In this paper, we propose a novel framework for transferring motions that reflect species-specific behavioral habits of animals to other species. Existing motion transfer methods mainly focus on human motions, focusing on skeletal alignment or style consistency, but overlook the preservation of unique behavioral habits of animals. To address this, we introduce a habit preservation module that includes species-specific habit encoders, and present a generative framework that learns motion priors that capture unique habitual features. In addition, we integrate a large-scale language model (LLM) to facilitate motion transfer to previously unseen species. We introduce a novel quadruped animal skeleton dataset, DeformingThings4D-skl, and verify the superiority of the proposed model through extensive experiments and quantitative analyses.
