4EV: Adaptive Video Editing with Spatial Temporal Dynamics and Motion Pathways

This paper presents 4EV, a novel end-to-end video editing model extending Stable Diffusion and T2I (Text-to-Image) frameworks to generate and edit videos with fine-grained motion dynamics, including object movement, path navigation, background transitions, and zoom effects.4EV uses spatial-temporal attention for smooth transitions and consistent object appearances, with attention map injection for precise alignment. Contributions include: (1) Motion4EV, a custom dataset of videos and prompts supporting precise motion control, generated through experiments on various motion dynamics to ensure diversity; (2) an enhanced 4EV architecture with attention blocks for improved temporal and spatial refinement; and (3) experimental results demonstrating superior CLIP scores and motion consistency compared to state-of-the-art models. These innovations establish 4EV as a scalable framework for advanced video generation and editing. The source code is available at https://github.com/npp058/videogen.