It finds the logo
The sparkle is matched against a rasterised copy of its own shape at every plausible size and position in both bottom corners — so it works whatever resolution you exported at.
Free · No sign-up · Nothing uploaded
Free to use
Hot
Remove the
watermark.
Images and videos. Drop in a picture from Gemini or Nano Banana, or a clip from TikTok, Seedance, Dola AI or Veo, and the logo in the corner is gone — reconstructed, not blurred or painted over. One at a time or a whole batch.
- Never leaves your browser
- Finds the logo by itself
- Video and images
- Bulk mode built in
100% private
Your images and videos are decoded, processed and re-encoded inside this page. Nothing is uploaded, nothing is stored — pull your network cable out and it still works.
This browser is missing the features the tool needs (Web Workers,
createImageBitmap and Path2D). Please use a
current version of Chrome, Edge, Firefox or Safari.
Select mode
Choose whether to process a single image or a whole batch.
Upload image
An image created with Gemini, Nano Banana or Google AI Studio.
Drag an image here or click to select
PNG, JPG, JPEG or WebP · up to 10 MB
Upload images
Up to 50 images, processed one after another so your browser stays responsive.
Drag images here or click to select
PNG, JPG, JPEG or WebP · up to 10 MB each, 50 per batch
This browser cannot process video yet — it is missing
. Chrome, Edge or Opera on desktop will
work; Safari and Firefox are still catching up on the video parts
of WebCodecs. The image tab works everywhere.
Select mode
One clip at a time, or a whole folder of them.
Upload video
A clip from Gemini or Veo, TikTok, Seedance or Dola AI — or anything else with a logo stamped in a corner. The tool works out which by itself.
Drag a video here or click to select
MP4 or MOV · up to 200 MB · audio is kept untouched
Upload videos
Queued and processed one at a time — two encodes at once finish later than two in a row and use twice the memory.
Drag videos here or click to select
MP4 or MOV · up to 200 MB each, 20 per batch
Reconstructed, not painted over
Most "removers" just smudge the corner. This one solves for the pixels that were actually there.
It measures the opacity
The background behind the logo is reconstructed from the pixels around it, which reveals exactly how transparent the overlay was — typically somewhere between 45% and 80%.
It undoes the blend
A watermark is just original × (1−a) + white × a. With
a known, that equation runs backwards and the original
colours come straight back.
It knows when to stop
Before touching a pixel it checks the match really behaves like a white overlay. If not, it tells you no watermark was found and hands your image back untouched.
Questions
It uses reverse alpha blending. The Gemini watermark is a semi-transparent white logo composited over the finished picture, so the blend is a known equation. Once the tool has located the logo and measured its opacity, it solves that equation backwards and recovers the original pixel values.
No. Every image is decoded, processed and saved entirely inside your browser. No image data is sent anywhere — you can disconnect from the internet after the page loads and it will still work.
Any MP4 or MOV with a logo composited into the picture — Gemini and Veo, TikTok, Seedance and Dola AI are the common ones. The Gemini mark is recognised outright, from an outline measured off real exports; everything else is found by looking for something that stays put while the picture behind it moves, which is what every stamped-on watermark is and what real footage never is. TikTok's mark moves between two spots every few seconds; the tool follows it and solves each position separately.
The Gemini mark is the one it recognises outright, and those clips go straight through. For anything else it shows you the areas it believes are the watermark, on a frame from your own video, and changes nothing until you say so — because "stays put while the picture moves" is also true of a face.
Because it has more evidence. With one picture, how transparent the logo is has to be inferred from the pixels around it. A video shows the same logo sitting on top of hundreds of different backgrounds, which turns that guess into a measurement — the opacity is whatever value explains every frame at once. It also means the tool can tell you honestly when it is not sure, rather than smudging the corner and hoping.
The audio is copied across untouched — it is never decoded or re-encoded, so it is bit-for-bit the original. The video has to be re-encoded, because the pixels changed; that pass is done at a high bitrate, but a re-encode is never completely free, so expect the same kind of quality cost as any other edit-and-export.
No. This tool only removes the visible logo. SynthID is an invisible watermark developed by Google DeepMind that is embedded in the image data itself. It is not visible to the human eye and is not affected by this tool.
PNG, JPG, JPEG and WebP, up to 10 MB per image and 50 images per batch. Results are always saved as PNG — saving back to JPEG would put compression artefacts straight onto the pixels we just reconstructed.
Where the watermark was at its most opaque, some of the original detail is genuinely gone — no amount of maths can recover information the blend destroyed. Those few pixels are rebuilt from their surroundings, which can read very slightly smoother than the rest of the picture. Everywhere else the recovery is exact.
Nothing happens to it. Before editing anything the tool checks that what it found actually behaves like a semi-transparent white overlay — consistent opacity, and brightening that stops dead at the logo's outline. If the evidence is weak it reports "no watermark detected" and gives your image back untouched, rather than stamping a dark star-shaped smudge into it.
Making the images too?
BatchGemini queues all your prompts at once and lets Gemini generate every image on its own — then saves them straight to your computer.
See BatchGemini →