With StreamLake, Kuaishou enters PaaS market to help enterprises make AI-driven videos

August 10, 2022 0 Comments

On August 10, Kuaishou, China’s leading short video platform, launched its new brand StreamLake, which offers Video+AI solutions for enterprises. StreamLake marks the company’s business expansion, from serving consumers to serving enterprises.

In the past year, Kuaishou has joined hands with question-and-answer website Zhihu, telecommunications operator China Unicom, CCTV’s video streaming platform Yangshipin, and smartphone maker Xiaomi to explore in the field of video and AI.

Kuaishou’s Chief Technology Officer Chen Dingjia, Yu Bing, Senior Vice President and Head of StreamLake Yu Bing, and Wang Zhongyuan, Vice President of Kuaishou Technology and Head of AI attended the press conference, while industry experts such as Li Dahai, Partner and CTO of Zhihu, Qin Jibo, Deputy General Manager of Unicom Technology, and Xi-Peng Li, General Manager of Development and Technology of NVIDIA Asia Pacific shared their insights.

The goal of StreamLakes is to “encapsulate all the technical details into an operating system for industry-wide video transformation, give users an interface with simple interaction and easy access,” says Yu.

With the global rise of short-video platforms, video as a form of content is becoming the most popular vehicle for connecting customers across industries. A recent report from Ericsson Mobility shows that mobile data traffic will grow 4.38 times over the next six years, with video traffic accounting for 79 percent of that total. Thus, it is crucial to make video creation and video streaming more efficient.

Over the years, Kuaishou has accumulated ample experience in the video and audio streaming industry, which StreamLake benefits from. For example, Kuaishou has developed its own form of video coding — KVC(Kuaishou Video Coding), which can optimize the decoding power consumption to essentially the same level as the H.265/HEVC, while saving 40% to 50% bandwidth.

According to Kuaishou, the KVC can balance between network speed and video quality, and drops the lag rate significantly by 49%. In addition, Kuaishou has developed a model which can accurately predict the “”viral-ness” of a short video, and compress the video to save bandwidth before it goes viral.

The company has also developed a cloud-based intelligent video processing SoC chip for live video-on-demand applications, which achieves 30% bitrate reduction over the classic open-sourced x265 medium at same quality, according to Kuaishou.

StreamLake’s services include three key parts — smart video creation, smart video interpreting and digital humans. StreamLake can search for video footages based on plain texts that users upload, and generate a complete video. For example, when users input a description of a lake, StreamLake can automatically pull out video footages of such nature view.

Furthermore, with StreamLake, short videos can be accurately interpreted with or without hashtags. A 64-dimensioned video interpretation would be automatically generated, and similar videos can be recommended to users based on that interpretation.

Lastly, StreamLake has established a strategic partnership with Unicom to jointly develop digital human solutions, including “communicative digital humans” which can serve as brand ambassadors, “service digital humans”which can act as live shopping hosts and “virtual avatar digital human” which would be user’s avatar in the virtual world. According to Kuaishou’s demonstration, the service digital humans can be activated without motion-capture suits, and can also operate on its own with pre-set process and trigger words.

StreamLake’s digital human live demonstration at Kuaishou’s press conference on August 10, in Beijing.

StreamLake aims to simplify the video streaming industry, “by turning core capabilities accumulated by Kuaishou such as AI, audio, video, algorithms into a modular and standardized service which opens to other enterprises,” says Yu.

Leave a Reply

Your email address will not be published. Required fields are marked *