CivArchive
    Amateur Nude Selfies - Z Image Base & Turbo - Z-Img Base & Turbo - v1.0
    NSFW
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined

    Designed to create two types of amateur selfies and removing the default Asian tilt of the model:

    • Mirror selfies with phone in hand

    • Front facing camera selfies

    Z-image understand this okay but has issues with nudity and "camera phone quality". This lora should get you closer. You can see the differences here: https://files.catbox.moe/2dvzae.mp4

    Base Prompts/Triggers:

    • "amat_nude, taking a mirror selfie"

    • "amat_nude, taking front facing selfie"

    See example images for more in depth prompting. Works with single or group photos. Should work for most clothing/undress states. Works well with almost any body type.

    Recommend around a 1.0 strength on base and between 1.0 to 3.0 on turbo.

    Working on some other versions with more control such as the age of photo, e.g. mid-2000's for control over the look.

    Description

    This version of the model was trained on the new Z-img base model and will work on both the Z Image Base & Turbo versions.

    FAQ

    Comments (7)

    fronyaxJan 29, 2026· 6 reactions
    CivitAI

    There’s no noticeable difference whether I use the LoRA or not. The output is almost identical, I'm using it on turbo and example prompts.

    TheAIDoctor
    Author
    Jan 29, 2026

    Try bumping up the strength for turbo. I've had a couple loras trained on base that have zero effect until I turn it up to 2.0

    45thJan 29, 2026

    Same thing for me

    CabaloniusJan 29, 2026

    @TheAIDoctor isn't that a sign of undertraining?

    TheAIDoctor
    Author
    Jan 29, 2026· 1 reaction

    @Cabalonius It works fine on base at 1.0 for me. I specifically trained it for base, it just happens to also work on Turbo as well with different settings. Since base just came out, it will take a little time to dial in all the training settings to find the best options.

    JellaiJan 29, 2026· 3 reactions

    @Cabalonius It's the standard result of training on Base, from people who have trained many types of loras. Loras trained on Base work on both, and work better on Turbo than being trained on Turbo, but they require about twice as much strength only on Turbo. Base works fine with 1 strength.

    TheAIDoctor
    Author
    Jan 29, 2026

    @fronyax @45th
    Here's an example showing that Lora does work and makes a difference. It removes the default Asian tilt of the model and in my opinion improves the breast shape/definition and general anatomy. I'll admit the results aren't extreme but it's definitely noticeable and become more clear as you increase the strength.

    https://files.catbox.moe/2dvzae.mp4