MotionClone is a training-free framework that allows for motion cloning from reference videos to control text-to-video generation. It utilizes a temporal attention mechanism to represent the motion in a reference video during video inversion and introduces a主 temporal attention guidance to alleviate the influence of noise or very subtle motions in the attention weights. Additionally, to assist the generation model in synthesizing reasonable spatial relationships and enhancing its prompt-following ability, a position-aware semantic guidance mechanism is proposed, leveraging the foreground rough locations from the reference video and original classifier-free guidance features.