Guiding AI-Generated Digital Content with Wireless Perception
Jiacheng Wang, Hongyang Du, Dusit Niyato, Zehui Xiong, Jiawen Kang, Shiwen Mao, Xuemin Shen
Abstract
Advancements in artificial intelligence (AI), and the surge in diverse training data, have facilitated AI generated content (AIGC). Despite high efficiency, the inherent instability of AI models poses challenges in creating user-specific content, especially when creating an avatar for a user. To address this issue, this article integrates wireless perception (WP) with AIGC and introduces WP-AIGC, a unified framework that leverages a user skeleton obtained by WP to guide AIGC, thereby generating the avatar that aligns with the user's actual posture. Specifically, WP-AIGC first employs a novel multi-scale perception technology to sense posture in the physical world and construct the user skeleton. Then, the skeleton and the user's requirements are conveyed to the AIGC, thereby guiding the creation of the avatar. Furthermore, WP-AIGC can adjust the computing resources allocated to perception and AIGC based on user feedback, thereby optimizing the service. Experimental results verify the effectiveness of the service. With limited computing resources, WP-AIGC achieves optimal QoS of 3.75 when four links are involved in perception.