VMC: Video Motion Customization Using Temporal Attention Adaption for Text-to-Video Diffusion Models

Figure 1. Using only a single video portraying any type of motion, our Video Motion Customization framework allows for generating a wide variety of videos characterized by the same motion but in entirely distinct contexts and better spatial/temporal resolution. 8-frame input videos are translated to 29-frame videos in different contexts while closely following the target motion. The visualized frames for the first video are at indexes 1, 9, and 17. A comprehensive view of these motions in the form of videos can be explored at our project page.