This repository provides training and testing code for Cold-Diffusion models applied to various image degradations including Gaussian blur, masking, resolution downsampling, snow effects, and color desaturation. The implementation extends DDPM (Denoising Diffusion Probabilistic Models) by removing the noise corruption step, allowing direct inversion of image transforms. Code builds on the lucidrains denoising-diffusion-pytorch repository and includes pretrained models for CelebA and AFHQ datasets.