Gemini Watermark Remover

Remove the Gemini sparkle badge from AI-generated images

Google Gemini stamps a small multicolored four-pointed sparkle (✦) into a corner of every image it generates or edits. The badge is a visible provenance marker: it tells a human looking at the file that the pixels came from a generative model. This tool reconstructs the pixels underneath that badge, which is useful when the sparkle collides with your composition - a slide background, a texture tile, a mockup, or an image you are going to crop and re-lay-out anyway.

Detection works on a colour property rather than a template. The sparkle is rendered in a saturated blue-to-magenta gradient, so the tool samples each corner region and looks for a tight cluster of pixels whose HSV saturation is far above the local background. That is why it finds the badge reliably on photographic, muted, or dark backgrounds and can miss it on an image that is already vividly coloured in that corner.

The fill step is directional ray-cast inpainting. For each masked pixel the algorithm shoots eight rays outward, walks each one until it hits an unmasked pixel, and averages those eight samples weighted by inverse-square distance - so near neighbours dominate and the patch inherits the local gradient instead of a flat colour. Practically:

• Flat or softly graded backgrounds (sky, studio backdrop, blur) reconstruct essentially invisibly
• Fine repeating texture (brick, foliage, fabric weave) comes back slightly smoothed
• A hard edge crossing the badge - a roofline, a face, a text baseline - will blur, because the algorithm interpolates rather than synthesizing structure

Does this remove SynthID?

No, and that distinction matters. Google applies SynthID, an imperceptible watermark embedded across the image's frequency content, in addition to the visible sparkle. SynthID survives cropping, resizing, JPEG compression, colour adjustment, and filters, and it is detectable by Google's own verification tooling regardless of what you do to the corner badge. Removing the ✦ changes the visual composition only; the image remains identifiable as AI-generated, and it stays subject to Gemini's usage terms and to any disclosure rules that apply where you publish. Use this for layout cleanup, not to pass generated work off as a photograph.

If the scan misses the badge, switch to Select manually and drag a rectangle over it - keep the box tight, since a larger mask means more pixels to interpolate and a softer patch. Upload JPG or PNG; output is always PNG so the fill is not re-compressed.

The whole pipeline is HTML5 Canvas running on your own machine. The file is read withFileReader, drawn to an off-screen canvas, and processed throughgetImageData - there is no upload endpoint, no account, and no copy of your image anywhere but your device.