Transition deadline: 2 December 2026

Mark your AI outputs machine-readably - before the Art. 50(2) deadline

The EU AI Act requires synthetic audio, image, video and text to be marked as artificially generated in a machine-readable way. Systems placed on the market before 2 August 2026 have until 2026-12-02 to comply (Art. 111(4), Regulation (EU) 2026/1744). New systems have no grace period.

EUR 29 one-time

Marking Readiness Pack: implementation snippet + self-audit checklist + deployer disclosure template. Instant download after payment.

Buy the pack - EUR 29

Secure checkout by Stripe. Instant download after payment.

What is in the pack

1. marking.py - stdlib-only Python, zero dependencies. JPEG: APP1 XMP packet with IPTC DigitalSourceType trainedAlgorithmicMedia. PNG: CRC-correct iTXt XMP chunk. Any other file: sha256-bound .art50.json sidecar. Idempotent re-marking, read-back verification, CLI with exit codes you can pipe into CI.
2. checklist.md - self-audit checklist mapped to the four Art. 50(2) properties: effective, interoperable, robust, reliable - with the exact commands and the evidence to retain.
3. disclosure-template.md - the Art. 50(4) deployer text notice, ready to fill and show to your users.

How it works

python marking.py mark input.jpg marked.jpg
python marking.py verify marked.jpg   # exit 0 = MARKED, 1 = NOT MARKED
python marking.py sidecar output.mp3  # hash-bound manifest + verify

Verified, not promised

The pack ships with 14 acceptance tests (test_art50.py), including the negative cases: unmarked files must fail verification, a one-byte tamper must break the sidecar hash, re-marking must never duplicate packets. The suite runs green (exit 0) on Python 3.14 with the standard library alone.

Delivery

After payment you are redirected straight to the download page - the three files plus the test suite, served from the public repository. No email wait, no account needed. MIT-licensed snippet: embed it in your own products.