Skip to content
GPU-assistedUpload required

Remove People from a Photo

A person occupies more of a photograph than the space they stand in. They cast a shadow across the ground, appear in the window behind them, block whatever they are standing in front of, and often supply the reason the picture is framed the way it is. Taking them out means dealing with all of that, and the leftovers — a shadow with nobody attached, a bench that does not line up — are what make an emptied photograph look edited.

Use this without the search next time. Prathom Workbench puts Prathom's tools in your toolbar.

Add to Chrome — free

Drop an image here, or click to browse

Up to 25 MB. Uploaded temporarily, deleted after processing.

GPU-assisted — results vary slightly between runs.

What it does

  • Figures removed and the location rebuilt behind them
  • Perspective and architecture explicitly preserved
  • A different reconstruction on every run
  • Roughly one megapixel output

How to use Remove People

  1. 1

    Count how much of the frame they cover

    A distant figure is a small repair. A person filling a third of the picture means a third of the picture has to be invented from scratch.

  2. 2

    Run the removal

    The photo is read at about one megapixel and repainted with the figures absent and the ground, background, and architecture behind them rebuilt.

  3. 3

    Hunt for what they left behind

    Shadows on the ground, reflections in glass and water, and structures that were partly hidden. These are the tells, not the space where the person stood.

How it works

Your photo is scaled so its total area is around one megapixel, keeping the original proportions. An image-editing model reads that copy along with a fixed instruction: remove the people and rebuild the ground, background, and architecture behind them so the place reads as empty and continuous, preserving the location, the perspective, the lighting, the shadows of remaining objects, and the composition, and adding no new figures.

The negative instruction names this job's specific failures — remaining figures, ghosted silhouettes, orphaned shadows, warped architecture, blurred patch.

The repaint runs at seventy five per cent strength, the highest of the removal workflows here, because a figure occludes far more than a small object does and the area behind it has to be genuinely built. The seed is not fixed, so every run produces a different reconstruction.

A person leaves four things

The outline is only the first of them, and it is the one the model handles best.

The shadow is second. A figure standing in sunlight lays a long dark shape across the ground, often several times their own area, and it does not resemble a person at all — it is a stretched gray patch on paving. Removed badly, it stays where it was, and the eye reads a shadow with no owner immediately even when it cannot say what is wrong.

The reflection is third. Shop windows, glass doors, wet pavement, still water, polished floors, vehicle panels. A figure removed from the pavement but still standing in the window behind it is a strange picture.

The occlusion is fourth and the most expensive. Whatever they were standing in front of is missing from your file entirely: the middle of a bench, three columns of a facade, half a doorway. That has to be constructed, and unlike grass or sky it has structure that must line up on both sides.

Checking a result means checking all four, and only the first is likely to be right on its own.

Coverage decides the outcome

There is a straightforward relationship worth knowing before you upload.

The fraction of the frame the people occupy is roughly the fraction of the result that is invented. Two walkers at the far end of a beach cover perhaps one per cent of the image, so ninety nine per cent of what comes back is a reconstruction of your photograph with a small local repair.

A crowd across the foreground of a plaza can cover a third of it. Now a third of the returned image is composed rather than observed, and it is composed to be plausible for a plaza rather than faithful to yours. Steps appear at a slightly different spacing. A doorway lands where a doorway would sensibly land.

Neither result is a failure of the tool. They are the same operation applied to very different amounts of missing information, and knowing which one you are asking for sets what you should expect.

The photograph changes meaning

Worth saying, because it is the reason some emptied images never look right no matter how clean the reconstruction is.

A photograph containing people is usually composed around them. The framing gives them room, the exposure is set for them, the moment was chosen because of what they were doing. Take them out and the composition is left pointing at nothing, with an oddly empty middle and a subject that has gone.

An empty street can be a fine photograph. It is a different photograph, and it is usually shot deliberately rather than made by subtraction.

If the result feels wrong while every individual region looks correct, this is generally why, and it is a framing problem rather than a rendering one.

Publication gate

This page ships once the workflow has been run against distant figures on sand, a crowd in front of a structured facade, a subject casting a long shadow, and a scene with figures reflected in glass, with each output checked for shadows and reflections left behind.

Examples

Distant figures in a landscape

beach.jpg - 4000x2600, walkers along a shoreline
prathom-remove-people.png - 1240x806, empty shoreline

The best case by a distance. Small figures against sand and water occlude almost nothing and sit on surfaces with no structure to reconstruct.

Tourists in front of a building

plaza.jpg - 3600x2400, a crowd before a facade
plaza-empty.png - 1224x816

The hard case. A facade has windows, columns, and courses of stone that must line up through the area a crowd was standing in.

Frequently asked questions

Can it remove one specific person and keep the others?

No. There is no selection here, and the instruction asks for people to be removed rather than a particular person. Whoever is in the frame is what it works on. For a photograph where some figures should stay, this is the wrong tool and a manual editor is the right one.

Why is there still a shadow on the ground?

Because a shadow does not look like a person, so it is not obviously part of what should go. The instruction names orphaned shadows in the negative prompt for exactly this reason, and it helps without solving it. A shadow with nothing casting it is the most common reason an emptied photograph looks wrong.

How much of the frame can the people cover?

As a rough rule, the more of the picture they occupy, the more of the result is invention rather than photograph. Distant figures are a small local repair. A close subject filling much of the frame means the model composes most of the scene, and what returns is closer to a picture of a similar place than of yours.

Should I run it more than once?

Yes, particularly on a crowded scene. The seed is not fixed, so each run rebuilds the background differently, and on a difficult photograph the variation between attempts is large. Generating three and comparing them also shows you which regions were reconstructed and which survived intact.

Is the AI Remove People tool free, and do I need an account?

No sign-up. The remove people workflow takes an image up to 25 MB, runs it on the GPU host and hands back a file you can download directly. Nothing is charged and nothing is stamped onto the output. Because no account exists, nothing keeps your history either, so save the result within the thirty minutes the job is retained.