Optimization of BatikGAN with Gradient Loss for Enhanced Batik Motif Generation

Batik, as an Indonesian cultural heritage, faces challenges in its preservation and development through modern technology. Previous Generative Adversarial Network (GAN) models, such as BatikGAN SL, have not been optimal in producing high-quality batik images. This study proposes various scenarios to optimize the BatikGAN model, focusing on improving the quality of the generated images. Experiments were conducted using the Batik Nitik Sarimbit 120 dataset sourced from Yogyakarta, Indonesia, which features 120 images from 60 different Nitik motifs. The addition of gradient loss resulted in images with finer details and sharper color gradations. This method showed the best results in producing new batik motifs of high quality. Optimizing BatikGAN with gradient loss proved effective in enhancing the quality of batik images, producing new motifs with sharp details and smooth color gradations, while preserving the authenticity of the batik motifs.