How to prove a photo existed before AI training

If your work is good enough that someone might one day claim you stole it from an AI image generator, you need a way to prove the photo existed before that generator's training cutoff. EXIF metadata won't help — it's a text field anyone can rewrite. A Google Photos timestamp won't help — Google can revise its own database and has no obligation to keep one for a decade. What does help is a cryptographic anchor to a public chain nobody controls.

The shape of the proof

A SHA-256 hash is a 64-character fingerprint of the file's bytes. Change one pixel and the hash changes completely. When you anchor that hash to the Bitcoin blockchain via OpenTimestamps, you publish — for anyone to see — a record that says: "this exact file's fingerprint existed on or before block N." Block N has a timestamp in UTC. That timestamp is enforced by ~50,000 independent computers running Bitcoin Core. Rewriting it would require redoing every block of mining work since then, which has never happened in sixteen years of Bitcoin history.

In practical terms: you drop a photo into Orphograph, the browser hashes it locally, the hash gets submitted to five independent OpenTimestamps calendars, and within about an hour the calendars fold your hash into a Merkle root that lands in a Bitcoin transaction. You walk away with a small receipt file. Five years later, anyone with your original photo plus that receipt can re-hash the photo, check the receipt against the live Bitcoin chain, and confirm the date — without trusting Orphograph or any other intermediary.

Why pre-AI proof matters now

Generative image models have published training cutoffs. Most major open-weights image models released through 2023 trained on data scraped before mid-2023; popular hosted models closed their datasets around the same window. If you can prove your image existed before those cutoffs, you defeat the most common bad-faith claim: that your work was generated by a model rather than fed into one. That claim shows up in stock-license rejections, in plagiarism callouts on social media, and in style-mimicry disputes between illustrators and AI service providers.

The defense isn't "I'm a real artist" — that's not falsifiable. The defense is a timestamp from a chain that started running in 2009.

The exact workflow

1. Export the photo at the resolution you want to defend. If you might publish at 4K, anchor at 4K. If you might publish a watermarked web-resolution version too, anchor that separately — they have different hashes.
2. Drop the file into the Orphograph drop zone. SHA-256 runs in your browser; the bytes never leave. You'll see a 64-character hash appear in seconds.
3. The hash is submitted to five calendars: a.pool, b.pool, alice, finney, and btc.catallaxy. Each is operated by a different party. You only need one to survive to verify.
4. Download your receipt JSON. Save it next to the original photo. Backups in two places — local drive plus a cloud bucket — cost nothing and inoculate you against drive failure.

What this is not

This is proof-of-existence, not legal evidence on its own. We don't claim court-admissibility, we don't act as a notary, and we don't certify your identity — only that some file with this exact content existed at the anchored time. For litigation, talk to a digital-evidence specialist who can place your timestamp inside a broader chain-of-custody. What we sell is the cheapest, most durable pre-existence anchor available to a one-person creative business.

FAQ