C2PA support lab: which tools did we parse ourselves?
For phones, cameras, and AI tools, what is the difference between a vendor sentence and a sample this site actually parsed?
The C2PA matrix separates a vendor statement, the official source, and whether this site parsed a sample. A missing parse stays missing. No AI percentage is filled in to hide that gap. Clean the copy in this browser before sending. The file stays on this device and is not uploaded to this site.
A marketing line about Content Credentials is not the same thing as a JUMBF box this site opened. Empty parse cells stay empty.
Short answer
Direct answer: a vendor claim and a parsed sample are different rows of evidence. This page does not fill a missing parse with a slogan.
- What the table is for: Six tools, phones, and cameras are listed only as far as the page’s sources go.
- What that does not prove: No percentage score is invented when a sample was not parsed.
- What you can do first: Open your own image in the local provenance viewer. Optional online checks stay opt-in.
🟢 Local sample verified · 1 November 2026 · Viewer and cleaner compared
Vendor statements read on 2026-10-18
The matrix
Published . Vendor columns paraphrase the linked pages. Our columns are measurements, and an empty measurement is written as Not verified.
| Product | What the vendor says | Official source | Our parse of a real sample | Viewer result |
|---|---|---|---|---|
| ChatGPT, Codex and the OpenAI API (images) | OpenAI says supported images carry C2PA metadata and a SynthID watermark. It points to openai.com/verify to check a file, and says uploads, edits, screenshots and format conversion can remove the metadata. | OpenAI Help Center OpenAI content provenance API guide | Not verified | Not verified |
| Adobe Firefly (Text to Image, Generate Image in Photoshop) | Adobe says it applies Content Credentials automatically to assets where 100% of the pixels are generated with Firefly. The credentials list actions, tool, app or device, date and issuer, and may also be stored in Adobe's Content Credentials cloud for recovery. | Adobe: Content Credentials overview | Not verified | Not verified |
| Google Pixel 10 and later, Pixel Camera (photos) | Google says Pixel Camera on Pixel 10 attaches Content Credentials to photos and reached Assurance Level 2 in the C2PA Conformance Program. The Pixel help page says videos are not supported and that credentials need an up-to-date, unmodified Android and a recent internet connection. | Google blog on Pixel and C2PA Pixel Camera Help | Not verified | Not verified |
| Samsung Galaxy S25 series (Galaxy AI edits) | Samsung's newsroom says the S25 series supports Content Credentials for content created and edited with generative AI. The list of covered features (Generative Edit, Sketch to Image, Portrait Studio, Drawing Assist, AI Sticker) comes from a press report quoting Samsung, not from a Samsung page. | Samsung Global Newsroom Android Authority (quotes Samsung) | Not verified | Not verified |
| Leica M11-P (camera, photos) | Leica says the M11-P signs each image when the function is on. The manual gives the menu path: Camera Information, Leica Content Credentials, Sign Content, On. | Leica M11-P instruction manual (PDF) Leica press page | Not verified | Not verified |
| Midjourney (images) | Midjourney's documentation describes a hidden ID tag in every delivered image and a checker at midjourney.com/verify. The page does not mention C2PA. We make no statement about whether a given Midjourney file carries a C2PA manifest. | Midjourney: Content Authenticity | Not verified | Not verified |
What we did parse: two public reference files
To show what the local scan reports, we ran two public test images from the c2pa-rs repository through the AI Image Origin Detector's parser (2026-10-18, structure scan only). They are test signers, not products from the table.
| File (SHA-256 prefix) | Size | Scan label | Claim generator | Actions | Creation type |
|---|---|---|---|---|---|
CA.jpge71bff58fc576408… | 1024×683, 166,864 bytes | C2PA / JUMBF structure detected | make_test_images/0.33.1 c2pa-rs/0.33.1 | c2pa.color_adjustments | unknown |
C.jpga2d14755db55de67… | 2048×1365, 132,518 bytes | C2PA / JUMBF structure detected | make_test_images/0.33.1 c2pa-rs/0.33.1 | c2pa.created | pure_synthetic |
Raw numbers: docs/evidence/runs/2026-10-18-c2pa-reference-parse.json in the repository. The scan does not validate signatures or the trust list.
How to add a row
A row turns from Not verified to a result when we hold a file made by that product and the parser output is on record with the file's hash. Check your own file now in the AI Image Origin Detector or the C2PA viewer. Both run in the browser.
Background: what C2PA is, Content Credentials, C2PA versus EXIF, C2PA overview. Standard: C2PA and its trust lists. Related lab: AI provenance labs.
FAQ
What does Not verified mean in this table?
We have not run one of our own samples from that product through the parser, or we could not confirm the claim from an official page. The cell stays Not verified until we have the file or the source.
Does a missing C2PA manifest mean an image is not AI-generated?
No. Uploads, screenshots, edits and format conversion can remove the manifest, and several vendors say so in their own help pages. The detector reports a missing manifest as inconclusive.
Does the local scan check the signature?
No. It finds the JUMBF and C2PA structure and reads the claim generator, tool and actions. It does not validate the signature or check the C2PA trust list. Official verification is a separate step on the AI Image Origin Detector.
Does this page cover removing AI marks?
No. This page only reads and detects provenance data.