How to Remove People from Photos
Updated Aug 14, 2026 · 7 min read
How AI people removal works (and why it sometimes fails)
Removing a person from a photo is two problems. First the tool must find the person precisely, including shadows, reflections and anything they are holding. Then it must fill the hole with plausible background. The People Remover uses a person detector for the first step and a generative inpainting model for the second.
That second step is the interesting one. The model does not copy nearby pixels like the old clone stamp; it predicts what the scene would look like if the person had never been there, using the surrounding context. Straight lines, repeating patterns and gradients are easy. Complex, unique detail behind the person is hard, because there is nothing to predict it from.
Knowing this tells you what to expect:
- A tourist in front of a plain wall or sky: near perfect.
- A person in front of a tiled floor or fence: very good, patterns are extended cleanly.
- A person covering half of a statue or a sign with text: the model will invent something, and it may be wrong. Consider keeping that person or cropping instead.
Step 1: Upload and review the auto-detected people
Open the People Remover and upload your photo. Within a couple of seconds every detected person is outlined with a numbered tag. Nothing is removed yet; this is a selection screen.
By default everyone is selected for removal. Tap the people you want to keep, typically yourself, your family or your model, and their outlines turn green. This keep-first approach is faster than selecting each stranger in a crowd of forty.
Two checks before you continue:
- Are partial people detected? A hand at the edge of the frame or a head behind a railing sometimes gets missed. If so, you will handle those with the brush in step 3.
- Are shadows included? The outline should extend to include the person's shadow on the ground. A removed person who leaves a shadow behind is the most common tell-tale sign of editing.
Step 2: Remove and inspect the fill
Click Remove. The tool processes all selected people in one pass, which is important: removing people one at a time can cause the model to reconstruct background from a person you are about to remove next. One credit covers the whole image regardless of how many people are removed.
Inspect at 100 % zoom, not fitted to screen. Look for:
- Ghost outlines. A faint silhouette where the person stood means the mask was too tight. Undo, and use the brush to expand the area by a few pixels.
- Repeated texture. If the same brick or leaf appears three times in a row, the fill leaned too hard on a pattern. Re-run; results vary slightly each time and a second pass is often cleaner.
- Broken lines. Horizon lines, kerbs and railings should continue straight through the filled area. Small kinks are fixable with a narrow brush stroke over just the line.
If a person was standing in front of your subject and overlapping them, expect the model to redraw part of your subject too. Check faces and hands especially.
Step 3: Fix stragglers with the manual brush
Switch to the Object Remover in Manual brush mode for anything the detector missed: a stray arm, a reflection in a shop window, a backpack left on a bench, or half a person cut off by the frame.
Brush technique that gives the best fills:
- Paint slightly larger than the object. Two to five pixels of margin lets the model blend edges. Too tight leaves a halo; too loose destroys detail you wanted.
- Paint the shadow and the reflection in the same stroke so they are removed together.
- Do not paint across strong lines if you can avoid it. If a person crosses a railing, paint up to the railing, run the fill, then paint the rest as a second job. The model handles each region with a clean boundary.
- Use a soft brush for grass, sand and water and a hard brush for architecture.
Each manual run costs 1 credit. The free preview lets you check the result before spending anything.
Special cases: crowds, reflections and real estate
Dense crowds at landmarks. If a crowd fills the lower third of the frame, removing everyone forces the model to invent a large empty plaza. Results are better if you remove the nearest, most distracting people and let the distant crowd blur into texture. Alternatively, crop the composition higher.
Reflections in glass and water. The detector finds people but not always their reflections. Paint reflections manually; a person removed from a bridge who still appears in the river below is an easy mistake to make.
Real estate listings. Removing the agent or a passer-by from an exterior shot is common and fine. Removing permanent features like power lines or a neighbouring building may breach disclosure rules in your jurisdiction; check before you publish.
Group photos where someone left. Removing an ex from a couples photo works best when they were at the edge of the group. If they were in the middle, the fill has to invent a background and the composition will look unbalanced; consider cropping instead.
Frequently asked questions
How many people can be removed at once?+
There is no hard limit. The auto mode detects everyone in the frame and removes all unselected people in one pass for a single credit.
Will it remove the person's shadow too?+
Usually yes, the detection outline includes ground shadows. Reflections in glass or water are not always detected; paint those with the manual brush.
What if the person is in front of my subject?+
The model will reconstruct the hidden part of your subject, which works for clothing and simple shapes but can look wrong on faces. Check the result closely and re-run if needed.
Does it work on video?+
The People Remover is for still images. For video, the Video Background Remover can isolate your subject and replace the entire background instead.