CivArchive

    For T2V, use the high and low noise models at 0.8 - 1 strength.

    For I2V, use the high noise model at 1-1.5 strength with the low noise model at 0.9 strength.

    With I2V, the high noise T2V model adds more motion, if the I2V version isn't doing enough.

    Start with some form of "a woman uses her breasts to titfuck a man". You can tweak it, like "an irish woman with red hair titfucks a man with her breasts". You can also try adding "she is kneeling between his thighs" if you're note having any luck.

    Follow that with "A squirt of thick white cum oozes out from the woman's cleavage, it coats the man's penis" for a cumshot. Increase the low noise CFG if it isn't working.

    Examples are all done with T2V or I2V lightning LoRAs, at 1-2 CFG with 6-8 steps.

    Description

    I tried training a high noise I2V LoRA, but the T2V version is much better. Use the T2V high noise lora at 1.5 strength for the high noise pass, with this I2V low noise lora at 0.9 strength for the low noise pass.

    FAQ

    Comments (4)

    KwijiboeAug 31, 2025· 2 reactions
    CivitAI

    Tip to anyone wondering why you get changing facial expressons and double chin: I think the issue is your cfg, it says cfg 2 is for cumshots. Most of us are using like cfg 3.5. This model seems to be trained on casetopia or something, but when a dude "cums" from a titfuck the female is usually looking down(double chin) and making a worried look incase the cum hits her eye or something. I could be wrong

    GsssyikSep 19, 2025· 1 reaction

    we got a scientist up in here, watch out

    MuggleWizardSep 1, 2025
    CivitAI

    Where is I2V high model?

    aimalias
    Author
    Sep 1, 2025

    Check the update note - I tried to make one but the T2V works much much better, so I didn't bother.