How to use
- Drop a photo onto the box or click to choose it (up to 6,000px per side). You can also paste a screenshot with Ctrl+V.
- Use Paint to brush over the person or object to remove. Change the size with the slider, grab large areas with Rectangle, and fix mistakes with Eraser or Undo (Ctrl+Z). Zooming in (up to 8×) lets you paint precisely.
- Click Remove. The first time only, a 28.1 MB AI model is downloaded and saved on this device; after that nothing is downloaded.
- If you don’t like the result, click Undo last removal. You can also keep painting on the result to remove more. When you are done, Download (the original format is kept; JPG is saved at quality 92).
Tools and options
| Tool / option | What it does | Range |
|---|---|---|
| Paint / Eraser | Paint the area to remove, or erase part of your painting | size 1–30% of the long side |
| Rectangle | Paints a whole dragged rectangle at once | |
| Pan | Drag a zoomed photo around (on touch, swipe) | zoom 1×, 2×, 4×, 8× |
| Grow painted area | Expands the painted area outward so no rim of the object is left | 0–8px, default 6px |
| Soften edge | Blurs the seam where the result is mixed into the original | 0–6px, default 3px |
Why big photos stay quick
The whole photo is not sent to the AI. Only the bounding box of your painting plus a 50% margin on each side is cut out, processed, and pasted back in place. For example, if you painted a 200×300px area, the processed region is 400×600px, and even a tiny spot gets at least 256×256px. If the region would extend past the edge of the photo it is shifted inward, and if the photo itself is smaller it uses the whole photo.
- If the long side of the region is over 2,048px it is scaled down proportionally, processed, then scaled back up. Painting all of a 6000×4000 photo is processed at 2048×1365px, so that part can lose a little sharpness.
- “Grow painted area” expands in a circle. A radius of 1px turns a single painted dot into 5 pixels (a plus shape); 4px gives 49, 6px gives 113 and 8px gives 197 pixels.
- Painting happens on a screen copy whose long side is at most 1,600px; at run time it is converted to original coordinates. A 6000×4000 photo appears as 1600×1067 on screen.
- After several removals, the earlier pixels kept for undo are limited to about 60 million pixels, newest first, so the oldest ones may be dropped.
When it works and when it doesn’t
Small objects, wires, dust and people against a simple background come out fairly natural. Removing something over a face, text or a busy pattern, or a very large area, can look unnatural. The model does not “recover” what was there; it looks at the surroundings and generates plausible content, so the result can differ from reality. Don’t use it where the photo has to stand as evidence. The model is MI-GAN, licensed under MIT.
Privacy and storage
The photo is processed only in your browser (in a web worker) and is never sent to a server. The model file is downloaded from this site and kept in browser storage (Cache Storage); its SHA-256 is checked when downloaded to reject damaged files. You can delete it any time with “Delete saved models”.
Related tools
To remove a whole background, use Background Remover; to enlarge a photo, use AI Image Upscaler; to hide a region instead, use Image Mosaic.
Your photos are processed in your browser and never sent to a server.