CivArchive
    Nipple cross piercing - LoCon (Lycoris)
    NSFW
    Preview 653531
    Preview 653377
    Preview 653371
    Preview 653373

    Sadly, it is a challenge to train or generate a piercing on nipple, clit, since the piercing always pollute the shapes of them. As you can see in generated examples, usually no nipples but only areole. I also try other similar models, they also don´t work that good.

    I cannot resolve these: no nipples but only areole, triple nipples, overfit,

    Training details see training notes below. Give me some suggestions.

    Example prompts: nipple piercing,

    Description

    For LyCORIS you should install additional extension in Automatic 1111

    FAQ

    Comments (5)

    buranj
    Author
    May 1, 2023
    CivitAI

    Training Notes

    Version: LoCon

    Base materials:

    1st concept 35 imagines with piercing, various person, gold or silver piercing

    2nd concept 5 imagines without to prevent pollution, but fails

    resolution: 768 * 768

    repeating times: 10

    Tag manipulation:

    1st concept add cross piercing to top, remove nipple piercing, piercing, cross,

    2nd concept put nipples to top

    Method:LyCORIS

    Model: grapefruitv41

    Parameters:

    network dim/alpha 64

    learning rate: all 1e-5

    conv dim/alpha 4

    Loss: average 0.06

    Lowest loss at 9. epoch (0.047)

    Video memory occupancy: 6.9-7.1G

    Training time: 5.9 hours, average 5.3 s/it

    buranj
    Author
    May 2, 2023

    Version: LoRA

    Base materials:

    40 imagines with piercing, various person, gold or silver piercing, replace some low-quality imagines

    no more 2nd concept.

    resolution: 768 * 768

    repeating times: 10

    Tag manipulation:

    move nipple piercing to the top, no remove, everything remain auto generated with auto tagger threshold 0.35

    Method:LoRA

    Model: anyLoRA

    Parameters:

    network dim/alpha 64

    learning rate: all 1e-5

    Loss: average 0.057

    Lowest loss at 9. epoch (0.051)

    Video memory occupancy: 6.9-7.1G

    Training time: 5.2 hours, average 5.1 s/it

    buranj
    Author
    May 6, 2023

    Iterative model: pick 60 of 160 generated to retrain, resolution 512*768. The parameters can be viewed by "show metadata" in the upper right corner of the LoRA thumbnail, learning rate 8.8e-6

    SayaMay 1, 2023· 1 reaction
    CivitAI

    It looks like it was trained with Koikatu. . .

    buranj
    Author
    May 1, 2023

    Not, but nearly

    LORA
    Other

    Details

    Downloads
    639
    Platform
    CivitAI
    Platform Status
    Available
    Created
    5/1/2023
    Updated
    6/11/2026
    Deleted
    -
    Trigger Words:
    cross piercing

    Files

    nipplecross.safetensors

    Mirrors

    CivitAI (1 mirrors)

    Available On (1 platform)

    Same model published on other platforms. May have additional downloads or version variants.