apps, the Chrome extension and shared looks.
What you still need to do
The label is invisible, and there is no option for a visible watermark or badge on the image itself. Metadata is also fragile: re-encoding, resizing or compressing an image — which many CDNs, image optimisers and upload pipelines do automatically — can strip it. So don't rely on the file alone.
- Tell shoppers in your UI. Put a short caption next to every try-on image, such as “AI-generated try-on”. It's honest, it sets expectations about fit, and it works however the image is served.
- Say it in the alt text too, so people using screen readers get the same information.
- Keep the file intact when you copy it.
resultUrllasts 24 hours. If you save results for longer, store the bytes exactly as downloaded rather than re-saving them through an image library, so the label survives.
<figure class="tryon-result">
<img src="RESULT_URL" alt="AI-generated preview of you wearing the Cropped denim jacket" />
<figcaption>AI-generated try-on — the real fit may differ.</figcaption>
</figure>This applies whichever way you integrate — plain HTTP, the TypeScript SDK, or your own UI built on the Next.js hook.

