Verify what the video really is.

It reads the video’s C2PA content credential — a cryptographically signed manifest that some AI generators (including BytePlus ModelArk, which serves Seedance) embed at generation time. The credential records which model produced the file (down to the exact SKU, e.g. dreamina-seedance-2-0-fast), when, and a cryptographic hash of the video bytes. We verify the signature, the certificate chain, and whether the file still matches its hash.

To forge a green (“issuer pinned”) result, an attacker would need the issuer’s private signing key. Producing a valid signature without it is what digital-signature schemes are mathematically designed to prevent (see the cryptography section below). What an attacker can do is self-sign a manifest with any claims they like — which is exactly why this tool treats claims from unrecognized issuers as self-declared, not fact (the amber state).

No — and it’s not proof of authenticity either. A credential is routinely dropped by ordinary post-processing done with tools that don’t carry Content Credentials forward: editing, adding effects, subtitles or overlays, color grading, re-encoding, exporting from a video tool, or re-uploading through platforms that re-compress. So a polished clip that is genuine AI output will still read “no credential.” What that means: this file, as delivered, can’t prove what model made it or whether it was altered. When a credential is present, it cryptographically proves which model made the file and that it hasn’t changed since — that is the artifact worth asking for. To check it, verify the original API file (the version before any post-processing), and ask your supplier for that file if you don’t hold it.

A no-credential result can’t tell you either way — it’s equally consistent with genuine model output that was post-processed and with a file that isn’t what it claims. For a finished showcase reel, though, no credential is the expected result: presentation steps — trimming, grading, captioning, re-encoding, exporting from an editor — typically strip the C2PA credential and any metadata-embedded AIGC label (a visible AI watermark, if one was burned in, still shows). Those labels survive reliably only on the untouched API file. So gray on a showcase clip isn’t a red flag on its own — but it isn’t a green light either. It does warrant a question in one case: if a file was handed to you as the raw original, or as a specific model tier you paid a premium for, and it still reads gray — ask your supplier for the credentialed original, the file this tool is built to verify. A supplier who can’t produce one can’t prove what they delivered.

Seedance 2.0 ships in multiple tiers — full, Fast, Mini — at very different prices and quality levels. The credential’s model_name field records the exact SKU that generated the file. If you paid for full 2.0 and the credential says -fast, you now know something your invoice didn’t tell you. A tier dictionary is on this page; we state facts about the file and leave conclusions to you.

No. Verification runs entirely in your browser via WebAssembly (the open-source c2pa-web library from the Content Authenticity Initiative). You can confirm this yourself: open DevTools → Network while verifying — you’ll see no upload request.

No — and that’s the point. C2PA verification is deterministic public-key math; any implementation reaches the same verdict. You can reproduce our result from the command line with the CAI’s open-source c2patool: run “c2patool your-video.mp4 --detailed” and inspect validation_state, the signature issuer, and the model_name assertion. The only judgment we add is which issuers we pin as recognized — and we disclose that list on this page.

No. We pin BytePlus’s public signing certificate — that is our statement that we recognize their chain, not their endorsement of us. We are a reseller with a direct BytePlus agreement; verification results are facts produced by public cryptography, not by any business relationship.

Yes — drop it in. If it carries a valid credential from a pinned issuer, the model_name tells you exactly which model and tier generated it. A file that claims 2.5 but carries a 2.0 SKU — or no credential at all — is telling you something. (We update our SKU dictionary the day new models ship.)

It cannot prove an uncredentialed file is fake (see above). It cannot tell you whether content is true, legal, or licensed — provenance is not rights clearance. It cannot yet prove who supplied a file to you (the credential proves which service generated it; supplier-level attribution is on our roadmap). And heuristics like “this looks upscaled” are signals, not verdicts — we don’t ship them yet.

Don’t accuse — ask. Post-production commonly strips credentials, so the first step is to request the credentialed original (the raw API output before any editing). A supplier selling genuine tiers can always produce it. Use the copy-paste request template below.