Spent the whole night trying to get characters into fixed poses, and finally figured out how these two work together. ControlNet handles the skeleton/pose constraints—OpenPose locks the movement in place; LoRA takes care of style and character traits. They don’t actually conflict, you stack them.
I was stuck wondering why my LoRA-only outputs had poses all over the place—turns out I wasn’t feeding it a skeleton map. And if you’re aiming for photo-real quality, those two alone won’t cut it; the sampler and high-res fix step also make a huge difference.
Wanted to ask you guys—when you’re doing fixed poses, do you usually set up the OpenPose skeleton yourself or just rip it from existing reference images? I’ve been extracting from images, and the fingers always end up messed up.