Wan2.1-Fun-14B-InP Controlnet support Native workflow - Native wan workflow with controlnet support
Description
Wan2.1-Fun-14B-InP Controlnet support Native workflow - Native wan workflow with controlnet support
FAQ
Comments (10)
Not bad, I added teacatch and sageattention to it, the generation time from 15 minutes went down to 7min 33sec. , although I am not sure if it doesn't need maybe a bit more stuff, or I need a bit different settings, because what it does, it replicates the clothes very well, even the background is decen, and the movements are perfectly grabbed from the video, but merges the faces together. I tried it by not changing anything in the workflow at first, that's why I know it takes 15 minutes, and than with and without lora, it still merges the faces. I mean I guess I could do like post face swap, but in theory I think if I tweak it with some extra stuff I could get the actual face. idk I am not super informed on these things. Oh also video card 5090 , so whoever reads this and has better results with either 5090 or other cards lemme know LOL
I kept it very simple, haven't added anything extra to make it less complex. Yes, you can add teacache, cfg zero star to make the output better and faster. For the image to video , the trick is to provide the reference image keeping the same pose as the first frame of the controlent video. So whatever reference image you are using tweak it a little bit with help of the first frame of openpose video or any other controlnet video whichever you are using. Then it would generate the video with exact same person, otherwise it just mimic the reference image for first few frames and then start following the controlnet video.
I got this error when run workflow with Wan3.1-14B-inp-Q5 gguf
KSampler
Given groups=1, weight of size [5120, 36, 1, 2, 2], expected input[1, 52, 11, 80, 52] to have 36 channels, but got 52 channels instead
Same issue. I guess it's not yet completely compatible with gguf models.
Use the control model, you can try this one https://huggingface.co/alibaba-pai/Wan2.1-Fun-1.3B-Control/blob/main/diffusion_pytorch_model.safetensors
@robinhud5738936 This is the 1.3B parameter model. We're searching for a 14B gguf model.
Two nodes are missing even after updating comfyui - CFG zeroStar and WanfuncontroltoVideo. Did you solve it someone, please? ^_^
Install from manager>install missing nodes> then select all missing nodes and install
@robinhud5738936 Don't exist
An error :bucket name len is between [3-63],now is 0" occured, how to solve it ?