CivArchive
    "Santa" (Spooky Milk Life) - killedmyself - v0.1
    NSFW
    Preview 51951115
    Preview 51951083
    Preview 51951091
    Preview 51951092
    Preview 51951080

    i fully meant to post this before christmas 2024, so now im posting it early for christmas 2025

    Post what you make down below! Tipping ⚡ images I like!

    Don't forget to leave a Like!

    Important information for this one

    • Outfit inconsistencies

    • Dataset is limited, but it works. Tested around 0.8, as i tried something different while training. can go up to 1.0, but i dont suggest going higher. likeness suffers if you go too low. i stayed above 0.7.

    -

    request a character | support so i can make more

    Tested at 0.8-1.0 strength (its best to keep it around 0.5 to start with).

    required triggers are

    Activator

    sml_santa,

    Character tags

    closed eyes, grey hair, wavy hair, hair between eyes, bangs, dark skin, gigantic breasts, heavy breasts, (plump:1.2), (thick thighs:1.35),

    Outfit

    santa costume, detached sleeves, bare shoulders, red leotard, cleavage, red thighhighs with white fur, santa hat, red gloves, belt, torn clothes,

    Extras

    gift box, sack, gift bag, dildo, sex toy, bell,

    Description

    FAQ

    Comments (5)

    SkurkgurkanFeb 8, 2025· 1 reaction
    CivitAI

    How did you deal with having such a limited dataset? Would love to train some rare characters myself, but I haven't been able to get good results with very few images. Great model!

    killedmyself
    Author
    Feb 8, 2025· 1 reaction

    im still tryna find the best method, but for characters that arent too complex, i usually chose a lower epoch. i used almost default settings besides flip aug and tag shuffling, so i dont actually do much besides prompting and using negatives where it suffers, and adjusting the lora weights when needed.

    also, general rule of thumb i try and follow: more repeats = solidifying aspects within the dataset (pose, style, stuff like that)

    hope this helps at least a little

    SkurkgurkanFeb 8, 2025

    @killedmyself Thanks, that does help! Do you crop the same image multiple times to add "new" data to the dataset (like using an upper body image, then cropping only the face to add to the dataset) or do you stick to what you have?

    killedmyself
    Author
    Feb 8, 2025· 1 reaction

    @Skurkgurkan for this lora, i did.

    but now, if i have at least a couple good clear images of the face, then i dont really have more than that and can focus on full body and whatever else i can get. so i stick with what i have, and use more than normal repeats during training if i need.

    SkurkgurkanFeb 8, 2025· 1 reaction

    @killedmyself Thanks for the answers! 

    LORA
    Illustrious

    Details

    Downloads
    240
    Platform
    CivitAI
    Platform Status
    Available
    Created
    1/16/2025
    Updated
    6/14/2026
    Deleted
    -
    Trigger Words:
    sml_santa,
    closed eyes, grey hair, wavy hair, hair between eyes, bangs, dark skin, gigantic breasts, heavy breasts, (plump:1.2), (thick thighs:1.35),
    santa costume, detached sleeves, bare shoulders, red leotard, cleavage, red thighhighs with white fur, santa hat, red gloves, belt, torn clothes,

    Files

    KMS_SML_Santa_IL-000003.safetensors

    Mirrors