Home Creators Posts Import Register Favorites Logout

Downloads

Content

Video - https://youtu.be/1CP5twTDtxw

Overview

Wan 2.2 Fun Control 14B gives us even more control as it supports things like depth, pose or canny ControlNet videos.

Even better still, from my point of view, is that you no longer have to pick between a Text2Video or Image2Video model for whatever it is you want to do, as this model supports both!

I've got so many models that it's hard to keep everything on my 2TB drive, but seeing as WAN Fun seems to do everything it may be an opportunity to free up some disk space and get rid of the earlier WAN 2.2 models ;)

This workflow has everything in one place, so you only need to enable the particular option(s) you want. Do be aware that if you want to run everything at once, you'll need a decent amount of RAM (128GB+ suggested).

VRAM-wise 24GB remains a decent starting point, but with block swap etc I'm sure it will work on less, just a bit slower!

Example workflow screenshot:

Personally, I only pick one option at a time, but the various choices available by default are:

Text to video

Video to video

Image to video (first, last, others)

Control video to video (depth default)

Control video to video (pose default)

Control video to video + flf to video (depth default)

Control video to video with reference image

Of course you can pop in whatever control video you like, or whatever pre-processor you fancy. I tested depth, pose and canny with others giving sub-optimal results (such as normals, etc)

There are plenty more combinations you can play with, so do let me know of any fun ones I've missed, but I think I've covered the major ones!

Looking at Kijai's 5B examples, swapping in the WAN Fun 5B model loaders should be possible too (as they look very similar), though I've not had the time to test as yet.

Workflow Info

All the model files you can grab from Kijai's Huggingface pages -

https://huggingface.co/Kijai/WanVideo_comfy_fp8_scaled

https://huggingface.co/Kijai/WanVideo_comfy/tree/main

Custom nodes I've tried to keep to a minimum, and you can install any custom nodes using ComfyUI Manager.

Main nodes:

ComfyUI Wan Video Wrapper - https://github.com/kijai/ComfyUI-WanVideoWrapper

KJ Nodes - https://github.com/kijai/ComfyUI-KJNodes

Video Helper Suite - https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite

RGthree - https://github.com/rgthree/rgthree-comfy

For Control Net stuff you can using things like:

Kijai's Depth Anything v2 - https://github.com/kijai/ComfyUI-DepthAnythingV2

And for more options (pose, canny, etc) -

ControlNet aux - https://github.com/Fannovel16/comfyui_controlnet_aux

Have fun! :)

Files

Comments

Danny

I was able to run your workflow for fun control which is the one I needed, however in many cases your workflow provide poor results because you are using depth anything for control, this make that not only the body of the person is being used as reference control but also their clothes are being used to generate the image, have you ever tried other models for pose estimation like YOLO-pose? Kijai is using vit pose wholebody which is something I didn't tried before but I'd like to know your opinion

Kirk Dudley

Is there a way to execute this workflow without having Sage Attention or Triton installed on my local machine? When I ran this workflow with the Text to Video option, it made it to the WanVideo Model Loader step and then stopped: "Can't import SageAttention: No module named 'sageattention'". I was able to get around this problem by selecting a different attention mode. However, it then stopped at the first WanVideoSampler model: "Cannot find a working triton installation". I can't see how to get around this stop. I assume I don't have either Sage Attention or Triton installed on my machine, so I looked into how to install them and it looks like a lot of work and might require reinstalling comfyUI which I'd rather not do.

М М

I cant make it work with the gguf version it gives me errors RuntimeError: Input type (unsigned char) and bias type (struct c10::Half) should be the same do you know how i can fix this