a simple workflow to freeze specific lora layers.
avoid to merge lora again and again.
Github:
Description
Comments (5)
Please add more examples how to use this. Can I merge two face loras? What layers for what? Thanx
No, it only for one lora at once. You could use the node mutiple times and try to adjust each lora weight on each layers. I don't know waht layers for what. You need to findout by yourself
Instead of a "Load Lora' node, or those Lora Loaders you just keep adding Loras, this one 'loads it RAW, untouched, in order to be manupulated', then, later, you chain it to the loaded model as the default ComfyUI operation. I'll be learning about manipulating those layers (text/math formulas/tokens/ whatever), in order to improve some Loras. Tons of testing. Thanks!
So, is this process completely random? It feels like surgically opening someone's skull and poking random parts of the brain to see what happens. Is there any logical way to approach doing this?
It is not. But it is very similiar to what you said. Based on a lot of test, you could guess what happen when you disable certain layers.