How about Ebsynth. about codef HOT 1 CLOSED

Ebsynth is indeed a commendable work based on propagation, and our work shares some similarities with it, but there are also significant differences: One key difference lies in the use of canonical images. In our method, the canonical image is learned during the process, which allows it to encapsulate more information from the sequence. In contrast, Ebsynth directly uses a frame from the video as the canonical image. Our method also shines in handling challenging cases such as water or smoke, where the motion cannot be accurately represented by optical flow. Thanks to the use of implicit representations, our method is capable of reconstructing these complex motions effectively.

While Ebsynth excels in scenarios involving rigid motion, our method offers a broader range of applications. I would suggest trying out both tools and deciding based on your specific requirements and the nature of the video sequences you're working with.

from codef.