CivArchive
    POV Doggy & Reverse Cowgirl - v1.0
    NSFW
    Preview 119730274
    Preview 119730273
    Preview 119730271
    Preview 119730277
    Preview 119730269
    Preview 119730275
    Preview 119730270
    Preview 119730272

    This lora depicts a woman having POV doggystyle or reverse cowgirl sex. It was trained utilizing Flux Klein 9b-base. The 1500 step lora was trained using 55 800x1200 portrait images. Even though the lora was trained for portrait orientation, it still works pretty well in most landscape resolutions as well.


    Example Prompts

    Sitting: pov vaginal doggystyle, a man inserts his penis into the woman's vagina, she is looking at the viewer over her shoulder eyes fixated on viewer, the mans thighs are barely visible at the bottom edge of the frame

    On Hands and Knees: pov vaginal doggystyle, a man inserts his penis into the woman's vagina, she is looking at the veiwer over her shoulder eyes fixated on viewer, the woman is on her hands and knees on a bed, the naked mans thighs are in view at the very bottom edge of frame, from above, the womans knees are spread wide to each side with her feet out of view off each side of frame


    As with other Klein generations there can be some anatomy issues with hands/feet which can be fixed with some quick prompting.

    This is my first uploaded lora so if you like it feel free to show some love. Looking to create some more loras in the future.

    Description

    FAQ

    Comments (3)

    zoom83Feb 27, 2026· 1 reaction
    CivitAI

    Samples?

    Velocity0
    Author
    Mar 3, 2026· 1 reaction

    Samples were posted, however, they have been marked as "flagged for review" by Civ since initial upload and are not being shown to users. Will try to update again to see if that fixes the issue.

    pmulefanMar 7, 2026
    CivitAI

    In view from above, on all fours, it is very diffcult to get the woman's lower legs to appear correctly, most of the time it is inside the bed even with different seeds, tested with i2i edit. If this is something you could improve in v2 would be great. Sitting pose works most of the time.

    Another thing I noticed, it also messes up with the head and body proportion, usually ends up with big head and small body, which is noticable in your examples, probably due to the training dataset that contained images with wide angle fisheye lens that distorts the perspective.