CogvideoX With ControlNet Pose For AI Animation

CogvideoX With ControlNet Pose For AI Animation - No More Boring Camera Scenes

Рет қаралды 8,733

Күн бұрын

Пікірлер: 34

@TheFutureThinker 3 ай бұрын

Workflow available in Custom Node examples/ folder Demo Used In This Tutorial www.patreon.com/posts/113667767 CogVideoX 5B In ComfyUI Better Quality - Local AI Video Model That Truly Works kzbin.info/www/bejne/rGqueISlltabodksi=GYd8qGsJXz2fmi4m CogVideoX 5B AI Video Model Updated With Img2Vid In ComfyUI kzbin.info/www/bejne/qaqZiXt8Z7F4nKssi=0waT9GIp7qAeNkhS ComfyUI-CogVideoXWrapper github.com/kijai/ComfyUI-CogVideoXWrapper Alibaba-pai CogVideoX huggingface.co/alibaba-pai

@Geffers58 3 ай бұрын

Your videos are very helpful. Is it posible to use CogVideo5b with video to video, but using a starting image rather than prompt? I had a try but I couldn't get it to accept the workflow. So like your fireworks, say, but your new fireworks were a matching firework image. Or more practically, perhaps source video is of three people doing stuff, your source image is 3 different people in roughly the same pose. It's pretty much the same I suppose as using a pose method against the source image.

@kalakala4803 3 ай бұрын

Great news! We are getting there. And really don't know why Gen 3 , Kling AI, they don't provide ControlNet. For this.

@TheFutureThinker 3 ай бұрын

Well, as we said before. Some gambling need in order to charge you.

@MilesBellas 3 ай бұрын

Interesting. There must be a way to create great videos from this that Benji laughed at with the 3 legged Wooly Mammoth! 😅

@TheFutureThinker 3 ай бұрын

So there will be 6 legs , 2 person on the screen

@TheEconoVision 3 ай бұрын

what is the difference of vvid2vid and pose2vid? both are using reference video

@TheFutureThinker 3 ай бұрын

Both are similar capture the motion. Vid2vid are more like a Lineart CN only

@Michael_H_Nielsen 3 ай бұрын

this is amazing. very many thinks. sugrestion for feedback. I advice you to make the worfklow look really simple to overview.

@TheFutureThinker 3 ай бұрын

Great suggestion! Cheers

@Michael_H_Nielsen 3 ай бұрын

this is magic

@MilesBellas 3 ай бұрын

The output video could be upscaled ?

@TheFutureThinker 3 ай бұрын

Yes it could

@wereldeconomie1233 3 ай бұрын

What is the min. vram for GGUF model?

@TheFutureThinker 3 ай бұрын

About 16 GB can be done.

@zinheader6882 3 ай бұрын

so this is like v2v animatediff?

@TheFutureThinker 3 ай бұрын

Yup yup👍

@crazyleafdesignweb 3 ай бұрын

The action scene is the most challenging for video. Even in real life video, its hard to take a good shot.

@TheFutureThinker 3 ай бұрын

Yes it is. That's why we have to breakthrough

@SeanieinLombok 3 ай бұрын

nice

@TheFutureThinker 3 ай бұрын

Almost there, this model have the potential

@MilesBellas 3 ай бұрын

COG LORA = 4 legged wooly mammoth!?😊

@TheFutureThinker 3 ай бұрын

Haha😂😂😂

@aivideos322 3 ай бұрын

Use depth anything v2 its soooooo much better than v1

@TheFutureThinker 3 ай бұрын

Yes v2 use for image gen. But for Cog, even you give it too detail it can't generate detail...

@aivideos322 3 ай бұрын

@@TheFutureThinker nice, I just downloaded cog to try, seems to copy many settings from animate diff. I like how they are working locally, as I refuse to use online tools. Hopefully they continue to update

@TheFutureThinker 3 ай бұрын

@@aivideos322 Yes, I really like the ControlPose idea from this version. As I shared previously, I don't like the gambling flow that I experienced with all existing AI Video generator out there. Gen 3 , Kling, Luma, and even MiniMax , yes , they have good quality with high parameters and cloud GPUs. But no control, that is a big draw back. The node setting concept , it does come a lot from AnimateDiff. Because KJ did a lot with AnimateDiff and other extend features of it. So I think when he create the code for this wrapper, he put a lot of that concept similar to AD. That is a good sign, as we play around with AD for a while.