There are problems with the quality and generation time of images synthesized by digital media. In order to boost system performance, this paper contributes network intelligence to the system, and conducts an in-depth study of digital media synthesis systems based on network intelligence by experimenting on image quality, video frame rate, image generation time, and video generation time aspects between GANs and self-encoder. The paper comprehensively analyzes the system design principles, advantages, disadvantages, and the potential improvement and optimization design direction for each model. The experimental results demonstrate that the evaluation value of the self-encoder algorithm model ranges from 0.84 to 0.93, and that of the GANs algorithm model ranges from 0.89 to 0.96. GANs perform better than the auto encoder algorithm model in terms of image quality and video frame rate based on testing practical functionality, generating high-quality, diverse image samples, and constituting continuous video frames that meet the visual interests and requirements of end users.