Abstract: Recent diffusion models have made significant advancements in generating lifelike videos from driving signals, including a reference character and a skeleton sequence. Nevertheless, these ...