Removing a watermark from your own work, and where the line is

What the mark is for

The corner logo is not a licence stamp and it is not there to stop you using the image. It is a disclosure. It exists so that a person looking at a picture can tell, without investigating, that a model produced it.

That matters because the default assumption about a photograph is that something happened in front of a camera. Generated media breaks that assumption, and it breaks it invisibly. The mark is an attempt to make the break visible at a glance.

It is a weak mechanism, and everybody involved knows it is weak. It survives only as long as nobody removes it. That is precisely why the stronger, imperceptible signal exists alongside it, and why the two should not be confused. That distinction is covered in SynthID and the visible watermark are not the same thing.

The cases that are straightforwardly fine

Most of the traffic to a tool like this is unremarkable, and it is worth saying so plainly rather than treating every user as a suspect.

• Assets in something you are building. A texture, a background, a placeholder, an icon. The image is a component of a larger work, nobody is being told it is a photograph, and a logo in the corner is simply in the way.
• Mockups and drafts. Concept work, storyboards, pitch material. The image is standing in for something, and its provenance is not a claim anyone is relying on.
• Personal use. A wallpaper, a print, something you made for yourself.
• Work you will disclose anyway. If the caption says the image was generated, the corner logo is redundant. The disclosure has been made, more clearly than a small mark would have made it.

What these have in common is that nobody is misled about what the image is. That is the thread worth holding onto.

The cases that are not

The same thread identifies the other side. Where someone is going to act on a belief about the image's origin, removing the marker of that origin is the harm, and the tool used to do it is beside the point.

• Presenting generated media as documentary. News, evidence, journalism, anything where the picture is offered as a record of something that happened.
• Depicting real people doing things they did not do. This is harmful regardless of watermarks, and removing the mark is an aggravating step rather than the offence itself.
• Submissions with rules about it. Competitions, stock libraries, client work, academic submissions and platforms that require AI content to be labelled. If the rules say disclose, stripping the marker is a breach whatever you think of the rule.
• Anything where authenticity is the value. Listings, profiles, reviews, claims about a product. If the picture is doing work that depends on it being real, it needs to be real.

None of this is specific to watermarks. Misrepresentation was already the problem before generated media existed. What has changed is how easy it has become and how hard it is to detect by eye.

What you agreed to

Generating an image through Google's products means accepting their terms, and those terms carry obligations about how the output is used and represented. They are worth reading rather than assuming, they differ between free and paid tiers, and they change.

That is a separate question from whether a given act is ethical, and the two do not always line up. Something can be permitted by the terms and still be a bad thing to do, and the reverse happens too. Both are worth thinking about, and neither substitutes for the other.

Removal does not achieve concealment

There is a practical point that cuts across the ethical one, and it is worth stating because it changes what is actually possible.

Taking off the visible mark does not make a file indistinguishable from a photograph. The imperceptible provenance signal is a separate system, it is embedded in the content rather than laid on top of it, and it is designed to survive the ordinary transformations that media goes through. This tool does not touch it, does not detect it, and has no mechanism that would act on it.

So anyone removing the corner logo in the belief that it launders the file's origin is mistaken about the technology, quite apart from being mistaken about the ethics.

Disclosure is becoming an obligation

The norm is moving from optional to required. The EU AI Act includes transparency obligations around synthetic content, with specific requirements for material that depicts real people and events. Several platforms require AI-generated media to be labelled and apply their own detection independently of whatever the file carries. More of this is coming rather than less.

Rules in this area are changing quickly enough that anything written here will date. If you are publishing at any scale, check the current position for your jurisdiction and your platform rather than relying on a guide.

The practical upshot is worth noting though: a label in the caption discloses far more clearly than a small corner logo ever did. If you are going to disclose, disclose properly, and then the mark genuinely is redundant.

Where this leaves the tool

This site does one narrow thing: it inverts a known alpha composite over a known region, in your browser, without the file going anywhere. It cannot tell what you intend to do next and does not try to guess.

That puts the judgement with you, which is where it belongs and also where it is unavoidable. The question worth asking before publishing anything generated is simple enough: is anyone going to believe something about this image that is not true, and does that belief matter to them?

If the answer is no, the corner logo was just clutter. If the answer is yes, the watermark was never the thing standing in your way.