CivArchive

    Adds a little jiggle and bounce to natural body movements. No caption is needed, but adding "jiggle" or "bounce" can add some extra movement.

    I highly recommend using both high and low noise versions for local generation. High noise affects larger movements, and low noise affects fingers, indentations, and subtler jiggle.

    On-site generation cannot use both LoRAs. I recommend high-noise if you just want bouncing and low noise if you want more subtle results.

    Description

    For Wan2.2! I recommend using both high and low noise models, but here is what each one seems to excel at:

    • High noise: bouncing, swaying motions, larger breast movements.

    • Low noise: Breast indentation, fingers, subtle jiggles and motion.


    No specific prompt needed, as the model is trained to have better breast physics in general. Bounce, jiggle, massage, squish, massage, etc. should all generate much more cleanly now.

    FAQ

    Comments (9)

    Alkorr64Feb 7, 2026· 1 reaction
    CivitAI

    awesome. Will there be a I2V version?

    ComfyTinker
    Author
    Feb 7, 2026· 1 reaction

    The T2V versions work for I2V as well. I'm not sure the on-site generator works well with Wan2.2 LoRAs, but for local generation I have used them for both I2V and T2V without issue.

    Alkorr64Feb 7, 2026

    @ComfyTinker oh ok, cool

    DennyDan84Feb 7, 2026· 1 reaction
    CivitAI

    Nice to see this on WAN 2.2. Thanks!

    lillblues78249Feb 9, 2026· 2 reactions
    CivitAI

    Very nice lora! I tested a few times with the wan2.2 version and found that it tends to smear and faint the nipples. Lower the strength on both high and low will help a little but not too much.

    ComfyTinker
    Author
    Feb 10, 2026

    My dataset included mostly clothed breasts, so that may be an unintended training. I can look to merge a diverse nipple LoRA for a future version.

    lillblues78249Mar 1, 2026

    Also noticed that the wan2.2 version tends to change the identity of character lora. The v7 wan2.1 version doesn't have this problem.

    ComfyTinker
    Author
    Mar 1, 2026

    @lillblues78249 Yeah the wan2.2 training is a whole different process I'm still working on learning. I'm doing some testing to see if I can figure out where that's trained.

    Opal766Mar 19, 2026
    CivitAI

    I got blury video when use this lora with T2V