For best outputs:

Sampler: DPM++ 2M Karras

Steps: 50, cfg 7, CLIP skip 2.

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

Pos prompt: it usually doesn't require if big weights nor quality keywords. Recommendable add the LoRa "FaceBeauty_qinglong_V3" with a weight of 1 at the end of your prompt (included in my example runs).

Neg prompt: "dark, shadows, shadowed face, out of frame, bad anatomy, small breasts, plain chest, underage, kid, child, loli, cub, worst quality, sketch, drawing, cartoon, text, signature, username, watermark, nsfw, nudity, topless, nipples, genitalia, aid210, boring_e621_v4", being the "aid210" and "boring_e621_v4" embeddings included in my example runs.

Note about the NSFW: first, obvious, remove the keywords "nsfw, nudity, topless, nipples, genitalia" from the neg. Now, the model is very able to NSFW, not only POV but actually able to draw the entire partner (side view), but it seems to have a hard time mixing species, like for example, 1 wolf with 1 cat.