Differences between same diffusion models from comfy-org repo
#38
by
AbsAnon
- opened
Hello Kijai, hello everyone. I've a question I asked myself for some time now. What exactly is the difference between, for example, the wan 2.2 i2v diffusion model found here with the _KJ at the end and the one found at the comfy-org repository?
Is it the same or is it different? Perhaps the KJ versions are made to make them compatible with the Kijai components in comfyui?
I would love to know.
These are compatible with both native ComfyUI workflows and the WanVideoWrapper workflows, however in the wrapper some layers are kept in the original precision which in some cases leads to better results.
The quantization method is slightly different as well, it's been somewhat experimental, that's why these are in separate repo.