CivArchive
    Imperfect_LoveXL (True 2.5D NSFW) Alpha 2 Edition - v18
    NSFW
    Preview 2902463
    Preview 2902455
    Preview 2902476
    Preview 2902475
    Preview 2902472
    Preview 2902458
    Preview 2902461
    Preview 2902459
    Preview 2902456
    Preview 2902457
    Preview 2902462
    Preview 2902460

    Please, if you like what I'm doing, support me down below!

    All images above were using the Mile high Styler and with minimal negatives. By using negatives related to styles or even just using LoRA, much greater style results can be achieved. Clip Skip -2

    https://www.buymeacoffee.com/tripleheadg

    Just like ImperfectXL this one is hit and miss. But when it hits, it hits hard! Also now capable of different styles.

    I've spent literally every waking hour, since releasing the previous version, perfecting this next release.

    I now merged my model with semirealNsfwXL_v10 and a couple more LoRA along with a bunch of my own models. This is like version 50 of subsequent merges. It was SO MUCH EFFORT to get the eyes decent as well as have blowjobs and the sex both in their at the same time without one mutating the other. Enjoy.

    I took nekoxlNsfwMix_v1 and merged it with Imperfect_XL and LoveXL along with a few LoRA to create the ultimate 2D, 2.5D and 3D NSFW workhorse.

    This one works very similarly to ImperfectXL except that I've managed to jigger it so that the default output is that of 2D/3D women instead of realism.

    Description

    FAQ

    Comments (11)

    2501Oct 12, 2023
    CivitAI

    Some of the faces are fugly XD, but good merge.

    Triple_Headed_Monkey
    Author
    Oct 12, 2023

    Yeah hopefully I can clean it up in a few days time :D

    Glad you like though!

    Triple_Headed_Monkey
    Author
    Oct 12, 2023· 2 reactions

    I'm not going to lie, I didn't want to mess too much with the faces yet in case it ruined the NSFW bodies lol.

    It's awkward because the best quality faces etc. come from my non-NSFW models, which means that merging with them doesn't produce the results you would expect. As the 2 models don't have a point of reference between their data.

    RogueLikelyOct 12, 2023
    CivitAI

    Only getting bad gens out of this. Grey noise. Can't seem to get anything out of it no matter what I do. 1024 x 1024, Euler, no loras, no VAE, no refiner, simple prompt. Nothing works.

    Triple_Headed_Monkey
    Author
    Oct 13, 2023

    Strange. Try setting clip skip to 2. I've had some odd issues with clip skip 1.

    RogueLikelyOct 15, 2023

    @Triple_Headed_Monkey  Sorry for the late reply, but I'm using ComfyUI and I get an error if I attempt to set the clip skip to anything but -1. Not sure if its because of the efficiency loader node I use or what. Is there a particular reason this model is set up to need clip skip on anything other than the default?

    Triple_Headed_Monkey
    Author
    Oct 16, 2023

    @RogueLikely Basically part of the merge process involved merging with the inpainting model which handles clip slightly differently. As a result of this it's just made clip skip -1 output weird fuzzy images of nothing. But otherwise improved the outputs in other aspects. Clip skip -2 is the minimum functional Clip layer.

    RogueLikelyOct 16, 2023

    @Triple_Headed_Monkey So you're saying I should set the Clip to -2?

    Triple_Headed_Monkey
    Author
    Oct 16, 2023

    @RogueLikely Yes indeed. Then you should get images out of the model.

    RogueLikelyOct 17, 2023· 1 reaction

    @Triple_Headed_Monkey thank you, and this indeed did work. One more question. It seems all the backgrounds are blurred, almost as if by an extreme motion blur. Would that be caused by something I'm doing?

    Triple_Headed_Monkey
    Author
    Oct 17, 2023

    @RogueLikely I'm not sure. Sometimes the model can produce camera effects. If you try prompting things like "Bokeh" and "depth of field" in the negative it can help though.

    Checkpoint
    SDXL 1.0

    Details

    Downloads
    459
    Platform
    CivitAI
    Platform Status
    Available
    Created
    10/11/2023
    Updated
    5/11/2026
    Deleted
    -