This repository provides a research implementation of Differentiable Augmentation (DiffAugment) for training Generative Adversarial Networks more efficiently with limited data. It includes PyTorch and TensorFlow implementations supporting StyleGAN2 and BigGAN architectures. The technique applies differentiable augmentation during training to stabilize GAN learning when only small datasets are available, enabling high-quality image generation from as few as 100 training examples without pre-training.