CivArchive

    Collection of my Caucasian female loras characters

    Faces are from https://thispersondoesnotexist.com/ some are just a single picture and others are a blend of a bunch (creating a face_model in reactor).

    for those who doesnt know what Caucasian is:

    About the term “Caucasian” in this LoRA

    Historically, the term Caucasian (or Caucasoid) was used in anthropology to describe a broad group of populations including all of Europe (not only Northern or Eastern Europe), as well as parts of Western Asia, Central Asia, and North Africa. This classification, proposed in the 18th–19th centuries by scholars such as Johann Friedrich Blumenbach, included Southern Europeans (e.g., Spanish, Italian, Greek) within the Caucasian category.

    Although modern science recognizes that race is a social construct and the term Caucasian is now outdated, it is still commonly used in a descriptive sense in creative works.
    For this LoRA, “Caucasian” follows the historical anthropological definition, which absolutely includes Southern European phenotypes. That means Spanish or Mediterranean women are just as Caucasian as Scandinavian or Slavic women.

    I provide this LoRAs for free and includes a wide variety of female characters that fall under the historical anthropological definition of the Caucasian group. The dataset is built from portraits I personally select — for example, from sources like thispersondoesnotexist.com — based on aesthetic appeal, not to match a single region or stereotype. This means you’ll see Southern European, Central European, Eastern European, and other appearances within the Caucasian spectrum. This LoRA is a collection of diverse characters, not a “Russian/Swedish women generator.” If you want a different or narrower interpretation, you’re free to train your own LoRA to fit your specific tastes.

    I'm also convinced any of this characters with the right prompt can look like a Russian or Scandinavian stereotipic blue eyes and blonde hair woman..

    ***** This LoRA is shared without warranties and with the condition that it is used in a lawful and responsible way. I do not support or take responsibility for illegal, harmful, or harassing uses. By downloading or using it, you accept that you are solely responsible for how it is used. *****

    Description

    FAQ

    Comments (21)

    wild_RPGAug 18, 2025· 1 reaction
    CivitAI

    Going to need an facial expression LORA to go with these. Since you've said they're trained on still photos from different angles but the same expression, they all end up a bit vapid and wooden with the same expression no matter what is prompted for (unless another LORA forces it - but right now that's basically just orgasm and BJ loras).

    Latest one seems to have a bit more range though.

    K3NK
    Author
    Aug 18, 2025· 1 reaction

    i've never encountered that problem, wan2.2 is pretty good in creating expressions.. im training this with reactor over a wildcarded prompt, letting SD create the expressions is the same as letting wan.. and wan2.2 is much superior in that.. idk why you saying this really, i dont see you posting anything to back up this claims..

    anyway thanks for the feedback i suppose..

    wild_RPGAug 18, 2025· 2 reactions

    K3NK Maybe I'm just loading my prompts up too much for the expressions to come through. I am putting together a storyboard post but getting frustrated with the lack of expressions >.< Sorry to vent.

    jincheng666Aug 19, 2025· 1 reaction
    CivitAI

    wan2.1 can do it?

    jincheng666Aug 19, 2025· 1 reaction

    There are so many lora, which ones are 2.1 and which ones are 2.2, how to distinguish them

    jincheng666Aug 19, 2025· 1 reaction
    CivitAI

    There are so many lora, which ones are 2.1 and which ones are 2.2, how to distinguish them

    K3NK
    Author
    Aug 19, 2025· 2 reactions

    i taged the ones for 2.1 as wan2.1

    CyberAImaniaAug 19, 2025· 1 reaction
    CivitAI

    I assume you train just LOW noise ? for wan2.2 ? just one file ?

    K3NK
    Author
    Aug 19, 2025· 1 reaction

    yeah, most of the times it does the work, if the character is complex a high model is recomended, but for faces only, low is enough imo

    4326369Aug 21, 2025· 1 reaction
    CivitAI

    love your work , is it possible to have one character lora to randomize faces each generations , it would make it easier to use

    K3NK
    Author
    Aug 22, 2025· 1 reaction

    why would you need that for? this ones are character loras, the intent is to produce the same face over and over.. xD i consider your approarch interesting nonetheless.. im curious now xD

    4326369Aug 22, 2025· 2 reactions

    @K3NK mainly for randomness , i really like being surprised, i use wild cards for outfits animations and locations and camera movements makes each generation unique but the only problem i have is repeted face

    K3NK
    Author
    Aug 22, 2025· 3 reactions

    @CLAAN3008 i get it, i myself like to use wildcards.. i guess doesnt hurt to try,, someone send me a dataset of different faces to trying this specifically, i have not trained it yet,, i might try later on, stay tunned ! im prolly releasein an anal fisting lora today!, just testing and fixing my damn workflow.. xD

    4326369Aug 22, 2025· 1 reaction

    @K3NK looking forward to it best of luck

    Grim0rAug 22, 2025· 2 reactions

    There is an addon for Comfy which allows you to use prompts to activate loras (like A1111) , rather than the lora loader so you can randomly select a different loras through wildcards.

    Not tried it myself yet, as only just recently started using Comfyui, but I think this does the job https://github.com/badjeff/comfyui_lora_tag_loader/

    K3NK
    Author
    Aug 23, 2025· 2 reactions

    @CLAAN3008 i tried training a lora with multiple faces, but it ends up blending them, maybe if you stack entire datasets, not only a dataset with 1 photo per girl.. but im not entirely sure.. captioning and so.. it might work, but probablly will merge the faces..

    4326369Aug 23, 2025· 1 reaction

    @K3NK understandable hope you can figure it out thanks for trying

    Grim0rAug 23, 2025· 1 reaction

    @K3NK I think if you put the datasets for each character in a different folder and have a clear distinct prompt for each one, that should reduce any feature bleed between the different characters, but i've not tried training a Flux lora, so don't know how different it is from Kohya

    ronikushAug 30, 2025· 1 reaction

    Merge them, simple

    Marsian_EngineAug 27, 2025· 3 reactions
    CivitAI

    Thank you for your work, we see very few face lora and these ones are particularly impressive. Are you considering sharing your full training workflow ? From thispersondoesnotexist to stable diffusion to wan 2.2 lora ?

    fashionableAug 28, 2025· 8 reactions
    CivitAI

    Each face looks too similar.

    LORA
    Wan Video 2.2 T2V-A14B
    by K3NK

    Details

    Downloads
    456
    Platform
    CivitAI
    Platform Status
    Deleted
    Created
    8/16/2025
    Updated
    5/29/2026
    Deleted
    2/21/2026
    Trigger Words:
    0lg4

    Available On (1 platform)

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