Upload camera-original
RAW or camera-original JPEG is retained untouched in cold storage. Web and thumbnail derivatives are generated as separate files — never overwriting the original.
Kelp Market · Agent-native content licensing
Photographers upload camera-original files with signed proof of gear, sensor, and capture time. AI labs license verified training data — every claim independently checkable, no server trust required.
ed25519 · SEAL records · PRNU sensor fingerprinting
The proof pipeline
RAW or camera-original JPEG is retained untouched in cold storage. Web and thumbnail derivatives are generated as separate files — never overwriting the original.
A single-use, short-lived QR challenge is issued and bound to your account. Photograph it with the same body and lens — serials are cross-checked against the image.
A signed SEAL record binds content hashes, gear identity, and capture time into an append-only proof ledger. The record is independently verifiable by anyone.
Verified sets are licensed over MCP — estimate, confirm, and receive signed grants. Every license is traceable to the underlying proof records.
Independent verification
Each SEAL record chains the content hash of the camera-original file, the verified gear identity, and the capture time into a signed, append-only ledger. Nothing about the claim lives in a database flag you have to take on faith.
Verification runs entirely client-side — in your browser, against the public record. No server, no database, no trust required. If the signatures check out, the image is authentic.
Open a sample proof →Verified training data
No datasets published yet — creators publish gear-verified sets from their dashboard.
Photographers & partners
For photographers
This can be you
Shoot, scan, publish. Gear-verified sets sell as licensed training data — you keep 50% of every license, per asset.
For archive owners
This can be you
Open a never-published back-catalog as attested, licensed datasets. Assets stay private — never posted, never resold as stock.
For AI labs
This can be you
License verified, legally defensible training data over MCP — every claim independently checkable, no server trust required.
For AI labs · The private archive
Frontier models were trained on what the internet published — scraped sites, stock libraries, social feeds. That supply is exhausted. Humanity has saved ~13.6 trillion unique photos, and Google Images indexes only ~136 billion of them. By conservative math, 80%+ of the world's photos and videos were never published anywhere online — they sit in camera rolls, hard drives, and archives, invisible to every crawler.
Kelp leaves public data alone. Partners open their never-published photos and videos as verified training sets. Assets stay private — never posted, never resold as stock — and move only as licensed datasets, each with signed, independently checkable proof of origin. Petabytes of unseen content, delivered over MCP.
$ estimate_training_set { type: "image", quantity: 8, keyword: "Windsurfing" }
quote 267c8f44 · 8 assets matched · $0.50/asset · total $4.00 · expires 1h
$ confirm_training_set { jobQuoteId: "267c8f44…" }
job 4d33c707 · queued · stream: /api/jobs/4d33c707/stream
$ # buyer pays the hosted invoice — the SSE stream reports settlement
event: { "status": "ready" } · 75s after payment
$ get_training_set { jobQuoteId: "267c8f44…" }
bundle datasets/4d33c707….tar.gz · 20 MB · LICENSE.md + manifest.json
+ 8 assets, each with a signed manifest and SEAL recordFor legal & compliance teams
Training on scraped or pirated images is a liability your acquisition process should eliminate. Every Kelp dataset ships with a signed, independently verifiable chain — camera-capture proof, a signed license grant, and fixed terms frozen at quote time — so your team can produce cryptographic evidence of licensed acquisition when an infringement question arises.
No trust lists, no embedded manifest chains, no validator disagreement: one fail-closed verifier and one public key, published in DNS. If any check fails, no verification badge is shown — ever.
Built for legal review — read the defense brief →