2017 may be the year the US starts a nuclear war because someone tweets something mean about Donald Trump, but at least we’re getting some weird selfie apps before we go. Earlier this month we had Meitu, which made you look like a terrifying anime character and possibly tracked your location data, and this week we have FaceApp, which uses neural networks to paste a smile on anybody’s photo and possibly steals your soul or something.
The app is iOS only and very hit-and-miss. You can see it did quite a good job with Trump up top, but that’s because his face is quite small in the photo, he doesn’t have a beard, and he’s looking straight at the camera. Add in any of these elements, and the results become much less convincing, as with Wesley Snipes below:
Adding smiles isn’t all the app can do, though. You can also make people old, young, male, female, and “hot.” (Like Meitu, this mainly makes you skin paler; an example of the often racist association of paleness with beauty.) The gender swapping is perhaps the most interesting feature, and often turns out some quite convincing results. But for some reason, you can only access it in “collage” mode, meaning the resulting images are quite small.
FaceApp isn’t anything more than a fun distraction, but it does demonstrate something we’ve written about before: how artificial intelligence is making it easier than ever to morph and manipulate photos.
As with Russian app Prisma, which uses an AI technique named “style transfer” to make selfies look like famous paintings, FaceApp is leveraging the power of neural networks; sending images to the cloud to transform them. With Prisma, its creators slowly improved the app by making the neural nets faster, adding more filters, and allowing the software to run locally on users’ phones. Then Google announced it had created similar software; then Facebook. It’s not impossible that FaceApp could make a similar journey to the mainstream, although modifying photos in this way is trickier than style transfer, and might be too taxing to work locally.
Yaroslav Goncharov, an ex-Yandex exec and CEO of the Russian company that created the app, Wireless Lab, told The Verge that an Android version of FaceApp is already in alpha and should be out soon. He said that the neural networks involved were trained by Wireless Lab “from scratch” and claimed that no other commercial products offered photo modifications as good. (Although there have been similar research prototypes, like Tom White’s SmileVector.)
Goncharov also confirmed that photos uploaded to the app are stored on the company’s servers to save bandwidth if several filters are applied, but get deleted not long after. And, he added, unlike the Meitu app, FaceApp doesn’t require any weird system permissions or track data like GPS. New filters will be coming to successive versions of the app, including some emulating photo effects like bokeh and different lighting setups.
If you try out the app yourself and create any good examples, do drop them in the comments below.