Image Watermark Remover – Erase Marks & Objects With Inpainting
About the Image Watermark Remover
This tool erases a selected area of a photo and reconstructs what should be underneath using inpainting, a technique that fills the gap by propagating color and texture inward from the surrounding pixels. You paint over what you want gone with an adjustable brush, the painted region shows as a red overlay, and OpenCV’s Telea algorithm rebuilds that area when you run the removal. It works entirely in your browser through OpenCV.js compiled to WebAssembly, so images are never uploaded.
A word on what this can realistically do. Inpainting is reconstruction from context, not generation from imagination. It has no idea what the photo depicts, so it can only extend patterns already adjacent to the area you erased. Small marks over sky, water, grass, blurred backgrounds, and flat walls disappear convincingly. A large logo across a detailed subject leaves a visible smear, because there is no correct answer for the algorithm to find. Working in small passes with a tight brush produces far better results than painting one large blob and hoping.
Please only use this on images you own or have permission to edit. Removing a watermark from someone else’s photograph does not transfer any rights to it, and doing so to reuse the image is copyright infringement in most jurisdictions. The legitimate uses are the common ones: taking your own watermark off your own archive, removing a burned-in date stamp, deleting a stray object or passerby from a photo you shot, and cleaning up dust spots or blemishes.
Sometimes the better answer is not to reconstruct at all. Marks sitting near an edge come off cleanly with the Image Cropper, which removes the region outright rather than guessing at a replacement.
The BigToolSite build team tested this against corner text marks on flat backgrounds, semi-transparent tiled overlays, date stamps over sky, and objects sitting on detailed textures, tracking where Telea reconstruction holds up and where it visibly smears, and confirmed that repeated passes operate on the cleaned result rather than reapplying to the original.
If you are protecting your own work instead of cleaning it up, the Image Watermark Tool handles the opposite job with text and logo placement.
Output is written as lossless PNG, which is large for photographs, so the Image Compressor is a sensible final step before publishing the result anywhere.
How to Use the Image Watermark Remover
- Wait for the status line to report that OpenCV is ready. The library is around 8MB and loads once per visit.
- Load your image with Upload Image (JPG, PNG, or WebP up to 10MB) or paste a direct link under Load from URL.
- Set Brush Size between 5px and 100px. The preview square below the slider shows the relative size. Use a brush just slightly wider than the mark you are covering.
- Zoom in before painting detailed areas. Use the + and − buttons, scroll the mouse wheel over the canvas, or click the fit icon to scale the whole image into view. Zoom runs from 10% to 500%.
- Paint over the area to remove. Click and drag, or use touch on a tablet. Everything covered in red will be reconstructed.
- Correct mistakes with Undo and Redo, or the Ctrl+Z and Ctrl+Y shortcuts. Twenty steps of history are kept. Clear Selection wipes the whole mask.
- Click Remove Watermark. The red region is replaced with reconstructed pixels and the mask clears automatically.
- Inspect the result at high zoom. If part of it still looks wrong, paint over the leftover area and run it again. Each pass works on the cleaned image, so corrections compound.
- Click Download to save the finished PNG, or New Image to start over.
Example Usage
Case that works well: a date stamp in the corner of an old photo, sitting over plain sky
Approach: zoom to 200%, set brush to about 20px, trace the digits with a small margin around them, run the removal
Result: the sky’s gradient continues across the gap with no visible seam, because the algorithm has consistent color on all sides to work from
Case that works poorly: a large diagonal stock-photo watermark across a person’s face
Why: the reconstruction has no facial structure to extend inward from, so it blurs skin tones across the erased band. No inpainting method solves this, and the honest answer is to license the image instead.
For a semi-transparent tiled overlay, work in sections rather than covering everything at once. Clean one tile, check the result, then move to the next. Small passes let the algorithm draw from clean surrounding pixels each time, while one large mask leaves it drawing from other damaged regions.
A practical trick for objects near an edge: check whether cropping removes it entirely first. Reconstruction always leaves some trace under close inspection, whereas cropping is exact.
Frequently Asked Questions
Only on images you own or are licensed to edit. A watermark marks ownership, and removing one from someone else’s work to reuse it is copyright infringement in most countries regardless of the tool used. Removing your own watermark, or a mark from a photo you took, is fine.
It treats the erased region as a hole and fills it by propagating color and texture inward from the boundary, following the direction of edges that run into the area. It reconstructs from surrounding context rather than generating new content.
No. OpenCV runs as WebAssembly inside your browser, so the image, the mask, and the result all stay on your device.
The erased area was probably too large or sat over detailed texture. Inpainting can only extend patterns that already border the gap, so complex regions have no consistent pattern to extend. Try smaller passes with a tighter brush.
The OpenCV library is several megabytes and has to download and initialize before any processing can happen. It loads once and stays cached for the rest of your session.
Not convincingly. Full-image overlays leave too little clean reference material, and the result will show obvious artifacts. Tools that promise otherwise are generating plausible-looking replacements, which is a different thing from recovering the original.
This uses a classical algorithm that interpolates from neighboring pixels, with predictable behavior and no server round trip. Generative models invent content that fits the scene, which handles large regions better but requires cloud processing and can produce details that were never there.
The undo history covers your brush strokes, not completed removals. Once a removal is applied, that result becomes the working image. Reload the original file if you want to start clean.
Yes. Painting works by touch, and pinch-free zoom is available through the toolbar buttons. A stylus gives noticeably more control than a fingertip for precise masking.
PNG, so no additional compression artifacts are introduced on top of the reconstruction.
