MediaTek and Kuaishou Announce New On-Device Video Generation Technology

TapTechNews July 4th news, MediaTek and Kuaishou jointly announced today the launch of an efficient on-device video generation technology.

This technology is the continuation and improvement of the video generation technology that first debuted at the 2024 Mobile World Congress. Combined with Kuaishou's AI model I2V (image to video) Adapter and the MediaTek Dimensity 9300 and Dimensity 8300 mobile platforms' AI computing power, it realizes the generation of dynamic videos from static images on the device side.

Through this technology, users only need to select pictures on the device, and the application can recognize the characters and scenes in the photos and generate natural and smooth videos, enhancing the creative expression and efficiency of video production.

I2V-Adapter is a lightweight adaptation module proposed by Kuaishou's AI team suitable for the StableDiffusion-based text-to-video diffusion model, specifically for the image-to-video generation task.

According to the introduction, the MediaTek Dimensity 9300 and Dimensity 8300 mobile platforms integrate an NPU (Neural Network Processing Unit) designed for generative AI, which can accelerate the operation of AI models. Combined with the adaptation and optimization of the Kuaishou I2VAdapter model, it realizes a faster video generation speed and at the same time has low power consumption characteristics.

According to TapTechNews' previous report, at the Dimensity Developer Conference held in May, MediaTek, in conjunction with partners such as Aliyun, Baichuan Intelligence, Transsion, ZeroOne Everything, OPPO, Honor, vivo, and Xiaomi, jointly launched the 'Dimensity AI Pioneer Program' to help developers create generative AI on terminal devices equipped with Dimensity chips.