Frame-Wise Guided Video Generation for Generative Inbetweening

Video inbetweening — generating intermediate frames between two given keyframes to produce smooth, temporally coherent video — is a longstanding problem in animation and video production, traditionally requiring skilled manual work or classical interpolation techniques that struggle with large motion or complex scene changes. This project proposes a frame-wise guided video generation approach for generative inbetweening, using deep generative models (e.g., diffusion-based video generators) conditioned on explicit frame-wise guidance signals to produce temporally coherent, high-quality intermediate frames between sparse keyframes.