<pre><code>Trigger word

`maid combatant`

Recommended weight

0.7 (range 0.6–0.9)

&gt; Suggestive (shiny bodysuit, transparent apron). No nudity in the samples.

A battle maid built out of porcelain and glass: a statue girl dressed in a fantasy maid uniform
made of transparent metal, a glossy painted bodysuit and a see-through apron.

The LoRA carries the full design — the porcelain skin and face protector, the transparent
frilled headdress, the reflective bodysuit, the shoulder plate, the red tie and gloves — and
keeps them stable across poses, framings and backgrounds.

### What you get
- Consistent silhouette and armour layout at any camera distance
- Material mix that reads clearly: glossy black bodysuit vs. transparent frills vs. porcelain skin
- Works in plain white-background portraits and in full scenes alike

### Notes
- The design is layer-heavy, so very wide shots (full scene, tiny figure) lose the headdress detail first. Use portrait / cowboy shot when the outfit matters.
- For a cleaner porcelain look, add `porcelain skin, smooth skin, high contrast` to the prompt.
- The trigger alone already pulls most of the design; the LoRA mainly adds material accuracy and consistency.

### Prompt template

```
maid combatant,porcelain skin, fusion of porcelain statue and maid, black frilled maid headdress made of transparent metal, shiny reflective painted bodysuit, gray black transparent maid apron, frilled shoulder pad, red tie, black gloves, cowboy shot, standing, looking at viewer, simple background, white background
```

Design tags used in training

```
porcelain skin, fusion of porcelain statue and maid, black frilled maid headdress made of transparent metal, shiny reflective painted bodysuit, gray black transparent maid apron, frilled shoulder pad, red tie, black gloves
```

**Generation settings (used for the samples)**
- Sampler: Euler a (euler_ancestral) / simple
- Steps: 28
- CFG: 7.0
- Resolution: 832×1216
- Negative: `worst quality, low quality, score_1, score_2, score_3, blurry, jpeg artifacts, watermark, text, signature, bad anatomy, extra fingers`

Training

123 images × 16 epochs = 1968 steps, network dim 32 / alpha 16, 1024 bucket. Trained on waiANIMA v10 Base10.

Licence

Free to use and merge. Credit appreciated but not required. Do not resell the LoRA file itself.</code></pre><p></p>

Files with SHA256: 9ee34edf43954a14572bba14798aa5aed076892415cd36ab8b8cd7133f8d8d39