What the AI marker in your product photo actually is
An image generated by one of the major hosted AI models is generally not anonymous. It leaves the model carrying provenance data that says a machine made it — some of it in the file's metadata, some of it woven into the pixels. Truest Lens passes that marking through to you unmodified, and this is what it consists of.
Two different mechanisms, often both present
C2PA Content Credentials are cryptographically signed metadata attached to the file. They record which tool produced the image and, where a tool chooses to record it, what happened to the image afterwards. Being metadata, they are readable by anything that knows the format — and strippable by anything that rewrites the file without preserving them.
SynthID is a watermark carried in the pixels themselves. It is designed to be imperceptible to a viewer and to survive the things that destroy metadata: re-encoding, resizing, cropping, a screenshot. It is detectable only with the matching detector, not by looking.
The two solve different halves of the same problem. Metadata is rich but fragile; a pixel watermark is durable but says less. An image that carries both is legible to a platform checking provenance and still identifiable after it has been through a marketplace's image pipeline.
What survives ordinary handling
This is the part worth knowing before you build a workflow around it.
Metadata does not survive much. Uploading to a platform that re-encodes images, taking a screenshot, or passing a file through an editor that does not preserve credentials will typically drop C2PA data. That is not usually malice — it is what most image pipelines have always done to metadata.
Pixel watermarks survive considerably more, which is the point of them. Moderate resizing, re-compression and cropping generally leave the signal detectable. Heavy degradation and aggressive editing can defeat it.
Neither mechanism is a guarantee, and neither is meant as one. They are provenance signals, not DRM.
Why a seller should want it left alone
The instinct on first hearing "your product photos are marked as AI-generated" is to want the mark gone. It is worth resisting, for three reasons.
Platform disclosure requirements are expanding, not contracting. Major marketplaces and social platforms have been adding AI-content labelling rules year over year, several of them keyed to exactly this metadata. An image that carries its provenance is an image already compliant with a rule that has not reached you yet.
Stripping it does not make an image undetectable — it makes it unverifiable. Detection classifiers exist independently of any watermark. A file with its provenance removed does not read as "photograph"; it reads as a file whose history someone removed.
And the disclosure itself is not the liability sellers fear. Buyers are not returning garments because a picture was AI-assisted. They return them because the garment that arrived was not the garment in the picture. That is a fidelity problem, and it is answered by not letting the model redesign the product — not by hiding how the image was made.
Where Truest Lens sits
We do not add marking of our own and we do not remove what the underlying model wrote. The image you download carries what came out of the model.
The related commitment is the one that actually protects a listing: the garment is not redesigned. Button placement, print direction, seam lines and the proportions of the cut are held as photographed, because those are what a listing is answerable for. A general-purpose image model will gladly make a piece more flattering; on a product page every one of those improvements is a misdescription.
Being open about how an image was made, and being accurate about what the image shows, are two separate promises. The second is the one that decides whether a buyer keeps the parcel.