Abstarct: Colorization of images refers to the prediction of color components in grayscale images. Variations in grayscale levels typically indicate the color of the image; thus, a grayscale image can be converted to its equivalent color image. Previous research has focused on colorizing grayscale images through object recognition in the scene, but these methods face the challenge of multimodality. Multimodality implies that each object can have multiple possible colors. In this research, two frxameworks are presented. The first proposed method introduces the hypothesis of selecting a harmonious color combination for the entire scene, ensuring that the colors are coordinated and aesthetically pleasing, resulting in a more natural outcome. To ensure color harmony, an improved version of a conditional generative adversarial network (GAN) called Pix2Pix is employed. Pix2Pix is a type of conditional GAN that maps an input image to an output image. The second proposed method is baxsed on selecting a harmonious color combination independently for each object in the image. While the first proposed method performs colorization by considering the color combination for the entire image, it does not colorize each object individually. To select a color combination for each object in the image, the improved Pix2Pix network, along with an attention mechanism, is used. The attention mechanism helps in selecting sharp variations and objects within the image. Additionally, the Lab color space is utilized due to its ability to distinguish the luminance component from the color component, and it can represent the widest range of colors. Two types of datasets, including compatible color elements and independent elements, were used. For each dataset category, 4000 images were used for training, and an additional 1000 images were used for the evaluation phase as a validation set. Our proposed method achieved significant results through quantitative and qualitative evaluations, producing final colorized images that are more natural and user friendly.