Image colorization provides significant data handling advantages by completely eliminating the need for labeling. This research introduces advancements in the process of converting grayscale images into color using modern machine learning techniques. By utilizing the Lab color space, where luminance (L) is processed separately from the color channels (ab), this research focuses on refining and developing the Dif-EDUNet model (a Diffusion model using ED-UNet) to address the challenges in image colorization. Our experiments, including training with datasets: Coco-Stuff, DIV2K, Places365, ImageNet and CelebA show that our results are very encouraging compared to the benchmark of the dataset in this issue. Additionally, the Weights & Biases (wandb) tool is employed to support the monitoring of the training process.