CivArchive
    💎 Arin (아린) SDXL LoRA V2 - Original Virtual Korean Model [SDXL] - v2.0
    NSFW
    Preview 146521064
    Preview 146517539
    Preview 146517537
    Preview 146517540
    Preview 146517544
    Preview 146517535
    Preview 146517545

    ### 🎀 Arin (아린) - Original Virtual Korean Model [SDXL]

    Arin is a 100% fictional AI-generated original character (non-existent virtual model).

    LoRA, a 100% user-made Korean female face sculpt based on official genuine SDXL Base 1.0. Specialized in fixing natural features and focusing on face lines.


    Type of action: Korean female facial sculpting


    🔑 arin_model

    ---

    # 💎 Arin (아린) SDXL LoRA V2

    A premium, photorealistic SDXL LoRA capturing the distinct elegance, natural beauty, Arin.

    Trained with a fully revamped V2 Vision-AI Pipeline on top of the pristine Stability AI SDXL Base 1.0, this version brings superior facial fidelity, lifelike skin texture, and seamless prompt responsiveness across any SDXL checkpoint.

    ---

    ## 🌟 What's New in V2?

    * 👁️ Vision AI Disentanglement (PromptGen-v2.0):

    Every training image was analyzed and captioned in detail using vision AI to decouple clothes, hairstyles, lighting, and backgrounds from her facial identity. You can now freely change outfits and hair without distorting her signature face.

    * ✨ 100% Native SDXL Base 1.0 Training:

    Zero checkpoint bias. 100% compatible with popular SDXL realistic models (RealVisXL, CyberRealistic, HelloWorld XL, etc.).

    * 💎 Preserved Trigger Word:

    Seamless backwards compatibility — your existing workflows and prompts work immediately with the same trigger: arin_model.

    ---

    ## ⚙️ Recommended Settings

    | Parameter | Recommended Value |

    | :--- | :--- |

    | LoRA Weight | 0.75 ~ 0.85 (Sweet spot: 0.80) |

    | Trigger Word | arin_model |

    | Base Models | RealVisXL V4.0 / V5.0, CyberRealistic XL, HelloWorld XL |

    | Sampler / Scheduler | DPM++ 2M Karras / Euler a |

    | Sampling Steps | 25 ~ 35 steps |

    | CFG Scale | 3.5 ~ 6.0 (Fooocus default 4.0 ~ 7.0) |

    | Clip Skip | 1 or 2 (Default 1 for SDXL) |

    ---

    💡 Pro Tips for Best Results

    1.Face Detailer / After Detailer: For wide-angle full-body shots, using ADetailer (or Fooocus Face Restore) with a weight of 0.6 ~ 0.7 will keep her facial features crisp and authentic even from a distance.

    2.Hair & Outfits: Because hairstyles and clothing were disentangled during training, explicitly adding long hair or short bob cut in your prompt will give you reliable control over her hairstyle!

    Enjoy generating and feel free to share your stunning creations in the gallery!
    V 1.0
    v 2.0

    Description

    # 💎 Arin (아린) SDXL LoRA V2

    A premium, photorealistic SDXL LoRA capturing the distinct elegance, natural beauty, Arin.

    Trained with a fully revamped V2 Vision-AI Pipeline on top of the pristine Stability AI SDXL Base 1.0, this version brings superior facial fidelity, lifelike skin texture, and seamless prompt responsiveness across any SDXL checkpoint.

    ---

    ## 🌟 What's New in V2?

    * 👁️ Vision AI Disentanglement (PromptGen-v2.0):

    Every training image was analyzed and captioned in detail using vision AI to decouple clothes, hairstyles, lighting, and backgrounds from her facial identity. You can now freely change outfits and hair without distorting her signature face.

    * ✨ 100% Native SDXL Base 1.0 Training:

    Zero checkpoint bias. 100% compatible with popular SDXL realistic models (RealVisXL, CyberRealistic, HelloWorld XL, etc.).

    * 💎 Preserved Trigger Word:

    Seamless backwards compatibility — your existing workflows and prompts work immediately with the same trigger: arin_model.

    ---

    ## ⚙️ Recommended Settings

    | Parameter | Recommended Value |

    | :--- | :--- |

    | LoRA Weight | 0.75 ~ 0.85 (Sweet spot: 0.80) |

    | Trigger Word | arin_model |

    | Base Models | RealVisXL V4.0 / V5.0, CyberRealistic XL, HelloWorld XL |

    | Sampler / Scheduler | DPM++ 2M Karras / Euler a |

    | Sampling Steps | 25 ~ 35 steps |

    | CFG Scale | 3.5 ~ 6.0 (Fooocus default 4.0 ~ 7.0) |

    | Clip Skip | 1 or 2 (Default 1 for SDXL) |

    ---

    💡 Pro Tips for Best Results

    1.Face Detailer / After Detailer: For wide-angle full-body shots, using ADetailer (or Fooocus Face Restore) with a weight of 0.6 ~ 0.7 will keep her facial features crisp and authentic even from a distance.

    2.Hair & Outfits: Because hairstyles and clothing were disentangled during training, explicitly adding long hair or short bob cut in your prompt will give you reliable control over her hairstyle!

    Enjoy generating and feel free to share your stunning creations in the gallery!

    FAQ

    LORA
    SDXL 1.0

    Details

    Downloads
    74
    Platform
    CivitAI
    Platform Status
    Available
    Created
    10/23/2026
    Updated
    11/2/2026
    Deleted
    -

    Files

    arin_model_v2.safetensors

    Mirrors

    CivitAI (1 mirrors)