Setting the LoRA weight to 1.5 seems to work better.
I was originally planning to delete it, but after giving it a quick test, I found it was actually pretty good. This is an earlier version.
KR2_3D_000000750.safetensors