Will the 256 Rank LoRA come?
#4
by
rossaai
- opened
Are you thinking of distilling the model into a LoRA of 256?
In fact, I haven't explore the ability of lora in control, as openpose is the hardest model to train in Controlnet series, so i am not sure whether the lora can realize the same thing. Or maybe I need to do some network Innovation to make it more pratical.
Yeah! The OpenPose is the hardest because it doesn't have enough information such as, canny and depth almost the whole image is the control, also the open pose is quite unpredictable because the poses overlap themselves in some perspectives. Have you tried densepose?