Split into 3 groups, the first is for image generation and straight forward.

The second group is for upscaling, it first goes through a colour correction, then Iterative Upscale of x 1.5 and then another colour correction as I find that tends to wash out the colours a bit otherwise. Feel free to change both colour correctors to suit your preference.

This then goes to the third group which has a face image loader for FaceID and a face detailer, there's also an additional prompt box called Eye Colour Correction as the eyes have quite often turned brown at this point. Use the box to prompt what colour you want, can also be used to add/alter other facial features. It then goes back to the Upscale group and runs through a x 2 Ultimate SD Upscale. Haven't run into memory issues on my 2080Ti with 11GB and going from a 896 x 1152 initial generation tp a 2688 x 3456 final output and it does a decent job of retaining details.

This is my first big workflow, I've learned through watching YouTube videos and reverse engineering other workflows, so it might not be the most efficient or elegant of flows, but it works well.