Patches, not a smudge
The gap is filled by copying pieces from elsewhere in the same photo, so brick stays brick and grass stays grass. Diffusion alone can only ever produce the smoothest surface that fits the edges.
Runs in your browser
Paint over what you want gone. The rest of the photo fills it in.
CleanSnap rebuilds the gap by copying patches from the picture itself, so texture and edges carry on across it. Nothing is uploaded, and nothing starts until you press the button.
- No upload, no account
- Full resolution
- Cancel any time
Drop a photo here
Or click to choose one. You can also paste from the clipboard.
JPG · PNG · WEBP · AVIF · GIF · HEIC
How it works
Three steps, and the heavy one waits for you.
- 1
Open the photo
It is decoded once and stays at full resolution. Nothing is changed and nothing leaves your device.
- 2
Paint over the thing
Brush, rectangle or ellipse. Zoom in as far as you need — the selection is stored at the photo’s real size, not the preview’s.
- 3
Fill it
Pick a method and press the button. You get a progress bar, a cancel button and the time it took.
What changed under the hood
Edges carry on
The fill order is decided by how much structure crosses each point of the boundary, so a line or a horizon enters the gap first and continues straight instead of being cut off.
Off the main thread
The work runs in a Web Worker in timed slices, so the progress bar moves, the page stays usable and Cancel actually stops it mid-way.
The selection is yours to shape
Grow it, shrink it, invert it, soften its seam — and re-run with a different method on the same selection without repainting a thing.
Nothing is uploaded
Decoding, editing and exporting all happen in your browser. No account, no model to download, and it works offline.
Undo that costs kilobytes
Only the rectangle that changed is stored, not the whole frame — so forty steps of history fit in the space one used to take.
Frequently asked questions
Is my photo uploaded anywhere?+
No. It is decoded, edited and exported inside your browser using canvas and a Web Worker. Nothing is sent to a server, there is no account, and once the page has loaded it works with no connection at all.
Is there an AI model behind this?+
No, and it no longer claims there is. The earlier version offered an "AI mode" that ran exactly the same diffusion as the normal one, just with more iterations. What is here instead is patch propagation: the gap is filled by copying real pieces of your own photo. No neural network, no download, and it is honest about what it does.
Why does the result sometimes still look wrong?+
Because the fill can only use what is already in the picture. If the thing you removed covered something unique — a face, text, a one-off object — there is nothing to copy it from, and the result is a plausible texture rather than the truth. It works best on repeating backgrounds: sky, walls, foliage, pavement, water.
What should I use each fill for?+
Patches for anything with texture or structure. Smooth for skies, plain walls and gradients, where it is both faster and better. Blur and pixelate do not remove anything at all — they cover it — which is what you want for a face or a plate and not what you want for a watermark.
The selection looks right but a faint outline is left. Why?+
Watermarks are usually semi-transparent and have a soft halo a few pixels wide that is easy to miss while painting. Raise "Grow selection" by two or three pixels so the halo is inside the gap and gets rebuilt too.
Can I stop a fill that is taking too long?+
Yes. Rebuilding scans the photo for matching patches, so a big selection genuinely takes time. The work is split into short slices in a background thread, which is what lets the progress bar move and the Cancel button take effect part-way through rather than only at the end.
Does it shrink my photo?+
Only past about 24 megapixels, and when it does it says so on screen with the original dimensions. The previous version quietly capped every image at 1920 pixels, so you downloaded something smaller than you uploaded without ever being told.
Which formats can I open and save?+
It opens JPG, PNG, WebP, AVIF, GIF and the HEIC that iPhones produce. It saves as PNG, JPG or WebP, with a quality slider for the two lossy ones — the old version always wrote PNG, which turned a small JPEG into a much bigger file.
Keywords
remove watermark from photowatermark removerremove object from photoinpainting onlinecontent aware fill onlineerase objects from picturesremove logo from imageremove text from photophoto retouch in browserfree watermark remover no uploadremove people from photosclone stamp online