Clear the Dola AI watermark from your generated images and videos. The mark is located in the corner and the area beneath it rebuilt from the pixels around it, right on your own device — no upload, no account, and no watermark of ours added to the result.
Nothing added yet.
Nothing to install, nothing to sign up for, and no queue to wait in.
Drag what you generated with Dola AI onto the drop zone, or click to browse. PNG, JPG and WebP images and MP4, WebM and MOV clips are all accepted, in portrait or landscape.
The bottom-right corner is scanned for the Dola AI label at a sweep of sizes. Whatever it finds is then rebuilt from the pixels surrounding it — once for an image, on every frame for a video.
Compare before and after, then download the cleaned PNG, or the cleaned video with its audio intact. Your original file on disk is never modified.
Dola stamps opaque text into the bottom-right corner of its output. That matters, because opaque is very different from blended. When a watermark is blended into an image — as Gemini's sparkle is — the original pixels are still mathematically present underneath, and our Gemini watermark remover recovers them exactly. When a watermark is opaque, whatever was beneath it is simply gone.
So the honest description of what this tool does is reconstruction, not recovery. First it finds the mark: the bottom-right corner is scanned at a sweep of candidate sizes, and each position is scored against a synthetic silhouette of the words Dola AI after a top-hat filter isolates bright strokes from whatever is behind them. This is the localisation strategy used by the open-source remove-ai-watermarks project (Apache-2.0), re-implemented here for the canvas instead of OpenCV. When the scan is not confident enough, the tool falls back to the fixed corner geometry Dola normally uses, calibrated separately for portrait and landscape.
Then it rebuilds. Every pixel inside the located rectangle is replaced by interpolating from the
untouched pixels immediately surrounding the box, weighting each contribution by how close it is —
the same technique the delogo filter uses, and the same one running inside Vynex
Studio. The desktop app calls ffmpeg; this page performs the identical arithmetic on a canvas.
Video runs the same removal on every frame. The clip is played once through a canvas while your browser re-encodes the cleaned frames, with the original audio carried across untouched, so a 60-second clip takes about 60 seconds. The corner scan runs on several frames spread through the file and the agreed position is applied to all of them, which keeps the patch from jittering.
In practice the results are very good, because Dola places its mark in a corner and corners are usually the least detailed part of a composition — sky, bokeh, a gradient, a wall. Over that kind of content the rebuilt patch is invisible. Over dense texture or a hard edge passing through the corner you may notice a soft area, and no tool can do better, because the information no longer exists.
Everything happens in your tab. Your files are never transmitted, stored or logged, so there is no retention policy to read.
PNG, JPG and WebP stills plus MP4, WebM and MOV clips, cleaned frame by frame with the original audio kept.
A silhouette scan across the corner locates the label at whatever size it was drawn, with a calibrated fixed position as fallback.
Nudge the padding outward if a faint antialiased edge survives on an unusual export.
Drop in a whole set of generations and process them in one pass, with a before/after preview for each.
No sign-up, no API key, no credit allowance, and nothing stamped onto the output.
Images come back as lossless PNG. Video comes back as MP4 wherever your browser can record it, and WebM where it cannot — either way with the original audio.
| File type | Supported | You get back | Notes |
|---|---|---|---|
| PNG image | Yes | PNG (lossless) | Only the rebuilt rectangle changes; every other pixel is untouched. |
| JPG image | Yes | PNG (lossless) | Saved as PNG so the clean result is not recompressed again. |
| WebP image | Yes | PNG (lossless) | Animated WebP is treated as a still frame. |
| MP4 video | Yes | MP4, or WebM | MP4 out on Chrome 130+, Edge and Safari; WebM on Firefox. |
| WebM video | Yes | MP4, or WebM | Decoded everywhere; audio is carried across untouched. |
| MOV video | Usually | MP4, or WebM | Depends on the codec inside — H.264 MOV plays in most browsers. |
| MKV, AVI | No | — | Browsers cannot decode these. Use Vynex Studio on the desktop. |
Shrink JPG, PNG and WebP files in bulk without leaving your browser.
Erase the Gemini sparkle watermark from AI images, fully offline.
Grab any video thumbnail in full 1280×720 HD — no API key needed.
Save YouTube videos as MP4 in up to 1080p, straight to your device.
Pull any video transcript as TXT, SRT or VTT in one click.
Vynex Studio is a free Windows app with the same removal built in — plus MP4, WebM, MOV and MKV written with ffmpeg, faster-than-realtime processing, recursive folder scanning, and bulk AI image and video generation.
Book a free 30-minute strategy call. We'll review your goals and match you with the right services.
We reply within 4 hours · No spam, ever