A simple Wan 2.2 I2V for ComfyUI that handles the steps correctly for the High and Low noise models. Wan 2.2 was trained to switch over from high to low at very specific points during generation that impacts the output. This workflow handles the correct timing.
The workflow requires you to manually install this node script: https://github.com/stduhpf/ComfyUI-WanMoeKSampler
The author of WanMoeKSampler has the wrong installation instructions. The correct command is
git clone https://github.com/stduhpf/ComfyUI-WanMoeKSamplerYou may need to install other models for this workflow.
This ComfyUI workflow works as of 8/24/2025
Description
FAQ
Comments (8)
TIL about those little alt-click reroute thingies. pretty sweet! (hold alt and left click on a noodle, it adds a point that you can move around to divert the noodle around things. shift click one of them and add another noodle linking to something else)
Can you provide the links for "WAN2.2-LowNoise_ImgVidv5_130edim64a64_14e4-000125.safetensors" and "WAN2.2-HighNoise_ImgVidv31_110edim64a64_1pt4e-4-000105.safetensors"? Thanks
@fborghi248 Sorry about that confusion. Those are models I'm training. It was not meant to be in the workflow. edit: fixed it right now.
Wan2.2 Pussy model has disappeared. Can you share the following files elsewhere
@fengyun321 It's still here. I think the site is having stability issues: https://civitai.com/models/1895314
I followed the instructions to a T, go this error: No module named 'sageattention'
TensorArt error:
Workflow does not contain a valid version. Zod error: Validation error: Required at "version"
worked for me!! needs ComfyUI+Sageattention