This study aims to generate high-resolution person images from hand-drawn stick figures. In similar studies, the face orientation of the generated person could not be determined because the heads of stick figures were filled with black, resulting in only front-facing images. We propose a method that generates images of person with various face orientations. To control face orientation, a cross-shaped curve is drawn on the heads of the stick figures. Two datasets were constructed using different types of stick figures to assess the impact of head representation. One dataset contained pairs of stick figures with a cross-shaped curve on the head and their corresponding person images, while the other had stick figures with black-filled heads instead. We assessed the performance of two models, each trained on a different dataset, using stick figures of the same type as those in their respective training datasets.