A PRISMA ‐Based Systematic Survey on Text to Image Generative Models in Multilingual Contexts

Abstract

ABSTRACT Purpose The main purpose of this study is to conduct a systematic literature review (SLR) on current trends in text‐to‐image synthesis, focusing on research questions related to datasets, generative approaches, evaluation metrics, and application domains. Special emphasis is given to multilingual pre‐trained models, particularly those handling Hindi language processing. Methods A systematic literature review (SLR) was conducted following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta‐Analyses) guidelines. Initially, 464 peer‐reviewed research articles on text‐to‐image synthesis were collected from top‐tier journals. After applying inclusion and exclusion criteria, 164 studies were shortlisted for analysis. Results The review highlights recent research streams, methodological trends, and existing gaps in the field. Generative approaches : Generative Adversarial Networks (GANs) are the most common ( n = 125), while diffusion models ( n = 29) are gaining popularity due to their ability to produce high‐resolution images, albeit with higher computational complexity. Datasets : The CUB‐200 dataset is the most frequently used ( n = 93). Evaluation metrics : The Inception Score is the most widely adopted metric across studies. Conclusion This SLR provides a comprehensive overview of text‐to‐image synthesis, identifying key trends, challenges, and opportunities for future research. The findings offer valuable insights for advancing generative deep learning in this domain, with specific recommendations for improving multilingual models and computational efficiency.