Remove Objects from Photos Online (AI LaMa)

What is object removal with AI LaMa

A classic editor requires a clone stamp, manual retouching, and hours of work. Here you just need a brush mask: you mark what to remove, and the LaMa model (Resolution-robust Large Mask Inpainting with Fourier Convolutions, WACV 2022) analyzes the surrounding context and synthesizes a plausible background texture, sky, wall, asphalt, grass, fabric.

LaMa handles large masks and repetitive structures (fences, tiles, facades) well. On filltools.com the pipeline is: soft mask dilation → inference → soft blending with the original so edges don't look "cut out".

How the tool helps

• Travel and architecture. Remove tourists near a landmark, keep the building or landscape in focus.
• Real estate and exteriors. Wires, antennas, trash bins, random cars in the foreground.
• Product and catalog photos. Dust, lint, price tags, extra fasteners in the background.
• Portraits and interiors. Outlets, cables, objects that "grow" out of the model's head.
• Restoration of minor defects. Spots, sensor dust dots, small scratches.

Scenarios (brush presets)

• General, a balance for most medium-sized objects.
• People, wider brush and larger radius: convenient for covering a figure entirely; better to remove one person per pass.
• Wires, thin brush along the line; go slightly onto the sky or facade so the mask "eats" the halo.
• Spots, small brush only on the defect: less risk of blurring the correct texture around it.

Presets don't lock the manual sliders, after choosing a scenario you can adjust size and radius.

LaMa vs OpenCV

LaMa is the main mode for complex backgrounds, large objects, and situations where naturalness matters. Telea / Navier-Stokes is classic fast inpaint: sometimes it wins on tiny defects when you need an instant fix without AI.

Step-by-step usage

1. Open or drag a photo into the workspace (very large frames are automatically downscaled for stable processing).
2. Choose a scenario and the "Brush" tool. Paint over the object in solid red, without "holes" in the mask.
3. Check the edges with the eraser. Enable LaMa (default) and click "Remove".
4. Evaluate the result. If the edge is sharp, slightly expand the mask and repeat. If the texture "smears", reduce the mask or try OpenCV on a small spot.
5. Save as PNG. For the next object, start with a new stroke on the already cleaned frame.

Tips for maximum quality

• The mask must fully cover the object; a thin unpainted edge often leaves a contour "ghost".
• Several short passes are better than one giant mask covering half the frame.
• On leaves, water, mesh, paint slightly wider, the model needs more context of the periodic texture.
• Don't mask extra sky or wall "just in case": too large a hole forces the network to invent structure with less accuracy.
• Zoom and the "Hand" tool help draw a thin wire precisely; hotkeys B / E / H switch between brush, eraser, and panning.

Privacy

You draw the mask in the browser. Inpainting is performed on the filltools.com server to process the request; the file is not published as a gallery and is not stored for showing to other users after completion.