Two kinds of watermark, one new tool

From today, VIPIX has a separate tool for a separate job: Watermark Remover. Not another button on an existing page, but a tool of its own, with its own editor.

This is not an ad. Three things: why the split was necessary, what the first version does, and what it cannot do yet — the last being the part most worth writing down.

First: which watermark do you mean?

A word problem that only exists in English. Watermark now names two things with almost nothing in common except the word.

The kind you can see

An overlay: a studio logo in a corner, a grid across a stock preview, a brand mark across the middle of a picture. Pixels laid on top, put there to be noticed. This is the kind our tool works on — the kind you mean if you can point at it.

The kind you can't

The other kind is a signal inside the file, meant never to be seen. The C2PA standard — the technical work behind Content Credentials — calls it a soft binding: "a content identifier that is either (a) not statistically unique, such as a fingerprint, or (b) embedded as an invisible watermark in the identified digital content." California's AI transparency law uses a related term, latent disclosure, and requires that it be "permanent or extraordinarily difficult to remove, to the extent it is technically feasible."

NIST files the pair as one design choice, and its wording is the shortest honest version: an overt watermark "can be perceived directly by the senses of a person who receives the content (e.g., a semi-transparent logo affixed to an image or video)", while covert ones are "machine-readable watermarks involving subtle perturbations of the content that are hard for humans to detect."

Opposite designs, then: a visible mark comes off with an eraser; an invisible one survives editing and has nothing on the surface to erase.

Watermark Remover works on the first kind. It is not a tool for stripping provenance data and is not built to defeat one. If the distinction is new, what actually changed in AI photo editing this year covers where the second kind came from.

The three photos people actually bring us

See whether the photo currently annoying you is one of these.

• A wedding shoot, paid for in full, and the studio still left its logo in the bottom corner.
• Product photos you commissioned, stamped with the photographer's mark, and now you need a clean copy to list.
• A photo you took yourself, which another site picked up and re-stamped with their logo.

One job: taking a mark off a picture you own. And none of them is "removing text".

Until this release, our Text Remover page tried to say both at once — its name said text, its description mentioned watermarks. A page doing two jobs does neither properly, so we gave it back to text and the marks moved here.

Why "text" and "a mark" are not the same job

We noticed this in our own site's search report. Across the whole list of queries bringing readers to VIPIX, the number of times anyone reached us by searching "watermark" or "logo" was zero — while searches for "remove text" existed.

One caveat before you carry that anywhere: it is a reading of our Vietnamese-language site's search data, in Vietnamese. It is not a US market figure, we have no US search data, and nothing about American search volume follows from it.

Read properly it is a gap, not bad news: people who need a mark removed do not say "text" — they type watermark, logo, copyright stamp, so they never met us at all.

Technically, the two jobs differ as well:

	Text	A mark (watermark)
Shape	thin, solid strokes, clear edges	often semi-transparent, blended in
Position	captions, corners, borders	corner, center, or tiled across the frame
Machine detection	it is text, so scanning works well	not text, so hard to detect automatically
What is behind it	usually flat	a face, a pattern, dense detail

A mark at thirty percent opacity lying across a patterned dress is not "text", and treating it as text is the surest way to get a blotchy result.

> Splitting the tools does not mean Text Remover was abandoned. It is still the right tool for captions, camera date stamps and signatures — and now it does that job without being asked to do another.

What the first version does

The shape of the work is simple: upload the photo, settle where the mark sits, remove it. The middle part is where the editor earns its keep, and it opens on the step that tries to do it for you.

Let it find the mark first

Signed in, the editor opens on its detection step. Press detect and the AI reads the picture and outlines the marked area — a translucent stamp, a corner logo, a copyright line. If there is lettering inside the mark it pulls that text into a box below, and trimming the words you want kept shrinks the outline around them instead of making you draw again.

Check the outline before removing anything. Detecting the mark and correcting the outline do not cost credits — run detection, adjust, run it again; nothing is charged until you press remove.

Brush over the mark

Paint over the area the mark covers. Brush size is adjustable, with an eraser for going outside the lines and a clear-all to start over.

Tap to select

Tap directly on the mark — convenient when it sits in one tidy place rather than spreading across the frame.

Then the AI lifts the mark off and rebuilds the background behind it, following the texture, color and light around it. It runs in the browser, on computer or phone, with nothing to install.

How we tested it, and why every before/after is ours

We watermarked our own images on purpose

We did not go looking for other people's watermarked photos to test on. We took images we hold the rights to and stamped them ourselves, with the image library VIPIX already runs, then asked the new tool to take the stamps off. That gives us the most valuable thing you can have when scoring this work: the unmarked original. No guessing what a patch should look like — the answer sits next to it.

The four groups we score against

Product shots on flat backgrounds, landscapes with sky, portraits, and images with a mark tiled across the frame. They are not equally hard, and we do not expect an even result.

> This article does not publish scores, deliberately. From how the fill technique works we can say the qualitative part now: the simpler the background and the smaller the mark, the cleaner the result. The before/after set goes in after the scoring round — we are not pasting a number in ahead of the measurement.

What it can't do yet

Three things, stated plainly.

Detection is not a guarantee — some marks you still have to point at. The automatic pass catches most of what people bring us, but a very faint stamp, or one sunk into a busy background, comes back outlined short. That is what the brush and the tap are for, and why neither is going away. Two edges of the same limit: detection sits behind signing in, so trying the tool on the page without an account gives you the brush only; and it reads what a mark looks like, never what it means — it cannot tell a logo you own from a logo you don't.

A mark tiled across the whole frame, over a busy background, is still a hard case. A repeated grid forces the AI to rebuild nearly the whole picture, so the outcome is decided by what lies underneath rather than by the grid. Over sky, water or a plain wall it rebuilds seamlessly; over a detailed facade it can come back with streaks. Worth zooming in before you download.

Which is why there are two levels of processing behind the remove button, not one. The standard level covers the ordinary work — a corner logo, a faint stamp over a simple background. A heavier level exists for the awkward cases, and they are the ones just named: a diagonal mark, a grid tiled across much of the frame, a mark lying over somebody's face. The heavier level costs more per run, and the price for the run you chose shows on the button before you press it.

Images only, not video. A screenshot of a video frame is fine; a video file is out of scope.

Whose photo is it?

The tool is for photos you have rights in — ones you shot, ones you bought, ones an old service stamped while you were the owner. Not for marks on other people's pictures. Stock previews are the obvious case: the mark signals a preview, and the license you hold decides what you may do with the file.

Two provisions are worth knowing by name and no more than that here: federal law on removing copyright management information turns on intent rather than on the act of erasing, and a transfer of copyright ownership is not valid unless it "is in writing and signed by the owner of the rights conveyed" — which is why paying an invoice is not the same as owning the copyright.

That subject has its own article: watermarks, copyright, and when you can legally remove a mark. This is general information, not legal advice.

Which tool to open

Logo, graphic or translucent stamp: Watermark Remover. Lettering only — caption, date stamp, signature: Text Remover is faster. The step-by-step walkthrough covers the awkward ones: how to remove a watermark from a photo you own.

If it is your image and you just need a clean copy, open Watermark Remover.

Frequently asked questions

Can it find the watermark for me?

That is the first thing it tries. Signed in, the editor opens on detection, outlines the mark and lets you correct the outline before anything is removed — and neither step costs credits. Faint marks and marks buried in detail can still come back outlined short, which is what the brush and the tap are for.

Will this remove Content Credentials or EXIF data?

Any edit-and-re-export can affect the data traveling with a file, and we have not measured what our export path does to it, so we will not claim an answer either way. If that data matters to your work, check one file yourself first.

How this article was made

• Written — seo-content produced the en-US version of /tin-tuc/vipix-ra-mat-cong-cu-xoa-watermark-ai under brief I18N-11, 2026-08-29.
• Reviewed — seo-review re-ran the §G publication gate on 2026-09-06 (VIPIX session s148) on an earlier revision and returned PASS-WITH-FIXES, and the next revision fixed the issues that verdict names. Verdict on file: reviews/en-US-regate-s148-verdict.md (earlier gates, 2026-08-29: en-US-T1-regate-verdict.md, en-US-T2T4-gate-verdict.md, en-US-T3-gate-verdict.md). Since then, the article's internal links have been updated, and on 2026-09-28 this note was corrected.
• Process — content drafted by the Vipix SEO team's LLM under a reviewed brief; edited and checked by the Vipix SEO team; no native-speaker contractor was engaged.