TITLE:
Deepfake and Synthetic Media: An Analysis of the Concept, Technology, and Regulation
AUTHORS:
Thiago Henrique Pessoa
KEYWORDS:
Deepfake, Synthetic Media, Generative Artificial Intelligence, Deep Learning, Digital Synthesis, Audiovisual Manipulation
JOURNAL NAME:
Journal of Computer and Communications,
Vol.14 No.10,
October
23,
2026
ABSTRACT: The advancement of generative artificial intelligence technologies has expanded the capacity for creating, synthesizing, and manipulating highly realistic audiovisual content. In this context, the term deepfake has come to be used to designate different forms of synthetic content, often associated with fraud, disinformation, and manipulation. This article proposes an analysis of the concept and technology used in the phenomenon, seeking to distinguish synthetic media, as a broader category, from deepfake, as a specific manifestation that can assume different purposes and degrees of manipulation. The analysis starts from the evolution of deep neural networks and examines architectures and models used in the generation and manipulation of content, with emphasis on convolutional neural networks, recurrent networks, Transformers, autoencoders, and Generative Adversarial Networks. Four recurrent modalities of manipulation are also analyzed: face swapping, facial reenactment, generation of talking faces, and editing of facial attributes, in addition to facial synthesis. From the selected literature, it is proposed to understand and identify the phenomenon through three dimensions: identity or reality of reference, intention of use, and granularity of manipulation. Finally, the article examines the Brazilian regulatory response, especially TSE Resolutions No. 23,732/2024 and No. 23,755/2026, arguing that regulation tends to focus on the effects, purpose, and risk of the content, and not exclusively on the technology employed. It is concluded that the distinction between synthetic media and deepfake is relevant to prevent the technology from being confused with the potentially abusive use that is made of it. It is argued that the identity or reality of reference and the granularity of the manipulation constitute the classification criteria, while the intention qualifies the risk associated with the content, without being a necessary condition to characterize it as deepfake.