Yesterday we announced that every Modelia image shows its certification. Today we open the file. Because our provenance marks are readable by any conforming verifier, anyone can inspect what a Modelia image carries; this post is simply the guided tour, taken before you take it yourself. And it covers the half of the story an inspector cannot see: what we decided to leave out, and why the exclusions are the part your procurement team will care about most.
The three things that travel with every image
Every image exported from Modelia carries, or is covered by, three layers.
- A signed C2PA manifest, embedded in the file. It asserts that the image is AI-generated, names the generating system and Modelia as the provider vouching for it, and when it was made and exported, independent timestamp. The whole block is cryptographically signed: change anything in the file afterwards and verification fails visibly. If the concepts are new, our plain-language C2PA explainer covers how manifests, signatures and trust work.
- An invisible watermark, woven into the pixels themselves. It carries a compact statement, enough to identify the content as AI-generated and trace it to our systems, and it is built to keep answering after social platforms strip the metadata layer away.
- An export record, held on our side. At every export we record the file's cryptographic hash and its provenance details, so there is always an authoritative answer to "was this made with Modelia, and when?", even for a copy that has been stripped bare. Think of it as the verifiable history of everything you have created, available to you while your account is active.
What is deliberately not inside
Provenance answers three questions: what is this content, who vouches for it, when was it made. It is not a diary of how you work. So the manifest in a Modelia image does not contain:
- No prompts, and no generation parameters. How you direct the model is your craft and your competitive edge. It stays with you.
- No client or brand identity. The manifest names Modelia as the signing provider, not whose campaign the image belongs to.
- No campaign context. Nothing about collections, launch dates, product lines or where the image will run.
- No source assets. Reference photos and input material you upload are not embedded in or described by the output's manifest.
The reasoning is simple once stated: published files travel. Anyone, including your competitors, can open the provenance of an image you post. A certification layer that leaked prompts or client relationships would be a liability dressed as a feature. Ours is designed so that a stranger inspecting your published file learns exactly one thing: that it is AI-generated content that Modelia vouches for, made at a certain moment. Your workflow stays invisible.
What we preserve of yours
There is a quieter design decision underneath: what happens to metadata that is already in a file when it passes through our pipeline. Early provenance implementations across the industry had a bad habit of overwriting existing metadata when writing their own. We read, merge and write: existing metadata in your uploads is preserved alongside the provenance we add, not replaced by it.
Check everything in this post
None of the above asks for trust. Export an image, open our public verifier, and read the manifest yourself; any independent C2PA inspector will show you the same fields, because the format is an open standard. What you will find matches this post, and what you will not find is the point of it.
The formal, versioned description of what our signature contains lives in our terms, and the practical picture of how duties split between Modelia and you as a publisher is on the compliance page. Tomorrow we publish the third piece of launch week: everything our certification does not do, in writing.
Modelia is the AI imagery platform where every export leaves certified by default: signed C2PA manifest, invisible watermark, verifiable by anyone. See how it works, or check any image in the free C2PA checker. Rules and data: The State of AI-Generated Content in Fashion 2026.