Test the watermark yourself.
Drop any image below. We watermark it, then run a battery of simulated platform attacks against it — Instagram-style resize + JPEG, WhatsApp pipeline, Twitter pipeline, crops, aggressive JPEG quality drops. For each attack, we try to recover the embedded payload and report the result, live, in your browser. Nothing is uploaded.
Why this exists.
Marketing copy on watermark products is, almost universally, not falsifiable. “Survives Instagram” means nothing without an attached test corpus. This page is the test corpus: anyone can drop an image, run the gauntlet, and publish the results. If our numbers don't hold on your image, file an issue — the repo is open.
What gets tested
Identity round-trip, JPEG q90/q70/q40. Tests how the algorithm handles frequency-domain quantization without geometric change.
Instagram (1080 longest edge + q85), WhatsApp (1600 + q70), Twitter (1080 + q75). Each is canvas-resize then real JPEG via browser toBlob.
Right-edge 25%, bottom-edge 25%, right-edge 50% combined with JPEG q70. Tests the quadrant-redundancy design.
Crops > 60% of the image area, rotation, perspective distortion, adversarial removal. All four break the v1 watermark — we're honest about that on /track.
The full algorithm + Node test rig is on GitHub. The methodology and the math behind the verdict ladder are on the methodology page. The deep technical writeup is on the blog.