CogVideo icon indicating copy to clipboard operation
CogVideo copied to clipboard

Published 20 hours ago • verified publisher icon THUDM

Simple question about CogVideoX-I2V model

Open KIMGEONUNG opened this issue 7 months ago • 5 comments
trafficstars

Hello, Thank you for the invaluable work. I have a simple question. Is CogvideoX-I2V model finetuned from CogvideoX-T2V model or is it trained from scratch?

KIMGEONUNG avatar Mar 27 '25 10:03 KIMGEONUNG

Hi, accroding to their paper, I think it is finetuned #751

OrangeSodahub avatar Mar 27 '25 10:03 OrangeSodahub

Yes, the I2V model is trained by doubling the channels of the T2V extension.

zRzRzRzRzRzRzR avatar Mar 29 '25 02:03 zRzRzRzRzRzRzR

@zRzRzRzRzRzRzR How you place the pretrained T2V parameters? The first 16 channels?

OrangeSodahub avatar Mar 29 '25 08:03 OrangeSodahub

Yes, then expanded to 16 channels.

zRzRzRzRzRzRzR avatar Mar 30 '25 03:03 zRzRzRzRzRzRzR

@zRzRzRzRzRzRzR What do you mean then expanded to 16 channels?......

OrangeSodahub avatar Apr 17 '25 10:04 OrangeSodahub