Diffusion Models for Image Generation

This paper will conduct a thorough review of 30 scholarly articles about Diffusion Models for Image Generation from 2021 to 2026. Four main categories of analysis will be covered: Foundations, Text-to-Image Generation, Controllable and Personalized Generation, and Efficiency and Applications. It is found that the diffusion models are no longer just a replacement, but they are becoming a new way of generating images of unprecedented quality by reaching the photorealism and text-image consistency level with preserved identity and spatial control. The methods studied include Denoising Diffusion Probabilistic Models, Latent Diffusion, Score-Based Stochastic Differential Equation frameworks, Classifier-Free Guidance, Consistency Distillation, and Flow Matching. Moreover, it is established that the development in this field went from academic proof-of-concept works to large scale applications used by millions of people. Thus, the purpose of this paper is to provide an academic overview for all the researchers and students interested in generative AI, computer vision, and creative technology.