This research examines the development and application of realistic audio-visual avatars in a variety of domains that improve human–computer interactions and improve user experiences. The study conducts a comprehensive comparison of diverse methodologies, datasets, and evaluation metrics employed in the generation audio-visual avatars. Additionally, it contributes to the field by improving a first-order motion model, which is based on the GRPGAN model. This improvement in the motion model is aimed at refining its accuracy and efficacy in generating realistic motions for the avatars. Such contributions pave the way for more human–computer interactions across various applications and domains. The improved audio-visual avatars acquire a highly effective percentage of realism, reaching 86.56% in the evaluation, according to the findings of the subjective evaluation.