Remove Gemini AI Watermarks
Remove visible Gemini watermarks from AI-generated images
Watermark Removal
Remove Watermarks
🧹
Drop images here or click to upload
Upload one or more Gemini AI generated images to remove their watermarks
PNG, JPG, WEBP — Multiple files supported — Ctrl+V to paste
How it works
Gemini AI adds a small visible watermark (the Gemini sparkle logo) to generated images using alpha compositing: each pixel is blended as watermarked = alpha × 255 + (1 − alpha) × original. This tool reverses that formula to recover the original pixels.
- 100% client-side — your images never leave your browser.
- Reverse alpha blending — mathematically recovers original pixel values.
- Adaptive detection — finds the watermark even at non-standard positions.
- Only removes the visible logo — does not affect invisible SynthID watermarks.
Credits
Watermark removal algorithm ported from GargantuaX/gemini-watermark-remover (MIT License), itself based on allenk/GeminiWatermarkTool.
Frequently Asked Questions
How does the watermark remover work?
Gemini AI adds a small visible watermark using alpha compositing — each pixel is blended with a semi-transparent white logo. This tool reverses the formula mathematically: original = (watermarked - alpha × 255) / (1 - alpha). It uses pre-captured reference images of the Gemini logo to know the exact alpha values at each pixel.
Does this remove invisible SynthID watermarks?
No. This tool only removes the visible Gemini sparkle logo in the corner of generated images. Google's SynthID is an invisible watermark embedded throughout the image at a level undetectable to the human eye. Removing SynthID is not possible with this approach.
Are my images uploaded to a server?
No. Everything runs 100% in your browser using HTML Canvas. Your images never leave your device. The processing uses JavaScript to read pixel data, apply the reverse blending formula, and output the cleaned image — all locally.
Why does the tool use two different reference sizes (48px and 96px)?
Gemini uses a 48×48 pixel watermark on smaller images and a 96×96 pixel watermark on images larger than 1024×1024. The tool automatically detects which size to use based on your image dimensions, and falls back to adaptive template matching if the watermark is at a non-standard position.
Can I process multiple images at once?
Yes. You can drag and drop or select multiple images, and the tool will process them concurrently (up to 3 at a time). When you have multiple processed images, you can download them all at once as a ZIP file.