Introduction
AI is all the rage these days. It has become an integral part of our smartphones and has seeped into many of the tasks we use them for – typing, talking, searching – but it really started with photography.
The big keywords were computational photography and image algorithms, and at some point, every manufacturer started bragging about its new AI Eraser that can remove distractions from your vacation photos. Google was seemingly the first with Magic Eraser, but now the industry has caught on, and we dare say it has even surpassed the Pixel in AI object removal.
Being naturally curious, especially when technology is concerned, we decided to pit the different AI erasers against each other to see which is best, but also to see if they’re even worth using.
After much testing around the office, we chose four contestants as our test subjects – the Samsung Galaxy S25 Ultra using Galaxy AI, the iPhone 16e with Apple Intelligence, the Google Pixel 9 Pro Xl with Gemini AI, and the Xiaomi 15, touting its own AI Erase 2.0.
These devices represent the best efforts of their respective companies, and you can expect the same results from their other high-end phones as well. Generally, midrange devices don’t do as well – probably due to their more limited hardware.
Samsung Galaxy S25 Ultra • Apple iPhone 16e • Google Pixel 9 Pro XL • Xiaomi 15
We spent a big chunk of time choosing the right photos for the job. We ended up with a selection of 20 images captured across a multitude of smartphones. That way the comparison is fair – every phone will use its AI on images captured with other phones.
The samples come in two groups – general shots with some distraction or other that ruins the photo – the typical example you’d use AI-powered object erasers on. A street sign ruining a vista, a construction crane between buildings, people blocking a famous landmark, etc.
Then we have the classic photobomber shot where someone has crashed in on your vacation pic and needs to be removed.
The other group of photos has tougher objects to remove – a hand in front of a face, sunglasses that need to disappear, a bottle of glass cleaner amidst bottles of spirits on a bar – we designed these shots to give the AI a challenge.
A word on the AI’s limitations
Before we look at the results, a word on each phone’s AI – what it needs to work and how it works.
Samsung
Samsung’s Generative Edit requires both an active internet connection and an enrolled Samsung account in order to work. It also resizes the manipulated image to 12MP (4,000 x 3,000px), and adds a watermark to let you know the photo was tampered with by AI.
One of the Galaxy AI’s strong suits is that it doesn’t have restrictions of working on humans – meaning it can edit parts of people in a photo – some of the other contestants can’t do that yet.
Samsung’s solution requires the least amount of taps – open the Gallery app, go to a photo, tap once, and you can draw around the obstacle you wish to remove.
Despite its limitations, Samsung’s Galaxy AI object erase is superior to the rest. It’s faster – measurably so, especially compared to the Pixel and Xiaomi. Samsung’s solution is also incredible at recognizing what you want to remove – just draw a loose circle around anything in a photo, and 9 times out of 10 the AI will recognize what you want to erase and mask it perfectly.
Apple
Apple’s Clean Up is part of Apple Intelligence, which until recently was limited to North America. It’s spread to more countries and regions recently, but it is still not available in China.
Clean Up works on iPhone 16e, iPhone 16, iPhone 16 Plus, iPhone 16 Pro, iPhone 16 Pro Max, iPhone 15 Pro, or iPhone 15 Pro Max; any iPad with A17 Pro or M1 and later; Mac with M1 or later, and requires iOS 18.1, iPadOS 18.1, or macOS Sequoia 15.1.
You’ll only need an active connection to download the machine learning the first time you set up Apple Intelligence, and it will work offline from then on.
You can find Clean Up in the Photos app, and you’ll need two taps before you can start drawing circles around distractions. The iPhone isn’t as good at recognizing what you want to delete – a simple person’s silhouette can require five or even six finger drawings for the phone to mask the entire thing.
Google’s Magic Eraser was the first of its kind, but it seems Google hasn’t improved upon it much since its announcement. Which is weird considering just how much airtime the feature gets in Pixel commercials.
Magic Eraser lives in Google Photos’ image editor inside the Tools section. It means it’s three taps away if you want to remove an object. Google Photos will sometimes automatically detect objects to remove in a photo and will prompt you to use Magic Eraser as a suggestion in the first screen of the image editor, but that happened only a handful of times during our testing.
Magic Eraser doesn’t require an active internet connection. Once you open a photo and select Magic Eraser, it will scan the photo and automatically mask anything it thinks is a distraction – this is handy sometimes – it found a crowd of people in one of the photos and deleted all of them without the need for us to do anything more.
Google offers a separate Pixel Studio, which, among other things, can erase objects in a photo. We tested it and saw identical results to the Magic Eraser that’s built into Google Photos.
If you need to manually select something to remove, draw around it and Magic Eraser will try and pick it out. It’s only average in this regard, needing multiple inputs to properly mask a human silhouette.
Xiaomi
Xiaomi’s solution lives in the proprietary Gallery and is called AI Eraser, AI Erase Pro, or AI Erase 2.0. It’s the hardest to get to – a whole four taps before you can finally start drawing around subjects or objects. It also requires an active internet connection, and it seemingly relies on servers in China because it’s sometimes very slow, and it even refused to work a few times.
Xiaomi’s object recognition isn’t good. Drawing around a person yielded poor results – the AI selected only a small part of the entire shape, and we needed to draw additional circles – sometimes up to 10 – to cover a simple human shape.
Selecting multiple objects is a tedious task, and the app would simply wipe all the masks clean at one point, and we needed to start from scratch.
