Synthetic Data Augmentation for Controllable Human Video Generation

A research paper on arXiv (2604.21291) explores the use of synthetic data augmentation to address the scarcity of large-scale, diverse, and privacy-safe human video datasets for controllable human video generation. The work proposes a diffusion-based framework enabling fine-grained control over appearance and motion, serving as a testbed to analyze how synthetic data interacts with real-world data during training. Experiments reveal complementary roles of synthetic and real data, aiming to bridge the Sim2Real gap. The research targets applications in digital humans, animation, and embodied AI.