CivArchive
    Micorsoft Lens - World FIRST Lens Lora - v1.0
    NSFW
    Preview undefined
    Preview undefined
    Preview undefined

    # Sebastian + Jessica v2 — Dual Character LoRA (Microsoft Lens)

    Dual-character LoRA for [Microsoft Lens](https://huggingface.co/microsoft/Lens-Base) — two recurring characters trained together in one adapter:

    - Sebastian — blonde man, long hair, Irish blue eyes, tuxedo

    - Jessica — redhead, long copper-red hair, Irish green eyes, purple dress

    Trained with [LoboForge LensTrainer](https://github.com/LoboForge/LoboForge-LensTrainer) — a config-driven LoRA trainer built for Lens-Base (flow-match loss, ComfyUI export, 16–24GB VRAM presets).

    Community model — not affiliated with or endorsed by Microsoft.

    ---

    ## Links

    | | |

    |---|---|

    | Download (weights) | [Hugging Face — lens-lora-sebastian-jessica-v2](https://huggingface.co/LoboForge/lens-lora-sebastian-jessica-v2) |

    | Trainer (GitHub) | [LoboForge/LoboForge-LensTrainer](https://github.com/LoboForge/LoboForge-LensTrainer) |

    | Trainer docs (HF Space) | [LoboForge/LoboForge-LensTrainer](https://huggingface.co/spaces/LoboForge/LoboForge-LensTrainer) |

    | LoboForge | [LoboForge.com](https://LoboForge.com) |

    ---

    ## Train your own Lens LoRAs

    If you want to train subject or style LoRAs on Lens-Base yourself:

    - GitHub: https://github.com/LoboForge/LoboForge-LensTrainer

    - Quickstart: python train.py configs/train_lora_lens_base_24gb.yaml

    - Docs Space: https://huggingface.co/spaces/LoboForge/LoboForge-LensTrainer

    - More from LoboForge: https://LoboForge.com

    The trainer handles flow-match training, checkpoint resume, mid-run samples, and ComfyUI-compatible diffusion_model.* export.

    ---

    ## Training (this LoRA)

    | Field | Value |

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

    | Base model | Lens / Lens-Base |

    | Dataset | 24 image/caption pairs, 1024×1024 |

    | Published checkpoint | 5000 steps |

    | LoRA rank / alpha | 16 / 16 |

    | Optimizer | AdamW 8-bit |

    | VRAM preset | 24GB (offload + caches) |

    ---

    ## Prompt format

    Full-sentence captions — no single trigger word. Describe both characters and the scene, similar to your training captions.

    Example:

    ```Sebastian is a blonde man with long hair and irish blue eyes in a tuxedo. Jessica is a redhead woman with long copper colored red hair and irish green eyes in a purple dress named. They are standing together facing forward and laughing```

    ---

    ## Inference (Lens-Base)

    | Setting | Value |

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

    | Steps | 50 |

    | CFG | 5.0 |

    | Resolution | 1024×1024 |

    ComfyUI: Load Lens-Base (or CivitAI Lens checkpoint) + this LoRA. Keys use the diffusion_model.* format from the trainer export.

    Lens-Turbo: 4 steps, CFG 1.0 — quality may differ from Lens-Base training.

    ---

    ## License

    PolyForm Noncommercial License 1.0.0 — non-commercial use only.

    Microsoft Lens base weights remain under their own license.

    Description

    # Sebastian + Jessica v2 — Dual Character LoRA (Microsoft Lens)

    Dual-character LoRA for [Microsoft Lens](https://huggingface.co/microsoft/Lens-Base) — two recurring characters trained together in one adapter:

    - Sebastian — blonde man, long hair, Irish blue eyes, tuxedo

    - Jessica — redhead, long copper-red hair, Irish green eyes, purple dress

    Trained with [LoboForge LensTrainer](https://github.com/LoboForge/LoboForge-LensTrainer) — a config-driven LoRA trainer built for Lens-Base (flow-match loss, ComfyUI export, 16–24GB VRAM presets).

    Community model — not affiliated with or endorsed by Microsoft.

    ---

    LORA
    SD 1.5

    Details

    Downloads
    14
    Platform
    CivitAI
    Platform Status
    Available
    Created
    6/5/2026
    Updated
    6/12/2026
    Deleted
    -

    Files

    lora_step_005000.safetensors

    Mirrors

    HuggingFace (1 mirrors)
    CivitAI (1 mirrors)