What a PDF watermark actually is
That distinction decides everything. A vector object can be deleted cleanly. Baked-in pixels cannot — they have to be reconstructed. Treating the two the same way is exactly why "remove watermark" tools so often return a page with blank holes where words used to be.
The four cases, and the fix for each
Header/footer text, "Draft", "Confidential", large diagonal marks — drawn as PDF text objects.
PDFnoted: PDFnoted removes the watermark object from the page content stream. Nothing underneath was ever covered, so the document comes back byte-clean and fully selectable.
The mark is baked into a flat image — scans, phone photos, CamScanner-style exports.
PDFnoted: PDFnoted runs OCR to rebuild a text layer, then reconstructs the pixels the mark covered from the surrounding layout, so the page stays searchable and copyable.
Approval stamps, "Copy" seals, signature images placed over the content.
PDFnoted: Stamp areas are detected and inpainted; where a stamp sits over a table, the ruling lines are re-drawn from the surrounding grid.
Invisible tracking marks, document metadata, hidden objects, embedded attachments.
PDFnoted: PDFnoted clears document metadata, hidden objects and attachments as part of the clean-up, before you send the file on.
How PDFnoted removes a watermark
- Parse the filePDFnoted reads the PDF object tree and page content streams to locate watermark objects, annotations and hidden layers.
- Classify the markEach mark is classified as vector text, raster, stamp or hidden metadata — because each type needs a different fix.
- Reconstruct the covered contentFor marks baked into an image, the covered region is rebuilt from the surrounding layout: text, table ruling lines and charts are redrawn.
- Verify and exportThe result is checked so text stays selectable and the layout is unchanged, then exported as a standard PDF.
Compared with the usual approaches
| Opaque eraser | Typical online tool | PDFnoted | |
|---|---|---|---|
| What happens to the text under the mark | Covered by an opaque block — unreadable | Often lost or flattened | Reconstructed, stays readable |
| Scanned pages and phone photos | Not handled | Rarely handled | OCR + reconstruction |
| Stamps and seals over tables | Blocked out | Blocked out | Inpainted, ruling lines redrawn |
| Hidden metadata and tracking marks | Not detected | Not detected | Cleared |
| Text stays selectable after cleanup | No | Sometimes | Yes |
| Where the file is processed | Varies | Uploaded to a server | Basic processing in your browser |
When you should — and should not — remove a watermark
Being straight about this matters more than it looks. Cleaning up your own documents is routine. Altering someone else's to change what it says is not.
✓ Legitimate uses
- Your own scanned contracts, receipts or internal reports
- Documents you have been authorised to process by the owner
- Templates or exports where your own tool added the mark
- Cleaning up marks before an internal review or filing
✗ Not legitimate
- Removing a mark to redistribute someone else's paid material
- Altering a document so it misrepresents what was agreed or approved
- Stripping a copyright notice you do not own
- Changing evidence or a signed record
Frequently asked questions
What is a PDF watermark, exactly?
A PDF watermark is any mark layered onto the page to signal status or ownership — "Draft", "Confidential", a company logo, a stamp, or a faint diagonal text. Technically it is either a separate object in the page content stream (a vector watermark) or pixels inside a flat image (a raster watermark). The distinction matters: an object can be deleted cleanly, while pixels have to be reconstructed.
Why does erasing a watermark usually ruin the text underneath?
Because most erasers paint an opaque rectangle over the mark. If the mark sat on top of a flat image — a scan or a photo — there is no separate layer to remove, so the same pixels hold both the mark and the content. Covering the mark therefore covers the content too. This is why "remove watermark" results often come back with blank patches where words used to be.
Can I remove a watermark from a scanned PDF or a phone photo?
Yes, but it needs a different method. PDFnoted runs OCR to rebuild a text layer over the scan, identifies the marked region, and reconstructs what the mark covered using the surrounding layout. The page comes back searchable and copyable instead of a flat image with a hole in it.
Will removing the watermark damage the layout?
The layout is preserved. For vector watermarks the object is removed and nothing else moves. For raster marks, only the marked region is reconstructed, and table ruling lines are redrawn from the surrounding grid so the structure does not collapse.
Is it legal to remove a watermark from a PDF?
It depends on the document and your rights. Cleaning marks off material you own or are authorised to process — your own contracts, scans, internal reports — is normally fine. Removing a mark in order to redistribute someone else's paid material, to strip a copyright notice, or to make a signed record say something it did not say, is not. When in doubt, ask the owner or your counsel.
Can I remove a watermark on a phone?
Yes. PDFnoted runs in a mobile browser — nothing to install. Upload the file, run the removal, download the clean copy.
Does the file leave my device?
Basic processing runs locally in your browser, so the document does not leave your device. Advanced AI reconstruction runs in the cloud; in that case only the pages that need repair are sent and they are not retained after processing.
What file types work besides PDF?
The workspace accepts PDF as the working format and can convert common inputs — images and Office documents — into PDF first, so you can clean a scan or a Word export in the same flow.
Clean up that PDF now
Upload the file, let the AI find the marks, and download a clean copy with the text underneath intact. Free to start, nothing to install.
Remove watermark now — free to startKeep going
Sources and further reading: ISO 32000 (PDF specification); Adobe PDF Reference; Schema.org structured data; Aggarwal et al., GEO: Generative Engine Optimization (arXiv:2311.09735). This page is a technical guide, not legal advice.