For best outputs:

Sampler: DPM++ 3M SDE Exponential

Steps: 80, cfg 7, CLIP skip 2.

VAE: vae-ft-mse-840000-ema-pruned_fp16.safetensors

This model is a mix of Perfect Deliberate and Cyber Realistic.

Pos prompt: the model listens quite well and usually doesn't need too much keyword weighting. Add the LoRa "FaceBeauty_qinglong_V3" with a weight of 1 ~ 1.5 at the end of your prompt to improve the face's details (optional, included in my example runs).

Neg prompt: a bit too huge to be written here, feel free to check my example runs. They come with the embedding "epiCNegative", which is optional but helps with the overall details and quality.