Malignant melanoma (MM) remains a leading cause of skin cancer mortality despite accounting for only 1% of skin cancers. Early detection and accurate classification are essential to improve outcomes. However, the necessity of relying on big datasets to obtain clinical-ready performing deep-learning models makes their training difficult. Generative Adversarial Networks (GANs) offer a promising solution by creating high-quality synthetic data for augmentation. This preliminary study focuses on developing a GAN-based framework for MM body lesion images, explicitly excluding areas like the face, palms, and soles due to their unique dermoscopic patterns. Utilising a StyleGAN3-t architecture with adaptive discriminator augmentation, the model generated synthetic images at a resolution of 512 \(\,\times \,\) 512 pixels, achieving a Fréchet Inception Distance (FID) score of 31.73 after 1,740 iterations. These results highlight the model’s ability to produce diverse, high-quality images comparable to real-world data. Further research will investigate different GAN models, improved metrics, and subjective validation via physician assessments. This development has the potential to minimize overdiagnosis and enhance clinical outcomes in melanoma treatment significantly.