MoCoDiff: A Controllable Autoregressive Diffusion Model for Expressive Motion Generation

Single content multiple styles ("walk"+Crouched style) ("walk"+Zombie style) ("walk"+Ninja style) Multiple content single style ("walk"+Airplane style)("jump"+Airplane style)("run"+Airplane style) Multiple content multiple styles ("walk"+Airplane style)("roll over"+No style) ("go upstairs"+Zombie style) ("go downstairs"+No style) ("sit"+Crouched style) Transition formulation actively suppresses drift, enforces smooth transitions across motion segments, and significantly improves temporal coherence over extended sequences. Experiments show that MoCoDiff achieves state-of-theart style fidelity, transition quality, and efficiency, while supporting flexible and interpretable multi-condition motion synthesis without retraining. The source code and model are available at https://github.com/ Xuehan0530/MoCoDiff-code.