ComfyPilotComfyPilot
模型库 / LoRA / Hotori (Neverness to Everness) - Anima LoRA
Hotori (Neverness to Everness) - Anima LoRA

Hotori (Neverness to Everness) - Anima LoRA

3060496

LoRA 13 次下载 1 喜欢 更新于 2026-11-22

底模:Anima

charactertrigger word异环hotori洵

下载模型 在源站打开

触发词

hotori

文件清单 线路:原始(huggingface.co) · 切换

下载后放进对应的 ComfyUI 目录。models/ 指 ComfyUI 的模型根目录。

权重 hotori-v4-000004.safetensors 175 MB 下载

说明

Hotori (Neverness to Everness) - Anima LoRA

作者:liomylio

Hotori / ホトリ

A character LoRA for Hotori from Neverness to Everness (异环).

Trigger word

hotori

Recommended settings

  • Anima base v1.0 (28 blocks) — weight 1.0
  • Anima-2.9B family (40 blocks) — weight 0.7 Steps 28–50, CFG 4–5.

Prompt

Positive: hotori, 1girl, solo, black hair, (hime cut:1.2), red eyes, pale skin, gold kanzashi hairpin, (floating purple halo:1.3), red spider lily hair ornament, white flower tassel, dark green ribbon, red-white tassel earrings, gold pendant with red gem, (black kimono:1.2), red floral pattern, (off-shoulder:1.2), red-white shoulder trim, white inner layer, (wide furisode sleeves:1.2), gold cord, white pom-pom, yellow tassel, (large red bow:1.2), white braided cord, red rope knot, (front slit to thigh:1.2), red-white hem trim, white thighhighs, red dot cuff, (white high-heeled geta boots:1.2), black sole, gold heel

Her features, in order of what matters

  • The floating halo — a pale violet ring above her head with a cross/pointer structure and star specks. It is her signature as an "Anomaly". Do not merge it with the hairpin; the hairpin is a separate gold kanzashi running horizontally through the top of her head.
  • Hime cut — blunt bangs, cheek-length side locks, waist-length black hair.
  • Red spider lily on the right side of her head, with white flower tassels and dark green embroidered ribbons.
  • Black kimono, red floral pattern, shoulders bare, white inner layer, red-white trim.
  • The red bow on her chest with a thick white braided cord hanging from it (shimenawa style).
  • White thighhighs with a row of red dots at the cuff.
  • The boots — this is the detail most models get wrong. They are not plain geta. White upper with red laces, black platform sole, gold heel. Geta toe, high heel back.

Known limits

  • She was trained mostly on standing showcase footage, so full-body poses beyond "standing straight" are weak. The LoRA knows her face and her costume, not her posture — expect to lean on your base model for dynamic poses.
  • The chest is not in the captions; it comes from the images. On some angles it flattens. If that matters, say (large breasts:1.2) yourself.
  • The front slit can collapse into a tube skirt. (front slit to thigh:1.2) in the positive and (long skirt:1.2), (tube skirt:1.1) in the negative helps.

Training

  • 322 frames from four source videos (two 1080p PVs, two 4K PVs, one 4K 60fps character showcase)
  • dim 32 / alpha 32, 5 epochs, lr 8e-5, sigmoid timestep sampling, on anima-base-v1.0 (28 blocks)
  • Captions carry identity + costume every frame; legs and boots only on frames where they are visible; props (umbrella, cup, sword) per frame
  • 8th epoch of a 5-epoch run — the last checkpoint before it starts over-baking

示例图