The requirement to embed watermarks in synthetic data transforms the immensely hard task of deepfake detection into a simple extraction of a watermark. In fact, the unauthorized use of synthetic samples is difficult when watermarks are embedded in all AI-generated data. The goal of our study is to generate realistic fingerprints that include transparent watermarks by combining the traditional watermarking with Generative Adversarial Networks (GAN). We embed a watermark into all training images, train GAN models, and study under which training hyperparameters the watermark is transferred from training to generated samples. Watermarks are embedded by a hybrid algorithm based on discrete cosine transformation, discrete wavelet transformation, and singular value decomposition. The pix2pix network is utilized to reconstruct realistic fingerprints from minutiae. The feasibility of the watermark transfer is demonstrated by assessing the imperceptibility of watermarks and their robustness to the GAN training as well as by assessing the realism of generated fingerprints and proving their identities.