Remove Object from Photo

How the fill is made

The tool uses LaMa, an open-source inpainting model built to fill large missing areas with texture that continues the surrounding scene. It works on a 512-pixel square, so each painted area is cut out together with the scene around it, filled at that size and scaled back.

Only the painted pixels are replaced, with a slightly softened edge so there is no visible seam. Everything else in the photo is the original, at its original resolution. Photos up to 6000 pixels on the long edge keep their size; larger ones are scaled down to 6000.

The fill is generated, not recovered. The model has never seen what was behind the object, so it invents a plausible continuation of the scene. On grass, sky, sand, walls and water that is usually indistinguishable from the real thing.

Where it works, and where it breaks

It works best when the object is small relative to the photo and sits on a background that repeats: a person on a beach, a bin on a pavement, a power line across the sky, a spot on a cheek.

It struggles when the hidden area has structure the model cannot guess: a person standing in front of a crowd, a sign covering a window frame, half of a face. Straight lines that continue behind the object, such as railings and horizons, can come back bent. A large painted area also leaves less real scene to copy from, so very large masks are refused.

For a difficult area, remove it in two passes: download the first result, upload it again and paint only what still looks wrong.

Use it on images you have the rights to

Removing a watermark, logo or photographer credit from an image you do not own or license can infringe copyright, and in many countries removing copyright management information is unlawful in its own right. Use the tool on your own photos and on images you are licensed to edit.