Angora Plush - Dense, Soft & Fluffy | Krea2 LoRA

Made for lush rabbit-angora knitwear: generous volume, a soft blended surface and that distinctive wispy halo. The aim is convincing angora texture and not ordinary wool with sharp hairs pasted over it.

This is a material LoRA rather than one specific outfit. Describe the garment you want: sweaters, turtlenecks, skirts, hotpants or something else. It works well with other LoRAs, although on some generations the fluff might bleed into other textures (such as cotton pants or whatever)

Usage

• Trigger: angoraplushlab

• Recommended strength: 1.0 is my preferred setting.

• Lower the strength if it starts overpowering the garment’s shape or your chosen style.

• Use natural-language descriptions and your checkpoint’s recommended sampling settings.

Examples:

angoraplushlab, a cute adult woman with a pink feminine bobcut wears a thick ivory angora turtleneck with generous soft volume, a dense plush surface and a diffuse wispy halo, with the knit structure largely obscured by the fluff.

angoraplushlab, a cute adult woman with a pink feminine bobcut wears a longsleeved angora crop top with a rounded neck and loose fit. the angora fluff is diffuse in its texture as all the fine fibers create a surface of deep soft fluff with only some tufts of fibers are visible, the angora has a diffuse halo around the garment. the angora outfit is so fluffy soft that the knit stitches are hidden in the lush fluff.

Training

Trained locally using 120 curated angora-garment photographs, with individual material-focused captions, 1024px aspect-ratio buckets, rank/alpha 32 and 1,785 training steps using Musubi Tuner.

The focus was material appearance across different garments and not reproducing a particular person or specific outfit.

Styles

Many styles work well. My personal favorites are Kodachrome, cinematic photography and candid photography. For style inspiration, see lumenastrum’s Clio Style Library. The prompt structure here also works really well.

My generation stack

For a similar starting point, I use local ComfyUI with Krea2 Raw FP8 → Turbo two-stage sampling, this angora LoRA at 1.0, SNOFS Krea v1.6 at 0.85, and SNOFS Photo Slider V2 at +3.0, plus an optional local Qwen prompt interpreter to assist with prompts

Sampling: Euler / Simple, stage schedules 52 / 12, 12% handoff, CFG 4 / 1. I’ve particularly liked 2MP, but yeh these are jsut my preferences, not required settings. The LoRA should be flexible enough to play around with