Generated results are distorted

[zzk88862]

Hi, great work
I found that the final result can only be in 480x720 resolution. When I input an 800x800 image, it causes significant distortion in my results.
input:

output:
https://github.com/user-attachments/assets/402877e3-1155-4400-9adb-41b1a859b91f

Are there any solutions to this?

tks

[wenqsun]

Thanks for your interest! I guess that the distortion may be caused by the crop and resize operation on the input image during the inference stage, and you can try to crop the image to a suitable height/width ratio to avoid severe distortion.

[zzk88862]

Thanks for your reply.
Do you mean that I must crop the image to 480x720 or maintain this aspect ratio?

btw
Now, CogVideoX-Fun supports multi-resolution, which will avoid distortion and maintain the original resolution.
Will you consider training these LoRAs for CogVideoX-Fun? I believe it will significantly improve performance.

Thanks.