CivArchive
    [Z-Image-Turbo] Asian Consistent Character | Instagram Influencer | Private Waifu - v1.0
    NSFW
    Preview 111966304
    Preview 111966212
    Preview 111966230
    Preview 111966239
    Preview 111966266
    Preview 111966281
    Preview 111966288
    Preview 111966300
    Preview 111966310
    Preview 111966352

    Here's a super consistent character LoRA I just finished ❤️

    Trained on the same dataset as my old WAN 2.2 version, but this time built on Z Image Turbo — the quality jump is actually insane, everything is cleaner, sharper, and way more reliable.

    You can literally throw this on any Instagram girl, streamer, celebrity, or your own private waifu and it'll stick to the face like glue. Works perfectly for making "your girlfriend on vacation" type series too.

    Best strength is 0.9–1.0 (I personally run it at 1 most of the time).

    Sometimes it can cook a little too hard if your prompt is vague, specify it on the prompt, and it should be fine. Super easy fix. Check out the sample prompts on the image for reference.

    Go wild with it, have fun! 🚀

    Description

    FAQ

    Comments (14)

    NicoGJNov 29, 2025
    CivitAI

    I don't understand how to use another person with this, is there a workflow where I can throw the dataset of the private waifu I want? Or how is it used?

    mobeldwaodoawNov 29, 2025

    You can’t. This is just a character LoRA. You can only use the person it was trained on. He meant that when you create a LoRA with the base model, the model will stick to your person.

    coolstrad
    Author
    Nov 30, 2025

    The point of this LoRA is just to create a consistent character so it can’t make any other woman other than the one on the picture. If you want to make different types of girl then you need to turn off the LoRA.

    Rixh666Dec 6, 2025

    @coolstrad can you make this woman's cleavage nicer and rounder, but not too big?

    coolstrad
    Author
    Dec 6, 2025

    @Rixh666 I think you can easily do that by lowering the LoRA weight and by prompting.

    newsukoi731Nov 30, 2025
    CivitAI

    The results are not good when used with the GGUF model.

    coolstrad
    Author
    Dec 1, 2025· 1 reaction

    I didnt use the GGUF model, I recommend that you use the BF16 version for the best results. Try remixing the the prompt from my images, ensure that you add the trigger word otherwise it wont work on any version.

    alois1887859Dec 9, 2025
    CivitAI

    what do you use to create a consistent dataset?

    coolstrad
    Author
    Dec 9, 2025

    Reference base face, qwen edit 2509, then refining.

    alois1887859Dec 9, 2025

    @coolstrad thanks. so you create all the images off of one face that was generated?

    alois1887859Dec 9, 2025

    and what did you use to refine the images?

    coolstrad
    Author
    Dec 9, 2025

    @alois1887859 Currently for newer models (private ones that I don’t upload on civitai) I use zit.

    coolstrad
    Author
    Dec 9, 2025

    @alois1887859 i already have a workflow for that, you can check it out on my profile

    alois1887859Dec 9, 2025

    @coolstrad cool, will check that out, thx a lot for the help :)