Remove Objects from Image

How it works, and what that means for your photo

Worth understanding first, because it tells you exactly which jobs will succeed.

The area you select is reconstructed from the pixels immediately around it, following the edges and textures that run into the region rather than smearing a flat average across it. Nothing is invented — the fill is built from what was already in the photograph.

So the question is always: is there enough around the object to rebuild from?

This works well on simple, continuous backgrounds. A person on a stretch of sand. A bin against a plain wall. A sign in the sky. A cable across grass. A blemish on skin. A logo on a flat surface. Anything sitting on sand, sky, water, lawn, tarmac, a painted wall or a plain floor comes out cleanly, and often perfectly.

It struggles on structured backgrounds. Brickwork, floor tiles, a bookshelf, a patterned fabric, text, a face, railings, a window frame — anything with a repeating or meaningful pattern the fill has to continue convincingly. There the patch will look soft or smeared, because reconstructing a brick course from its neighbours is genuinely hard without understanding what a brick is.

That is the honest difference between this and the generative AI removers. They can invent plausible bricks. This rebuilds from real pixels — which is faster, private, unlimited, and better on plain backgrounds where invention is not needed.

Keep the rectangle tight, and work in stages

The technique that decides your result, and it is the opposite of what people instinctively do.

Selection is a rectangle, so a tall thin object like a lamp post drags in a lot of surroundings if you box the whole thing at once. Take it out in sections instead. Several small removals up the length of the post, each with a tight rectangle, will beat one large box every time — because each small patch has a short distance to bridge and plenty of nearby material to work from.

The same applies to anything irregular. A person is not a rectangle: box the head and shoulders, then the torso, then the legs, rather than drawing one box around all of them and asking the tool to rebuild a person-sized hole.

Leave a little margin around the object — one or two pixels of the object left behind is far more visible than a fraction more background replaced. But only a little: every extra pixel inside the rectangle is real photograph you are discarding.

Undo is real, so experiment

The removals are recorded as a list and replayed, rather than burned into the image one at a time. Which means:

Undo takes back the last removal cleanly. Reset all puts every object back and returns you to the original photograph. Nothing is committed until you download.

So there is no reason to be cautious. Try a removal, look at it at full size, undo it and try a tighter rectangle if it is not right. That loop is how good results happen with this kind of tool, and it is why the undo matters more than it sounds.

The blend control, and when to move it

Blend strength decides how much the patch is smoothed into its surroundings. The default of 8 suits most photographs.

Raise it when the patch is visible as a slightly different shade — a common outcome on a gradient like a sky, where the background is not one colour but is changing gradually. More blending hides the seam.

Lower it when the surroundings have texture you want to keep: sand, gravel, foliage, fabric weave. Heavy blending turns texture into a smooth patch, and a suspiciously smooth area reads as retouched just as clearly as a badly filled one. Less blending keeps more of the reconstructed grain.

If neither setting looks right, the rectangle is usually the problem rather than the blend — try a tighter selection or several smaller ones.

Check the result at full size

The one habit worth keeping. A patch that is invisible in a preview can be obvious at 100%, and if the photo is going to be printed or viewed large, that is where it will be seen.

Look for a soft rectangular area where the object was, a repeated scrap of texture, or an edge that stops short and restarts. Any of those means a tighter rectangle or a different blend, and both are one click away because of the undo.

What you get back

The output is a JPEG at high quality, which suits photographs. A transparent PNG comes back with white where the transparency was.

Everything happens on your own device. No upload, no account, no watermark and no resolution limit — which is the practical difference from the free tier of most AI removers, where the output is watermarked, capped in size, or requires signing up.

Related tools

Crop Image is sometimes the better answer: if the thing you want gone is near an edge, cropping it out is cleaner than any reconstruction. Pixelate Image hides something rather than removing it, which is the right choice for a face or a number plate where the point is privacy. Remove Background from Image removes the entire background instead of one object.