Apple Intelligence Clean Up: How your iPhone erases a stranger from a photo

Picture a photo from a weekend walk. The light is lovely, your friends are laughing, and right behind them stands a stranger in a bright orange jacket. You like everything about the shot except that jacket. On a compatible iPhone, you can open the picture in Photos, tap Edit, choose the Clean Up tool, and touch the stranger. A moment later the orange jacket is gone, replaced by pavement and a bit of hedge that look like they were always there.

That small moment hides a fair amount of work. The phone has to figure out which pixels belong to the person you touched, decide what the background behind them probably looked like, and paint something new that matches the light and texture of the rest of the photo. All of this happens on the device itself, not on a distant server.

What happens after you tap the object

When you tap or drag your finger over something in the photo, the tool first works out the outline of what you selected. It looks at the colors, edges and shapes around your tap to guess where the object ends and the background begins. This is the same kind of pattern recognition that helps your phone sort photos by faces or scenes, just aimed at a smaller target.

Once the object is separated from its surroundings, a second model takes over. This one is a generative fill model, a type of image generator trained to complete missing parts of a picture. It looks at the empty space you created and at the pixels around it, then produces new content that fits: a stretch of wall, a patch of grass, a continuation of a shadow. The result is meant to look like the object was never there.

The whole process runs on the iPhone's Neural Engine, the part of the chip designed for this kind of machine learning work. Apple says the photo is not uploaded anywhere for Clean Up, which matters if the picture shows your home, your family or anything else you would rather keep private.

Why the result sometimes looks off

The fill model is guessing, and guesses are not always right. If the background behind the object is complicated, like a busy bookshelf or a crowd of people, the new pixels may not line up perfectly. You might see a slight blur, a repeated pattern, or a shadow that points the wrong way. These are not signs that something is broken; they are the visible edge of a model doing its best with limited information.

The tool also depends on the hardware. Clean Up is part of Apple Intelligence, which requires a recent iPhone with enough memory and a fast enough chip to run the model locally. Older or entry-level phones do not have the capacity, so the feature simply is not offered there. That is a hardware limit, not a settings problem you can fix.

How it compares to similar tools

Clean Up is not the only eraser in town. Google Photos offers Magic Eraser, and Samsung Galaxy phones include an Object Eraser. All three aim at the same job: remove something from a photo and fill the gap convincingly. The differences tend to show up in how the tools handle tricky edges, how much control they give you, and where the processing happens. Apple's version stands out mainly for keeping the work on the device.

• Tap once on a clearly separated object, like a sign or a single person, for the cleanest result.
• Circle with your finger when the object is small or partly hidden behind something else.
• Check the filled area at full size before saving; small flaws are easier to spot when zoomed in.
• Avoid using Clean Up on faces you plan to keep, since the model may redraw features imperfectly.

None of these tools is a magic wand. They are useful for tidying a photo, not for rewriting what happened. If you remove a person from a picture and then present that picture as a record of an event, you are changing the story, and the edit may still be detectable by someone looking closely. The tool is best used for small distractions, not for pretending a scene was different from what it was.

Back to that orange jacket. A few taps later, the stranger is gone and the photo looks like the one you meant to take. The phone did the work quietly, on its own chip, and handed you back a picture that fits the moment you remember. Knowing how it got there helps you decide when to trust the result and when to leave the scene as it was.