How to Remove the Gemini (Nano Banana) Watermark From an Image

If you have generated an image with Google’s Gemini models, you have probably noticed a small four-point star tucked into the bottom-right corner. That “sparkle” is the visible Nano Banana watermark, and it is the part most people want gone before they use an image they created. This guide walks through how to remove it cleanly and, just as importantly, what it does not remove, so you know exactly what you are working with.

A quick naming note first. Nano Banana is Gemini 2.5 Flash Image, and Nano Banana Pro is Gemini 3 Pro Image. Both add the same style of corner sparkle, and both also carry Google’s imperceptible SynthID signal. Those are two different things, and treating them as one is where most confusion starts.

#Three layers hide on a single Gemini image

It helps to stop thinking about “the watermark” as one object. A Nano Banana output can carry up to three separate signals, and each is handled in a completely different way.

The sparkle is the easy layer. The other two are the reason a “watermark remover” that only edits the corner can leave you with a false sense of a clean file. If your goal is genuinely to change what a detector reports, read our deeper walkthrough on how SynthID removal actually works before you decide.

#Removing the visible sparkle

The sparkle sits in a predictable spot, which gives you a few options depending on how much of the frame you can sacrifice and how busy the background is. There is no single “best” method. The right one depends on your image.

1. Crop it out. If the corner does not hold anything important, trimming the bottom-right is the fastest, most reliable fix. You lose a sliver of the frame but keep every remaining pixel untouched.
2. Reverse-alpha-blend on flat areas. Because the sparkle is a semi-transparent overlay, tools that know its shape and opacity can subtract it and recover near pixel-perfect results on plain skies, walls, or solid colors.
3. Inpaint or regenerate a busy corner. Over texture, faces, or fine detail, the region has to be reconstructed. This is object-removal work, and the result should always be inspected at full size.

A dedicated Nano Banana watermark remover automates the detection and reconstruction step so you do not have to mask the corner by hand. If you are working across several Google models, the more general Gemini watermark remover handles the same sparkle on both Nano Banana and Nano Banana Pro output.

#Why a clean corner is not a clean file

This is the part worth slowing down for. Once the sparkle is gone, the image looks free of any AI marking, but the imperceptible SynthID signal is embedded in the pixels themselves, not in the corner. Editing the bottom-right does nothing to it. Similarly, any C2PA or EXIF metadata written at export time stays put unless you strip it separately.

SynthID is designed to survive cropping, filters, and lossy compression, so the ordinary edits that erase a visible logo tend to leave it intact. Removing it, if that is even your goal, means regenerating the image through a diffusion process, and even then no tool can promise a detector will return “not detected.” We would rather tell you that plainly than sell you certainty that does not exist.

#How to verify what is actually left

Do not take any tool’s word for it, including ours. Check the file yourself so you know which layers are still present.

• Run the image through Google’s SynthID Detector or the Gemini app to see whether the embedded signal is still flagged.
• Inspect provenance at contentcredentials.org/verify to see any attached Content Credentials.
• Open the file’s metadata to confirm whether C2PA, EXIF, or XMP tags remain.
• View the repaired corner at 100% zoom before you publish, so you catch any reconstruction artifacts.

For a fuller picture of how these tags travel with a file, our guide to image metadata and Content Credentials covers what each one records and why re-encoding is not proof of removal.

#What Delete SynthID does with a Gemini image

Delete SynthID works on all three layers in one pass: it reconstructs the visible sparkle, strips file metadata, and regenerates pixels to reduce the embedded signal. It handles JPEG, PNG, WebP, AVIF, HEIC, and HEIF files up to 20MB, and your first image is free so you can inspect the output before committing. What it will not do is guarantee that every detector, everywhere, returns a clean result. That promise is not one any honest tool can make.