Diffusion models have significantly advanced the fields of image, audio, andvideo generation, but they depend on an iterative sampling process that causesslow generation. To overcome this limitation, we propose consistency models, anew family of models that generate high quality samples by directly mappingnoise to data. They support fast one-step generation by design, while stillallowing multistep sampling to trade compute for sample quality. They alsosupport zero-shot data editing, such as image inpainting, colorization, andsuper-resolution, without requiring explicit training on these tasks.Consistency models can be trained either by distilling pre-trained diffusionmodels, or as standalone generative models altogether. Through extensiveexperiments, we demonstrate that they outperform existing distillationtechniques for diffusion models in one- and few-step sampling, achieving thenew state-of-the-art FID of 3.55 on CIFAR-10 and 6.20 on ImageNet 64x64 forone-step generation. When trained in isolation, consistency models become a newfamily of generative models that can outperform existing one-step,non-adversarial generative models on standard benchmarks such as CIFAR-10,ImageNet 64x64 and LSUN 256x256.