Xingyun (alias) is a tunnel worker who lives in a small rural Chinese town. He has never picked up any video editing skills and certainly isn’t into any of those fancy new technologies like artificial intelligence or augmented reality. But that didn’t stop him from creating an amusing short video clip that sees notorious Japanese monster Godzilla blasting his “atomic breath“ into the air at a mountainside construction site. The video has been widely viewed on social media and earned Xingyun thousands of followers.
Hollywood’s eye-popping visual effects are usually incubated in high-end visual studios by professional artists with budgets in the millions. But now, thanks to social video app Kuaishou and its plug-and-play tools, even rural Chinese with cheap smartphones are quickly and easily producing short videos featuring Hollywood-style special effects.
Founded in 2011, Kuaishou is one of the most popular multimedia mobile apps in China for creating entertaining short-form videos. The company’s user growth has skyrocketed along over the last five years, and today more than 200 million people around the world use Kuaishou every day, with 30 percent coming from first- and second-tier cities. In comparison, popular Los Angeles-based social platform Snap just surpassed 200 million daily active users this quarter.
Kuaishou is not so different from other mobile video social apps: Users shoot in either selfie or short video mode; then add visual effects, AR filters, and royalty-free music clips to make the content fun and entertaining; and finally share their creations on the Kuaishou app for everyone to see. Kuaishou has a well-established recommendation engine and multi-modal understanding system that can match videos with interested users based on likes, follows, viewing history, etc.
Kuaishou is now a 25 billion-dollar unicorn backed by Chinese tech giant Tencent. The company has 8,000 employees and has set an ambitious goal of reaching 300 million daily active users before the 2020 Spring Festival. A global version of the app, “Kwai,” is popular in Brazil, India, Turkey, South Korea, Russia and Vietnam.
The best part of Kuaishou is its wide range of filters, stickers and AR effects, aka “Magic Expression.” Users can create a dancing video with a cute Memoji-like filter atop their heads, swap faces with Hollywood movie characters like Caesar the intelligent ape from Rise of the Planet of the Apes, or of course go full Godzilla. Some 15 million videos are created and shared every day on the Kuaishou app. The platform has more than 5000 content creators whose followers have surpassed one million.
Kuaishou Founder Hua Su is a former Google and Baidu engineer who sought to integrate Hollywood’s special effects into a user friendly video production process to deliver a novel experience. In 2016 Kuaishou launched its Y-Lab research facility (later changed to “Y-Tech”), led by Wen Zheng, who graduated from Stanford University with a PhD degree in computer science. Y-Tech’s goal is to become another Google X and contribute to cutting-edge R&D results in AI, AR, and 5G.
The Y-Tech team’s major research efforts are on developing fundamental algorithms ranging from facial keypoint detection to semantic segmentation. The efforts have spawned various successful applications, among them a facial swapping effect that gained traction on social media last year which enables transposing portraits into 3D augmented virtual characters.
Wireless Lab’s FaceApp recently went viral when its aging filter was applied to celebrity photos. About a year back, Kuaishou rolled out a similar aging filter for videos, the “Kuaishou Time Machine.” The filter predicts how users will look up to 60 years in the future, demonstrating the aging process year-by-year. The app was a well-developed combination of algorithms including fast dynamic facial capture, timing-based facial deformation, hair segmentation and dynamic aging. It had a profound impact on many users who commented it had impressed upon them the value of their youth.
Y-Tech Semantic Segmentation Team Head Congli Song told Synced that his wife is a big fan of Kuaishou’s magic expression. “As we understand the algorithm behind the application, I don’t think this is that magical. But they find it unbelievably fascinating, and especially love to see people filmed with various face filters. Sometimes I feel quite a sense of accomplishment.”
Moreover, Kuaishou visual effects can be processed on low-end smartphones that cost only US$150-200, which is an important consideration as a substantial portion of the company’s users are country folk who cannot afford flagship smartphones. Y-Tech built its apps to accommodate low-end devices by developing a home-grown deep learning inferencing engine YCNN, which performs neural network operations on the edge and deploys models through quantification to speed up inference without compromising accuracy.
Another solution used to smoothly run AI applications on cheap smartphones is model compression. Kuaishou this year introduced two novel techniques on energy-constrained compression for deep neural networks. The two research papers are a joint collaboration between Kuaishou’s Seattle lab and the University of Rochester, and have been accepted by top AI conferences CVPR and ICLR.
Director of Kuaishou Seattle AI Lab Ji Liu told Synced that their research has one main gist: equally enabling everyone to record their life moments. “Unlike many academic papers which are actually done offline and never factor in hardware limitations, we are conducting experiments online with real-time data processing… the technique has to work on even the cheapest smartphone.”
Founder Su is an advocate of algorithms and machine learning who firmly believes AI should be used for the good of all humans. For him, this is what Kuaishou is doing — making it possible for people like Xingyun to create imaginative and amusing videos even without tech know-how or expensive hardware. “No matter what kind of technology we do, we should finally use it to improve happiness.”
Journalist: Tony Peng | Editor: Michael Sarazen