The neural network that makes smile

How the FaceApp project from Yandex, which is popular in Russia and Asia, is developing and why it has not yet won the trust of investors.

The application for processing photos using neural networks FaceApp began to rapidly gain popularity in Russia in early 2017. The FaceApp service allows you to add a plausible smile to a gloomy person in a photo, to turn a man into a girl, and anyone who wants to dream up how he will look in old age.

On April 20, FaceApp managed to get to the first place in the list of the most popular Google Play Store apps for Android in Russia, the founder and CEO of the company that developed the Wireless Lab application Yaroslav Goncharov said on his Facebook page. In the Russian App Store, the application ranked first on April 15, 2017, and continues to hold its position on April 24.

Goncharov himself worked until the end of 2013 at Yandex, where he headed the department of mobile platforms. He got to the Internet company in 2011, when Yandex bought SPB Software in St. Petersburg, which also specialized in the creation of mobile products. The transaction amount was $ 38 million.

In SPB Software Goncharov held the position of technical director, and, according to him, he became interested in neural networks long before this technology became popular. “I trained the first neural network about ten years ago, although it, of course, was much simpler than those used in FaceApp,” Goncharov recalls.

Leaving Yandex at the end of 2013, he created his own company, Wireless Lab, which also specialized in mobile applications.

FaceApp appeared in the App Store – originally the application was available only to iPhone owners – at the end of January 2017. Then about the application that allows you to add a smile to the gloomy photo, wrote the leading Western publications – from TechCrunch to The Verge .

But it’s only in April that the FaceApp really got viral. The creator of the application notes that several factors contributed to this at once. On the one hand, it was possible to significantly improve the quality of photos that the user receives after being processed by neural networks.

Photos of Russian singer Phillip Kirkorov were processed at FaceApp during the show “Evening Urgant”

On the other hand, on February 15, a version for Android was released, which is very important for viral distribution, Potters notes. This allowed more people to add a smile or “turn” on the photo into a person of the opposite sex.

“Everyone writes programs”

2016 has become a real flourish of applications that convert user photos based on artificial intelligence (AI) technologies.

At the beginning of 2016, in just three months, the application of Belarusian developers MSQRD gained global popularity. In March of the same year, Facebook bought it. The amount of the transaction, according to sources of RBC magazine, was about $ 85 million.

A little earlier, in September 2015, another project with Russian-speaking founders, Looksery, a Ukrainian startup that allowed adding filters to user photos, bought Snapchat for $ 150 million.

And in the summer of 2016, the boom of photo-processing applications using neural networks began in Russia. Appearing on the App Store on June 11, the Prisma application , which transformed user photos into the style of famous works of art, in just nine days was the most downloaded in ten countries (Russia, Belarus, Moldova, Estonia, Kyrgyzstan, Latvia, Kazakhstan, Uzbekistan, Ukraine and Armenia) .

To repeat the success of the application of its former employee – the founder of Prisma Alexey Moiseenkov – Mail.Ru Group released two applications based on neural networks: Vinci for photo processing and Artisto for video.

The idea of ​​FaceApp itself appeared in May 2016. Then they began to develop the application. It took nine months to create the first version of the product. Existing developments and expertise from past machine learning projects were actively used in the development, Goncharov notes. Four people work full time on the FaceApp team. Three more are partially involved in the project.

FaceApp processes the image so that it remains a photograph, Goncharov explains the difference from the competitor Prisma. According to him, the team is trying to achieve photo-realism, regardless of changes in the photo content, adding a smile or changing a person’s gender and age in the image.

The developers specifically polled users to see whether photos processed with FaceApp are perceived as pictures or as real pictures. So far, the application has managed to convince users of the latter.

FaceApp tries to achieve realistic images, but still confuses painting and photography

Technologically, FaceApp differs in the neural network architecture, which processes the image, Goncharov notes, but does not go into details.

According to the founder of Prisma Alexey Moiseenkov, FaceApp’s neural network analyzes the image, laying it out into significant features. Developers find signs that distinguish an old man from a young, smiling from a non-smiling, man from a woman. Thus, they know which elements of a neural network most often distinguish a smile from a calm facial expression, he explains.

Then, in the photo, a group of signs is determined that is responsible, for example, for the smile zone, and only the values ​​of this group are changed to typical values ​​for the average smiling person. So only the smile changes in the photo, explains Moiseenkov.

Prisma, on the other hand, completely changes all the elements of a user photo, but in accordance with the style image analyzed by the neural network. “Our task is to change everything, but in such a way that the structure of the image itself does not break. And for FaceApp, find a specific element in the picture structure and change it, ”explains the differences.

“You are not surprised that everyone is writing programs or creating applications for mobile phones?”, Goncharov answers when asked if there are too many AI-based photo apps.

Neural networks are just technology, today all projects are somehow created on the basis of AI and neural networks, explains the creator of FaceApp.

The main thing is that the effect that end users get is different. The user can easily use two applications (FaceApp and Prisma) at the same time to get different effects when processing photos.

Goncharov calls Meitu the main competitor of his application , but Facebook and Instagram have to fight for users, who actively invest in the development of user photo processing technologies and selfies. Also among the competitors, he singles out the Israeli company Lightricks , which specializes in creating products for photo processing.

In August 2015, Lightricks raised $ 10 million from the Israeli Carmel Ventures Foundation. Its main product, a paid Facetune selfie app , at that time had 4 million users, each paying $ 4 to use it.

The founder of another art photo editing application, Mlvch , Anton Pronin agrees with Goncharov. In his opinion, there can not be too many projects where a computer imposes unusual effects on a photo. And most likely only they will soon win the attention of users, Pronin notes.

Soon, when the general excitement around the AI ​​will pass, only the content will remain important for the user: in this case, this is the effect that the application creates, and it is different for each product, the founder Mlvch is sure.

His words are confirmed by the global trend. According to a PwC Money Tree Report study , in 2016 the volume of investment in companies somehow related to artificial intelligence grew by 16% to $ 705 million, and the number of transactions increased by 22% to 71 (US statistics only).

Investor interest in AI is growing in spite of the general trend towards a decline in investment activity in the United States. In the same year, the number of investments in monetary terms fell by 20% to $ 58.6 billion, and the number of closed transactions fell by 16% to 4.5 thousand, the PwC study noted.

AI “undermines” not only b2c services, but also “traditional” industries: automotive, financial services and healthcare, said PwC partner Anand Rao in the study.

Russian Meitu

Now there are seven filters in the FaceApp application: the user can add two types of smiles to his selfie, make himself a man or a woman, an older or younger version, and just become more beautiful in the photo.

According to Goncharov, by mid-April 2017, FaceApp downloaded several million times. But more representative is the daily installation rate of the application, he notes. For example, only on Tuesday, April 18, FaceApp downloaded 700 thousand times. “I would like to keep this pace for the next few weeks, mainly due to new countries,” the creator of the application notes.

However, investors polled by doubt that the application will succeed. Applications of one function that offer users only “decorations” quickly bother people, and the viral effect subsides, even if initially the project managed to reach a large audience, notes Artem Inyutin, TMT Investments investment director.

“It is unlikely that people will use the application with one or two functions more than several times,” explains the investor. To keep the user for a long time, the application must meet basic needs – for example, in communication, he said.

“Remember Instagram, it started as an application for publishing photos, and today it is an instant messenger and a tool for users to communicate with the business,” says Inyutin. Stickers and filters need to be “strung” on a basic function that guarantees the use of the application every day, he explains.

There are three factors that shape the success of a mobile product: the virality of its distribution, the mass potential market and the ability to retain users for a long time (retention), explains the creator of FaceApp.

If you are lucky enough to create a product that meets all three factors, congratulations, you made Facebook

Yaroslav Goncharov

founder and CEO of FaceApp

FaceApp now meets only two criteria: it is a viral product that is in demand on the mass market, admits Potters.

The application already has effects designed for the viral distribution of photos: for example, becoming a person of the opposite sex. But there is a “basic need”: a general improvement in the portrait. “Everyone wants to look a bit better on a selfie,” Goncharov explains.

It’s not going to make your own social network (for example, Prisma, which launched an analogue of the Instagram tape in this photo application), Goncharov is not going to – a too high barrier to entry and unpredictable results. “It’s easier to try to become a new Meitu, than to make Facebook or Instagram,” he believes.

The Chinese company Meitu, which in December 2016 placed its shares on the Hong Kong Stock Exchange and is currently valued at $ 6 billion, produces mobile phones, each of which is equipped with tools for improving photos. In 2008, the company introduced the first eponymous application for improving selfies in “cute” styles. In Russia, Meitu gained short-term popularity in January 2017.

So far, the most popular FaceApp filters are a smile, an aging effect, a transformation into a woman and a “prettier selfie,” Goncharov notes. The latter function is particularly similar to the Meitu filters, which made a fortune on the Chinese love for “more beautiful” photos.

In Asia, FaceApp had a “very powerful launch,” says Goncharov. The application managed to enter the top most downloaded applications on the App Store in Japan, Hong Kong and China, says the creator of FaceApp. According to the AppAnnie service, the application ranked first in the Apple app store in 12 countries (primarily in Russia and the CIS countries). In 29 countries, mainly in Europe – entered the top 10.

Own way

How much has been spent for the development of FaceApp for almost a year, Goncharov refuses to say: it was the internal development of the Wireless Lab for the company’s money, so the costs for the application were not considered separately. The company has no third-party investors.

On the monetization of the product Potters only thinks about. In the long run, the most organic monetization model is branded filters. This model successfully uses Snapchat. However, in order to work effectively, a large and stable daily audience is needed, Goncharov notes.

For now, FaceApp only offers its users the opportunity to disable advertising for $ 4 and remove the FaceApp logo from the processed photo. She already took advantage of several thousand people, says Goncharov.

For the further development of the project will need third-party investments, recognizes Potters. Now FaceApp needs several million dollars to further improve the technology and create a full-fledged selfie editor based on neural networks. Goncharov is negotiating with several funds, but is not yet ready to talk about the results of the negotiations ..

“Investment is, unfortunately, not a very attractive project,” says Nikolay Davydov, managing partner of Gagarin Capital (investor Prisma). FaceApp makes processing of photos on servers, and “Prisma or Facebook can“ on time ”repeat these effects directly on a mobile device,” thereby saving on computing power, according to Davydov.

With the development of computer vision technology, such projects will appear more and more often, the investor is sure. “A year and a half ago, only a super professional team that understands mathematics, neural networks and can write complex code, could do such a project, and now it can be assembled out of the box,” explains Davydov.

Soon, any user will be able to create such models by moving blocks with the mouse in the constructor, and the next step will be neural networks that write and train new neural networks themselves, he summarizes.

However, the founder of Mlvch Anton Pronin is sure that there is a great demand for teams capable of making high-quality products using machine learning technologies. And it depends only on the ingenuity of the FaceApp team whether it will be possible to build a working business, he said. The Mlvch team itself is now creating boxed image processing solutions using neural networks for various media agencies.

“I am sure that the guys will receive a lot of offers for the purchase, including from Western corporations. Here is a question of price and compliance goals, ”adds Pronin.

Despite the fact that the preferred option is independent business development with the involvement of third-party investments, Goncharov notes that the company is ready to consider including a takeover.

Back to top button