CivArchive
    Preview 27189117
    Preview 27189443
    Preview 27189847
    Preview 27189961
    Preview 27190307

    thx to GraXXXho - 10k Buzz


    if you want more , eg only grab no suck or side view, positions... -> give me buzz for training
    so thx for the buzz, i can train more :*

    ALL training is based on handpicked images (without watermarks). Promt was created with tuned joycaption and corrected manually
    trained with 1024pix, if you notice noise raise up resolution up to 1800pix


    every lorra does what it says
    -> *** ~25steps *** EULER AND (Schedule) DDIM or SGM Uniform or Beta
    HINT: GGUF dont work that good (rise up weight), and they are very slow, take one of my suggested fp8 models
    AND for more complex poses i suggest to use a distilled model with CGF ~2
    if you Raise up "CFG Scale" 2 with neg-prompt: "photo, photograph, realistic" you can force anime/cartoon style (only on special models distilled or use my distilled lora)

    play with ALL ages, but i dont train forbidden stuff

    rimjob(v2) - pls read my prompts on info symbol on images (i), it is only that working what my images shows, DPM++2M and SGMUniform works well

    cowgirl(anal) - seems okay, sometimes body twisted, call mans hair and beard to make him more visible, no cum

    doggy-anal - one of five are okay - its a complex pose, try to add some more details to the scene, may increase disitilled CFG to 5,

    missio_v2 - more luck with good poses, trained on natural language

    dildo_v1 - first shot, read my prompts (is not that easy to train 4-6 positions at once)

    cowgirl(vulva) - sometimes weight 0.7 better than 1. start with my prompts and change it slowly

    reverse cowgirl(anus) - sometimes bodies are fused or limbs are missing, but otherwise quite okay, start with my prompts! -> for free folks ;) ... (i have trained vagina penetration but it doesnt mater FLUX dont know)

    noblow - 1 of 5 are quite ok, cum works not that good, (training is very confusing)

    noblow_v2 - you need more explained prompts to get a good image but may be its more felxible (trained on more epoches and joy-captioned) maybe dont call "man", 1 of 3 are okay

    missio - start with my prompt (important: view from below or view from above), iam not that happy, only 1 of 5 are good
    use -> new checkpoint

    side view - play with it (dont trained big/small)

    pov_V2 - simple prompt, is more trained, so u can use it stand alone, in some cases weight 1.2

    pov_V1 - simple prompt, in some cases i use the loras two times for better images
    eg: <lora33:0.8> <lora15:0.4> or vs
    play a little with the weight

    (i dont trained, pubic hair and text and whatever he/she sucks ^^)

    keywords for blowjob_v1/v2: blowjob, pov, deepthroat, small penis, big penis, huge penis, sucking deep, mout stuffed, grab penis, long big black penis, lot of cum on face


    you can play with all ages and animals, cum on face only sometimes work, for black penis may add "black mans" and "white skinned woman"
    hybrid seems heavy, because no weights in FLUX... tell me if you have an option


    Euler: with simple, SGM Uniform, Beta -> i suggest DDIM
    steps: ~25 if you have power raise up to 40
    1024x704 / 704x1024 or 1200*800 / 800*1200 - most times portrait
    weights like pony for prompt words ":1.6" dont work well


    Be careful with publications and creations with well-known personalities or minors. I assume no responsibility!

    Description

    only no blowjob
    grabs, licking,bit cum (force it with "lot of creamy")
    -> use noblow_v2

    FAQ

    Comments (12)

    32BitshifterSep 1, 2024· 10 reactions
    CivitAI

    There's no reason to be using Flux over realistic Pony models if you're going to hit it in the head and drop the quality to pre-SDXL days.

    sevenof9247
    Author
    Sep 1, 2024

    be patient ;)

    FLUX is out 2 weeks OMG and you are not a newcomer ^^

    32BitshifterSep 1, 2024

    @sevenof9247 Flux has been out for a month so that is only 250% off, and right back at you; why 6 loras of wasted effort? Be patient.

    JanetSep 1, 2024

    what's up, the dataset was low quality or what?

    sevenof9247
    Author
    Sep 1, 2024

    @32Bitshifter oc iam with you ... but push it a bit ^^

    sevenof9247
    Author
    Sep 1, 2024

    @Janet maybe ... i have no idea what is more important, caption or image and one training cost 3000buzz ^^ and one day time .. so its hard to get 5 versions and compare them .. so i dont do that at the moment

    plkSep 1, 2024· 6 reactions

    He's not wrong. Every Flux LORA should at least put out what SDXL is capable of, and surpass it, but people are in such a hurry to be the first that they're putting out very poor LORAs. I've also seen people putting out LORAs of stuff that Flux is already well capable of, which is just another example of the problem with Flux LORAs. What people need to do is have the right mindset to actually use better and more training datasets, but we're instead seeing datasets being used for Flux that are even lesser than those used for SDXL.

    az420Oct 9, 2024

    TBH Pony Realistic doesn't cut it, it has the rough idea of what real looks like but those faces :S
    I recently switched back to Pony after playing with Flux for a while and... realism is horrible! Pony is passe y'all! Time to move forward, and complaining about quality this early on? He's putting in effort, show some appreciation!

    rando2048Oct 22, 2024· 1 reaction

    @az420 FWIW, if you refine at .6 and highres at .4-.5 with a nsfw sdxl model for both, you can get better pony images. Sadly you still loose a lot of creativity from normal SDXL models since the core generation is still done with pony. It does fix the face issue to a large degree though. You can also [from:to:0.6] for prompt words that don't do well in one or the other checkpoint.

    az420Oct 22, 2024

    @rando2048 interesting, I'll try that out.

    az420Oct 23, 2024

    @rando2048 what's a good nsfw SDXL in your opinion?

    TwinklingSquidSep 1, 2024· 4 reactions
    CivitAI

    Thank you for the NSFW models! They're performing similar to how Pony and SDXL models performed at first, but we have to start somewhere and if there are no toys to experiment with, people will stop trying to train the model!