Training GANs in low-data regimes remains a challenge, as overfitting often\nleads to memorization or training divergence. In this work, we introduce\nOne-Shot GAN that can learn to generate samples from a training set as little\nas one image or one video. We propose a two-branch discriminator, with content\nand layout branches designed to judge the internal content separately from the\nscene layout realism. This allows synthesis of visually plausible, novel\ncompositions of a scene, with varying content and layout, while preserving the\ncontext of the original sample. Compared to previous single-image GAN models,\nOne-Shot GAN achieves higher diversity and quality of synthesis. It is also not\nrestricted to the single image setting, successfully learning in the introduced\nsetting of a single video.\n