Object & Blemish Eraser
Paint over what you do not want and let the AI fill it in from the surroundings, or heal small spots instantly without any download. Everything runs on your device, and the rest of your photo stays untouched.
Runs on your device. Nothing is uploaded.
Fill method
AI fill for objects and people. Small spots are healed instantly without the AI until it has been loaded.
The AI loads with your first AI erase: about 29 MB once (WebAssembly build), then kept in this browser. Spot heal needs no download.
Mask edges
Output
Saved at the photo's full resolution as name-erased.png. The file is written again from the pixels, so EXIF metadata (camera, GPS location) is not copied.
No file at hand?Photo: Yellowstone National Park, public domain
How to use the Object & Blemish Eraser
- Drop a photo, paste one with Ctrl+V, or click to choose. You can add several and switch between them; each keeps its own edits while the page is open.
- Mark what should disappear. Paint with the Brush (set its size, or press [ and ]), drag a Box, or draw around it with the Lasso. Include the object's shadow and any reflection. Unmask paints parts back out, Clear mask starts over, and Show mask hides the red overlay.
- Choose the fill method. Auto uses the AI for objects and people and heals tiny spots instantly while the AI is not loaded yet; AI fill always uses the model; Spot heal fills from the surrounding pixels without any download, which is ideal for dust, blemishes and small scratches.
- Press Erase. The area around the mask is filled and blended back into the full-resolution photo with a soft edge; Expand mask and Blend edge under Mask edges control how far around your marking it reaches. The line under the photo says which method filled it and how long it took.
- Check the result with Compare (drag across the photo), zoom with Ctrl+wheel or a pinch, and hold Space to pan. Erase again over anything left behind; Undo takes back a whole erase and brings its mask back so you can adjust it.
- Download the photo at its original size in the same format (JPG, PNG or WebP), or pick another format under Output. With several edited photos you can save them all as a ZIP.
What it does well
Your photo stays your photo. The fill is computed on a crop around the marking and blended back into the full-resolution image with a soft, adjustable edge. Pixels outside that edge are not changed at all, so the rest of the picture keeps every detail and the saved file has the same size as the original.
Two ways to fill. The MI-GAN model rebuilds what was behind people, objects, signs and text from the scene around them. Spot heal, a classical fast marching fill with matched grain, removes dust, blemishes and small scratches in milliseconds without downloading anything, and Auto picks it for tiny marks while the AI is not loaded.
Built for real retouching. Brush, box and lasso selections, a mask you can show or hide, separate areas filled in their own crops, long marks such as wires filled in full-detail tiles, repeated erases with undo and redo of whole erases, a before/after slider, zoom and pan, and several photos at once with a ZIP download.
Frequently asked questions 8
Are my photos uploaded?
No. The fill runs in your browser, on your graphics card through WebGPU where available or on your processor through WebAssembly. The page downloads the AI model and its runtime from this site; your photos and masks never leave your device. Your usage history, if you keep one, records only settings such as the fill method and output format, never images or file names.
Which AI model is used, and what is its licence?
AI fill uses MI-GAN, an inpainting network by Picsart AI Research published at ICCV 2023, in its 512 px version trained on the Places2 scene dataset. Its code and its weights are released under the MIT licence. It runs with ONNX Runtime Web (MIT licence). The files are served from this site and checked with a SHA-256 hash before use. Spot heal is a classical fast marching method (after Telea) written for this tool and needs no model at all.
How big is the first download?
Spot heal downloads nothing. The first AI erase downloads about 42 MB with WebGPU (a 27 MB runtime plus the 15 MB model) or about 29 MB without it. The files are kept in your browser's storage, so later visits start without downloading; if you have used the Background Remover here, the runtime is already stored and only the 15 MB model is new. The Fill method section shows how much is stored and has a button to delete it.
How fast is it?
Spot heal takes milliseconds for a blemish and well under a second for a larger mark. On a typical laptop an AI erase of an object takes about half a second on the graphics card and about two to four seconds on the processor; the very first erase also prepares the model, which adds a few seconds (plus the one-time download). Long marks such as wires are filled in several tiles, so they take a little longer per erase.
Does the rest of my photo change?
No. Only the marked area plus the Expand mask and Blend edge distances is replaced; every pixel beyond that keeps its exact original value, and the photo keeps its full resolution. The saved file is encoded again, so a JPG or WebP is recompressed at the quality you choose (92 by default), while PNG stays lossless. EXIF metadata such as camera details and GPS location is not copied.
What does it handle badly?
Large areas on detailed or regular textures (crowds, brick walls, text, patterned fabric), because the AI has to invent a lot and may leave a smudge or a repeated texture; erase big objects in parts and repeat where needed. Faces, hands and other things that must look exactly right cannot be rebuilt. Objects that touch the edge of the photo, or whose shadow you did not mark, can leave traces. Spot heal blurs anything larger than a small spot; use AI fill for that.
Why is a large area softer than a small one?
The model works on a 512 by 512 pixel square around your marking. Small and medium marks fit at full detail. A long thin mark such as a cable is walked in several full-detail tiles. A big, wide area is filled from a reduced copy and scaled up, so it looks softer; the tool tells you when that happens. Erasing it in two or three parts gives a sharper result.
Which formats can I open and save?
It opens JPG, PNG and WebP, plus GIF (first frame), BMP and AVIF where your browser can decode them, up to 50 megapixels. It saves in the format you opened (PNG for formats browsers cannot write) or in JPG, PNG or WebP if you choose one. Transparency in PNG and WebP files is kept.
More free tools
- Background Remover Remove or replace photo backgrounds with AI that runs in your browser. Full-resolution PNG/WebP cut-outs, colour, blur or image backgrounds, brush fixes.
- Face Blur & Pixelate Find faces automatically, even small ones in group photos, and blur, pixelate or cover them. Draw over plates or badges. Batch, ZIP, no upload.
- Screenshot Stitcher Join scrolling screenshots into one full-page image. Overlaps and sticky headers are removed automatically. PNG, JPG, WebP or PDF.