Ulos is a traditional woven textile of the Batak people. Ulos are difficult to design as skilled laborers must ensure that colors are aligned the right way. Ulos weavers usually select colors to match the design that has already been created. While AI-driven deep learning technologies have been able to automate the production of textiles, making sure that the original pattern structure is retained while utilizing colors consistent with the local culture remains challenging. In this paper, we developed UlosColor-GAN, a Conditional Generative Adversarial Network (cGAN) that would color Ulos patterns. The model implements a U-Net generator with skip connections to keep important details in the patterns, and a PatchGAN discriminator to mimic the texture. The model is trained using 153 images, which come from three major types of Ulos: Harungguan, Puca, and Sadum. The result indicated that the model is able to keep the pattern structure very well with the high Structural Similarity Index Measure (SSIM) of 0.9929. The average colorfulness score is 152.4561, which means that the model can produce strong traditional colors. When we put together the resulting designs and compared them with the original hand-colored designs, most of them looked pretty similar. This research is a solution to standardize the design process. It contributes to traditional weaving communities and enables digital heritage conservation.