Everything, including the awkward parts

Yes, completely. There is no account, no trial, no credit system, no per-file charge and no watermark on the output. The processing happens on your own device, so there is no server cost to pass on and no reason to meter it. Use it as often as you like.

Drop the file onto the page, or press the button and pick it. The sparkle is located automatically by matching it against the known Gemini logo shape, then removed, and you get a before/after slider to check the result before downloading. It takes a second or two and there is nothing to configure.

Yes. Video is handled frame by frame, so motion and detail survive instead of being blurred over. The mark sits in the same place on every frame, so it is located once and then removed throughout. Frame rate is preserved and the original audio is carried across. A clip takes roughly its own length to process, because it is rebuilt in real time.

Your files, no — never. They are read and rebuilt inside your browser, so there is nothing about them for anyone to collect: not the image, not its name, not its size, not whether a watermark was even found. The site may count page views to see which pages are useful, and that is the whole extent of it. See the privacy page for exactly what is and is not recorded.

Not during normal use. The file is read with the browser File API, decoded to a canvas, processed by JavaScript on your machine and handed back as a download. Load the page, disconnect from the internet, and watermark removal still works.

There is exactly one exception, and it never happens on its own. Some videos use a codec your browser cannot decode at all — H.265/HEVC above all, which is what most modern phones record. Nothing in the browser can open those files. When that happens you are offered a button to convert that one file on our server; it is re-encoded to H.264, sent straight back, and deleted as the response finishes. If you decline, nothing is transmitted and you are told how to re-encode it yourself. The choice is always explicit — there is no silent fallback.

No, and you should not trust any tool that says it does. SynthID is Google's invisible watermark, spread statistically across the whole image rather than drawn in one corner. This tool removes the visible sparkle logo only. The invisible provenance signal stays in the file and remains detectable. Anyone claiming "zero traces" is describing the logo, not the provenance data.

Blurring destroys the detail the watermark was sitting on and leaves an obvious smudge that reads as a cover-up. Cropping changes your composition and aspect ratio. Reverse alpha blending is different: because the logo was composited with a known formula, that formula can be solved backwards to recover what was underneath, at the original sharpness.

The watermark was drawn with result = α·logo + (1 − α)·original. Rearranged, that gives original = (result − α·logo) ÷ (1 − α). Given the alpha map for the mark, the pixels underneath fall straight out of the arithmetic. Where alpha approaches 1 the division becomes unstable, so those few core pixels are filled by diffusion from the recovered ring instead.

Images as PNG, JPG, WEBP and AVIF; video as MP4, MOV and WEBM. Sources the detector knows by name include Gemini, Nano Banana, Imagen, Whisk, Veo and Google Flow. Anything else can still be handled in manual mode by brushing over the mark.

Only what your own device can hold in memory. There is no server quota, so the practical ceiling is your RAM and your browser's canvas limits. Long 4K videos are the realistic constraint — those are processed frame by frame and will take roughly as long as the clip itself runs.

Re-encoding uses MediaRecorder, and the container it can write depends on your browser. Chrome and Edge will produce MP4 with H.264 where the codec is available and fall back to WEBM/VP9 otherwise; Safari prefers MP4. The tool picks the best container your browser reports and tells you which one it used before you download.

That depends on you, not on the tool. Removing a watermark from your own generated output for layout or print reasons is ordinary work. Passing AI-generated media off as a real photograph or real footage is not, and in several jurisdictions disclosure is now a legal requirement rather than a courtesy. Google's generative AI terms also have their own say. Check both before you publish.