Seeking Assistance with Artifact Issues in Video Extension using Mora's Method

[encounter1997]

Dear Authors,

We have recently been working on extending videos using the Mora framework. Specifically, we have been following the techniques described in Mora's paper and implementing the stable-video-diffusion-img2vid-xt-1-1 (SVD) model. Our approach involves using the last frame of a preceding video segment as the input for the subsequent segment, leveraging the SVD model for video extension. However, we encountered significant artifacts in the generated videos. These issues seem to worsen as the number of video extension iterations increases, leading to a cumulative accumulation of errors and more pronounced artifacts. An example of the generated video is shown below.

https://github.com/lichao-sun/Mora/assets/30790888/cbad2760-eed0-4ba7-aa1e-d3b5702b7f9b

We are wondering if you have employed any additional techniques to address the aforementioned artifact problems.

Thank you very much for your time and consideration. We look forward to your valuable guidance and assistance.