The paper proposes a new neural network architecture based on conditional generative adversarial networks for font style generation. This neural network can extract information about what character is expected to be output, what global and local stylistic features the complementary typeface has, and what additional non-graphic information about the typeface (stylistic descriptors) is available. In our approach, we leverage the PANOSE-1 typeface classification code, a widely used standard for describing the visual characteristics of typefaces, as a source of stylistic descriptors. This code is embedded in the metadata of most fonts, and a method for processing it and integrating it into the extracted stylistic properties was implemented. The quality of font style transfer conducted via the proposed neural network was evaluated using seven metrics, namely FID, mFID, SSIM, PSNR, ERGAS, RASE, and RMSE-SW. A comparison with FTransGAN showed that the model proposed in this study achieves higher scores for all of the aforementioned metrics except for mFID.