Mint with 3 Outfits Support

Trained on a curated mix of 28 images—blending official 2D illustrations and 3D game models.

This model is designed to put Mint in different outfits.

Key Features

• 3 Built-in Outfits: Supports Default, Loungewear, and Bikini out of the box.

• High Versatility: Clean outfit tagging allows custom clothing using (alternate costume:1.3).

• Anima Research Included: Features detailed comparison notes and training benchmarks below.

Downside: At Dim 16, it ended up learning the art style too.

Another LoRA (Outfit 1 only)⇩

https://civarchive.com/models/2996345/mintnteoutfit1

Quick Reference

LoRA Weight: 0.8 – 1.0

Character Trigger:

mint \(nte\), cat girl, cat ears, animal ear fluff, long hair, multicolored hair,streaked hair, aqua hair, braid, cat tail, pink eyes, large breasts

Outfit1: Default

black gloves, hairband, (cat face on hairband:0.8), white over shirt, black dress, black corset, black skirt, white thighhighs,thigh strap, black boots, fanny belt pouch, gold aiguillette,

Outfit2: Loungewear

loungewear, x hair ornament, see-through clothes, neck bell, white dress shirt, pink bikini,striped underwear,baggy clothes

Outfit3: bikini, [specific tags]

aqua bikini,hat,white frilled bikini, sun hat ,navel, white thigh strap

Custom Outfits: (alternate costume:1.3)

Simillar LoRA⇩

https://civarchive.com/models/2935268/mint-neverness-to-everness

Outfit1 Key Points A and B

single outfit models outperformed 3outfits models.

As a side effect, having too much 3D source data caused the output style to look more like 3D.

Comparing Anima and IL

There is no difference between them.

Surprisingly, Anima performed just as well as IL even though the text encoder wasn't a perfect match.

Turns out, as long as the dataset quality is solid, training just the U-Net is more than enough.