Snitch Tracker
local · no upload · no signup
§ Catch the snitch

Find out who leaked it.

Give a different invisibly-watermarked copy to each person. Every copy looks identical to the original — no QR, no visible mark. When one leaks, drop it below and we'll tell you which recipient's copy escaped, with the full timestamped receipt.

§ What this survives

Three independent invisible signals are embedded into each copy, each tuned to a different attack class:

  • Direct file shares (email, AirDrop, Signal-as-document, e-discovery): the lossless LSB layer catches the leaker.
  • JPEG re-compression down to quality ~40: the block-DC and DWT layers survive.
  • Resize from 0.5× up to 2× (covers Instagram, WhatsApp, Twitter, Telegram, Signal's default image mode): the DWT+DCT+SVD layer survives.
  • Partial crop up to ~50% of the image area removed from any single edge: the quadrant-redundant embedding keeps a recoverable copy intact.

Combined attacks recover in our synthetic 512×512 test suite, including the realistic Instagram pipeline (1200×800 PNG resized to 1080 longest edge + real JPEG q85 with chroma subsampling). Real-world content varies with image type — run the rig at /lab against your own material before treating the synthetic numbers as a guarantee. The full method is on the methodology page.

Limits we're honest about: crops > 60% (most of the image removed), rotation more than a few degrees, and adversarial-removal attacks where someone reads our open code and engineers a stripping pipeline. Those break it.

§ When you'd use this
Group chat leak

You shared a sensitive screenshot in a friend group of three. It ended up on Twitter. Mint one watermarked copy per friend before you share; when a leak shows up, drop it here and the variant id tells you which friend.

Creator content (OF / Fansly)

Per-subscriber watermarked copies, free, browser-only. When content leaks, the variant id points to the subscriber. No $30/mo SaaS required — full creator workflow.

Confidential draft → PR firm

Send three watermarked copies of a pitch deck to three PR firms. When the deck appears on a competitor's TikTok the next week, you know which firm to never work with again.

01
Produce

Mint one watermarked copy per recipient.

Drop the image to protect
next: name your recipients · PNG, JPG, WebP · up to 25 MB
Each recipient gets a copy that looks identical to the original but carries an invisible per-recipient fingerprint. Everything happens in your browser — the image bytes never reach us.
02
Verify

When one leaks — drop it here.

Drop the leaked image here
we extract the invisible per-recipient watermark
Runs locally in your browser. The image bytes never touch our servers — only the recovered variant ID is looked up.
Invisible

The watermark cannot be seen. Each copy looks pixel-identical to the original you uploaded.

Three-layer fingerprint

LSB for lossless re-shares, block-DC for JPEG, and DWT+DCT+SVD for platform pipelines that re-encode AND resize AND partially crop.

Local

Watermarking and detection both run in your browser. The image bytes never reach our servers — only a 32-byte hash does.

The robust layer is a browser-side port of the ShieldMnt invisible-watermark algorithm (DWT+DCT+SVD on the chroma channel), patched with quadrant redundancy and a CRC-validated recovery search. Read the methodology for the algorithm and the empirical robustness numbers.