Reangle-A-Video is an integrated framework for generating synchronized multiview videos from a single input video. Unlike mainstream approaches that train multiview video diffusion models on large-scale 4D datasets, our method reframes the multiview video generation task as a video-to-video transformation by leveraging publicly available image and video diffusion priors. Reangle-A-Video operates in two steps. First, it synchronously fine-tunes an image-to-video diffusion transformer in a self-supervised manner to extract view-invariant motion from a set of distorted videos. Second, it warps and fills the first frame of the input video with different camera viewpoints using DUSt3R, following inferred temporal cross-view consistency guidelines, to generate a multiview-consistent starting image. Extensive experiments on static view transfer and dynamic camera control demonstrate that Reangle-A-Video outperforms existing methods, offering a novel solution for multiview video generation. Code and data will be made public.
