MotionS introduces a novel approach to motion stylization by embedding cross-modality style prompts and cross-structure skeleton motions into a canonical space. The pipeline utilizes techniques like cross-modality style embedding and topology-encoded tokens to achieve flexible and generalizable motion stylization. Extensive experiments demonstrate the effectiveness of MotionS in preserving content, fidelity to style prompts, and generating diverse results. Comparisons with baselines highlight the superior performance of MotionS in handling various skeleton structures and multi-modality cues for animation creation.
翻譯成其他語言
從原文內容
arxiv.org
深入探究