I've been asked several times how to do this, so I'm publishing workflow for use in daily practice.
✨In the first group we will load the basemodel and LoRa and increase their temperature, patch perturbed and self-attention (you can skip all or some of the steps after experimenting):
✨In the second set of nodes we do the sampling and upscaling using a model (get one from: https://openmodeldb.info/models/4x-Nomos8kDAT for example):
✨In the third (the most important for this workflow) we do the refinement itself, first the whole body (person), then the face and finally the hands:
👉As you can see - the models were able to correctly identify all the parts we wanted to refine ...
🤞Good Luck Generating Nice Images ;)