For best outputs:

Sampler: Euler a.

Steps: 30, cfg 7.5, CLIP skip 2.

VAE: kl-f8-anime2_fp16.safetensors

Pos prompt: you can keep it simple, model listens well and usually doesn't require of big weights. Also, it doesn't need quality keywords like "masterpiece" or "best quality" ect.

Neg prompt: "EasyNegative, BadDream, (UnrealisticDream:1.2), (worst quality, low quality, normal quality, bad quality, nsfw:1.4), lowres, bad anatomy, blurry, black-white, bad anatomy, disfigured,malformed, blurred, bad proportions, bad proportions, haunch, gross proportions, fused hand, bad hands, missing fingers, extra digit, fewer digits, malformed hands, fused fingers, one hand with more than 5 fingers, one hand with less than 5 fingers, extra breasts, poorly drawn breasts, heavy breasts, missing breasts, fused breasts, bad breasts, multiple breasts, uncoordinated body, liquid body, long neck, huge thighs, huge calf, disappearing arms, disappearing thigh, disappearing calf, disappearing legs, missing limb, extra arms, extra thighs,liquid tongue, fused ears, bad ears, extra ears, missing ears, bad hairs, poorly drawn hairs, fused hairs, fused cum, big muscles, ugly, bad face, fused face, poorly drawn face, cloned face, big face, long face, bad eyes, fused eyes, poorly drawn eyes, extra eyes", being the embeddings "EasyNegative" and "BadDream" included in my example runs.

Checkpoint Trained.