CivArchive
    Cum in mouth - CIMzz - Z-imageturbo test1
    NSFW
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined

    ZImageTurbo Info:

    This is a test model. I've updated it today after a few more trial runs. the tag, cum swirl, fails miserably with ZImage, creating a magical loops of white stuff...don't use it unless you're a cum bender.

    Otherwise, the tags listed below for the SDXL version still work well, although "from the side" only seems to turn the focus by about 30 degrees vs a solid 45deg on SDXL.

    STR 0.8-1.5 works well depending on the subject. Sometimes it's too much, sometimes too little.

    *************

    SDXL Info:

    I wasn't happy with the outcome from other models in this aspect, at least with SDXL Base. So here is my Cum in mouth lora. Trained for women, haven't tested with male generation. The strength is subtle to avoid face alteration. Masking was used to avoid background modification.

    STR: 1.0 - 3.0

    Works with EULER and UniPC. I haven't tested others.

    Best used with a model aware of cum already, like the suggested models. It struggles to apply this concept to SFW models partly due to the masked training. I don't think it applies the concept properly if the model isn't capable of generating a good pose first.

    Also, this seems to fail to add cum to her tongue if her tongue isn't sticking out. I'm testing some updates in training to solve this.

    Activation Word: CIMzz

    Trained on Positional Words:

    from the front

    from the side

    from above

    from below

    on her back

    closeup

    Trained on Concept Words:

    cum in mouth

    cum on tongue

    cum drip

    cum on lips

    penis visible

    lipstick

    tongue out

    Description

    Quick test with dataset on z-image. Seems to mutilate nipples, which is weird.

    additional tag: cum swirl

    for that texture that girls crave.

    THIS IS A WORK IN PROGRESS....I have 0 idea how this model, zimage, works yet.

    FAQ

    Comments (4)

    wyxzddsjj919Nov 29, 2025· 1 reaction
    CivitAI

    Thanks for sharing, which node does this LoRa need to be loaded? Can I use COMFYUI's default LoRa load node? Or do you need a dedicated LoRa loading node like Nunchaku? Are there any special requirements for schedulers and samplers?

    Gericho222
    Author
    Nov 29, 2025· 1 reaction

    sample images have metadata in them. I believe you can load one into ComfyUI to bring up the workflow. It's a modified version of this: https://civitai.com/models/2169035/z-image-turbo-workflow?modelVersionId=2442551

    wyxzddsjj919Nov 29, 2025

    @Gericho222 THX

    bevajon407743Dec 10, 2025
    CivitAI

    Did you use the training adapter or 'de turbo'?
    And if you used the training adapter, which version did you use, v1 or v2?
    Thanks!