Gemini Watermark Remover

Removes the visible semi-transparent Gemini and Nano Banana sparkle from the bottom-right corner of an image. Gemini blends a white mark with a known transparency map, so the pixels underneath are calculated back instead of being painted over. Everything runs in your browser: the file is never uploaded.

Drop images here or click to choose You can also paste from the clipboard: Ctrl+V / ⌘V. PNG, JPEG, WebP · up to 25 MB · up to 10 files

Processing happens in your browser — images are not sent to our server.

What is removed and what stays. The tool removes only the visible mark. The invisible SynthID watermark is woven into the pixels of the whole image and stays: Google can still tell that its AI made the picture. How to check SynthID. Metadata (C2PA, XMP, EXIF) is copied to the new file by default; untick the box and the browser saves the picture without it, but that does not remove SynthID either. Check C2PA.

Use it responsibly. Process your own images. Do not pass off a generated picture as a real photo, and follow the rules of Google and of the platforms where you publish: many require AI content to be labelled.

How to Remove the Gemini Watermark from an Image

Images from Gemini and Nano Banana get a semi-transparent white sparkle in the bottom-right corner. Gemini adds it with alpha blending over a fixed transparency map, so the original pixels under the mark can be calculated back with a formula instead of being painted over. The tool finds the mark on its own (size and margin depend on the resolution), removes it and shows a before/after comparison. If the mark is not found, you can place the frame by hand and switch to filling the area from its edges. Everything runs in your browser with no upload.

Reverse alpha blending with Gemini mark transparency maps for 36, 48 and 96 px

Automatic detection: known Gemini layouts and resized copies

Manual frame by mouse or finger, plus edge-based fill-in

Before/after slider with a zoomed corner view

Up to 10 files at once, PNG, JPEG and WebP up to 25 MB; save as PNG or JPEG

Files are never uploaded: processing happens in the browser

How to Use

Drop your Gemini images, choose files or paste from the clipboard
Check the corner with the before/after slider and adjust the frame if needed
Pick a format and download the file without the mark

FAQ

No. Only the visible sparkle in the corner is removed. SynthID is an invisible mark spread across the pixels of the whole image; it stays, and Google can still tell the picture was made with Gemini.

The mark is not erased or painted over, it is subtracted: Gemini mixes white into each pixel in a known proportion, and original = (pixel − α·255) / (1 − α) gives back what was underneath. On an untouched PNG from Gemini only rounding error remains, about 1–2 brightness levels out of 255.

The tool still finds the mark in a resized copy, but restoration is no longer exact: scaling mixed the mark with neighbouring pixels, so a faint outline may remain. The best result comes from the original file downloaded from Gemini. For leftovers, use Strength or the fill-in mode.

By default, yes, when you save to the same format (PNG → PNG, JPEG → JPEG): EXIF, XMP, IPTC and C2PA blocks are copied from the original. After the pixels change, the C2PA signature no longer matches the image. Untick the box and the browser saves the file without metadata; SynthID still stays.

The exact method only fits the Gemini / Nano Banana mark. The fill-in mode technically works on any small area, but use it on your own images: removing someone else's authorship marks without permission is not allowed.

How to remove the Gemini watermark, step by step

To remove the Gemini watermark without a trace, the file matters more than the tool. Reverse alpha blending restores the pixels exactly only when the image has not been changed since it was generated. Work in this order:

Save the image from Gemini with the download button, not a screenshot. A screenshot rescales the picture and smears the mark into the neighbouring pixels.
Do not send the file through messengers or social networks before processing: WhatsApp, Telegram, Instagram and others resize and recompress images.
Drop the files on the page (up to 10) and wait until each one says “removed”.
In Corner zoom, move the slider and look at the spot where the mark was.
Download the result. Choose PNG for photos with smooth gradients, JPEG when the file goes on a website and size matters.

If the original is gone and you only have a smaller copy, the tool will still find the mark, but the result will be approximate. More on that below.

Where Gemini puts the mark and how big it is

The visible watermark is a white four-pointed sparkle in the bottom-right corner. Its size and distance from the edges depend on the image resolution. Google does not publish these numbers; the table shows what open-source projects measured across thousands of files. Layouts change over time, so the tool does not rely on the table alone: it searches for the sparkle by its shape.

Image	Mark	Margin from right and bottom
Small sizes, about 512 px per side	48 × 48 px	32 px
About 1024 px (1K)	48 × 48 px; some older files use 96 × 96 px	32 px; older files 64 px
2K and 4K (2048 px and up)	96 × 96 px	64 px
Some 2025–2026 formats	36, 48 or 96 px	96 or 192 px

Nano Banana is the name of the image model inside Gemini, so a Nano Banana watermark remover and a Gemini watermark remover solve the same problem: the mark is identical. If the picture was scaled down, the size and margin shrink in proportion, and the tool scans for them along the diagonal of the corner.

How to read the result

Every file gets a status that tells you whether anything needs to be done by hand.

“Gemini mark found: 48×48 px, 32 px from the right…” means the mark matched a known layout. Usually nothing else is needed.
The same with a note about resizing or recompression means the sparkle was found at a non-standard size, or part of its shape is still measurable after removal. Check the corner zoomed in.
“Found with low confidence” means the image has not been changed; the frame sits where something looks like the sparkle. Place it exactly over the mark and processing runs as soon as you let go.
“No visible Gemini mark found” means there is no sparkle-shaped mark in the corner. This happens with files generated through the API or already cleaned. If you can see the mark, place the frame manually.

What to do if an outline remains

A faint light or dark outline in the shape of the sparkle means the file was rescaled or heavily compressed as JPEG. Try these in order:

Adjust Strength: raise it by 0.02–0.05 for a light outline, lower it for a dark one.
Turn on Adjust frame and nudge it by 1–2 pixels; when you release it, the frame snaps to the sparkle.
On a plain background (sky, wall, gradient), switch to Fill in by sparkle shape. On a busy background fill-in leaves a blurry patch, so keep reverse blending there.

Reverse blending or fill-in: which to choose

	Reverse alpha blending	Fill-in (inpainting)
What it does	Subtracts the white using the transparency map	Fills the area with colours from the surrounding pixels
Original Gemini file	Restores the pixels almost losslessly	Unnecessary: detail under the mark is invented
Resized copy	Good, a faint outline is possible	Good on a plain background
Texture, fine detail, text	Keeps them	Blurs them
A mark that is not Gemini’s	Does not fit	Works on a small area

Why not just use a healing brush

An eraser, clone stamp or content-aware fill does not know what was under the mark, so it borrows similar pixels from around it. On a flat background nobody notices; on a face, grass or text it leaves a smudge. The Gemini mark is different: every pixel under it is mixed with white in a known proportion α. Example: the original pixel had a brightness of 100 and α is 0.4 at that spot. After blending it becomes 0.4 × 255 + 0.6 × 100 = 162. Reversing it: (162 − 0.4 × 255) / 0.6 = 100, exactly the original value. The same calculation runs for every channel of every pixel under the sparkle, which is why fine detail, noise and texture survive.

The method has one requirement: the exact α map and the exact position of the mark. Once the image is scaled down, the sparkle bleeds into neighbouring pixels and the formula becomes an approximation.

Batch processing and phones

You can add up to 10 files at once. They are processed one after another; switch between them with the thumbnails, and each keeps its own frame and method. Download all saves the files in sequence; your browser may ask once for permission to download multiple files. Only the bottom-right corner is recalculated, so processing is quick on a phone too; very large files (over 16 megapixels) may fail to save on older iPhones, in which case use a computer. There you drag the frame with a finger and resize it with the round handle in its corner.

Format and metadata when saving

PNG keeps the result lossless but the file is larger. JPEG at quality 90–95 looks the same to the eye and is several times lighter; if the image is going on a website, run it through the image compressor as well. WebP originals are saved as PNG.

The Keep metadata box copies EXIF, XMP, IPTC and C2PA blocks from the original when the format stays the same. Google adds Content Credentials (C2PA) to some images, and once the pixels change the signature no longer matches the picture, so a verifier will report the file as edited. You can see what the file carries before and after in the C2PA checker. If you do not need the metadata, untick the box or strip it separately with the EXIF remover.

What stays in the image after the mark is gone

The visible sparkle is only one of Gemini’s two marks. The other, SynthID, is spread across the pixels of the entire image and is designed to survive cropping, compression and edits. Removing the visible mark does not remove it, and this tool does not try. How to check for SynthID, and what can be seen without Google’s detector, is covered on SynthID Checker; for a broader look at AI signals in a file, use the image watermark detector.

In practice, removing the mark makes sense when it gets in the way of a design: a banner, a cover, a slide, an illustration for an article. It will not hide where the image came from, and that should not be the goal. YouTube, TikTok and Meta require realistic AI content to be labelled or label it themselves, and Google’s rules, like those of most platforms, prohibit passing off a generated image as a real photo.