ReCapture
Generative video camera control from user videos
CommonProductVideoVideo GenerationCamera Control
ReCapture is a method for generating new videos and unique camera trajectories from a single user-provided video. This technology allows us to regenerate the source video from completely different angles with cinematic-level camera movements. ReCapture generates noisy anchor videos with new camera trajectories using multi-view diffusion models or depth-based point cloud rendering. It then refines the anchor video through our proposed masking video fine-tuning technique to produce clean and temporally consistent re-angled videos. The significance of this technology lies in its ability to leverage the strong priors of video models to regenerate visually pleasing and temporally coherent videos from approximate inputs.