For best outputs and usage:

Sampler: Euler a.

Steps: around 30, cfg 6 ~ 9, CLIP skip 2.

VAE: none required.

Pos prompt: always begin your pos with "(score_9, score_8_up, score_7_up)".

Neg: "(score_4, score_3, score_2, score_1), poor quality, low quality, bad quality, ugly, bad feet, bad hands, unaestheticXL2v10", being the "unaestheticXL2v10" embedding included in my example runs.

What changes between v3.5 and v4.1: overall quality, way better listening.

Checkpoint Trained.