Remove Gemini watermarks in bulk

There is no enforced limit. Processing happens on your machine, so the practical ceiling is your available memory. Three images are worked on at a time and the rest queue, which keeps peak memory flat regardless of how many you drop in.

No. There is no paid tier, no account and no daily quota. Bulk is usually where hosted tools charge, because bulk is what costs them server time — that cost does not exist here.

No, and that does not change with volume. There is no upload endpoint on this site at all. Open your browser's network tab during a batch run and you will see no request carrying image data.

Yes. They all carry the same four-pointed sparkle, and the watermark's size and inset are resolved per image, so mixed resolutions and mixed models in one batch are fine.

Yes — Download all triggers every finished file. Your browser may ask once for permission to download multiple files. Each keeps its original name with -clean.png appended.

They are marked as having no watermark found and left untouched. Handing back an unmodified file labelled as cleaned would be misleading, so the tool says so instead.

Select them all at once — the picker is multi-select and the drop zone accepts a whole folder's worth in one go. Twelve 1024×1024 images finished in about 18 seconds on a laptop. Three are processed at a time and the rest queue, so a folder of two hundred behaves the same as a folder of ten, only for longer.

Slightly, per image, because three run concurrently and share your CPU — but total throughput is far higher than one at a time. The detector's calibrated alpha maps are also built once and reused for the rest of the batch, so every image after the first skips that setup entirely.

Yes. The mark's size and inset are resolved per image against Gemini's output catalogue, so a 1024×1024 render with a 48 px mark and a 2048×2048 render with a 96 px mark can sit in the same batch without any setting to change.

No. This tool solves for one specific overlay using its known alpha map, so it only works on the Gemini and Nano Banana family. It is not a universal watermark remover, and on any other mark it reports that nothing was detected rather than damaging the image trying.

It should not. A decoded 2048×2048 image is about 16 MB of raw pixel data, so an earlier build that started every dropped file at once could exhaust memory on a big batch. Processing is now capped at three concurrent jobs; peak heap across a twelve-image run measured 157 MB.

None. Batching changes only the scheduling, not the arithmetic — each image goes through exactly the same reversal it would get on its own, and exports as lossless PNG at the original resolution.

No enforced limit — the constraint is your own memory. A decoded 2048×2048 image is roughly 16 MB of raw pixel data before the engine's working buffers, and three are held at once, so very large images simply mean a slower queue rather than a rejected upload.