Google's policy does not prohibit removing the visible Gemini watermark. The company's Generative AI Prohibited Use Policy never mentions watermarks at all, and since August 2026 Gemini ships a Media watermark setting that lets you switch the visible mark off yourself.
What the rules do target is a different act: presenting an AI-generated image as though a person made it. Taking the mark off and misrepresenting where the image came from are two separate things, and this page keeps them apart.
Two marks, not one
Almost every argument about "removing the watermark" goes wrong here, so it is worth separating them first.
- The visible mark. The white sparkle logo in the corner of a Gemini-generated image. It is ordinary pixels, composited over your picture. This is the only thing this site's tool touches.
- SynthID. An invisible pattern written into the pixels by Google DeepMind. You cannot see it, and it survives normal editing.
- C2PA metadata. Provenance information stored in the file itself, readable by other software.
Removing the first one does not remove the other two. That is not a limitation we invented — it is how Google designed it, and the company said so explicitly when it made the visible mark optional.
What Google's Prohibited Use Policy actually says
The policy that governs generative AI in Google's products was last modified on 17 December 2024. Read it and you will find no mention of watermarks, watermarking, or removing a mark. It is a list of things you may not do with the output — dangerous activity, compromising security, harmful content, and so on.
The clause that actually bears on this question is in the fourth section, and it is about provenance rather than pixels:
Misrepresenting the provenance of generated content by claiming it was created solely by a human, in order to deceive.
That is the line. Not "the mark must stay on", but "do not lie about where the image came from". A picture with the mark taken off and a picture you claim a photographer shot are treated very differently by that sentence, even though both are watermark-free.
Google made the visible mark optional in August 2026
On 14 August 2026 Google announced that the visible watermark is now a user setting rather than a fixed feature. Josh Woodward, the company's VP for Gemini, described the reasoning in the announcement post:
We're striking a balance here between creative control and safety: while the visible watermarks are now optional, invisible SynthID watermarks and C2PA metadata are still being used for transparency.
The control lives in Settings > Media watermark, applies to the Nano Banana, Omni and Lyria models, and was rolled out across Gemini and Google's video editor, Flow. In other words, Google's own position is that a Gemini image without a visible mark is a normal, supported outcome — as long as the invisible layers stay in place.
One practical consequence: if you have not generated the image yet, you do not need a remover tool at all. Turn the setting off before you generate, and you get a clean file directly. This page's tool exists for the other case — images you already have, generated before the setting existed or with it still switched on.
Where it does get risky
None of the following is about the mark itself. They are about the situations around it.
- An image you did not generate. If the picture came from someone else, the mark is not yours to take off, and the usual copyright and permission questions apply exactly as they would to any other edit.
- A claim of human authorship, where that claim is load-bearing. Editorial work, commercial deliverables, competitions, academic submissions, journalism. This is where the provenance clause above starts to matter, and it matters whether or not a mark is visible.
- Places with their own house rules. Stock libraries, ad platforms and newsrooms increasingly require you to disclose AI generation regardless of what the file looks like. Those rules bind you separately from anything Google says.
- Images that imply something about a real person. Putting a real, identifiable person into a generated scene and presenting it as real is a different order of problem from tidying up a logo in the corner.
What this tool changes, and what it leaves alone
Our tool rewrites the pixels inside the watermark box and copies everything else straight from your original file. It removes the visible mark. It does not touch SynthID, it does not strip C2PA metadata, and it does not alter anything else in the picture. You can read the full explanation of the method, or just run it on an image and compare the result yourself.
This is not legal advice
Everything above describes published documents — Google's policy, Google's announcement, and how the tool behaves. It is not a legal opinion, and we are not your lawyers. Rules differ by country, by platform and by what you intend to do with the picture. If the answer matters for money, a contract, or a client, ask a qualified professional rather than a website.
Sources
- Google, Generative AI Prohibited Use Policy — policies.google.com/terms/generative-ai/use-policy (last modified 17 December 2024). Section 4, clause 5 quoted above.
- Josh Woodward, VP for Gemini, announcement post on X, 14 August 2026 — quoted above.
- Google DeepMind, SynthID documentation — ai.google.dev/responsible/docs/safeguards/synthid.
- Our own tool's behaviour, measured on this site: only the pixels inside the watermark box are rewritten.