Visible mark vs SynthID
Google adds two marks to Gemini images. The visible sparkle is pixels in a corner, which an eraser can remove like any other logo. SynthID is an invisible watermark spread through the whole image (Google DeepMind describes it as designed to survive common edits like cropping, filters and compression). Removing the sparkle doesn’t remove SynthID, and Google’s tools can still recognise the image as AI-generated. That is by design, and it is a good thing.
Why it erases cleanly
The sparkle is small and sits near the edge over background, which is the easiest case for patch-based inpainting. Use the on-device eraser first. If the mark crosses a detailed edge, AI fill redraws just that spot.
Other AI image watermarks
The same Find AI watermark button looks for visible badges from other generators, such as Meta AI’s label or a corner logo, on this page. Erase them the same way.
Your own content
Use these tools on files you made or have the right to edit. A watermark often identifies the creator or tells viewers that something was made with AI. Removing it from someone else’s work, or to pass AI content off as real, can break copyright law and platform rules.
Questions
How do I remove the Gemini watermark from an image?
Open the image here, press Find AI watermark (or brush the sparkle), then press Erase. It takes a few seconds and runs on your device.
Does this remove SynthID?
No. SynthID is invisible and spread across the whole image. Only the visible sparkle is removed, and the image remains detectable as AI-generated by Google’s tools.
Is it allowed to remove the Gemini watermark?
Read Google’s generative AI terms for your use. Don’t remove it to present AI images as real photos. Many platforms require AI content to be labelled.
Does it work for Nano Banana images?
Yes. Nano Banana is the Gemini image model, and its downloads carry the same visible sparkle.