Gemini Watermark Remover: Clean Copies of Your AI Images

Which watermark layer this removes — and which it doesn't

Gemini images actually carry up to three marks. The first is the visible one: a semi-transparent logo blended into a corner of the image at a known size and margin. The second is SynthID, Google's invisible watermark woven into the pixel patterns themselves, designed to survive cropping, compression and editing. The third is a metadata label — fields in the file that say the image was made with Google AI.

This tool removes the first layer only: the visible logo. SynthID is left exactly as it is — removing it is neither possible with this technique nor something this site offers. The metadata layer disappears as a side effect of how the download is produced, and you can verify that with the metadata remover on our homepage. In short: the corner logo is gone, but the image remains identifiable as AI-generated by SynthID detectors. We think that's the right trade.

Exact restoration, not AI repainting

Most watermark removers are inpainting tools: they erase a rectangle and ask a model to imagine what might have been there. That works on photos of skies and walls, but it invents detail — and on precise artwork it smudges.

This tool does something different. Gemini applies its logo with standard alpha blending: each watermarked pixel is a known mix of the logo pixel and your original pixel. Because the logo artwork, its opacity map, and its position follow fixed patterns, that equation can be run backwards — subtract the logo's contribution, divide by the remaining opacity, and the original pixel comes back out. Nothing is guessed. Pixels outside the watermark area are copied through untouched.

The engine is the open-source gemini-watermark-remover project (MIT licensed), which maintains a catalog of Gemini's watermark sizes, positions and opacity maps across output resolutions — including the layouts Google introduced in mid-2026 — and validates every removal before accepting it.

Use the file exactly as Gemini saved it

The detector anchors on Gemini's original output dimensions: it knows which watermark size and margin to expect at each resolution the model produces. That's what makes exact reversal possible — and it's also the main way removal can fail. If an image was cropped, resized, screenshotted or run through a chat app that recompressed it, the anchor no longer lines up and the tool will report that no watermark was found rather than damage your image guessing.

So the golden rule is: feed it the file exactly as Gemini or AI Studio saved it. If you've already edited your image, go back to the original download, remove the watermark first, then edit.

Nothing to upload, nothing to trust

Like the photo tools everywhere else on this site, this runs entirely in your browser: the page ships the removal code to your machine, and your image never crosses the network. You can verify that claim in ten seconds — open DevTools, watch the Network panel, and process an image. Zero uploads.

That matters more than convenience. Unreleased product mockups, client work, images with your prompts' fingerprints — none of it should take a round trip through a stranger's server just to lose a logo. Here it doesn't.

A note on using AI images honestly

Removing the visible watermark from an image you generated, for use in your own deck, mockup, video or post, is an aesthetic choice. Removing it to pass AI output off as human work, as photographic evidence, or as someone else's art is a different act — and it mostly won't work anyway, because SynthID and platform-level AI detectors don't rely on the corner logo. Google's terms for the Gemini API and consumer apps also have their own rules about provenance disclosures, and they're worth a read if you're publishing at scale.

Our position is simple: we'll clean up the corner of your image, and we'll be straight with you about what stays detectable. How you present your AI images is on you.