AI Object Remover

AI Object Remover erases things from photos: the stranger behind you, power lines across a sky, litter on grass, a watermark of your own, the ex in the group shot. Paint over what should go, press remove, and the area is regenerated from its surroundings by LaMa — a published inpainting network running entirely in your browser. The result becomes your new working photo, so you can erase one thing after another and download when the picture is clean.

Why the erasing looks right (and when it will not)

The model does not smear neighbouring pixels over the hole the way clone-stamp tools do — it regenerates the missing area so that edges continue, textures repeat believably and lighting stays consistent. Skies, grass, walls, water, pavement and foliage come out excellent. But it invents plausible background, not truth: whatever was genuinely behind the object — text, a face, a licence plate — will be an educated guess, because the camera never recorded it. No object remover on the web can do otherwise, whatever their marketing says.

Small masks beat one giant mask

The network works on a 512-pixel window cut around your marks, which keeps small and medium removals at nearly full local resolution even in a 12-megapixel photo. Mark half the frame at once and that window must zoom out to contain it, so the fill comes back softer — the page tells you when that happened. Erasing three objects in three passes is deliberately the intended workflow, and each pass costs about thirty seconds of your own device's compute.

The first click downloads the brain

The weights are about 200MB — the price of a network that can genuinely reconstruct backgrounds. They download once from a public CDN with a real progress bar, your browser caches them, and every later removal — today or next month — starts instantly. That download is also your privacy guarantee: because the model lives in the page, your photo has no reason to leave your device, and it never does.

Nothing is uploaded. Tools on this page that use a neural network download the model to your browser once, cache it, and run it on your own machine from then on. That is why there is no account, no queue, no watermark and no daily limit — there is no server doing the work, so there is nothing for us to meter and no copy of your photo anywhere but your device.