K-pop :D

If you like it, consider buying me a coffee, I'm poor.

A few details:

It was trained on Stable Diffusion 1.5 as a base model, so it might work on most checkpoints without many alterations, my favorite checkpoint for this was Real Dream Legacy SD1.5, RealisticDigital produces some good results too.

Generation parameters:

Clip Skip: 2
Best weight: 0.95

Sampler: DPM++ 2M Automatic/Karras, 35 steps

Adetailer: mediapipe_face_full, mediapipe_face_mesh_eyes_only, hand_yolov, all with detection at 0.85

HiResfix: 2x, denoise 0.25, steps 15, using 4x_NMKD-Siax_200k as upscaler.

img2img (optional but it'll wield the best results)

Just rescale,

Ultimate SD upscale: 2x, using 4x NMKD as well, no types.

I realized now that it'd be better to share the entire process of what do I do to achieve my results.