Drop an image here
or
PNG · JPG · JPEG · WebP · HEIC
Up to 1600px on the long side. Larger images get resized.
How it works
The browser loads your image, you paint a mask over the watermark, and an inpainting algorithm fills those pixels using surrounding colors. We use the Telea fast-marching-method inpainting (Telea 2004, An Image Inpainting Technique Based on the Fast Marching Method) — a deterministic algorithm that runs locally. There is no neural network, no GPU, and no server roundtrip.
What this can remove
- Logos, text, and icons layered over flat or near-flat backgrounds
- Small watermarks on uniform areas (sky, walls, paper, plain gradients)
- Date stamps, signatures, and similar overlays on quiet backgrounds
What this cannot remove
- Watermarks covering faces, fine textures, or sharp edges — the algorithm samples nearby pixels and has no idea what is behind the mask
- Watermarks baked into the source photo (no untouched pixels to sample from)
- Multi-layer or repeating watermarks where every layer overlaps busy content
- DRM-protected or scrambled images
Privacy
Your image is decoded, edited, and re-encoded entirely inside your browser. We do not upload it. We do not log it. Open DevTools → Network and you will see zero requests containing your image data. The only network traffic is loading this page itself.
FAQ
Does the output keep the original file format?
Yes. PNG in → PNG out. JPG in → JPG out. WebP in → WebP out. The extension and quality settings are preserved. One exception: HEIC input is exported as PNG, because browsers can decode HEIC but cannot encode it.
What about HEIC photos from iPhone?
Modern Chrome (109+) and Safari (17+) decode HEIC natively. If your browser cannot, we tell you and ask you to convert to JPG first — no upload, just a local file conversion tool.
Why does auto-detect miss some watermarks?
Auto-detect marks the four corner regions of the image, because most watermarks live in a corner. Real-world watermarks are designed to survive simple detection. For best results, switch to manual mode and paint over the watermark directly.
How is this different from AI inpainting?
AI inpainting tools (LaMa, MAT, Stable Diffusion inpaint) need server-side models hundreds of megabytes in size. This is a deterministic 2004 algorithm that runs locally in milliseconds. It is better for simple watermarks on flat backgrounds and worse for complex scenes.
What does it cost?
Nothing. No account, no payment, no email.
Is there a file size limit?
Browsers run into memory limits around 4096px on the long side for a Canvas pipeline. Images above 1600px are downscaled before processing to keep things responsive. Output is always at the (possibly downscaled) processing resolution.