Remove objects, text and watermarks from a photo
Paint over a watermark, caption, date stamp, stray person, wire or clutter. Your browser fills the background with the MI-GAN model on your own device, and you download a PNG or JPG the same size as the original. No account, and the photo never leaves your device.
Canvas
Drop, paste or choose a photo
The photo is read and processed on this device only, never uploaded. iPhone HEIC photos work too.
Steps
- Choose or drop a photo
- Paint over what to remove (a little wider than it)
- Erase, then download when it looks right
How to use it
- Make the brush a little wider than the object and include its shadow and edges.
- If one pass leaves traces, paint and erase again on the result; you can mix both methods.
- The first smart fill downloads the model and runtime from this site (about 39 MB); after that the browser caches them. The work runs on your device and can take a few seconds on a phone.
- iPhone HEIC photos open directly; if your browser can't read them, this page decodes them locally (a 2 MB decoder, loaded once).
- Ctrl/⌘ + Z undoes the last stroke, or the last erase when nothing is selected.
- [ ] change the brush size.
- Painting needs a mouse, trackpad or touch screen; keyboard-only painting isn't supported yet.
Privacy
Your browser reads the photo into memory and does the work locally. The photo is not sent to this site or to any third-party server and is not written to browser storage; closing or reloading the page clears it. The smart-fill model, its runtime and the HEIC decoder are downloaded from this site to run in your browser, and those downloads contain no photo or selection.
This page uses no cookies and no ads. The host, Cloudflare, processes network information when you visit and may keep basic visit and performance statistics; these never include your photo or selection.
Downloads are re-encoded from the canvas and carry none of the original's EXIF data (such as the GPS location).
How it works, sources
Smart fill: A. Sargsyan et al., “MI-GAN: A Simple Baseline for Image Inpainting on Mobile Devices”, ICCV 2023 (code and model, MIT licence including the weights). It uses the authors' ONNX pipeline (Places2 512 model), run by ONNX Runtime Web 1.20.1 in your browser's WebAssembly; pixels outside the selection are left untouched.
Quick patch: A. Telea, “An Image Inpainting Technique Based on the Fast Marching Method”, Journal of Graphics Tools 9(1), 2004 (DOI).
Ported from OpenCV's cv::inpaint(INPAINT_TELEA) source (inpaint.cpp, read 2026-10-21): the selection is grown by 2 pixels and the inpainting radius is 5 pixels.
What erases cleanly
- Captions, date stamps and small watermarks over sky, water, sand, grass or a wall.
- Clutter that covers a small part of the frame: a bin, a sign, a wire, a stain on the ground.
- Small glare spots, dust and scratches: quick patch is fastest and needs no model download.
- The camera's date stamp on your own photos, or an old watermark you added yourself.
What won't erase cleanly, or shouldn't
- A selection covering a big part of the frame: the fill can only borrow from what's outside it, so the bigger it is, the more it blurs or repeats.
- Faces, IDs and written content: the fill is a plausible guess, not what was there.
- Watermarks tiled across the whole picture: the selection is too large; get the original instead.
- Someone else's copyrighted photo: removing the owner's watermark can break copyright law. Use this only on images you have the right to edit.
Tested guides
Each guide runs this page on a photo where we know exactly what was under the mark, reports the error and time for both fill methods, and opens its example here in one click.
- How to remove text from an image: three backgrounds tested
- How do watermark removers work? What they actually do to the pixels
- How to remove glare from a photo: when erasing works, by spot size
- AI object remover test: how big an object can you erase?
- How to remove a shadow from a photo (and why erasing it is usually wrong)
- How to remove a date stamp or your own watermark from a photo