CivArchive
    CumHereV1 Flux Cum LoRa (clothes, hair, non-facials) - CumHereV1
    NSFW
    Preview 28218261
    Preview 28217434
    Preview 28217426
    Preview 28220803
    Preview 28243722
    Preview 28220362
    Preview 28295491
    Preview 28295885

    Since there are already a few really great facial cum LoRAs out there, I wanted to focus CumHereV1 training on a variety of non-facial cumshots. It is intended to produce realistic cum on clothes (sweaters, panties, swimsuits, etc.), breasts, tummy, hair, legs, feet, and more. I did include some Anime and cartoons in the training data but illustrated images produced are hit or miss.

    You do not need to include words like "white cum" because it already understands that cum is white. CumHereV1 was not trained on facials but it does randomly drop a load on faces at times.

    I am still testing to see how well it plays with other LoRAs. Please leave a comment if you have feedback or find that it doesn't work in a specific scenario.

    Usage:
    Weight of 1 seems to work well most of the time. Adjust the weight down to .8 if you get too much cum but if you set it too low the cum starts to look unnatural.

    The LoRA understands what cum is so no trigger word is needed. It usually understands the concept of "a lot of cum" so you can add that phrase if you want more. You will get best results by stating very specifically what you want:

    • There is cum on her ass and panties

    • There is cum all over the front of her swimsuit

    • There is cum on her tummy

    • There is a lot of cum on her breasts and nipples

    • There is cum all over her body

    I've tested the LoRA with several Flux variants in Forge, SwarmUI, and ComfyUI:

    • flux1-dev-fp8

    • flux1-dev-fp16

    • flux1-schnel - you may need to play with the weight.

    These resolutions have been tested extensively but it seems to do well at other resolutions as well:

    • 832 x 1216

    • 896 x 1152

    • 1024 x 1024

    Other notes:

    Training was about 7 hours on an RTX 4090, for 4000 steps on just 34 images with tightly controlled dimensions across 4 buckets, Learning Rate 2e-4, and very detailed natural language captions.

    Description

    FAQ

    LORA
    Flux.1 D

    Details

    Downloads
    6,557
    Platform
    CivitAI
    Platform Status
    Available
    Created
    9/7/2024
    Updated
    4/22/2026
    Deleted
    -

    Files

    CumHereV1.safetensors

    Mirrors

    CivitAI (1 mirrors)
    TensorFiles (1 mirrors)
    Torrents (1 mirrors)
    Other Platforms (TensorArt, SeaArt, etc.) (2 mirrors)