The software, called Voila AI Artist, uses AI algorithms to upload a portrait and produce images in four filter styles — a 3D cartoon (Disney-style), a 2D cartoon, a Renaissance painting and a comic character. Voila also has a database of celebrity photos, and a search on the App will show you how the face changes look.
Unfortunately, the user can’t edit the image, adjust the eyes, mouth or hair, change the color or contrast, and so on. Compared with the laboratory model, it is obvious that the packaged application has the advantage of being simple and easy to use. Foreign media even described the App as spreading rapidly like a “virus”.
Roughly three months after its launch, Voila’s IOS version has topped the free charts in several countries and regions, beating out formidable rivals like TikTok, Instagram and Snapchat.
On Android, Voila has been downloaded more than 10 million times on the Google Play store and is in the top 10 charts in 26 countries and territories. The official Facebook account celebrated reaching 20 million users in a post on June 13.
A mysterious facilitator
Voila AI Artist, whose parent company is Wemagine.AI, was registered in London in January this year and launched the App less than two months later. Beyond that, however, there is little more detailed information about the company. Fox Carolina, however, said she had contacted the company: Wemagine.AI says it is based in Canada, but its team is small and works remotely around the world.
According to Company Check, the Company’s founders were Lim Eliska and Wilson Tjoa. Both lived in Indonesia frequently and were under the age of 35.
We also configured Google Cloud Platform to delete photos and photo related information within 24-48 hours of the last time we used the app.
We may also disclose your information in other circumstances: to our subsidiaries and affiliates. To contractors, service providers and third parties that support our business.
But Voila admits that it does collect personal information about users, including websites visited, Cookies, phone models and photos, and provides it to advertisers it partners with. This usage data may include your Internet protocol address (such as IP address), browser type, browser version, service pages visited, time and date visited, time spent on those pages, device identifiers, and other diagnostic data.
When you use a mobile device to access a service, this usage data may include information such as the type of mobile device you use, mobile device UNIQUE IDENTIFIER, IP address of the mobile device, operating system, type of mobile Internet browser used, unique device identity, and other diagnostic data. We also collect data such as IP address, device model, screen resolution and operating system, session duration and your location with the assistance of Google Analytics. Based on this data, we analyze your needs and interests and improve our services.
When you use the free version of the App, we will display ads in the App. These ads are provided by partners and may be targeted based on your use of the App or your activity elsewhere online.
Besides, what if you don’t want to see ads? Obviously you just have to pay for the membership. Voila says members will be able to speed up photo processing and get rid of watermarks on in-app ads and exported images. Prices are £2.49 a week, £4.99 a month or £25.99 a year
Adversarial generative network
Since the release of “Generative Adversarial Networks” in 2014, it has been widely used, and is now commercially viable. GAN is composed of a generative network and a discriminant network. The generated network takes random samples from latent space as input, and its output should imitate the real samples in the training set as much as possible.
The input of the discriminant network is the real sample or the output of the generated network. Its purpose is to distinguish the output of the generated network from the real sample as much as possible. The generative network should deceive the discriminant network as much as possible. The two networks compete with each other and constantly adjust the parameters. The ultimate goal is to make the discriminant network unable to judge whether the output results generated by the network are real.
Among them, Nvidia’s StyleGAN produces images that are very realistic, and by modifying each level of input in the network separately, it can control the visual features within that level, from rough poses and facial shapes to fine hair colors.
In addition to lifelike people, StyleGAN can also be used to generate other animals, cars and even rooms. Its updated version, StyleGAN2, fixes artifacts and further improves the quality of generated images.