Fot best outputs:

Sampler: DPM++ 2M Karras.

Steps: 35, cfg 7, CLIP skip 2.

VAE: sdxl_vae.safetensors.

This is a Versatile type of model, which means capable of quite decent photo realism and very nice anime, knowing of many sort of art styles and different artists.

Pos prompt: not very high weights required usually. Quality keywords add a little bit more of quality.

Neg prompt: can be left empty, or add what you don't want to see. For example, for photorealism "Compression artifacts, (asymmetric face), nsfw, nipples, breasts out, topless, face bad art, worst quality, low quality, plastic, fake, bad limbs, conjoined, featureless, bad features, incorrect objects, watermark, logo, BeyondSDXLv3, epiCNegative", where "BeyondSDXLv3", and "epiCNegative" can be found in my realistic example runs; for anime, for example, this one would do "realistic, real photo, photography, photo realistic".