follow this link for v2 guide :
https://www.reddit.com/r/StableDiffusion/comments/1rafyfb/nice_sampler_for_flux2klein/
I have been testing non-stop to find the layers that actually allows for changes but preserve the original details those layers I pasted below are the crucial one for that, and main one is sb2 the lower it's scale the more preservation happens , enjoy!!
custom node with more in depth information can be found here:
https://github.com/shootthesound/comfyUI-Realtime-Lora
full discussion and Q&A can be found on this thread:
here
after more and more testing I found a great combo with this node
the sampler is already a built in sampler inside of comfyui, and the custom sigma can be found in this package :
https://github.com/capitan01R/ComfyUI-CapitanFlowMatch
coming soon v2 (more powerful and precise) not yet available.. same node but with control over each tensor for more surgical control; eg 201 tensors here: