The two things people call an AI watermark
They are unrelated, and confusing them causes most of the disappointment in this category.
- The visible overlay — a semi-transparent logo or badge drawn in a corner. It sits on top of the image. The original pixels are still underneath, merely blended.
- The invisible provenance signal — an imperceptible fingerprint embedded in the content itself. There is no layer to peel off, because nothing was layered on.
A third marker also exists in the wild: Content Credentials, a signed provenance record built on the C2PA standard. It is metadata with a signature, and tampering with it invalidates the chain.
Side by side
| Model | Visible overlay | Invisible signal | Removable here? |
|---|---|---|---|
| Gemini (chat & app) | Yes — corner badge | SynthID | Overlay only |
| Nano Banana / Gemini 2.5 Flash Image | Yes — same badge | SynthID | Overlay only |
| Nano Banana 2 / Gemini 3.1 Flash Image | Yes — same badge | SynthID | Overlay only |
| Veo, Veo 3, Veo 3.1 | Yes — same badge | SynthID | Overlay only |
| Sora | Configurable | Provider-specific | Not handled |
The important column is the first one. Every Google model in that table stamps an identical overlay, which is why one tool handles all of them and why there is no need for a separate “Nano Banana remover” — the watermark being removed is the same watermark.
Why Google’s models share one overlay
Consistency, mostly. A single badge design applied across the image and video product lines means one geometry to document, and it is why Image Studio and Video Studio cover the whole range without needing a per-model selector. When Google has shipped Nano Banana 2 and Veo 3.1, the overlay has not changed, so neither has the removal method.
The invisible side is handled separately. Google states that all Nano Banana output carries SynthID, and that Gemini can check for it and report which parts of a video carry the signal — useful for verification, and the reason to ask Gemini rather than a remover when the question is whether a mark is present.
What “removable” actually means
For the visible overlay, it means arithmetic. A mark composited with a known alpha value satisfies
so the original is recoverable wherever α is below 1. Nothing is synthesised, which is why the output does not smear. The geometry detection problem is the hard part →
For SynthID, no equivalent exists. Google describes it as built to survive re-scaling, re-colouring and compression, and notes there is still a chance it will not be detected after heavy alteration. That is a statement about degradation, not an invitation. The full SynthID reasoning →
Sora, briefly
OpenAI’s approach differs enough that it is not handled here, and saying so plainly is more useful than a claim. Other tools target Sora output, and there is a real market for it. This site does not attempt it, does not pretend to, and will not take your file to find out.
Working out what your own file has
The model name is a decent clue but not a reliable answer, because the same family ships with different defaults and vendors change them. A three-step check that does not depend on guessing:
- Look for a visible badge. If there is a corner logo or sparkle mark you can see and point at, that is the recoverable layer. If there is not, there is nothing for an overlay remover to do, and a tool that claims otherwise is describing a different product.
- Ask the model provider. For a Google file, Gemini will check for SynthID and read Content Credentials if present, and it reports which parts of a video are marked. That is a better source of truth than any third-party page, including this one.
- Check the file’s history. This is the step people skip. A screenshot, a re-save, a crop, or an upload through a messaging app or social platform will have re-encoded the file. The visible badge will still be there, but the recoverable signal beneath it may already be gone, and the failure looks like a broken tool rather than a damaged input.
That third point is the most common reason a result disappoints. The mark is visible in every case, so its presence tells you nothing about whether the underlying pixels survived. Only the original export does.
Why vendors use two systems at once
The pairing is not redundant, it is aimed at different problems. The visible badge communicates to a human at a glance that an image is synthetic. SynthID communicates to software, at scale, without altering how the image looks or how large the file is. Removing one does nothing to the other, which is exactly why a tool can credibly offer the first and never the second.
It also explains the tone of most vendor documentation on the subject. Google describes SynthID as a protection measure, and its Terms of Service prohibit removing its branding or bypassing protective measures. Those are separate obligations, and satisfying one has never satisfied the other.
How to check your own file
- Ask Gemini directly whether your file carries SynthID or Content Credentials. It reports which parts of a video are marked, and it is the authoritative answer for a Google model.
- If there is also a visible badge, that is the part this site’s tools address.
- If your file has been screenshotted, re-saved or uploaded through a platform, say so before trusting any result — a re-encode may already have destroyed the recoverable signal.
A word on legality
This page describes what is technically present in a file. It is not advice that removing it is permitted. Google’s Terms of Service prohibit removing or altering its branding, and that can apply to material you otherwise own. The sourced legal analysis is here, and it is worth reading before using any of this commercially.