CogVideoX-5B: Top AI Video Model on ComfyUI? Features, Parameters, Video Repair, Long Video Creation

Recently, Tsinghua University released the CogVideoX-5B series, a large video model that claims to be the best open-source "AI video generation" solution. After testing it myself, it seems the claim is true. Both text-to-video and image-to-video results are impressive. On top of that, Alibaba has launched the CogVideoX-Fun series based on CogVideoX, maintaining similar video quality while unlocking resolution limits and adding the ability to generate videos from keyframes.

In this video, I demonstrate videos I created using the CogVideoX series models. I introduce the features and functions of the models, showcase how to install the plugin on ComfyUI, and explain the details of image-to-video and text-to-video generation. Additionally, I share tips on using AnimateDiff to optimize and repair videos, including a trick for using the last frame of one video as the first frame of the next to create longer, more cohesive videos. I also compare the Fun series models and their effects, and conclude by summarizing the pros and cons of CogVideoX.

[cvactika]

This channel is one of the few on YouTube that contain only useful information without any BS! I signed up! ❤

[Radarhacke]

Nice! But why is the t_tile length and overlap setting is missing in my CogVideo Sampler? And why take it about one hour to generate one Video on my 16 GB 4070 ti Super?

[zonghengbaihezongheng]

Whether or not to do an integrated package of text-born video, graphic-born video and video-born video, big benefits？

[yovanvujacic]

Fantastic video!!! any resources on how to prompt ? Its difficult sometimes to get the right motion

[zuzuumaibam]

where to get the 1xDeJPG upscale model?

[blofu]

Very good tutorial, thank you! For some reason there are no .safetensors in the loadCLIP node for me? How do i go about getting/downloading these?