CivArchive
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined

    Make em droopy

    V4_Anima: Trained on full release

    V3_Anima: Preview 3

    V2_Anima: trained on the new dataset. preview 2

    V1_Anima: trained on anima preview 2


    V4.1_Klein_9b: Trained for longer, more consistent and higher quality output. it seems like the stretch marks are here to stay, though they appear less and are far less of an eyesore than in previous versions and do appear more natural, no more thanos chin titties.

    V4_Klein_9b: Trained on the new dataset. stretch marks are largely subdued but not totally gone, with a more natural look when they do appear.


    Klein really wants to fixate on the stretch marks and blemishes, i had to fully denoise everything in gimp and lightly blur some images to get this to a point where I'm at least kind of happy with it. I do still recommend pairing with nipplediffusion it gives the nipples a bit of an upgrade and compensates for where the lora is lacking. but overall the lora performs much better on its own.

    V3_Klein_9b: increased the rank. output is higher quality. this maybe isn't as big of an upgrade as i thought, some gens make it seem better and then others make it seem worse. i probably amplified both the good and bad of the dataset by increasing the rank. still try it out though if you want to.


    V2_Klein_9b: mostly fixed the stretch marks issue.

    V1_Klein_9b: seems to have really overtrained on the stretch marks, i find nipplediffusion
    to help with this though. otherwise it works very good.


    V1_ZiB: base


    V4_ZiT: No more teardrop nipples.

    V3_ZiT: Can now be prompted for small, medium, large sizes as well as flattened, tubular, rounded. much better overall quality, a little closer to my flux version. it won't go to the extremes of my 2 previous attempts but it gets close and i'll work on improving that for my next release, was mainly trying to get the quality up with this one.


    V2_FLUX: Trained on a much larger dataset, and then merged with v1. recommended weights are 0.10-0.80. It has a pretty decent sliding effect with lower weights focusing more on smaller or more normal/natural breasts and higher weights making em hit the floor. Enjoy.

    Description

    increased the rank

    FAQ

    Comments (3)

    pmulefanMar 30, 2026
    CivitAI

    just curious, for Klein9B v1 v2 v3 LoRA, is all images in your dataset has the nipples and areolas facing front? It looks not correct, with the long saggy floppy hanging breasts, the nipples and areolas should be facing downwards. Prompting the areolas to face downwards does not do a thing.

    Soup_Fusion
    Author
    Mar 30, 2026

    the dataset consists of images with them all facing forward yes, i actually prefer this look even if it's not the typical representation. Plus some real, long sagging breasts do have them a little higher on the breast. maybe I'll create a separate lora where they're facing down, trying to focus on getting the quality of this one up though, mainly the low quality stretch marks.

    pmulefanApr 2, 2026

    @Soup_Fusion looking forward to the nipples facing downwards version, thanks.

    LORA
    Flux.2 Klein 9B

    Details

    Downloads
    338
    Platform
    CivitAI
    Platform Status
    Available
    Created
    3/30/2026
    Updated
    6/13/2026
    Deleted
    -