How It Started vs How It's Going: GANs to Diffusion Models

Past few months have been trailblazing for Generative AI, but itโs also fascinating how far we have come along with LLMs and diffusion models. Above is my very basic CNN-based generative adversarial network (GAN) from 2019 trying to draw a house after 1200 epochs compared to a pre-trained Stable Diffusion model from SageMaker JumpStart with no fine tuning and default inference parameters.