Apple, Google, and Samsung are competing on multiple fronts—from phones, to health apps, to smartwatches—and that rivalry extends to AI tools as well. All of these tech giants offer AI-powered object removal features in their mobile photo editing apps, which ostensibly allow you get rid of an unwanted person, a tree, a hand, or anything else that’s ruining your shot, with only a tap or two on the screen.
However, just because Apple, Google, and Samsung all offer AI object removal doesn’t mean the tools are of the same standard. There’s been a fair bit of chatter online about how these features stack up against one another, so it feels as though a head-to-head object removal test is in order.
These AI tools all work in the same way. Obviously, the tools don’t know what’s really behind the object being removed—but thanks to a combination of clues in the surrounding pixels and the image training built into their underlying models, they can make a reasonably good guess. How good that guess is then determines the quality of the end result.
This test covers three tools: Clean Up (in Apple Photos on iOS), Magic Eraser (in Google Photos on Android), and Object Eraser (in the Samsung Gallery app on One UI). These tools all appear as options when you open images in the relevant apps. For testing purposes, I’ve borrowed some great photos from Taan Huyn, Tolga Ahmetler, and Marek Piwnicki.
(A quick note on the comparison images: There’s some variation in resolution because of how I saved them using various devices; it has nothing to do with the AI tools in question, so you can mostly ignore the sharpness of the pictures.)
Which AI tool is best at removing a lamp from a photo?
Credit: Taan Huyn/Unsplash
Our first photo challenge—removing the lamp in the image above—is relatively simple in some ways, but complex in others. There are some useful guidelines for the AI in terms of the wall and the window, but there are also shadows to deal with (including the shadow of the lamp on the wall behind it, which AI isn’t smart enough to know it needs to remove as well).
First up was Apple Clean Up. The tool did a fine job of identifying the lamp as I scribbled over it—I like the glow effect—and once the removal process was complete, the back wall was filled in nicely. Most of the window pane work looked okay too, but there were a few weird artifacts left as evidence of an object removal.
Next was Google Magic Eraser. The lamp wasn’t identified as a suggestion for something to erase, but selecting it was pretty straightforward and accurate in terms of the boundaries. Unfortunately, while most of the job was done well, Magic Eraser left evidence of the lampshade up at the top, like a ghost of what had been there before.
Finally, I tried Samsung Object Eraser. Here the object selection process was a bit more fiddly—it’s more difficult to zoom and pan in Samsung Gallery, and the automatic selection was less accurate. However, the removal job was the best of the three, with both the wall and the windows accurately filled in.
Credit: Lifehacker
Which AI photo tool does the best job removing a person from a photo?
Credit: Tolga Ahmetler/Unsplash
Our second photo challenge is removing the walking man from this street scene, and it’s a tricky one—the AI has to imagine most of the white car behind our primary subject, as well as fill in the road and its markings. What’s more, it’s a darker scene, with the foreground in focus and the background getting increasingly blurry.
With Apple Clean Up, the tool’s automatic selection wanted to clear away the cars in the background rather than the man in the foreground, but selecting him was easy enough. The AI fill results were patchy, and got worse further back in the scene, with the car at the back turning into a pixelated mess.
Over to Google Magic Eraser, which also wanted to erase people in the background rather than the foreground. It had some trouble picking out the man against the dark and wet street—and it then had difficulty both removing all of the man and replacing the pixels properly. It borrowed pixels from the wrong part of the image on the whole, and the tool couldn’t even carry on the yellow line consistently.
As for Samsung Object Eraser, again the selection process was a bit awkward, but ultimately it did a perfect job of selecting the male figure against the background. It then did really well at removal: It was the only AI to actually end up with a realistic looking car and road after the man had gone. While it’s not perfect, this would probably fool most people at first glance.
Credit: Lifehacker
Which AI photo tool is best at removing a bicycle from a photo?
Credit: Marek Piwnicki/Unsplash
This third photo challenge, removing the bicycle from this scene, offers a lot of information for the AI to use in terms of the surrounding field. The barely visible fencing behind the bike adds to the difficulty, so I was interested to see if these tools would try to continue the fence, or just ignore it and go with the grass.
With this one, Apple Clean Up needed a few goes to get the whole of the bicycle selected—possibly because it doesn’t appear to be one single object. When that was done, Clean Up was reasonably proficient at filling in the blanks, though both the greenery and the fence it created showed inconsistencies.
Google Magic Eraser also needed several attempts to get the entirety of the bike selected. It tried to fill in the fence as well as the field, and while the results weren’t a disaster, there were obvious errors—some of the bike wheel spokes are still there, for example. It was maybe slightly better than Apple Clean Up, but not by much.
As with the man in the street, Samsung Object Eraser was the best at making an accurate selection first time around, and better at putting something convincing in the space that was left. Both the field and the fence were filled in an accurate, realistic way, and this could easily pass as an unedited photo.
Credit: Lifehacker
The final verdict: Samsung’s tool performs best
The first thing to say is that your mileage may vary with these sorts of tasks—as you will see if you look at similar comparison challenges from around social media and the internet at large. Every photo has different characteristics, and every AI tool has its own approach in dealing with object removal.
In terms of my tests, Galaxy AI and Samsung’s Object Eraser are the clear winners. The tool seems to take more time over the replacement pixels, and does some actual image generation rather than just copying and pasting pixels from elsewhere in the image—as you can see with the car on the street. That’s just an AI imagining of a car, not the actual car, but it makes for a photo that looks genuine.
Apple and Google did okay, and were about on a par with one another. The selections are easy to make (easier than with the Samsung, actually), and some more straightforward removal tasks are handled perfectly well. However, these AI removal features do tend to struggle with complex operations where there are fewer clues about how the background should look.
Ultimately, it’s unlikely that anyone’s going to change their phone solely because of the AI object removal tools it offers, but it’s interesting to chart the progress these three companies have made with their AI offerings. It’s also worth noting that Google Photos and its Magic Eraser are available on all Android and iOS devices, which puts pressure on Apple and Samsung to keep up (or stay ahead).
