Why artifact in lora fine-tune i2v model?

Open Madaoer opened this issue 11 months ago • 1 comments

I finetuned the CogVideo i2v 5b model using lora, and after training for a while, the generated videos have a white line in the middle.

https://github.com/user-attachments/assets/02e68e12-d08d-4d6a-8976-2311cf16d629

Feb 19 '25 03:02 Madaoer

We are currently unable to reproduce this problem. Are all your videos like this?

Feb 20 '25 05:02 zRzRzRzRzRzRzR