How to Remove the Gemini Watermark (Nano Banana) from Images

What is the Nano Banana watermark?

Every image generated by Google Gemini (including Gemini Advanced and Google AI Studio) is stamped with a visible watermark before download. It is a small semi-transparent logo shaped like two bananas, roughly 48x48 or 96x96 pixels depending on image resolution. Google calls it the "Nano Banana" mark internally, though most users just see a faint yellowish smudge in the bottom-right corner.

The watermark is applied at the moment you click "Download." The image on your screen inside the Gemini chat is clean. But the PNG or JPEG file you actually save has the banana composited onto the pixel data using alpha blending. The blending factor is low enough that the mark is barely visible on light backgrounds, but it shows up clearly on dark or solid-color areas.

This is separate from SynthID, Google's invisible watermarking system. SynthID embeds statistical patterns into the pixel noise of the image. It is undetectable by the human eye and survives resizing, cropping, and JPEG recompression. You cannot remove SynthID without degrading the image significantly, and frankly, it does not affect how the image looks. This article only covers the visible Nano Banana watermark.

Why is the watermark there?

Google adds it for provenance. When an AI-generated image circulates online, the Nano Banana mark signals that the image was machine-made. It is part of Google's responsible AI policy, similar to how DALL-E images once carried the rainbow stripe and how Midjourney metadata tags images as AI-generated.

Whether you agree with mandatory visible watermarks on AI art is a separate debate. In practice, many users need clean images for mockups, presentations, personal projects, or social media posts where a random banana logo in the corner is not acceptable.

Method 1: Manual removal in Photoshop or GIMP

The brute-force approach. Open your image in an editor and paint over the watermark.

Using the Clone Stamp tool

1. Open the watermarked image in Photoshop (or GIMP, which is free)
2. Select the Clone Stamp tool (shortcut: S in Photoshop)
3. Alt-click a clean area near the watermark to sample the texture
4. Paint over the banana logo carefully, matching the surrounding pixels
5. Export the result

This works, but it is slow (2-5 minutes per image), and results vary. On textured backgrounds like grass, fabric, or clouds, clone stamping can produce visible repetition artifacts. On flat color backgrounds it is faster, but you are still doing manual labor for every single image.

Using Content-Aware Fill

Photoshop's Content-Aware Fill is a step up. Select the watermark region with a lasso, then go to Edit > Content-Aware Fill. The algorithm tries to reconstruct the area from surrounding context. For complex backgrounds this often works better than clone stamping, but it can still hallucinate texture patterns that look wrong on close inspection.

The main downside of any manual method: you are guessing what the original pixels looked like. The result is always an approximation.

Method 2: Online watermark removers

Several web services advertise AI-powered watermark removal. You upload your image, their model inpaints the marked region, you download the result. Some popular options:

• Watermarkremover.io -- uses inpainting ML models. Free tier limits resolution to 720p. Paid plans start at $10/month.
• Aiseesoft Free Watermark Remover Online -- browser-based, no signup. Quality is inconsistent on detailed backgrounds.
• Apowersoft -- similar inpainting approach. Adds its own watermark to free exports (ironic).
• Cleanup.pictures -- clean interface, decent results on photos. Struggles with illustrations and flat art.

These tools share the same fundamental problem: they do not know what the original image looked like under the watermark. They guess, using surrounding context. On a photo of a forest, the guess might be good enough. On a logo, diagram, or illustration with precise geometry, the inpainted area will look off.

There is also the privacy angle. You are uploading your image to a third-party server. For personal projects that might be fine, but if you are working with confidential mockups or client work, that is a concern.

Method 3: Reverse alpha blending (the math-correct approach)

Here is the thing most people miss: the Nano Banana watermark is not destructive in the way a stock photo watermark is. Stock photo watermarks deliberately destroy pixel data by tiling a huge pattern across the entire image. The Nano Banana mark is a small semi-transparent overlay, applied with a known blending formula.

When Google composites the watermark, it does standard alpha blending:

output = original * (1 - alpha) + watermark * alpha

Where alpha is the watermark opacity at each pixel (derived from the mask), watermark is the watermark color (white, RGB 255), and original is the pixel you actually want back.

Since the watermark mask is known (it is the same banana shape every time), you can solve for the original pixel algebraically:

original = (output - alpha * 255) / (1 - alpha)

That is it. No guessing, no inpainting, no machine learning. Pure arithmetic. You take the watermarked pixel value, subtract the watermark contribution, and divide by the remaining weight. The result is mathematically exact, limited only by 8-bit integer rounding (which costs you at most 1 unit per channel per pixel -- invisible to the human eye).

This is called reverse alpha blending, and it is the technique Banana Clean uses.

Why this only works for Nano Banana

Two conditions make reverse alpha blending possible:

• The watermark mask is static. Google uses the same banana shape at the same position for every generated image. If the mask changed per image or was randomized, you would need to detect it first -- a much harder problem.
• The alpha values are low. The watermark opacity peaks around 15-20%. At higher opacities (say, 80%), the original pixel data would be so attenuated that integer rounding errors would compound and produce visible banding. At 15%, the math stays clean.

For SynthID, reverse alpha blending does not apply. SynthID is not an overlay; it is a perturbation of the image's frequency-domain structure. Different problem entirely.

Banana Clean: automatic removal on download

Banana Clean is a Chrome extension that removes the Nano Banana watermark automatically. No extra steps -- you generate an image in Gemini or AI Studio, click download, and the file you receive is already clean.

How it works under the hood

The extension runs as a content script in MAIN world on gemini.google.com and aistudio.google.com. It intercepts the fetch call that downloads the generated image by matching the URL pattern (lh3.googleusercontent.com/rd-gg). When a matching response arrives, the extension:

1. Decodes the image into an ImageBitmap
2. Draws it onto a temporary Canvas
3. Reads the pixel data with getImageData()
4. Applies reverse alpha blending using the known watermark mask (48px or 96px, selected based on image dimensions)
5. Encodes the cleaned canvas back into a PNG blob
6. Returns the clean blob as the fetch response

The entire process runs locally in your browser. No server calls, no uploads, no external API. Processing takes under 100ms for a typical Gemini output image.

Accuracy

Because reverse alpha blending is mathematically exact (minus 8-bit rounding), the cleaned image is nearly identical to what the image would have been without the watermark. The maximum per-pixel error is 1/255 per color channel. That is below the threshold of human perception.

Compare this to inpainting tools, which might reconstruct the general color and texture of the area but will always introduce subtle artifacts -- wrong texture direction, slightly off color gradients, or blurred edges.

What about different watermark sizes?

Gemini uses two watermark resolutions. Images up to around 1024px on the long side get the 48x48 mask. Larger images (from AI Studio's higher-resolution outputs) get the 96x96 mask. Banana Clean detects which one to use based on the image dimensions and applies the correct mask automatically.

Comparison: which method to use

Method	Time per image	Quality	Privacy	Cost
Photoshop / GIMP	2-5 min	Varies (approximation)	Full (local)	Free (GIMP) or $23/mo
Online removers	30-60 sec	Varies (inpainting)	None (upload required)	Free or $10+/mo
Banana Clean	<100ms (automatic)	Lossless (reverse math)	Full (local, no upload)	15 free, then $4.99 once

Frequently asked questions

Does this work on images from Google AI Studio?

Yes. AI Studio uses the same Nano Banana watermark as Gemini. Banana Clean intercepts downloads from both gemini.google.com and aistudio.google.com.

Does removing the visible watermark remove SynthID too?

No. SynthID is an invisible watermark embedded in the statistical properties of the image. It survives crops, resizes, and format conversions. Removing the visible Nano Banana mark does not affect SynthID at all -- they are independent systems.

Is this legal?

Removing a watermark for personal use is generally allowed. Google's Terms of Service for Gemini do not prohibit watermark removal. That said, if you use an AI-generated image commercially, check the applicable terms for your use case. This is not legal advice.

Will Google change the watermark and break Banana Clean?

Possible. If Google changes the watermark shape, position, or opacity, the mask would need to be updated. The extension is actively maintained -- when Google last updated the watermark resolution from 48px to 96px for high-res outputs, a new mask was added within days.

Can I use this on Firefox or Safari?

Currently, Banana Clean is a Chrome extension only (works on any Chromium-based browser: Chrome, Edge, Brave, Arc, Opera). Firefox and Safari are not supported yet because the extension uses Manifest V3 and the MAIN world content script injection, which Chromium handles differently from other engines.

What happens after the 15 free cleans?

The extension shows a prompt to unlock the full version for $4.99 (one-time payment, lifetime access). Payment is processed through NOWPayments (cryptocurrency). After payment, you receive a license key that unlocks unlimited cleanings across all your Chrome profiles via chrome.storage.sync.

Quick setup guide

1. Go to the Banana Clean page on Chrome Web Store
2. Click "Add to Chrome"
3. Open gemini.google.com (refresh if already open)
4. Generate any image and click the download button
5. The downloaded file is already clean -- no banana

No configuration needed. The extension activates automatically on Gemini and AI Studio pages. You can toggle it on/off from the popup if you want to keep the watermark for a specific download.