Photographer
Photographers
Your portrait, re-uploaded without your terms attached.
ProveConsent · Media Consent Registry
REGISTRY SNAPSHOT
REV. 2026-10-05
REV. 2026-10-05
A chain-of-custody ledger for media consent
The record thatsurvives the edit.
Choose how your photos, videos, and audio may be used. Register the permission, trace it back from an edited copy, and keep a record when consent changes.
Studio portrait
Photographer
Musician
From creation to revocation
One file, six momentsFollow a single illustrative file through its whole lifecycle — registered, chained, shared, checked, and withdrawn.
Studio portrait
Not yet registered
Start with the original
An illustrator finishes a piece in their own studio. Before sharing it anywhere, they decide what other people may do with it.
studio-work.jpg · permission record does not yet exist
Studio portrait
Registered · Level 1 · active
Set the terms, sign them
They register the file's content-derived identity and sign a W3C ODRL policy — commercial use, AI training, derivative works — with a wallet or email-derived key. No new account, nothing installed.
ISCC:KAD-7XY2-M4P6-R8TA · signed
Original
Shared copy
Cropped · recompressed · renamed
The copy that circulates
A platform crops it, strips the EXIF, re-encodes it, and reposts it under a new filename. The pixels are what's left — everything else is gone.
post-final-v2.jpg · no metadata
Shared copy
Exact match · consent active
Resolved back to the original terms
Anyone — a platform, a buyer, the creator themselves — can check the edited copy against the registry. It resolves by content, not by filename or metadata.
- ✓ Display allowed on whitelisted platforms
- ✕ Commercial use not allowed
- ✕ AI training not allowed
Studio portrait
Consent revoked
Withdrawn, permanently on record
The creator can withdraw consent at any time. The revocation itself becomes a new, permanent, timestamped ledger entry — never a silent deletion of what came before.
Block #2 · revoked · 2026-10-04
Illustrative walkthrough using a stock photo, not a live verification result. Matching depends on the file and the edits made to it; a fingerprint match is not a guarantee that every altered copy can be identified.
Whose work travels without them
Use casesAny file that leaves your hands can lose its terms along the way. These are the four we hear about most.
Musician
Musicians
A clip lifted into someone else's track, credit stripped.
Model / talent
Models & talent
A face used to train a model you never agreed to.
Illustrator
Illustrators
Work scraped into a dataset with the byline removed.
Why metadata isn't enough
The problem
Step 01 — Original
File leaves the camera
EXIF intact, terms attached, owner known.
Step 02 — Platform
Upload & re-encode
✕ metadata stripped on ingest
Step 03 — Re-upload
Copy circulates
✕ no trail back to original terms
✓ the pixels themselves are unchanged
✓ the pixels themselves are unchanged
Embedded metadata
- ✕ Stripped on upload by most major platforms
- ✕ Gone entirely from a screenshot or re-encode
- ✕ Nothing left to trace back to original terms
The content itself
- ✓ A crop or recompression barely moves it
- ✓ Resolves back to the registered consent record
- ✓ Can't be removed without destroying the file
How the ledger works
Mechanism
L0
Total Prohibition
Private personal use only. Opts out of all display, dissemination, AI, or likeness extraction.
L1
Personal Showcase
DefaultPublic display for organic viewing. Prohibits monetization, AI crawl, copy, or likeness harvesting.
L2
Commercial Monetization
Allows commercial use on approved platforms subject to mandatory smart-contract royalties.
L3
Full Rights & AI
Allows unrestricted use, redistribution, and commercial dataset training on opt-in media.
—
Custom Terms
Power usersSet monetization and AI training independently instead of picking a preset combination — e.g. non-monetized but AI-trainable, a combination none of the four presets cover.
Setting terms during registration looks like this — four preset consent levels plus independently configurable custom terms, each a W3C ODRL policy signed with a wallet or email-derived key.
01
Register
The rights holder sets terms — commercial use, AI training, derivative works, monetization split — as a W3C ODRL policy, then signs it with a wallet or email-derived key. Fingerprints are computed from the file itself, client-side.
02
Chain
Every action — registration, amendment, revocation — is appended as a SHA-256-linked block in a real Firestore-backed ledger, then anchored to the public Sigstore Rekor transparency log: an independent, permanent record outside ProveConsent's own database. Each entry's hash depends on the one before it, so earlier history can't be quietly edited without breaking the chain.
03
Match
An uploaded copy is resolved back to its original registration by what it is — perceptual and content-derived fingerprints — not by filename, embedded tags, or metadata that a platform may have already discarded. In our own measurements against this matching code: 95–100% of lightly re-encoded or resized copies, 80–90% of heavily cropped or recompressed ones, with 0% false matches across unrelated images in the same test.
04
Revoke
Consent can be withdrawn at any time. The revocation is itself a permanent, timestamped ledger entry — never a silent deletion of the record.
Versus a C2PA manifest: a manifest travels attached to the file, so it's gone the moment a platform strips it or a screenshot is taken. ProveConsent is built to complement that approach, not replace it — resolution and revocation status come from the content itself, so the record is still reachable after the manifest isn't.
Methodology for the match-rate figures above: measured directly against this app's own production matching code (real perceptual hashing and real content-derived embeddings — never a simulated stand-in) across 7 real photographs and 10 realistic transformations (JPEG recompression, resize round-trips, center crops, and brightness/contrast shifts, at light/moderate/heavy severity), plus every cross-pair among those 7 photos for the false-match figure. A small sample from an early-stage product, not a large-scale study — re-measured whenever the matching code changes.
R&D in progress, not yet in the live matching above: a prototype technique that tries several simulated crop alignments plus 9 overlapping regions per registered photo lifts heavy-crop matching from 80–90% to 100% in the same test, and separately catches 14–100% of tight, off-center crops (e.g. zoomed into just one detail of a photo) depending on how closely the crop lines up with a stored region — averaging around 40% across the 5 crop positions tested, with no change to the 0% false-match rate. Benchmarked, not shipped: rolling it into live matching needs a registry schema change this hasn't gone through yet.
Sign with what you already hold
Identity
Solana wallets
EVM wallets (MetaMask)
Email-derived identity
No prior Web3 experience required
Put your work on the record.
Register a file, set its terms, and know the record is still reachable — long after the metadata isn't.