[Return] [Bottom]

Posting mode: Reply

BBCode
(for deletion)
  • Allowed file types are: gif, jpg, jpeg, png, webp, webm, mp4
  • Maximum file size allowed is 25000 KB.
  • Images greater than 255 * 255 pixels will be thumbnailed.
  • 98 unique users in the last 30 minutes (including lurkers)




    https://civitai.com/models/926992/rbqinori-style-lora-for-illustrious

    tag:rbqinori,guro,blood
    optional:disembodied___,severed___,disembodied___ in/on ___,etc.
    You can download training text set in civitai and find some tags.
    Viewing last 50 posts
    >>
    >>6623
    Thank you, your stuff is awesome
    >>
    >>
    >>6348

    I want another meal to eat. Your mother died there is your Cannibalism by Alsina Dumitrescu died
    >>
    >>
    >>
    Wow 😍

    >>7139
    >>7140
    >>
    >>5791
    Can you share me prompt
    >>
    >>8519
    it has exif metadata, You can click it to view original img and drag it on
    PNG info in webui
    >>
    >>8532

    1girl, komi shouko, komi-san wa komyushou desu, midrbqinori,
    long hair, black hair, large breasts, white bra, diagonal-striped skirt, red skirt, cumdrip, guro, quadruple amputee, blood, bloody arm stump, bloody leg stump, hanged, asphyxiation, rope around neck, unconscious, (tears), open mouth, blank eyes, torn clothes, head tilt, after rape, chalkboard in background, writing on chalkboard, tally, bruised stomach, bruise, exposed bone,

    , masterpiece, best quality, amazing quality, very aesthetic, absurdres,

    <lora:rbqinori_v4:1>
    >>
    your LORA are amazing.

    >>8532
    some of your latest pics do not have metadata, at least no metadata i was able to find
    >>
    >>7139
    can you post the metadata for this pic?

    Also some other pormpts that work well with lazyman.

    ideally could yo ushare the tagged training dataset, so i can modify that to find the best prompt?
    >>
    >>8706
    lazyman, uzumaki kushina, naruto \(series\), blue eyes, red hair, long hair, hairclip,

    1girl, tongue, solo, open mouth, tongue out, toenails, teeth, makeup, toes, barefoot, feet, veins, half-closed eyes, rolling eyes, decapitation, guro, severed head, severed foot, severed head hanging, severed foot hanging, ((cum on feet)), rope around ankle, blood, cum, cum on severed foot, five toes, outdoors,



    , masterpiece, best quality, amazing quality, very aesthetic, absurdres,


    Datasets txts only
    https://mega.nz/file/nJcBzJSY#qgq_rb6XSk4MDM5gk2AMsfI5662-cCv9WMnXcxpInu8
    >>
    >>
    >>8830
    new version, again.
    v8
    https://civitai.com/models/1298072

    better concepts
    >>
    >>8831
    MUCH better severed pussy, (used as 'disembodied pussy')
    >>
    >>
    >>8832
    This looks amazing! Can't wait to test it out.
    >>
    >>
    >>
    >>8831
    What do the different suggested style tags do?
    >>
    Amazing. I never managed to make a lora with a working disembodied pussy (i kept tagging it as as "severed pussy" and "severed cunt") maybe this is why
    >>
    What do youbuse as the base model? NoobAi or Illustrious?

    And what differences have you seen when training on one or the other?
    >>
    >>8843
    they simply have different style(as original art style changes), you can check the sample ones with lana's mother. You don't have to use them if you want.
    those tags also makes different style on stumps and injury, bruise etc.

    >>8846
    I used to use "severed pussy", just like you. But lora can't distinguish well between "severed pussy" and "severed breast".

    >>8849
    I used ilxl 0.1. Its datasets were too large for me to handle on my local PC.
    I haven't had much chance to test it on other base models. I once trained with different datasets on NoobAI VPred, but the results were pretty disappointing.
    >>
    >>8829
    it's awesome. Not very stable. but 1/4 come out great.

    >>8831

    this one weights 800MG. have you experiemnted with double that amount, 1.4 GB to see if ti fit more concepts?
    >>
    What do you use for the initial tagging run of the picture?
    I still use Waify diffuser 1.4

    anything better came out recently?
    >>
    >>8897
    I haven't tried. I think 1.4GB is too much for lora, but I'll give it a try later.

    >>8908
    I'm not sure if this is what you're asking, but I use wd-vit-tagger-v3.
    I tag a lot manually, so I think it doesn't matter that much.
    >>
    zoe
    >>
    >>
    >>
    >>
    File: 6-2.png
    (7.45 MB, 1920x2400)
    7809882
    Here’s my feedback after testing it myself:

    This LoRA(v8.) can generate most of the common guro concepts like amputation, hanging, and severed limbs or heads. It also handles some more niche concepts, such as disembodied pussy or severed limbs inserted into the ass/pussy. The common themes tend to be more stable in text-to-image generation, while the more specialized ones require a bit of luck or inpainting to get right.

    One issue with packing so many concepts into a single LoRA is that it can get quite noisy: using a single "guro" tag may trigger a bunch of unrelated elements that mess up the composition. To work around this, I usually apply a low weight (around 0.2–0.3) to "fish out" the layout I want, then use i2i/ControlNet to refine and stabilize the result.

    Another feature is that this LoRA doesn’t alter the base art style if you don’t use the style trigger (e.g., “rbquinori”), which makes it very convenient if you’re only aiming to apply the concepts without changing the aesthetics.

    Great work on training such a concept-rich LoRA. There are still plenty of concepts I haven’t explored yet, and I’m definitely planning to keep using it.

    Here are some examples of my Pixiv works using this LoRA, with several other guro LoRAs(Ayaswan,PDV6_Guro_Refine,YB). ControlNet, i2i, and many layers of inpainting are used, so it's probably not reproducible with the metadata embedded in the images.
    >>
    File: 7-2.png
    (7.19 MB, 1920x2208)
    7538677
    >>
    File: 8-2.png
    (6.83 MB, 2208x1920)
    7166580
    >>
    File: 8-3.png
    (7.13 MB, 2208x1920)
    7475153
    >>
    File: 3-1.png
    (7.01 MB, 2208x1920)
    7351268
    >>
    This is insanely potent lora, that can work wonders. I wasn't trying hard, because i didn't have any particular idea in mind, so i just ran some text2img without controlnets, inpaintings and other stuff, and still got a very good results!
    >>
    >>
    >>
    Is there tag list for xingzhe lora?
    >>
    I have mixed feelings seeing the art of this great master being used. However, I see that the results are approaching Bansye, but they leave much to be desired because his human creativity still can't quite reach its full potential.Each image is just a plagiarism of the great master's style.However, he must be paid.
    >>
    >>10205
    >>10206
    >>10207
    These are amazingly sexy. Those last two especially! Looks like the start to a good story.
    >>
    >>5279
    Could you reupload this? Civitai deleted it.
    >>
    >>13221
    Not him, but it looks like it was uploaded to seaart too https://www.seaart.ai/models/detail/9b7664bf3e0cff8949fe3e469bce484a
    >>
    bro,can you train this rbqinori lora for anima model?really want it,please
    >>
    >>8829
    Is there a tag set for Xingzhe's lora? I can't find it anywhere
    >>

    [Top]

    Delete post: []