Generative Adversarial Networks(GAN) are widely used in graphics conversion problems. This project aims to explore the overall performance of different GAN models when translating photos of objects into black-white sketches. We proposed con-CycleGAN, which conditions a CycleGAN with the categorical label by introducing label projection as input of residual block. Experiments are performed on pix2pix, CycleGAN, and con-CycleGAN, with Sketchy Database. By comparing the performance of pix2pix and CycleGANs, we found that the effect of loss function which accounts for reconstruction error is significant in dealing with sketch synthesizing problems based on Sketchy Database. On the other hand, by comparing the performance of CycleGAN and con-CycleGAN, our experiment indicated that the label-conditioned utility of con-CycleGAN has not achieved an expected improvement in performance. But in general, CycleGAN and con-CycleGAN are both able to generate identifiable sketches based on images with clearly distinguishable chromatic contour. However, when dealing with images with unclear chromatic contours, the performance is a bit unsatisfactory. The project will analyze all the phenomena that appear in the experiments and propose possible solutions for future work.