From 2D Archives to 3D Avatars: Automation and Optimization of Photorealistic Synthetic Humans in the Broadcast Industry

Throughout history, the idea of creating artificial humans has been a constant in the collective imagination, fascinating different cultures and eras. Not only does the interest in this topic continue, but thanks to the possibilities offered by today's technological landscape, it is possible to create 3D avatars with a remarkable degree of photorealism and likeness. This study, conducted in close collaboration with RAI, aims to investigate and optimize an advanced workflow for the creation of photorealistic 3D avatars. The broadcast industry is constantly evolving, and viewers are increasingly demanding engaging and interactive contents with accessibility at the forefront. The challenge is to give a second life to Rai's huge archive of 2D footage of the subject by creating 3D avatars from 2D images, carefully analyzing the technical and creative challenges encountered during the creation process. A critical aspect of this research is the identification and implementation of solutions to automate the workflow of creating synthetic humans, with the main goal of significantly reducing the need for human intervention and ensuring an efficient and flexible creation process that can operate at scale. Subsequently, an evaluation of the generated synthetic humans was performed, both from a subjective and objective point of view, focusing on fidelity and similarity to the original subjects. This study represents a first step towards a future where the creation of photorealistic avatars will be highly automated, enabling easy and widespread integration of synthetic humans into the broadcast and metaverse ecosystems. This will have a significant impact on visual storytelling, revolutionizing the way audiences interact with media content and opening up new possibilities for visual storytelling.