Prove where AI content comes from

0byte stamps every AI generation with a fingerprint and a public record — who made it, with which model, and when. Anyone can check it, even from a screenshot.

See it survive

Screenshot it. Crop it. It still knows.

Watermarks wash off and manifests get stripped on upload. 0byte’s fingerprint is part of the image itself — so the origin survives the trip.

01Generated
An AI-generated portraitStamped

Fingerprinted by 0byte the instant it’s made — and written to the public registry.

02Exploited
EXIFC2PA manifestLocationOriginal file

Screenshotted, cropped, re-compressed. Every tag a watermark or manifest needs is gone.

03Origin
Provenance confirmed

Recovered from the screenshot alone — and cryptographically provable.

Made byAcme
ModelImagen 3
CreatedMarch 14, 2026
On recordPublic registry
Fingerprint match99%
Cryptographic proof
signatureEd25519 ✓
inclusion proofverified ✓
log position1,247,892,103
merkle root0xc7f1…a2b4
The Problem

Every existing approach breaks in the same place.

They attach trust to the file. The internet strips the file.

how provenance is handled today
An AI-generated image
01Image
02Watermark · metadata
03Platform
04Detectors
  1. 01

    The image

    A model generates an image, and it spreads across the open internet.

  2. 02

    Trust is embedded

    A watermark or C2PA manifest is written into the file to record its origin.

  3. 03

    Platforms strip it

    On upload, platforms re-encode the file — and the embedded proof is discarded.

  4. 04

    Detection is a guess

    With only raw pixels left, detectors can estimate the source, never prove it.

The proof has to live outside the file, based on what the content looks like. That’s 0byte.

The stakes

The gap is already expensive — and the regulatory clock is running.

Features

Proof that survives the real internet.

One fingerprint. One public record. Verifiable by anyone.

01Fingerprint

A fingerprint, not a label

0byte reads what the content looks like and turns it into a fingerprint. Screenshots, crops, color shifts and re-uploads still match — and your file stays byte-for-byte identical.

input · the full imagederiveperceptual fingerprinta3f9 c4e1 7b02 d9f4 …128-bit · not the imageSURVIVES CROPS · RE-ENCODES · SCREENSHOTS
02Signed proof

Proofs that can’t be faked

Each proof is signed at creation by whoever made the content. Alter a single detail and the signature breaks — tampering is provable, not deniable.

SIGNED · ED25519 · TAMPER-EVIDENT
03Ledger

A record that can’t be rewritten

Proofs land in an append-only public log. Records are only ever added, never edited or deleted — not even by us. Anyone can audit the full history.

#04 · newest#00 · oldestAPPEND-ONLY LOG
04Verify

One call checks everything

A single call aggregates every signal — the registry match, any surviving watermark, and a detection score — into one verdict, with the evidence behind it.

FINGERPRINTSIGNATURELEDGERverify()✓ VERIFIEDconf 0.99VERIFIED BY 0BYTE
05SDK

One line of code

Drop it into your pipeline in one line. Stamping runs in under 50ms, in the background, and hands you back a cryptographic proof of origin.

stamp.py# stamp at the moment of creationproof = zbyte.stamp(image)→ 0b_3f2a8c91 ✓signed · logged · 41msONE LINE · <50MS
Use Cases

Built for both sides of the trust problem.

Whether you create AI content or have to judge it — the answer comes from the same public record.

AI model providers

Stamp every generation at the source — proving the model, the moment, and the maker, without ever exposing how it works.

Newsrooms & media

Settle whether an image, video, or clip is AI in seconds — and trace it back to its exact origin, with a record you can publish.

Developer platforms

Add provenance to your product in a single call — across every model you ship, even the open-source ones. No cryptography to learn.

Creators & rights holders

Register your work and prove it was yours first — the moment it’s reposted, scraped, or claimed by someone else.

Why 0byte

Every approach, side by side.

Each existing system solves part of the problem. 0byte is built for the part of the journey where the others break — and when their signals survive, our verification surfaces those too.

Survives social media upload

C2PA labelsNo — stripped on upload
WatermarksMostly
AI detectorsn/a — guesses after the fact
0byteYes

Survives screenshots & re-encoding

C2PA labelsNo — already gone
WatermarksOften degraded
AI detectorsn/a
0byteYes — built and trained for it

Covers any AI model, including open-source

C2PA labelsOnly signing providers
WatermarksOnly the vendor's own models
AI detectorsOnly generators it was trained on
0byteYes — anyone can stamp

Modifies your content

C2PA labelsAdds metadata bytes
WatermarksAlters pixels (invisibly)
AI detectorsNo
0byteNo — byte-for-byte identical

Verifiable by anyone

C2PA labelsNeeds a C2PA-aware viewer
WatermarksOnly the vendor can check
AI detectorsVendor API only
0byteYes — free, public, no account

What a match gives you

C2PA labelsSigned claim (when it survives)
WatermarksVendor classifier score
AI detectorsA probability
0byteCryptographic proof: who, what model, when

Admits where it fails

C2PA labels
Watermarks
AI detectors
0byteYes — see below
The honest limit

Heavy AI regeneration of content defeats every system on this table — 0byte included. We’re the only column that says so out loud.

FAQs

The questions we get most.

Straight answers — including what 0byte can't do.

Still have a question? Talk to us

0byte is the origin protocol for AI content. When content is generated, it gets a fingerprint and a digitally signed proof of origin in a public registry. When anyone needs to check content later, one call matches it against the registry and returns who registered it, when, and with which model — with the evidence to back it up.

Make it verifiable. Forever.

Free to verify, no account needed. Stamp your first proof in five minutes.

Building something bigger? Compliance questions?

Talk to us