The “Bold Glamor” and “Teenage Look” filters on TikTok that make your face look like that of a top model with perfect skin and contours have been going viral for several weeks now. But what technology is behind it? An ingenious piece of artificial intelligence, it turns out.
Earlier similar filters mostly used ‘overlays’. A kind of virtual mask that they automatically put over your face to make it look nicer, for example. Or they project the camera image of your face onto a 3D model and adjust that model – by making the jaws of the virtual model less pronounced, for example, yours look like that too. That gives an interesting effect, but is not nearly as impressive as the new filters that have been available for about a month.
LOOK. Five people test the controversial TikTok filter that gives you a ‘perfect’ face: “Everyone has pores, don’t hide it”
Artificial intelligence
‘Bold Glamour’ used to be controversial but nice effect artificial intelligence. But how exactly? Technology site The Verge early on TikTok for a response and technical explanation, but did not receive one until today. The company keeps its -perfectly sprayed- lips tight.
On February 22, the company did launch an update to Effect House, a tool developers use to create effects and filters for users on TikTok. With this you can already adjust the lips in filters, or hide eyebrows, for example. Chances are that the same technology is used in the ‘Bold Glamour’ filter. TikTok does not immediately admit that those new effects use AI.
‘Generative Adversarial Networks’
We actually owe the new filter to the American Ian Goodfellow. He now works at Apple and is (and we don’t use that term lightly) genius in the field of artificial intelligence. He invented the so-called Generative Adversarial Networks together with some colleagues about eight years ago.
In short, such a GAN is an artificial intelligence that generates things itself. Think of images, for example. But we are a bit ahead of the facts, because of course it takes a while before you get that far. Traditionally, for a GAN you start by teaching an AI something: you teach a computer, very simply, what a beach ball is by showing those thousands of pictures of beach balls. Eventually you get an AI that can say ‘this is definitely a beach ball’ or ‘this is probably not a beach ball’.
A GAN then uses a second AI to set up a kind of competition with the first. By generating random pixels all the time and making pictures. He shows it to the first AI and asks “Did I make a beach ball now?”, a bit like a child who asks his parents with every scribble: “Look, mom, isn’t this you?” and a parent who then lovingly makes it clear that he has five fingers on each hand, not two as the child drew.
The AI will get better just like a child
After a while, both the child and the AI will get better at producing the right output: the child will start drawing hands with five fingers, the AI will ‘know’ after a while how to design one beach ball after another.
Fake
In this way, artificial intelligence learns to generate images itself from scratch. And the same technique is used behind the scenes in AI filters for social media. In the case of the filters we are talking about, you first have to show AI a lot of lips, eyelashes and eyebrows that really exist. For example, you could tell AI ‘this is what the perfect lips look like’, and then have it projected onto your own face.
The AI will really adjust every pixel of the image or video it sees for that. Not like the traditional first generation of those filters, by just putting a mask over it. So it is a much more advanced way of working. Although TikTok has not yet admitted that to this day, mainly because this type of filter and the technique used is also controversial.
Not ‘out of the blue’
For example, you may wonder what the ‘perfect’ lips are. And why we let the choice of that depend on the bosses of a social network. In the past, women were presented with the curves of Palema Anderson, which sometimes caused complexes. But now also what they could look like if they met perfection.
Moreover, generating ‘out of nothing’ is also not entirely correct. Of course, as already mentioned, thousands of images and photos of lips, eyebrows and teenage faces have been put into the computer. And of course they really existed. But where did they come today? No one is saying that TikTok didn’t just take the faces of the most popular models and influencers on the network to train their computer model, so to speak. For example, without realizing it, with such a filter we increasingly resemble the ‘perfection’ dictated by the profiles that get the most clicks and views.
Read also:
Does Lieven Scheire use AI to predict the identity of the ‘Masked Singers’? “We already had Herman, Marc, Timmy and Anouk right” (+)
1 in 4 consider themselves addicted to their smartphone: how do you get rid of it? Our tech journalist Kenneth Dée gives tips
“Children used to want to be firefighters or astronauts. Now vlogger or influencer”: this is how you get more out of social media (+)
LOOK ALSO. TikTok users stunned by new realistic ‘Bold Glamour’ filter
Free unlimited access to Showbytes? Which can!
Log in or create an account and don’t miss a thing of the stars.
