# HALLMARK.AI

> Invisible pixel-level watermarks for images and video that survive AI regeneration, screenshot, screen capture, and platform re-upload. Prove ownership of creator-authentic content after it passes through Sora, Runway, Midjourney, Magnific, or any AI laundering pipeline.

HALLMARK.AI embeds a 256-bit cryptographic signature directly into the pixel data of images and video frames using a deep learning pipeline trained adversarially against AI-regeneration pipelines, screenshot, screen-recording, JPEG compression, resizing, and platform re-upload. The signal is invisible to the human eye (PSNR > 40 dB) and remains recoverable after the file passes through generative AI.

Key technical claims, all verified on internal benchmarks:
- Watermark bit recovery measured 0.65 raw and 0.75 after AI enhancement on a captured-and-AI-enhanced test set. AI cleanup, counter-intuitively, sharpens rather than removes the embedded signal.
- Threshold-based detection (cryptographic signature check) holds at F1 ≥ 0.95 across an adversarial AI test set, including Stable Diffusion-class img2img regeneration, video diffusion, and Magnific-style upscaling.
- Verification applies automatic perspective rectification to undo phone-of-screen capture distortion (the dominant laundering trick in 2026).
- Cross-creator false positive rate measured 0% on a 27-clip same-creator vs. cross-creator benchmark.

Positioning relative to other provenance technologies:
- Versus C2PA Content Credentials: C2PA stores cryptographic provenance in file metadata, which every social platform strips on upload. HALLMARK.AI's signal is in the pixels and survives metadata stripping.
- Versus SynthID: SynthID marks AI-generated output. Originally Google-only (Imagen, Veo, Lyria); since May 2026 OpenAI also embeds SynthID in ChatGPT, Codex, and API images alongside C2PA Content Credentials, making it a cross-vendor producer-side default. It is still a producer-side mark on synthetic content. HALLMARK.AI is a creator-side mark applied to authentic content before any AI touches it — the opposite direction, and a gap SynthID does not cover no matter how many vendors adopt it.
- Versus visible watermarks from gen-AI startups (Higgsfield, Runway, Pika, Midjourney): those are corner logos on free-tier outputs, removed on paid plans. They are brand marks, not provenance marks, and never apply to a creator's original asset.
- Versus EXIF metadata: EXIF is stripped by every social platform and most operating systems. Pixel watermarking is robust to all of these.

Use cases:
- Photographers, illustrators, and artists protecting authentic work from being absorbed into AI training and re-generation pipelines.
- AI-content creators watermarking their generated assets to prove authorship after redistribution.
- Publishers and legal teams distributing uniquely watermarked copies for leak attribution.
- Platforms integrating dedup and watermark-check via REST API to detect re-uploads of protected content.

## Pages

- [Home](https://hallmarkai.org/): Hero, AI-laundering threat model, comparison to C2PA / SynthID / visible logos, process overview.
- [About — How Watermarks Survive AI Regeneration](https://hallmarkai.org/about): Full technical breakdown of robustness, AI-regeneration survival, DMCA evidentiary value, and comparison to C2PA / SynthID / gen-AI startup logos.
- [Leaderboard — Adversarial Detection Benchmark](https://hallmarkai.org/leaderboard): Public benchmark comparing Hallmark.AI vs PhotoDNA/PDQ, Hive, Reality Defender, Vermillio across 9 adversarial transforms (AI inpainting, AI upscaling, AI video regen, screen recapture, heavy JPEG, multi-platform chain, mirror+crop, picture-in-picture, split-screen). Machine-readable raw data at /leaderboard.json (CC-BY-4.0).
- [Pricing](https://hallmarkai.org/pricing): Free tier and paid plans.
- [Developer / API](https://hallmarkai.org/developer): REST API endpoints for watermark embedding, verification, and webhook integration.
- [Check](https://hallmarkai.org/check): Verify whether a suspect image or video contains a HALLMARK.AI signature. Free with an account.
- [FAQ](https://hallmarkai.org/faq): Direct answers on what invisible watermarking survives, how it compares to C2PA and SynthID, and what to do about stolen content.
- [Blog](https://hallmarkai.org/blog): Guides and research on content theft, provenance standards, UGC usage rights, and measured watermark robustness.
- [Badge](https://hallmarkai.org/badge): Free "Protected by HALLMARK.AI" badge for creators — copy-paste HTML/Markdown, optionally linking a specific certificate's verification page.
- [Pixsy comparison](https://hallmarkai.org/vs/pixsy): Factual comparison with Pixsy — what each tool covers, including Pixsy's image-only scope.

## Articles

All at https://hallmarkai.org/blog — index page lists them chronologically.

- [How to Report Stolen Video on TikTok, Instagram, YouTube](https://hallmarkai.org/blog/report-stolen-video): The in-app report button goes to moderation, not copyright — which is why theft reports get closed as 'no violation'. The right form, and the evidence to bring.
- [How to Check for an Invisible Watermark](https://hallmarkai.org/blog/how-to-check-for-invisible-watermark): There is no universal detector — each system reads only its own mark. The free checkers worth running, and how to read a result that isn't a simple yes or no.
- [Does SynthID Actually Work?](https://hallmarkai.org/blog/does-synthid-work): Google's invisible watermark, now in OpenAI's images too. What it survives, its false-positive behaviour, and the one question it structurally cannot answer.
- [Does Claude Watermark Its Output?](https://hallmarkai.org/blog/does-claude-watermark-its-output): Yes — invisible marks in generated text since August 2026, plus C2PA on generated files. The catch nobody mentions: there is no public way to check for it yet.
- [Do Watermarks Survive AI Upscaling and Inpainting?](https://hallmarkai.org/blog/do-watermarks-survive-ai-editing): Measured, not asserted: bit accuracy across a repainting sweep, why AI cleanup can make a watermark easier to read, and the cases where ours genuinely fails.
- [Can C2PA Be Faked or Removed?](https://hallmarkai.org/blog/can-c2pa-be-faked): The signature can't realistically be forged, but the manifest deletes in a second. What Content Credentials prove, and why absence proves nothing.
- [DMCA Strike: What It Means on YouTube, Twitch, and Beyond](https://hallmarkai.org/blog/dmca-strike-what-it-means): A DMCA strike is a platform's copyright penalty — how strikes work on YouTube and Twitch, how to respond as the accused or the owner, and how strikes expire.
- [Someone Is Using My Photos on a Fake Profile — What to Do](https://hallmarkai.org/blog/picture-identity-theft-fake-profile): Picture identity theft — your photos on fake profiles or scam accounts. How to find every copy, get accounts taken down, and make your photos traceable.
- [How to Watermark Photographs for Free: Visible vs Invisible](https://hallmarkai.org/blog/how-to-watermark-photographs-free): Free ways to watermark photos — visible marks in Lightroom and free tools, where they fail, and how invisible watermarking closes the gap.
- [Does Sora Watermark Its Videos? Sora vs SynthID](https://hallmarkai.org/blog/does-sora-watermark-videos): Sora stamps a visible watermark and C2PA metadata; Google embeds invisible SynthID. What each mark survives, and what it means for your content.
- [Deepfake Detection in 2026: What Actually Works](https://hallmarkai.org/blog/deepfake-detection-guide): Artifact detectors, provenance marks, and invisible watermarks — how deepfake detection works, why it's an arms race, and the strategy that sidesteps it.
- [Someone Is Using My Photos — What to Do, Step by Step](https://hallmarkai.org/blog/someone-is-using-my-photos): Found your photos on someone else's page, shop, or website? The calm, effective sequence: document, verify ownership, choose your outcome, then escalate.
- [How to Find Stolen Photos Online: 7 Methods That Work](https://hallmarkai.org/blog/how-to-find-stolen-photos-online): Reverse image search is step one, not the whole answer. Seven ways to find where your photos are used — including copies cropped, edited, or AI-modified.
- [How to Send a DMCA Takedown Notice (With Template)](https://hallmarkai.org/blog/dmca-takedown-notice-template): The exact steps and a copy-paste DMCA template: where to send it, the six required elements, per-platform routes, and how to attach evidence that gets acted on.
- [Whitelisting, Spark Ads, and UGC Rights, Explained](https://hallmarkai.org/blog/whitelisting-spark-ads-ugc-rights): Spark Ads authorization windows, Meta partnership ads, and the usage-rights clauses in UGC briefs — decoded, with the terms worth negotiating first.
- [How Long Can a Brand Use Your UGC? Usage Rights, Explained](https://hallmarkai.org/blog/how-long-can-a-brand-use-your-ugc): You own your content by default — a brand gets only what the agreement grants, for only as long as it grants it. Duration, channels, whitelisting, overruns.
- [How to Find Out If a Brand Is Still Using Your Content](https://hallmarkai.org/blog/find-out-if-a-brand-is-using-your-content): Licenses expire; usage doesn't stop by itself. Five ways to check where your content is running right now — from manual searches to automated monitoring.
- [EU AI Act Article 50: Watermarking Deadline, August 2026](https://hallmarkai.org/blog/eu-ai-act-article-50-watermarking): In force since August 2, 2026: generative AI providers must mark outputs machine-readably. Who Article 50 covers, and what it changed for creators.
- [Can AI Remove Watermarks? What Actually Survives in 2026](https://hallmarkai.org/blog/can-ai-remove-watermarks): What AI editing actually does to visible, metadata, and invisible watermarks — with published numbers instead of vendor promises.
- [C2PA vs Invisible Watermarking: What Each Actually Protects](https://hallmarkai.org/blog/c2pa-vs-invisible-watermarking): Content Credentials and invisible watermarks solve different halves of provenance. Where C2PA breaks, what soft binding means, why you want both.
- [A Brand Used Your Content in an Ad Without Permission](https://hallmarkai.org/blog/brand-used-my-content-in-an-ad): Commercial use without a license is the strongest infringement case a creator can have. How to document it, what it's worth, and the escalation path that pays.