For best outputs:

Sampler: Restart.

Steps: 30 ~ 60, cfg 7 ~ 8, CLIP skip 2.

VAE: blessed2_fp16.safetensors / kl-f8-anime2_fp16.safetensors

Pos prompt: add the LoRa's "FaceBeauty_qinglong_V3" (weight 1) and "add_detail" (weight 0.7) at the end of your pos (optional but they add some extra detail). Both are included in my example runs.

Neg prompt: too long to be pasted in here (feel free to use the example runs') but it uses the embeddings "By bad artist -neg, "bad-hands-5" and "ng_deepnegative_v1_75t", included in my example runs.

What changes from v0.12 to v0.15: input_image feature.

Checkpoint Merge.