MediaTek and Kuaishou have joined forces to bring generative video (Gen-AI) capabilities to smartphones powered by Dimensity 9300 and Dimensity 8300 chipsets. Kuaishou's image-to-video conversion technology leverages the powerful 7th generation NPU of the Dimensity mobile platform to generate dynamic videos based on image prompts directly on smartphones. The technology continues the video generation innovations that debuted at Mobile World Congress (MWC) 2024.
Kuaishou I2V-Adapter is a lightweight plug-and-play diffusion module based on Stable Diffusion. The module can convert still images into dynamic videos without modifying the original structure and pre-trained parameters of existing text-to-video generation (T2V) models. In-house benchmarks have demonstrated that I2V-Adapter produces image-to-video results comparable to Stable Video Diffusion, but with only 1% parameter training and 18% data usage, with significantly reduced power usage, bandwidth, and storage requirements. In addition, the decoupled design allows the solution to work seamlessly with modules such as DreamBooth, LoRa, and ControlNet to realize customizable and controllable image-to-video generation.
In recent years, the global short video market has grown rapidly, opening up new avenues of content creation for users and businesses and creating further expansion possibilities. The industry's demand for innovative and efficient video production technology is ever-increasing. The Kuaishou app now boasts 379.9 million daily active users and an annual e-commerce transaction volume of over RMB 1 billion.
With this in mind, MediaTek and Kuaishou developed this handy video generation tool to foster a thriving short video ecosystem and foster creativity. The two companies will continue to deepen their collaboration on generative AI tools with a focus on on-device processing.
