# 黄晓烟 / Huang Xiaoyan
A character LoRA for Huang Xiaoyan from Wang Pai Yu Shi (王牌御史).
## Trigger word
huangxiaoyan
## Recommended settings
- anima-base-v1.0 (28 blocks) — LoRA weight 1.0
- Anima-2.9B family (40 blocks) — LoRA weight 0.7
Steps 28–50, CFG 4–5. Works on both families; on the 2.9B side the block remap is handled by the Anima-2.9B LoRA patch.
## Prompt
Positive:
masterpiece, best quality, huangxiaoyan, 1girl, solo, black hair, long hair, red eyes, ahoge, green gourd hair ornament, (long white taoist robe with blue cloud pattern:1.3), blue trim, red waist sash, red dudou, bare legs, (blue shoes:1.2), dynamic pose, (messy flowing hair:1.2), (translucent rim light:1.3), (floating particles:1.2), cinematic composition, depth of field, wide angle lens
Negative — this line matters:
(pants:1.4), (trousers:1.4), (socks:1.2), (stockings:1.2), (short robe:1.3), worst quality, low quality, blurry, bad hands, extra fingers, deformed
Without the negative the model keeps inventing trousers and socks. She wears a dudou, the robe, and a pair of blue shoes — nothing else. The legs are bare by design, so don't put bare thighs in your negative or you will fight yourself.
## Two outfits in the dataset
1. White cloud-pattern taoist robe (+ red dudou, blue shoes)
2. School uniform (white shirt, dark pleated skirt, red ribbon)
Name the one you want. Left unspecified, it drifts toward an open robe.
## Training
- 160 frames from the anime. Captions carry identity tags only: huangxiaoyan, 1girl, solo, black hair, red eyes, ahoge, large breasts
- dim 32 / alpha 32, 8 epochs, lr 8e-5, sigmoid timestep sampling
- Trained on Anima-2.9B (40 blocks) first, then block-remapped to 28 blocks so the same file also loads on anima-base-v1.0. The 12 inserted blocks' deltas were dropped in the conversion; the 28 original blocks are carried over unchanged.
## Known issues
- The face can come out softer than the anime. Add (sharp eyes:1.2), serious expression if she looks too sweet.
- Side slits sometimes open too wide. Add (closed robe:1.2).

