CivArchive
    LongYue V4 - Mature MILF LoRA (Pony SDXL) - v1.0
    NSFW
    Preview 134288663
    ### Overview                                                                                                                                             

    LongYue(龙玥)is a mature MILF character — the female protagonist of the mature-themed novel "Matron System" (熟女名器系统).

    This LoRA is trained on Pony Diffusion V6 XL at 1024×1024. It captures her signature look: short black bob cut, piercing blue eyes, red lips, pale

    skin, soft waist, thick thighs, big breasts, and a naturally curvaceous mature figure.

    Works with both SFW and NSFW prompts.

    ### Trigger Word

    ly_longyue

    ### Recommended Settings

    | Setting | Value |

    |---------|-------|

    | LoRA weight | 1.0 (or 0.8-0.9 for softer presence) |

    | Base Model | Pony Diffusion V6 XL |

    | Resolution | 1024×1024 |

    | Sampler | Euler Ancestral |

    | Steps | 30 |

    | CFG Scale | 5.0 |

    ### Prompt Template

    Positive:

    ```

    score_9, score_8_up, score_7_up, ly_longyue, 1girl, solo,

    mature woman, 40yo, milf, curvy, big breasts, thick thighs, soft belly,

    short black hair, bob cut, blue eyes, red lips, red lipstick, pale skin

    ```

    Negative:

    ```

    score_4, score_5, score_6,

    child, young, loli, kid, teenage, source_anime, anime,

    skinny, thin, petite, bony, collarbone bones,

    huge breasts, gigantic breasts,

    abs, toned, muscular, fit, athletic,

    long hair, very long hair, deformed, bad anatomy

    ```

    ### Tips

    - Weight 1.0 for strongest character presence; 0.8 for natural look

    - Seed 42 gives the most consistent face — good for batch production

    - Add nude or specific clothing after the trigger word

    - Remove clothing tags from prompt when generating nude

    - simple background / plain background helps clean character isolation

    ### Training Details

    Base model: Pony Diffusion V6 XL

    Framework: Kohya sd-scripts (sdxl_train_network.py)

    Dataset: 30 images, 1024×1024, repeats=3

    Steps: 1800 (30×3×20)

    LoRA dim/alpha: 16/8 | LR: 1e-4, cosine scheduler | Optimizer: AdamW8bit

    Final loss: 0.083

    ```

    Description

    LORA
    SDXL 1.0

    Details

    Downloads
    23
    Platform
    CivitAI
    Platform Status
    Available
    Created
    6/20/2026
    Updated
    6/29/2026
    Deleted
    -

    Files

    longyue_v4.safetensors

    Mirrors

    CivitAI (1 mirrors)