"Nano Banana" is the nickname that stuck to Google's image model behind recent Gemini image generation. Its outputs carry the same style of visible corner mark as other Gemini images, and it comes off the same way.
Removing it
Open the tool, drop the image in, and let detection find the bottom-right corner mark. If a fragment survives — most often when the mark sits over busy detail rather than flat background — switch to the patch tool, draw a loop around what is left, and drag onto a clean part of the same surface.
Everything runs in your browser. The image is never uploaded, which is the practical difference from most alternatives.
Why the variants matter
Different output sizes place the mark at different scales — 48 px up to 1024 px, 96 px beyond it — and a detector tuned to one size does poorly on another. That is the usual reason a removal tool "misses" a mark that is plainly visible to you — not that the mark is exotic, but that it is a size the detector was not calibrated for. When automatic detection misses, the manual path is not a workaround so much as the reliable route: you can see the mark, so you can select it.
What is still in the file
The visible mark is one of two things Google applies. The other is SynthID, an invisible statistical watermark spread across the whole image, which is untouched by removing a corner logo. Stripping the file's metadata does not remove it either, because it lives in the pixels. Worth knowing before assuming a cleaned image reads as camera-original anywhere.
Background on both kinds: how AI image watermarks actually work.