Description

This LoRA is trained for -17 and similar Illustrious XL checkpoints. It is intended to recall the character identity and the Xiexiang outfit while keeping enough flexibility for portrait, full-body, seated, and elegant showcase prompts. Cross-model behavior on Pony, realistic XL, or non-Illustrious checkpoints is not guaranteed.

Recommended Usage

Basic Prompt

<lora:zheya_xiexiang_recommended_e000008:0.7>, zheya_xiexiang, zheya (arknights), 1girl, solo, no text, looking at viewer, upper body, orange red hair, blue eyes, side braid, hair flower, white wide sleeves, simple background

Full Body Prompt

<lora:zheya_xiexiang_recommended_e000008:0.7>, zheya_xiexiang, zheya (arknights), 1girl, solo, no text, full body, elegant pose, floral chinese dress, blue dress, teal skirt, white wide sleeves, black ankle boots, soft lighting

Negative Prompt

low quality, worst quality, lowres, bad , bad hands, bad fingers, extra fingers, missing fingers, extra limbs, disfigured, deformed, blurry, jpeg artifacts, text, watermark, logo, signature, arknights logo, chinese text, english text, caption, artist name, bad face, bad eyes, multiple girls, 2girls, cropped head, out of frame, wrong hair color, black hair, blonde hair, short hair, red eyes, missing orange red hair, missing blue eyes, missing side braid, missing hair flower, wrong outfit

Practical Notes

• Start from 0.7.

• Use 0.6 if a pose needs more prompt freedom.

• Use higher weights only when identity or outfit recall is too weak.

• Avoid 1.0 as the default if it makes the flower basket, lanterns, or outfit details too rigid.

• Keeping zheya_xiexiang, zheya (arknights), 1girl improves identity stability.

Training Summary

• Base checkpoint: WAI--Illustrious-17.safetensors

• Recommended epoch: 8

• Training images: 100

• Caption files: 100

• SFW / samples: 100 / 0

• Training setup: SDXL LoRA, UNet-only, rank 32, alpha 16, dropout 0.05, AdamW8bit, bf16, 10 epochs.

• Final average loss: 0.0973