This is a pruned version of my current workflow i've been testing, people keep asking me to share it, so here it is. I tried to put up those notes around to give some instructions.
Not beginner friendly.
Download links for detection/cnet model in one of the notes.
Latent version: second Ksampler does latent upscaling. So if you use a model that can't handle it, use the pixel version instead.
Pixel version: second Ksampler does pixelspace upscaling.
(NOTE: Looking at the version i have of the pixel version, the second sampler has cfg 5.0 normally)