CivArchive
    Thigh sex / thighjob - v0.1
    NSFW
    Preview 1265144
    Preview 1265142
    Preview 1265140
    Preview 1265146
    Preview 1265145
    Preview 1265185
    Preview 1265180
    Preview 1265179


    Support my work on Patreon and Ko-Fi and get access to tutorials and exclusive models.
    Rate and leave a like if you enjoyed it, and follow for new models!


    LoRA trained to generate thigh sex.

    Trigger: 1girl, 1boy, thigh sex, penis

    Description

    FAQ

    Comments (5)

    GaspatcherJun 24, 2023· 7 reactions
    CivitAI

    I love this position, pee-pee feeling crushed like that.

    BixBit11Jun 26, 2023
    CivitAI

    May i ask how in the world are you able make so many models in such a short time? I'm really impressed. When i'm working on a model, the tagging is what takes ages for me. You seem to be doing this so quickly, yet with top results. Might you be hiding the secret formula ?

    datSatoJun 26, 2023· 1 reaction

    You can use something like "Grabber" to scrape images from boorus in bulk (or hand select images) - and you can configure it so that when you save the images using Grabber, all of their booru tags are automatically saved into a matching .txt file.

    From there, you can use Automatic1111's built in tag editor to prune or edit tags in bulk as needed.

    You can also just let WD tagger interrogate all your images and add tags automatically, but IMO the results are more hit or miss.

    psoft
    Author
    Jun 26, 2023

    @BixBit11 , it's very basic tbh. the suggestions from @datSato are definitely good. preparing a dataset can get tedious, so anything you can use to streamline the process is better. and using Grabber or similar to get tags directly from a booru website is better.

    once I get an idea I get a dataset of 100-400 images, often manually since I'm picky. no cropping!

    then I use WD1.4 auto tagger in kohya's gui, and I add/prune tags with BooruDatasetTagManager. tagging is important so you should make a effort to see if there aren't wrong tags. the bigger the dataset the less likely a sloppy tagging will have a big influence.

    i train using kohya gui, and I start with very similar parameters. I do a 512x512 training always, and sometimes a 768x768.

    then during training of a lora, I'm simultaneously doing another dataset prep.

    having a 4090 also helps since each lora takes between 20min-1h.

    but I have to say that I only the ones I'm quite happy with, which is probably less than 15% of what I have.

    datSatoJun 26, 2023
    CivitAI

    Pretty good. It's still not 'perfect', but, I think I prefer it over the other thigh sex/sumata models I've come across so far.

    LORA
    SD 1.5

    Details

    Downloads
    15,203
    Platform
    CivitAI
    Platform Status
    Available
    Created
    6/24/2023
    Updated
    6/11/2026
    Deleted
    -
    Trigger Words:
    1girl, 1boy, thigh sex, penis

    Available On (1 platform)

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