CivArchive
    Casting couch illustrious - v1.0
    Preview 61170900
    Preview 61170889
    Preview 61170903
    Preview 61170902
    Preview 61170899
    Preview 61170901
    Preview 61170881
    Preview 61170886
    Preview 61170890
    Preview 61170895
    Preview 61170892
    Preview 61170887
    Preview 61170888
    Preview 61170894
    Preview 61170891
    Preview 61170882
    Preview 61170893
    Preview 61170884
    Preview 61170883
    Preview 61170896

    Casting couch illustrious,Thanks C0rn_Fl4k3s https://civarchive.com/user/C0rn_Fl4k3s/images?sort=Newest to explain to me how to make lora, with my image set

    Description

    FAQ

    Comments (6)

    settima_aiMar 3, 2025
    CivitAI

    Hello, I have a problem with this LORA: until I place a girl in the prompt (Lum from Urusei Yatsura) works all good. But when I try to place a boy (Ataru from Urusei Yatsura) it will include a girl in the scene, or will give a female body with boobs to the male and so on... I dont think is a problem of WAI Checkpoint. May be there were only girls in the dataset and the LORA believe there MUST be a girl on the scene?

    arrssenne
    Author
    Mar 3, 2025

    My datasets are always created without any humans precisely to avoid having a person in the image without specifically requesting it. So it's not because of the LoRA. Maybe lowering its weight to 0.6 or 0.8 could make it easier to control, or trying another checkpoint might help.

    settima_aiMar 3, 2025

    @arrssenne ok, thanks!

    LazmanMay 19, 2025

    @arrssenne Nice! Appreciate that foresight. Most people just put girls/women in even on loras that have nothing to do with humans at all, and thx to the extreme bias in all the base models, those loras will push females/female features into everything. I don't personally think it's right to push your bias onto others, but our society is such that people think I hate women just cuz I don't want to see them naked..

    LazmanMay 19, 2025

    @settima_ai There's a 'homo' lora that leans more towards males, though it's mainly for cartoon/anime, you can sorta push 2.5D with it. The base models have been trained with like, 90% images of naked/half naked women though, so even the gay models (what few there are) will often push females. Make sure you add 'female, woman, 1girl, boobs, breasts, tits, vagina, puffy nipples, to the negatives, but don't overdo the negatives or it'll start pushing that stuff.

    Even if you overemphasize males in the positive prompt, it'll go in reverse if you do that too much. Also play with CFG. higher or lower will give you desired results depending on the combinations of other factors. I typically stick between 3.5-7.2 myself.

    Also, avoid the words 'beautiful, anatomy' at all costs. At least 'anatomy' on it's own I know it bad, since most things tied to that are female. I've gotten away with using 'bad anatomy' in my negs though.

    it also sometimes helps to use male related loras to push males. But you've gotta be careful of the ones you use, since a lot of people these days seem to like men with pseudo breasts for some reason.. massive nipples and all, or even futas..

    I've personally got every form of 'female' in my negative tags on this site so that I mostly avoid loras that only display females in the samples, or males with overtly female parts. If a lora only has male samples, it's a better bet to be good, or like this guy, has no humans in samples.

    But, AI is purely random, no matter how well you fine tune it, at some point you'll get one micro-molecule of a thing off in the generation, and spend an entire night fighting female generations while not prompting for them at all.

    LazmanMay 20, 2025· 2 reactions
    CivitAI

    decided to try this for myself, one of my first times using a background lora, and it works fairly well, even below 1.0 str. I did find it somewhat difficult to spawn more than 1 character on the couch (about 1/5-1/8 chance), though given the randomness of AI, it would sometimes give several with multi characters in a row.

    I honestly think that's just the fault of AI in general though. I suppose 6-7gb models lack the precision to do mutli characters on command without perfect prompt/setup, and worse when trying different character types.

    But as I said, model works good, I did have to add the camcorder manually though, via 'camcorder on a tripod, camera facing couch' prompt

    LORA
    Illustrious

    Details

    Downloads
    1,266
    Platform
    CivitAI
    Platform Status
    Available
    Created
    3/3/2025
    Updated
    5/11/2026
    Deleted
    -
    Trigger Words:
    C4st1ng c0uch

    Files

    C4st1ng_c0uch-i.safetensors

    Mirrors

    HuggingFace (1 mirrors)
    CivitAI (1 mirrors)
    TensorArt (1 mirrors)

    Available On (1 platform)

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