Frequently Asked Questions

An invisible watermark is a signature embedded directly into the pixels of an image or video. It is imperceptible to the human eye, but a detector can read it back and confirm who marked the file — even after the file has been edited, compressed, or re-uploaded. Unlike a visible logo, there is nothing to crop out or blur.

HALLMARK.AI uses a neural network to spread a 256-bit cryptographic signature across the entire frame as tiny pixel-level adjustments. Because the signature lives in the pixels themselves — not in metadata or a corner logo — it travels with the content through screenshots, platform re-encodes, and most edits. A matching detector network recovers the signature and verifies it against our registry.

Not visibly. The embedding makes sub-perceptual pixel adjustments — you can compare the original and watermarked versions side by side on our How It Works page and flip to an amplified difference view to see where the signature actually lives.

Metadata (EXIF, IPTC, C2PA manifests) is information attached alongside the file — and most social platforms strip it on upload. A watermark is embedded inside the pixels, so it survives the stripping. Metadata answers questions when everyone cooperates; watermarks answer them when they don't.

Yes — surviving screenshots and screen capture is a core design goal. Because the signature is spread across the frame's pixels, a screenshot carries it along. Our public benchmark page shows detection rates across screenshot, re-compression, and other transforms.

AI editing degrades watermarks, and heavy regeneration can destroy any watermark — no vendor can honestly claim otherwise. HALLMARK.AI is engineered so the signature survives realistic AI edits: upscaling, relighting, enhancement, and partial inpainting. In our published tests, the watermark remained detectable after commercial AI upscaling and after inpainting of up to roughly half the frame. Full re-generation from scratch is the hardest case, which is why we pair watermark detection with content matching that doesn't depend on the watermark at all.

The signature is embedded redundantly across the frame, so crops, platform compression, and re-uploads generally leave enough of it to detect. Extreme combinations (a tiny crop of a heavily re-generated frame) reduce confidence — our benchmark page publishes exactly how detection holds up transform by transform, rather than asking you to take our word for it.

Yes. Video is watermarked frame-level and verified against keyframes, so it survives re-encoding, clipping, and platform re-uploads. Video detection also handles common theft patterns like cropped verticals, picture-in-picture, and split-screen reaction formats.

C2PA attaches a signed provenance manifest to the file — valuable when platforms and tools preserve it, but it is metadata, and most social platforms strip it today. An invisible watermark is C2PA's recommended 'soft binding': it lives in the pixels and lets you re-identify the content after the manifest is gone. They are complements, not competitors.

SynthID marks content generated by Google's own AI models, so Google can label its outputs. HALLMARK.AI marks your original content — photos and videos you created — so you can prove ownership and find where they travel. Same core technology family, opposite direction: SynthID says 'an AI made this'; HALLMARK.AI says 'this is yours.'

No — and be skeptical of anyone who claims it can. A watermark doesn't prevent scraping or training. What it gives you is detection and evidence: if your marked content resurfaces — reposted, edited, or AI-modified — you can prove it started as yours and act on it.

Document first, act second. Save the URL, take screenshots with timestamps, and capture the account name. If the copy carries your watermark or matches your original, that is your ownership evidence. From there the usual path is: a takedown request on the platform (all major platforms have copyright forms), a DMCA notice to the site's host, or a direct license conversation if it's a brand using creator content commercially.

A recovered watermark is strong technical evidence that the file originated from you — it documents provenance in a way a takedown reviewer or opposing counsel can verify. It doesn't replace copyright registration (which unlocks statutory damages in the US), but it materially strengthens takedown notices and licensing claims. This is general information, not legal advice.

Use the free checker — upload any image or video and it reports whether a HALLMARK.AI signature is present. No account required, and files are processed in memory, never stored.

Article 50 of the EU AI Act requires providers of generative AI systems to mark AI-generated content in a machine-readable way, with obligations applying from August 2, 2026. Korea's AI Basic Act, effective January 2026, has a similar labeling requirement. These rules target AI providers rather than individual creators — but they are pushing the whole ecosystem toward machine-readable provenance.

Watermarking (MARK) is free, forever. TRACK — continuous monitoring that scans for your content across platforms and alerts you with evidence — is $29/month. There's a 14-day money-back guarantee on initial purchases.

Files are processed in memory and not persisted — what we keep is the watermark record (the signature and its registration), not your content. We don't train models on user content. Details are in the privacy policy.