when combine regular lora with lcm, the output is quite "lighter" than the output from lora with sd, any thoughts? do i need to finetune lora? or there is some parameters need to pay attention to?
I also encountered a similar situation. When I retrained LCM based on 48 million data points, I found significant differences in style changes between LCM+Lora and LDM+Lora.
when combine regular lora with lcm, the output is quite "lighter" than the output from lora with sd, any thoughts? do i need to finetune lora? or there is some parameters need to pay attention to?