🎩 Sparxie (Honkai: Star Rail) — Anima — default outfit + Candy-Coated Waves skin 🐇

Hi Everyone, hopefully you will like this LoRA, Sparxie is my favorite HSR character so I gave my extra to make her look good in rendering! If there are any errors, bugs, let me know I will then release v2 version.

Sparxie, the white-haired Sparkle SP version, trained for Anima with full text-encoder training — so her design holds together under complex prompts, unusual angles and outfits that never appeared in the training data.

Two outfits under one trigger. The trigger alone gives her default look; one extra word switches to the Candy-Coated Waves skin. No outfit tags needed for either.

Particular attention went to the details most versions lose:

• the card-suit face marks under her eyes, rendered as actual suits, but often the diamond shape changes into other shapes..

• the club brooch on the choker and the card-suit trim on the sleeves

• rear and side views — the back of the kimono and the skin's off-shoulder top are correct from behind, not guessed

Trained at 1024px on zoomed in-game reference captures of both outfits plus curated community art.

⚙️ Usage

Trigger: sparxie \(honkai: star rail\)

Skin toggle: add candycoatedwaves

Strength: 1.0

Base: Anima Base v1.0 / waiANIMA

Steps: 25–30

CFG: 4–5

Sampler: er_sde / simple

sparxie \(honkai: star rail\) alone → red card-motif kimono, sailor collar, white frilled skirt, mini top hat with rabbit ears, thigh strap, rabbit charm.

sparxie \(honkai: star rail\), candycoatedwaves → black off-shoulder top, red plaid pleated skirt, red heart sunglasses on her head, studded belt. Write the toggle as one word, exactly as shown.

No extra tags required. Hair, twintails, eyes, face marks and the outfit all hold on their own. You don't need to describe her to get her.

💡 Prompt tips

Suggested negative:

worst quality, low quality, blurry, watermark, signature, text, extra fingers, human hands, deformed hands

Other outfits work. She was trained with a small wardrobe set so named garments override the default: white t-shirt, denim shorts, pajamas, bikini, one-piece swimsuit, hoodie. If the top hat tags along and you don't want it, add hat, top hat to the negative. For swimwear add facing viewer so the base model doesn't default to a rear view.

Things that respond well: portrait, close-up (the face marks show best here), from side, profile, from behind, looking back, cowboy shot, outdoors, city street, day, halftone, pink background for a poster look.

🔧 Troubleshooting

Output looks like a game screenshot. Add 2d to the negative or anime style, illustration to the positive. The in-game captures are tagged 3d in training, so it is directly negatable.

The skin shows up on a plain prompt / the kimono shows up with the toggle. Check the toggle word is spelled as one word: candycoatedwaves. Strength 0.9–1.0 — lower strengths weaken the toggle before they weaken the default.

Face marks or pupils look simplified. They render most reliably in portraits and close-ups; at full-body framing they are a few pixels and any model will simplify them. Use portrait, close-up or upscale with a face detailer.

📝 Version notes

v1 - the detail pass. Trained at 1024px on zoomed in-game captures of the face, eyes and waist for both outfits, plus tight face crops from high-resolution art. Compared to earlier versions: the face marks are now the correct card suits on the correct sides (red heart under her right eye, white diamond under her left; the skin swaps the diamond for a small pale ribbon), the waist charm is a proper white rabbit, the sleeve trim shows card suits, the butterfly motif in her pupils is present in close-ups, and the skin toggle no longer drags the top hat along.

Every pose, framing, expression and background tag is separate from her identity in training, so all of them stay fully steerable without losing her.

Feedback and bug reports welcome — they genuinely improve the next version.