Disclose Stamp — @qadar-ai/disclose-stamp

A publisher-side CLI that embeds machine-readable AI-content marking into your image files at publish time, so the marking travels with the asset even after it's downloaded — a C2PA manifest carrying the IPTC DigitalSourceType (trainedAlgorithmicMedia / compositedWithTrainedAlgorithmicMedia). It complements the on-page Disclose widget (the visible EU AI Act Art. 50(4) disclosure).

Transparency tool, not a compliance guarantee. Stamp adds a machine-readable contribution alongside provider-side marking; it does not by itself discharge the provider's Art. 50(2) duty.

Local-only / privacy. Stamp runs entirely on your machine. Your assets never leave it — nothing is uploaded to Qadar, and signing is local.

Install & use

Requires Node ≥ 22.

# stamp every JPEG/PNG/WebP/AVIF under content/ as AI-generated
npx @qadar-ai/disclose-stamp "content/**/*.{jpg,png,webp,avif}" --level ai-generated

# one file, a specific level, write to a dist dir
npx @qadar-ai/disclose-stamp hero.png --level ai-modified --out dist/

# sign with your own (trusted) certificate instead of the self-signed default
npx @qadar-ai/disclose-stamp img/*.jpg --level ai-generated --cert chain.pem --key key.pem

# also bake a visible EU label badge into the pixels
npx @qadar-ai/disclose-stamp hero.png --level ai-modified --badge \
  --badge-position top-right --badge-size lg --badge-tone black

Levels (consistent with the widget taxonomy): ai-assisted, ai-generated, ai-reviewed, ai-modified. A per-file sidecar hero.jpg.disclose.json ({ "level": "ai-generated" }) overrides --level.

| Level | IPTC DigitalSourceType | | --- | --- | | ai-generated, ai-reviewed | trainedAlgorithmicMedia | | ai-assisted, ai-modified | compositedWithTrainedAlgorithmicMedia |

Formats: JPEG, PNG, WebP, AVIF. Options: --level, --cert/--key, --out, --overwrite, --badge (+ --badge-position/--badge-size/--badge-tone), --dry-run, --help.

Visible baked-in badge (--badge)

Composites the official EU label icon (matching the widget's badge — same size ladder and positions) into the image pixels, so a human-visible mark travels with the file:

• --badge-position top-left (default) · top-right · bottom-left · bottom-right · center
• --badge-size sm (27px) · md (38) · lg (53) · xl (74) · 2xl (104) · 3xl (146)
• --badge-tone black (default, dark pill) · white (light pill)

The badge family follows the level (AI · AI GENERATED · AI MODIFIED). The badge is composited before the C2PA manifest, so the manifest hashes the final pixels.

Known limit: a pixel badge survives re-encode but is destroyed by cropping and permanently alters the asset — so it is opt-in. The machine-readable C2PA/IPTC marking (always on) is the durable part.

Signing: self-signed vs trusted

By default Stamp generates a local self-signed certificate chain. The resulting C2PA marking is structurally valid and fully machine-readable — but provenance viewers will show the signer as "unverified" because the certificate isn't on the C2PA trust list.

Untrusted signer ≠ invalid AI assertion. The "unverified signer" note is about who signed (identity), not about the AI-disclosure content. The DigitalSourceType marking is present and readable regardless. To remove the note, sign with a trusted certificate via --cert/--key.

(DE) Nicht vertrauenswürdiger Signierer ≠ ungültige AI-Kennzeichnung. Der Hinweis „Signierer nicht verifiziert" betrifft nur die Identität des Signierers, nicht den Inhalt der AI-Kennzeichnung. Die DigitalSourceType-Markierung ist vorhanden und maschinenlesbar. Mit einem vertrauenswürdigen Zertifikat (--cert/--key) entfällt der Hinweis.

What survives

The embedded C2PA + IPTC marking is present after a download and local open, and major platforms (TikTok/YouTube/Meta/LinkedIn/Pinterest) read C2PA at upload and can re-label. Most social platforms re-encode and strip embedded metadata, so surviving a social re-upload needs durable watermarking — that is a planned v2 (not in this release).

API

import { generateSelfSignedSigner, stampBuffer, readStamp } from '@qadar-ai/disclose-stamp';

const signer = await generateSelfSignedSigner();
const stamped = stampBuffer(pngBuffer, 'image/png', 'ai-generated', signer, '0.1.0');
const info = await readStamp(stamped, 'image/png'); // { hasManifest, title, digitalSourceType }

Publishing (maintainers)

Publishing is gated and manual — the Publish disclose-stamp to npm workflow only runs via workflow_dispatch and defaults to a dry run. It uses npm OIDC Trusted Publishing (no long-lived token). npm requires a package to exist before a trusted publisher can be configured, so the first publish is manual:

1. First publish (creates the package). From stamp/ with dist built, sign in with 2FA and npm publish — this creates @qadar-ai/disclose-stamp on npm. (OIDC/CI cannot create a brand-new package.)
2. Add the Trusted Publisher. At npmjs.com/package/@qadar-ai/disclose-stamp/access → Trusted Publisher: Organization/user qadar-ai · Repository disclose · Workflow filename publish-stamp.yml · Environment name npm-publish · Allowed action npm publish (must be explicitly selected for configs created after 2026-05-20).
3. Every later publish is CI-only. Push a stamp-vX.Y.Z tag, then dispatch the workflow with dry_run=false. No token; npm authenticates via OIDC. Provenance attaches automatically once the repository is public.

License

MIT © Qadar AI. Built on the Content Authenticity Initiative's C2PA tools (@contentauth/c2pa-node, Apache-2.0/MIT), @peculiar/x509, and sharp.