This repository provides an unofficial PyTorch implementation of the Palette paper, which uses conditional diffusion models for image-to-image translation. The implementation includes U-Net architecture with attention mechanisms adapted from guided diffusion, encodes gamma conditioning, and supports distributed training (DDP) with EMA. The project covers three main tasks: image inpainting on CelebaHQ and Places2 datasets, uncropping, and colorization, with evaluation metrics including FID and IS scores.