Generative AI in Depth: A Survey of Recent Advances, Model Variants, and Real-World Applications

TL;DR

This survey provides a taxonomy-driven synthesis of GANs, VAEs, and diffusion models, including their hybrids, focusing on architecture, training dynamics, variants, and real-world impact. It highlights how innovations such as WGAN-GP, LSGAN, InfoVAE, and latent diffusion extend performance and controllability, while diffusion models offer stable training at the cost of slower sampling. The paper also discusses ethical considerations, including deepfakes, IP, and fairness, and outlines persistent challenges like training stability, mode coverage, speed, and alignment with ethical guardrails. By mapping methods to applications across vision, healthcare, robotics, and environmental modeling, it serves as a forward-looking resource for researchers and practitioners to navigate rapidly evolving generative AI responsibly.

Abstract

In recent years, deep learning based generative models, particularly Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and Diffusion Models (DMs), have been instrumental in in generating diverse, high-quality content across various domains, such as image and video synthesis. This capability has led to widespread adoption of these models and has captured strong public interest. As they continue to advance at a rapid pace, the growing volume of research, expanding application areas, and unresolved technical challenges make it increasingly difficult to stay current. To address this need, this survey introduces a comprehensive taxonomy that organizes the literature and provides a cohesive framework for understanding the development of GANs, VAEs, and DMs, including their many variants and combined approaches. We highlight key innovations that have improved the quality, diversity, and controllability of generated outputs, reflecting the expanding potential of generative artificial intelligence. In addition to summarizing technical progress, we examine rising ethical concerns, including the risks of misuse and the broader societal impact of synthetic media. Finally, we outline persistent challenges and propose future research directions, offering a structured and forward looking perspective for researchers in this fast evolving field.

Figures (4)

• Figure 1: We categorize generative models into two branches based on Likelihood approaches. GANs use implicit density and are classified into six branches based on two perspectives: (1) loss functions and regularization techniques, like batch and spectral normalization, enforcement of Lipschitz continuity, gradient penalty, and least square loss function; and (2) architectures of generators and discriminators: Laplacian pyramid. VAEs use explicit density and are classified into two branches based on use of ELBO and MMD techniques. Hybrid GAN-VAE approaches categorized into two branches based on using adversarial learning and learned similarity metrics. DMs are categorized into three branches based on their main formulations: DDPM, SDE, SGM.
• Figure 2: Schematic illustration of the Generative Adversarial Network (GAN) architecture, showcasing the interplay between the generator $(G)$, which synthesizes new data samples from latent space $(Z)$, and the discriminator (D), which distinguishes between real data $(X)$ and generated samples $(G(z))$.
• Figure 3: The type of directed graphical model under consideration. Solid lines denote the generative model $p_{\theta}(z)p_{\theta}(x|z)$, dashed lines denote variational approximation $q_{\phi}(z|x)$ to the intractable posterior $p_{\theta}(z|x)$. The variational parameters $\phi$ are learned jointly with the generative model parameter $\theta$.
• Figure 4: Overview of a hybrid network. Combination of VAE with GAN by collapsing decoder and generator into one.