As oppose to visible-light images, thermal images can be captured even under low illumination. However, thermal facial images are difficult to recognize by human viewers due to the large modality gap between the visible and the thermal domains and the lack of details in the thermal domain. In this project we translate a thermal facial image into a visible-light facial image while maintaining identity preserving features and creating a natural-looking visible-light image. For this aim, we examined several methods. Finally, we used a generative adversarial network (GAN) that uses details from the thermal domain to generate a visible-light image. We used several thermal-visual facial databases, created for similar tasks, and preprocessed them in order to create a single unified database. In addition, we created a small database using FLIR ONE camera. Following an extensive training on the unified database, we performed transfer learning to the FLIR ONE database, and created an algorithm to translate a thermal facial image into a counterpart image in the visible domain.