Invisible Watermark vs Metadata (EXIF): What Survives a Re-Upload

Short answer: Metadata (EXIF, XMP, PDF document info) is a label attached to the file. It's precise while the original file travels intact, but a screenshot, a re-export or many upload pipelines drop it. An invisible watermark lives in the pixels themselves, so part of it can survive those steps, with lower confidence. Using both gives you a strong signal when the original file leaks and a fallback when only a copy of the pixels does.

Last updated: October 12, 2026. Practical guide, not legal advice.

What metadata actually is

Every photo and PDF can carry extra fields next to the content:

EXIF: camera model, date, sometimes GPS location, written by cameras and phones.
XMP: a more flexible format used by editing apps for authorship, copyright and custom fields.
PDF document info: title, author, creator app and similar fields.

Metadata is useful because it's exact and machine-readable. Anyone can open the file and read it. That's also its weakness: anyone can edit or remove it with free tools, and a lot of software removes it without being asked.

What happens to metadata on the way out

What happens to the file	Metadata	Invisible watermark in the pixels
Original file forwarded as-is	Kept	Kept
Screenshot of the file	Gone (the screenshot is a new file)	Partly kept, if enough of the image is visible
Re-saved or re-exported as a new JPEG	Often dropped, depends on the app	Usually kept, at lower confidence
Uploaded to a social network or chat app	Many platforms strip most of it	Depends on how much the platform resizes and compresses
Someone deletes it on purpose	Gone in seconds	Has to be destroyed by visibly damaging the image

The pattern is simple: metadata is all or nothing, and it's the first thing to disappear. A watermark woven into the image degrades gradually instead.

Markedfile

Start today

Imatag

Talk to sales

Why you still want metadata

Metadata isn't useless just because it's fragile. When someone forwards the original file, which happens often with PDFs and email attachments, a signed metadata mark is the cleanest evidence you can get: it identifies the copy directly instead of being inferred from a pattern.

The key word is signed. A plain author field can be typed by anyone. A mark signed with a secret key can't be forged convincingly by someone who doesn't have that key.

Why you also want an invisible watermark

Most leaks don't travel as the original file. People screenshot a page, crop a photo or upload a copy to a group chat. In all of those cases the metadata is gone, and the only thing left to examine is the image itself. If the copy you find still contains enough of the pattern, detection can point back to the copy it came from.

Mark

Leak

Detect

How Markedfile combines the two

Every file you mark gets both layers:

A signed metadata mark. In images it's written to XMP, EXIF or PNG text fields; in PDFs to XMP and the document info. It's signed with a key unique to your workspace.
An invisible pattern in the image or page content that ties the copy to the identifier you chose.

When you run detect on a file you found, Markedfile checks both and reports a confidence score. As reported scores (not guaranteed detection): about 99% for the exact file, 95% from the signed metadata mark alone, and about 81% for a re-saved JPEG or a screenshot, where only the visual pattern is left. A match identifies a copy, not intent, and is not legal proof.

Limits to know

Size: roughly 256 px of a detailed area is the approximate minimum. Images and video frames under 96×96 px are rejected; everything else uses the same pattern.
Rotation: most 90° rotations are handled. Only a 90° rotation combined with strongly uneven stretch truly fails; an upright stretched image still detects.
PDF screenshots: best effort, not guaranteed. As an approximate minimum, a ticket needs about 200 pt, a photo about 160 pt, a slide about 320 pt, and a text page about the top half. Dense text alone is often not detected. Older PDF marks aren't compatible with the current invisible mark, so re-mark older PDFs.
Video (Business and Enterprise only): inputs are MP4, MOV, WebM and MKV, up to 10 minutes and 100 MB. The download is MP4. The same pattern is written on every frame. It still reads after H.264 re-encoding up to CRF 23, a trim, a slow zoom, or a downscale to 1280×720 or 854×480, even with the metadata stripped. A picture-only match stays at or below 85% confidence. Signed metadata adds confidence when it is kept. Deleting the record ends detection. Free, Starter, and Pro have no video and no API.
Nothing is 100%. A match identifies a copy, not who shared it or why, and it isn't legal proof on its own.

A quick test you can run

Mark a photo or a one-page PDF.
Download the marked copy and run detect on it: you should see the highest confidence, because the metadata is intact.
Take a screenshot of the same file and run detect on the screenshot: the metadata is gone, so the score now comes from the pattern alone.
Try a smaller crop until it stops matching, so you know the practical limit for your own files.

Where Markedfile fits

Markedfile is self-serve forensic watermarking for images, PDFs and, on Business and Enterprise, video. Plans: Free; Starter $9.99; Pro $19.99 (no active-file cap); Business $49.99 (shared workspace, video, API); Enterprise on a custom quote. Free, Starter, and Pro have no video and no API. The API is POST /api/v1/mark and POST /api/v1/detect on Business and Enterprise (docs).

It was built because teams needed a way to mark files before sharing them (PDFs, photos, identity checks) and no self-serve tool did it.

For more on the visible side, see visible vs invisible watermarks and can you trace a screenshot back to its source.

FAQ

Does removing EXIF remove the watermark?
No. Stripping metadata removes the signed metadata mark, but the invisible pattern is in the pixels, not in the metadata fields.

Can I just add my name to the EXIF copyright field?
You can, and it's harmless, but it disappears with a screenshot or many uploads and anyone can edit it. It says who claims the file, not which copy leaked.

Is metadata ever enough on its own?
When the original file is forwarded, a signed metadata mark gives a strong result (95% on its own as a reported score). It's the screenshots and re-uploads where you need the pattern.

Does the invisible watermark change how the image looks?
It's designed not to be noticeable in normal viewing. If you also want a deterrent people can see, add a visible watermark on top.

Next steps

Mark one file, then compare detect results on the original and on a screenshot. Plans are on pricing.

Mark each copy before you share it

Create a free account. You can mark 5 files and see whose copy leaked.

Create account See how it works