In this research, we provide an architectural and performance analysis of advanced generative models for text-to-image & text-to-video generation, evaluating models for visual quality and computational efficiency. We analyse leading models—DALL·E, Stable Diffusion [6], and FLUX [2]—for image generation, and VideoGAN, Lumiere-3D [3], and MoCoGAN [4] for video generation [1]. This paper analyses diffusion paradigms that create coherent images through iterative denoising, as well as adversarial models that generate creative outputs through training generative adversarial networks. To conduct a quantitative analysis, we leverage key metrics including Fréchet Inception Distance (FID), Inception Score (IS), Structural Similarity Index Measure (SSIM), Peak Signal-toNoise Ratio (PSNR), Learned Perceptual Image Patch Similarity (LPIPS), and Video Multi-method Assessment Fusion (VMAF). In addition, this paper provides trade-offs for each model regarding inference time, memory consumption, training time, energy efficiency, and scalability. We identify challenges related to maintaining temporal coherence during video synthesis, emphasizing methods that utilize temporal loss functions, attention mechanisms, and hybrid models that leverage the pluses of both diffusion and adversarial architectures. The results show that models like DALL·E and Lumiere-3D are most effective for photorealism and scene detail, Stable Diffusion provides the greatest stylistic flexibility, and MoCoGAN provides the smoothest motion transitions.