Diffusion Models Beat GANs on Image Synthesis

ABSTRACT

We show that diffusion models can achieve image sample quality superior to the current state-of-the-art generative models. We achieve this on unconditional image synthesis by finding a better architecture through a series of ablations. For conditional image synthesis, we further improve sample quality with classifier guidance: using gradients from a classifier to guide a diffusion model during sampling. We achieve an FID of 2.77 on ImageNet 256x256, substantially outperforming the current state-of-the-art GANs (like BigGAN). This work proves that likelihood-based models can scale to high-fidelity, high-resolution image synthesis without the training instabilities associated with adversarial networks.