Back to All Resources

Dreamix: Video Diffusion Models are General Video Editors

This paper presents the first diffusion-based method capable of performing text-based motion and appearance editing of general videos. By combining low-resolution spatio-temporal information from the original video with synthesized high-resolution information, the method significantly boosts fidelity. The authors introduce a new framework for image animation and demonstrate extensive qualitative and numerical experiments showcasing the remarkable editing ability of their method.