How to remove a person from a photo

Five methods, what each one needs before it will work, and the question that decides which of them is right for your picture.

The short answer

On an iPhone 15 Pro or newer, use Clean Up in the Photos app. On anything else with the photo in Google Photos, use Magic Eraser. If the photo is a file on a computer, or you do not want it going to Apple or Google, do it in a browser tab.

All five methods below do the same thing underneath. You mark the person, a model generates something that plausibly continues the surroundings, and it is pasted over the mark. What separates them is what they demand of you first and what they hand back afterwards.

Device and pricing rows are from Apple, Google and Adobe's own published requirements, checked on 25 September 2026. The measured rows are further down and are marked as measured.

1. Clean Up, on an iPhone

Open the photo, tap Edit, tap Clean Up, then circle or tap the person. It is the fastest of the five when it is available to you.

The catch is which iPhone. Clean Up is part of Apple Intelligence, which needs an iPhone 15 Pro, an iPhone 15 Pro Max, or an iPhone 16 or later. An iPhone 14, or a standard 15, does not have it and no update will add it. Apple also notes that devices bought in mainland China do not get it.

Worth knowing about where the picture goes: Apple offers a Fast setting that runs on the phone and a High Quality setting that uses Private Cloud Compute, which is Apple's servers. The second one is a network round trip, and Apple applies daily limits to the server-backed features.

2. Magic Eraser, in Google Photos

Open the photo in Google Photos, tap Edit, then Tools, then Magic Eraser. It offers suggestions first and you can also brush or circle anything it missed.

This is the most widely available option by a distance. Google made it free for all Google Photos users in May 2024, on Android and on iPhone, with no monthly cap. The requirement is a 64-bit processor and Android 8 or iOS 15, which is almost every phone still in use.

Two things to keep straight. Magic Eraser is not Magic Editor. The save limit people run into belongs to Magic Editor, the larger generative tool. And Magic Eraser works on photos in Google Photos, so using it means the picture is in Google Photos, which is a decision about the picture rather than about the edit.

3. Generative Fill, in Photoshop

Select the person with any selection tool, click Generative Fill, leave the prompt empty and click Generate. You get three variations and can keep cycling.

It is the best result of the five on a difficult background, and it is the only one that gives you alternatives to choose between rather than one answer. It is also the only one that costs £20 a month, and it sends the selection and its surroundings to Adobe's servers to do it. For a photograph you were going to retouch anyway it is the obvious tool. For one picture on a Tuesday it is not.

4. A browser tool, and the thing to check first

Search for this and you will find a dozen sites that will do it in a tab with no account. They work. The question nobody asks in the first thirty seconds is what size the photo is when it comes back.

We put one 1600 by 1067 image through the four highest-ranking ones on 25 September 2026, in Chrome 153 on macOS 27, and read the canvas dimensions out of each page rather than off its marketing copy.

Cleanup.pictures is the straightforward one. Drop a photo bigger than 720 pixels and it stops you with a modal offering Continue with SD or Try Pro 3 days free, at $5 a month. Magic Studio gives you 600 by 400 and does not mention it, which you find out when you open the file you saved. Photoroom keeps your resolution and does not charge, and it is the one rival on this list we cannot beat on that axis.

All four post your photograph to a server. For a picture of a street that is nothing. For a picture of your child, a friend who did not consent to being uploaded, or the ex-partner you are removing, it is the whole question.

5. A browser tool that keeps the photo

An inpainting model is a file, and a browser can download a file and run it. That is the entire trick behind the object remover on this site: 26.8 MB of weights arrive once from floi.dev, your browser caches them, and every fill after that is arithmetic in the same tab the photo is already open in. There is no endpoint to post an image to, because none was written.

The honest trade is quality on hard backgrounds. What floi ships is MI-GAN, a model designed to run on a phone, and it is not as strong on rigid repeating structure as the larger models behind the server tools. Measured below.

What we measured, and where it fails

Three synthetic test images, generated by a committed script so anybody can reproduce them byte for byte, each chosen to break the method in a different way. Chrome 153, macOS 27, MacBook Pro M1 Max, 25 September 2026. The full notes are in docs/research/object-remover.md in the repository.

The third row is the one to carry away, and it applies to every method on this page rather than to one of them. An inpainter generates a plausible continuation. It does not recover what was underneath. On grass that is indistinguishable from the truth, because grass carries no information and one convincing patch of it is as true as another. On a line of text, a number plate or a face it is a fabrication that looks like the right sort of thing.

How long it takes, and on what

Six runs on the same machine, warm, with the model already compiled. Timed from the press of the button to the picture changing, so this is the whole wait rather than the model's share of it. The result was flatter than expected.

A 10.7 megapixel photo costs the same as a 1.7 megapixel one, and a mark covering half the frame costs the same as a blemish. Both follow from the same design decision: the model is never shown the whole photograph, only a square window around what you marked, capped at 1024 pixels. Your photo's size changes how much memory the tab needs and nothing else.

Starting up is separate. With the weights already cached it is 284 ms from hovering the panel to ready, of which 218 ms is the runtime compiling the network. The first visit has to fetch 26.8 MB first, and how long that takes is your connection.

Getting a good mask, whichever method you use

The model does not choose what to remove. You do, and the fill is only ever as good as the shape you gave it. Four habits that do more for the result than switching tools.

1. Paint a little wider than the person

   A mask that stops exactly at the outline leaves the outline behind, because the ring of pixels just outside still carries their colour and the model matches it. Overshoot by a few pixels. Tools that offer a grow or feather control are doing this for you.

2. Include the shadow and the reflection

   The commonest reason a removal looks wrong is a person-shaped shadow still lying on the ground, or a shape still in the window behind them. Nothing will spot those for you.

3. Zoom in before you paint, not after

   An accurate mask needs an accurate view. If the tool only shows you a fitted preview, every stroke is approximate, and this is exactly what the 512 pixel proxy in the table above costs you.

4. Work in passes

   Two people overlapping, or a person against something complicated, go better in three small marks than one big one. Each pass gives the next one better surroundings to continue from.

When not to do this at all

Three cases where the right answer is something other than erasing.

Sources

• Apple, Requirements to use Clean Up in Photos. Checked 25 September 2026.
• Google, Google Photos expands AI editing features to all users. Checked 25 September 2026.
• The four browser tools were run directly, on the same image, on 25 September 2026. Method and hashes in docs/research/object-remover.md.