Remove Objects from Photos — Free, Private, Full Resolution

How object removal works here — and what “Smart Fill” actually is

Removing an object from a photo is an inpainting problem: the pixels the object covered are gone, so something has to reconstruct what plausibly stood there from the pixels around the hole. This page does that reconstruction with a classical, published algorithm — Telea inpainting (Alexandru Telea, 2004) — run at several scales. The fill walks inward from the edge of your mark, nearest pixels first, and colours each one as a weighted average of its known neighbours, where the weights follow the direction the surrounding image structure is travelling. A horizon line, a roof edge or the gradient of a sky therefore continues through the filled area instead of smearing across it. For large marks, the fill first runs on a shrunken copy of the region — so the broad structure is decided while the shape is simple — and that result guides the full-resolution pass.

It is important to say plainly what this is not. It is not a generative AI model of the kind server tools run on GPUs: it cannot invent a window, a face or a brick pattern that the surrounding pixels do not already imply. What it does — skies, seas, walls, lawns, beaches, roads, and the small-to-medium clutter that ruins otherwise good photos — it does well, instantly, privately and at full resolution. Where a big neural model would beat it, on large objects over busy detail, the honest answer is in the expectations box under the editor and in the tips below. A tool that tells you its limits is worth more than one that hides them behind the word “AI”.

Remove an object in under a minute, step by step

1. Add your photo. Drag it in, browse, or paste it straight from your clipboard. It opens at editing size in the workspace, but the original full-resolution pixels are kept untouched underneath — editing size is only for your screen. 2. Mark what should disappear. Paint over the object, person, text or blemish with the brush, making the mark slightly bigger than the thing itself, and cover its shadow too — a shadow left behind is the most common reason a removal looks fake. Flat, rectangular things (signs, stamps, captions, plates) are faster with the Rectangle tool. 3. Pick a fill mode. Smart Fill for general work; Smooth Fill when the background is a plain sky, wall or skin and any grain would show. 4. Clean, inspect, repeat. Press “Clean marked area”, compare with the before/after slider, and if a stubborn patch remains, mark just that patch and clean again — the second pass works from your already-cleaned photo. 5. Download at full resolution. PNG keeps every pixel; JPG is smaller for sharing. Nothing was uploaded at any point, so there is nothing to delete afterwards.

What Smart Fill removes well — and where it struggles

Removes well: tourists and photobombers standing against sky, sea, sand, walls or lawns; power lines and cables against the sky; date stamps and captions; signs and text on plain surfaces; skin blemishes and dust spots; litter, bins and stray props in product photos; an ex at the edge of a group photo when the background behind them is simple; sensitive details like a licence plate or a house number before you share a photo publicly.

Struggles, honestly: a large foreground object covering a third of the frame; anything where the hidden background contains specific detail the viewer expects — faces, hands, text, railings, brick courses, fabric patterns; strong reflections and shadows of the removed object (mark the shadow as well and it usually goes too); and repeating patterns that must line up exactly, like fences or tiles. The professional workaround for all of these is the same: work in smaller passes. Remove the object in two or three overlapping chunks and the fill has real pixels nearby to learn from at every step, instead of guessing across a desert.

The mistakes that make removals look fake

• Marking too tightly. A mark that hugs the object leaves a halo of its edge pixels for the fill to learn from — and it will faithfully rebuild the halo. Overflow the object by a few pixels on every side.
• Forgetting the shadow and the reflection. The object leaves; its shadow stays, pointing at nothing. Cover both in the same mark.
• One giant pass over a busy background. Split it. Two smaller cleans beat one big one almost every time.
• Judging at editing size. The workspace is a screen-sized preview. Download and look at the full-resolution result before deciding it failed — the feathered composite often looks better at real size than in the preview.
• Expecting it to invent detail. If a person stands in front of a shop sign, no honest tool knows what the sign said. Crop, choose another photo, or accept a soft patch — a generative server tool will invent letters; they will be wrong letters.

Why this page can be free at full resolution when others charge

Upload-based removers pay for a GPU every time someone presses “erase”, so the free tier has to be a teaser: cleanup.pictures, the category leader, caps free exports at 720px and sells the full-resolution version as a subscription — its own FAQ says so. This cleaner runs the same class of job on the one computer that is already paid for and already trusted with the photo: yours. No server cost per erase means no teaser economics — the full-resolution download is simply the default. The trade is real and worth naming: a data-centre neural model will out-fill this page on the hardest ten percent of photos. For the other ninety percent — and for any photo you would rather not upload to a stranger’s server at all — the on-device fill is the better tool. If metadata is your actual problem rather than objects, use the dedicated EXIF & metadata cleaner instead; if the cleaned photo then needs shrinking for the web, the image compressor keeps it sharp while the kilobytes fall away.