I got worse result than the previous 3 k-samplers solution
#5
by lvsemi - opened
After replacing the previous 3-pass k-samplers setup (3 steps high without lightx2v + 2 steps high with lightx2v + 2 steps low with lightx2v) with the current model (2 steps high + 2 steps low), I can feel a significant decline in creativity. Increasing the high noise steps barely made a difference.
Another problem is using this model alongside other LORA significantly kills other LORA's motion.