CivArchive
    MiaoMiao Harem MOSS - BASE v1.1
    NSFW
    Preview 147102558
    Preview 147102460
    Preview 147102466
    Preview 147102455
    Preview 147102457
    Preview 147102456
    Preview 147102465
    Preview 147102472
    Preview 147102462
    Preview 147102464
    Preview 147102467
    Preview 147102461
    Preview 147102463
    Preview 147102477
    Preview 147102471
    Preview 147102469
    Preview 147102468
    Preview 147102476

    MiaoMiao Harem MOSS 正式发布!

    这次不是简单“换个底模”,而是把「喵喵后宫」那套审美、构图与二次元气质,迁移到 MOSS-Image-Base 生态上,做一次真正面向新架构的重塑——更能写、更能控、更能排版,也更适合做海报/封面/卡面/带字画面等“有设计感”的二次元成图。

    MiaoMiao Harem MOSS | 二次元美学 × 文字渲染 × 布局控制

    MOSS 系列的核心强项之一是文字渲染与 Layout(布局)控制: 你既可以像以前一样用自然语言/标签写法快速出图,也可以进阶用 JSON Layout 把“文字/物体/位置”写清楚,让画面更可控、更像在做设计稿。

    你可以把它理解为: “更懂泥的 MOSS 工作流入口 + 喵喵后宫的审美”。

    IMPORTANT:食用指南(超重要)

    这是 MOSS 模型,推理方式与 SD1.5/SDXL checkpoint 不同: 请使用 MOSS 对应的 Diffusers pipeline / ComfyUI 支持方式来跑。

    推荐推理参数(官方推荐向) Sampler:Euler(Flow-matching Euler)

    Steps:建议 ≥ 50(默认 50) CFG:默认 3(推荐范围 3–6;短 prompt 可适当提高)

    Shift:默认 5 SLG / Skip Layer Guidance:推荐 uncond skip,skip layer = 22(0-index;22/24/26 都可能有效)

    分辨率:推荐 768x1024/1024x1024 / 768x1280/1280x1280/960x1536/1536x1536;也可从 512×512 二次采样到 2K 级别 尺寸注意:宽高请使用 16 的整数倍

    提示词写法建议(兼容三种流派) 自然语言:适合氛围/镜头/剧情感 Tag 串:适合精确控元素(建议顺序:Quality → Score → Meta → Year → Character → Series → Artist → General) 混合写法:最推荐(又稳又自由)

    prompt

    masterpiece, best quality,very aesthetic, ultra-detailed, high contrast,

    Negative prompt(建议)

    extra toes,lowres, (bad), text, error, fewer, extra, missing, worst quality, jpeg artifacts, low quality, watermark, unfinished, displeasing, oldest, early, signature, artistic error, username, bad feet,scan, [abstract], english text, shiny hair

    MiaoMiao Harem moss 训练时加入了专属数据集用来验证正常使用不影响

    A dedicated dataset was included during the training of "MiaoMiao Harem moss" to verify that its intended, normal use remains unaffected.

    ⬅️企鹅1群 ⬅️Telegram⬅️企鹅2群⬅️企鹅3群⬅️企鹅4群⬅️企鹅5群

    Description

    BASE1.1 版本 优化了整体的画风平衡 但是本系列依然 属于美学模型 对于其他风格的 lora 肯定是有影响的 还有需要注意的是 不管是1.0 还是1.1 版本在在画面描述文字的时候要降低 重绘幅度 也就是去噪强度

    Sampler:Euler

    Steps: 30–50(追求最高质量可尝试 50 steps)

    CFG Scale: 3.5(可略低于常规 Anima 2B,建议从 3.5–4.5 起步实验)

    推荐分辨率:

    768x1280/1280x1280/960x1536/1536x1536

    Positive Prompt(默认):

    masterpiece, best quality,very aesthetic, ultra-detailed, high contrast,, {此处添加提示词 / Add Prompt here}

    Negative Prompt:

    extra toes,lowres, (bad), text, error, fewer, extra, missing, worst quality, jpeg artifacts, low quality, watermark, unfinished, displeasing, oldest, early, signature, artistic error, username, bad feet,scan, [abstract], english text, shiny hair

    感谢大家一直以来的支持与宝贵反馈!

    Thank you all for your continuous support and valuable feedback!

    ( MiaoMiao Harem moss 训练时加入了专属数据集用来验证正常使用不影响)

    (A dedicated dataset was included during the training of "MiaoMiao Harem moss" to verify that its intended, normal use remains unaffected.)

    FAQ

    Checkpoint
    MOSS Image

    Details

    Downloads
    2,708
    Platform
    CivitAI
    Platform Status
    Available
    Created
    10/28/2026
    Updated
    11/4/2026
    Deleted
    -

    Files

    miaomiaoHaremMOSS_baseV11_int8.safetensors

    miaomiaoHaremMOSS_baseV11_fp16.safetensors

    miaomiaoHaremMOSS_baseV11_int8.safetensors