This is my simplified workflow that I use with Tower13Studios amazing embeddings and models. It is also compatible with CivitAI automatic metadata population.
I used to run ComfyUI on CPU only as I did not have an nVidia graphics card.
Thus I have used many time and memory saving extensions like tiled (en/de)coders and kSamplers.
Another important extension was Preview Chooser, which meant I didn't waste time with images I did not like.
Now I have a 3060TI, not the best and only 12G VRAM but still love the time and memory saving extensions.
I also wanted a way to upload my images to CivitAI without worrying about not having attribution to the great models and resources that I have used. (Thank you for all your work, buy them a Kofi)
I have had many iterations of this workflow and I have stripped it back to the bare basics.
Update: V2 of the workflow thanks to @so_simple's suggestion below. I have also rearranged the face detailer to come after the upscale, I think this produces a better result (I was having some issues with certain sampler/scheduler combinations). Attached as v2 of the workflow.