Motion Style Transfer

Motion style transfer aims to change the stylistic characteristics of a motion sequence while preserving its underlying content, a crucial task in animation and gaming. Recent research focuses on improving the realism and controllability of transferred styles, employing diffusion models and transformer-based architectures to achieve fine-grained manipulation, including disentangling content and style, and even controlling style at the level of individual body parts. These advancements address limitations of previous GAN-based approaches, leading to more natural and expressive results with applications in virtual characters, interactive gaming, and data augmentation for action recognition.