A survey paper by Nonghai Zhang and Hao Tang reviews over 490 papers on text-to-image synthesis published over the past decade, with representative advances covered through 2026. The authors examine how three dominant paradigms—GANs, autoregressive models, and diffusion models—have been applied to generate images from text, and they also analyze the shared components that have driven progress.

Rather than a purely chronological account, the survey groups recent work by problem area, including controllability, personalized and identity-preserving generation, instruction-based image editing, preference alignment with human and AI feedback, efficiency, safety, and consistency in content and spatial relationships. It also covers unified multimodal models that combine understanding and generation, and it summarizes commonly used datasets and evaluation metrics.

The authors provide a comparative discussion of the three paradigms and their trade-offs, and they consider potential applications within AI-generated content (AIGC). They close by identifying challenges and forward-looking research opportunities, making the survey a broad reference for the field's trajectory and open problems.