Free · In your browser · No upload
Remove the Gemini watermark
Detects the visible Gemini logo in the corner of AI-generated images and mathematically reverses it.
Need more than one operation? Open the full editor to combine tools and reorder your edits.
This is not an AI eraser and it does not invent replacement pixels. Gemini composites its logo onto the image using known transparency values, which means the operation is reversible arithmetic: given the blended result and the alpha map used to blend it, the original pixel values can be solved for directly.
Because it is a calculation rather than a guess, the result is either correct or it does not happen at all. The tool matches the watermark against calibrated maps of both the current and legacy logo profiles, and if neither correlates well enough it tells you no watermark was found rather than quietly altering the corner of your picture.
How it works
- 1
Load an image generated by Google Gemini showing the logo watermark in its bottom-right corner.
- 2
Click Remove watermark. The tool identifies which watermark profile and size is present, then reverses the blend across that region.
- 3
Compare the corner against the original. Areas where the logo was almost fully opaque are the hardest to recover, since little of the underlying image survived there.
- 4
Download as PNG to avoid re-compressing the restored pixels.
Frequently asked questions
How does it work without AI?
The watermark is composited with the formula result = alpha × logo + (1 − alpha) × original. With the alpha map known, that equation can be rearranged to solve for the original. It is straightforward arithmetic on the pixels, which is why the restored area matches the surrounding image rather than looking like plausible invented detail.
Does this remove the invisible SynthID watermark?
No, and deliberately so. Only the visible corner logo is affected. Google’s SynthID marking, which is embedded imperceptibly throughout the image to identify it as AI-generated, is untouched and remains detectable.
Why does it say no watermark was found?
The detector looks in the specific corner position where the watermark is placed, at a size derived from the image dimensions. If the image has been cropped, resized, or heavily re-compressed since it was generated, the watermark no longer sits where the formula expects. It also reports this on images that were never watermarked.
Will the result be perfect?
Usually very close, but not always. Where the logo was nearly opaque, very little of the original signal survived underneath, and dividing by a small remaining fraction amplifies rounding noise. Those pixels can come out slightly off, most visibly against a smooth background.
Is it acceptable to use this?
That depends on what you are doing and is your call to make. It is genuinely useful for cleaning up images you generated yourself, and the provenance marking that identifies AI content remains intact either way. You are responsible for complying with the terms of whatever service produced the image.