CivArchive
    Yuutenji_Nyamu_MOSSimage_lora - v1.0
    Preview 149532785
    Preview 149532788
    Preview 149532789

    【模型简介】 祐天寺若麦 Yuutenji Nyamu|BanG Dream! Ave Mujica 鼓手 底模 MOSS-Image 训练集 300 张(Live2D 原生渲染 + 官方立绘,透明底) 触发词:Yuutenji Nyamu 推荐参数:LoRA 权重 1.0,30 步,CFG 4.0,采样器 euler / sgm_uniform,竖幅 1024×1368

    【常服】 Yuutenji Nyamu, short ash-purple bob hair with bangs, magenta eyes, narrow sharp eyes with a slight upward tilt at the outer corners, she wears a grey ruffled off-shoulder top with thin spaghetti straps and four small silver buttons on the bodice below the ruffle, black high-waisted mini skirt and thin black choker worn snug around the middle of her neck, black flat shoes with a thin ankle strap.

    【演出服】 Yuutenji Nyamu, short ash-purple bob hair with bangs, magenta eyes, narrow sharp eyes with a slight upward tilt at the outer corners, she wears a black vertically striped vest with gold buttons, dark red lace bustier, black high collar with a red gem, three-strand pearl necklace, cream ruffle peplum at the waist, black split tailcoat panels, burgundy shorts, garter straps with gold clips, sheer dark stockings, red ankle-strap heels, sheer grey sleeves with black cuffs and long dangle earrings, the necklace has three strands, every pearl the same size and evenly spaced, and it sits inside her black high collar so only the front part shows.

    【注意事项】

    1. 服饰细节请按上面的语序用。扣子要紧跟 "top" 当定语、裙子放后面;塞到裙子后面会被画到裙子上。

    2. 半身要写明 "the frame cuts her body cleanly at the waist",负向加 "multiple people",否则方形画幅会堆出好几个人。

    3. 全身是弱项(训练集只有 13 张远景全身)。要出全身请写 "seen from far away so that her entire body from head to shoes fits inside the frame",画幅建议 1408×1872。

    4. 不要写表情词,交给模型自己决定,硬写容易崩脸。

    5. 演出服的耳坠在标注里只覆盖 10.9%,要出耳坠必须显式写 "long dangle earrings"。

    【V1.1 更新说明】

    本次为重新训练,并非在 V1.0 基础上微调,权重完全重新生成。

    1. 训练管线迁移 V1.0 使用在线训练服务;V1.1 改为本地 Next-Trainer(sd-scripts)全流程训练, 数据集、检查点、每阶段采样预览全部本地可控,可复现。

    2. 常服前襟银扣 V1.0 经常默认画出 5 颗。V1.1 实测 4 张样本全部为 4 颗, 且位置均落在荷叶边下方的前襟上,没有跑到裙子上。

    3. 演出服配饰 三股珍珠项链清晰可数,只从高领前方露出一段;领口红宝石与长耳坠稳定出现。

    4. 文件兼容 已转为 PEFT / DiffSynth 通用键名。同一份文件可直接用于 ComfyUI 与线上平台, 加载时 178 个 patch 正常挂载,无需再做格式转换。

    训练参数:6000 步|dim 16 / alpha 16|lr 1e-4|底模 MOSS-Image-SFT|训练集 300 张 触发词:Yuutenji Nyamu

    Description

    ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

    Yuutenji Nyamu · MOSS-Image LoRA

    ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

    【 提示词 / PROMPTS 】

    ■ 常服 · Casual

    Yuutenji Nyamu, short ash-purple bob hair with bangs, magenta eyes, casual grey off-shoulder set, she wears a grey ruffled off-shoulder top with thin spaghetti straps, four small silver buttons on the bodice below the ruffle, no buttons on the ruffle or at the neckline, a thin black choker worn snug around the middle of her neck, black high-waisted mini skirt, black flat shoes with a thin ankle strap

    ■ 演出服 · Stage Costume

    Yuutenji Nyamu, short ash-purple bob hair with bangs, magenta eyes, stage tailcoat costume, she wears a black vertically striped vest with gold buttons, a dark red lace bustier with a row of small gold buttons down its front, a black high collar with a red gem, a three-strand pearl necklace, every pearl the same size and evenly spaced, worn inside her black high collar, only the front part of the necklace shows and the rest is hidden behind the collar, a cream ruffle peplum at her waist, burgundy shorts, black tailcoat panels, garter straps with gold clips, sheer dark stockings, sheer grey sleeves with black cuffs, long dangle earrings, red ankle-strap high heels

    【 简介 / About this LoRA 】

    中文

    祐天寺若麦(Yuutenji Nyamu)的 MOSS-Image LoRA,用 300 张图训练,覆盖两套主要服装:

    灰色露肩荷叶边上衣配黑短裙的常服,以及黑红配色的燕尾舞台服。

    触发词是 "Yuutenji Nyamu",必须写在最前面 —— 底模本身不认识这个角色。上面两段提示词

    已包含发型、瞳色与全套衣物细节,可直接复制使用;接动作、景别或背景时,用逗号隔开追加在后面。

    已知特性:

    · 常服胸口的银色小纽扣默认会排满五颗。

    提示词里写 four 只是弱偏置,数量主要跟着种子走。

    English

    A MOSS-Image LoRA of Yuutenji Nyamu (Nyamu Yūtenji), trained on 300 images and covering her two

    main outfits: the grey off-shoulder casual set and the black-and-red stage tailcoat costume.

    The trigger word is "Yuutenji Nyamu" and must come first — the base model does not know this

    character. Both prompts above already contain the hairstyle, eye colour and the full outfit

    description and can be pasted as-is; append any pose, framing or background after a comma.

    Notes:

    · The casual top's silver buttons default to five across the bodice. writing "four" in the prompt is only a weak bias and the seed dominates.

    【 推荐参数 / Recommended Settings 】

    底模 / Base model moss_image_sft_int8_convrot

    文本编码器 / Text encoder qwen3.5_9b_w4a8_gptq (type: moss_image)

    VAE flux2-vae

    采样 / Sampling 30 steps · CFG 4.0 · euler · sgm_uniform · denoise 1.0

    LoRA 权重 / Weight 1.0

    尺寸 / Resolution 1216×1616 全身 · 1216×1216 半身

    负面提示词 / Negative 模糊, 低质量, 变形, 错误文字, 乱码, 水印

    ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

    LORA
    MOSS Image

    Details

    Downloads
    49
    Platform
    CivitAI
    Platform Status
    Available
    Created
    11/19/2026
    Updated
    11/22/2026
    Deleted
    -

    Files

    Yuutenji_Nyamu_MOSSimage_lora_c1-st6000.safetensors