Pulchra Fellini (Zenless Zone Zero) – Illustrious LoRA

Character LoRA of Pulchra Fellini from Zenless Zone Zero, trained on her default outfit.

Usage

• Base model: Illustrious XL. Trained and tested on waiIllustriousSDXL v1.7.0.

• Trigger word: pulchra_fellini

• Recommended weight: 0.8–1.0 for her default outfit, around 0.7 for outfit changes.

Her feline features, blonde hair, green eyes, ponytail and tail are generally learned by the LoRA, so you don't need to prompt for all of them every time.

Her mechanical face mask can be included or omitted depending on the desired appearance.

Prompt tags

Appearance (use these to reinforce her look):

furry, furry female, body fur, yellow fur, animal ears, blonde hair, long hair, tail, ponytail, green eyes

Default outfit:

shirt, jacket, belt, black jacket, choker, shorts, mask, fingerless gloves, navel, breasts, thigh strap, thigh pouch, long sleeves, bare shoulders,

Changing her outfit

The LoRA is fairly flexible with alternate outfits.

For outfit changes:

1. Lower the LoRA weight to around 0.7.

2. Describe the new outfit clearly in the prompt.

3. If parts of her original outfit still appear, add the default outfits tags to the negative prompt.

4. Keep her appearance tags when necessary to preserve her feline features and character identity.

Her feline features generally remain recognizable even when changing her outfit or removing the mechanical mask.

Limitations

• Small details on the mechanical face mask may vary between generations.

• Some outfit accessories may simplify or shift slightly.

• The thigh pouch may appear on either the left or right thigh, rather than consistently staying on the same side.

• Other Illustrious-based checkpoints may reinterpret some outfit details or change the overall art style.

• The mechanical mask may need explicit prompting when using alternative outfits.

Recommended negative prompt

Adding 3d to the negative prompt is recommended to help reduce unwanted 3D-like appearances and maintain a more consistent 2D anime illustration style.

Generation settings used in testing

• Primary checkpoint: waiIllustriousSDXL v1.7.0

• Alternative checkpoint: Hassaku XL v1.3 Style A

• Sampler: Euler a, Normal schedule

• Steps: 30

• CFG scale: 6

• Resolution: 768x1280

• Recommended LoRA weight: 0.8–1.0 for default outfit, around 0.7 for alternate outfits

• Hires. fix: R-ESRGAN 4x+ Anime6B, upscale x1.5 (1152x1920), 20 hires steps, denoising strength 0.35

• ADetailer: Optional face refinement, tested with denoising strength 0.25

For evaluating character fidelity, I recommend testing without Hires. fix or ADetailer first.

Training info

50 images, 5 repeats, 10 epochs (2,500 steps), trained with kohya_ss. Final epoch used.

• Training resolution: 1024

• Network Rank / Alpha: 32 / 16

• Optimizer: AdamW8bit

Credits

Character: Pulchra Fellini from Zenless Zone Zero © HoYoverse.

Alternative checkpoint: Hassaku XL (Illustrious) v1.3 – Style A, created by Ikena.

Special thanks to Ikena for creating Hassaku XL. A preview image was generated using this checkpoint to demonstrate compatibility and alternative visual styles.

This is an unofficial fan-made LoRA, not affiliated with or endorsed by HoYoverse.