Know where your photos go, and who let them out.
An invisible mark in every copy you send out. When one leaks, a scan names whose copy it was, even from a screenshot.
- Independent timestamp on every registration (RFC 3161)
- Five interface languages
- An open lab: try to beat it yourself
One mark. Four kinds of leak.
Pick a field. Every card is one real situation and what the scan says afterwards.
Wedding previewThe preview that went public
Previews sent to the couple and two vendors. One shows up online early. The scan names the copy.
See the full case
Design conceptThe concept a competitor saw
A pitch went to three people. A rival had it a week later. Now you know which one.
See the full case
Repost accountThe account that “found” it
Your photo, cropped and filtered, on someone else’s feed. The scan shows it’s yours.
See the full case
Real estateThe photos of a listing are copied into a fake advert on another portal.
The scan says: whose photo it is and which copy the scammer started from.
See the full case
What came back from real files.
Scanned by hand through the app, 4 and 5 October 2026. The misses are listed too.
A screenshot of a Facebook post, on a computer and on a phone, sent on through WhatsApp.
ReadA phone photo of a laptop screen, with a hand over half the picture, from a distance and at an angle.
ReadAn AI tool that erased a tree and repainted the sky: read from the grass, three bits corrected.
ReadA full AI regeneration of the same composition: the picture recognised (72 % match), the mark gone.
Recognised
Numbers we can stand behind.
From our own test series and real chains, on small samples; the lab lets you repeat any of it. Where there is no number yet, there is none here either.
- 98/100
- Reads after 25 attacksEnhanced level, four photographs, every attack in the catalogue. The two misses were a blurry phone photo at a steep angle.
- 118/120
- Right recipient namedCopies issued to three recipients, ten chains each. Not one wrong name; two were unreadable.
- 0
- False positives305 scans, nine unmarked originals, our whole stress series: never a name that was not there.
- 99.3–99.8 %
- Of the picture unchangedSimilarity of the marked copy to the original (SSIM, Enhanced level). The difference is in the pixels, not on them.
- 66/66
- Reads on MaximumEleven photographs, six attacks the calibration never saw: JPEG 70, half size, screenshot, 50 % crop, Facebook feed, blur.
- 3
- Protection levelsStandard, Enhanced, Maximum. The level is fitted to each photo, and the app tells you how visible the mark would be before you choose.
The mark in the copy · An independent timestamp · Anchored in Bitcoin
Mark. Send. Scan.
Three steps, done once. From the first upload to the scan that names the copy.
dipmark is open by invitation while we test it.
We’re working with a small group of photographers, studios, agencies and companies that share sensitive images. Tell us what you share, and we will open an account for you.
Straight answers.
What the mark changes, what it survives, who can see your images.
Not to the eye. The mark adapts to each image and stays out of sight on smooth areas like sky and skin.
Every copy you send out should know who got it.
Access is by request while we test. Tell us what you share, and we open an account for you.