[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.
  • 103 unique users in the last 30 minutes (including lurkers)




    First
    [1]
    Last

    All of my Stable Diffusion 1 loras is LoHa, so need LoCon extension for WebUI. ComfyUI, as i know supports it from the box.
    >>
    >>
    Pillar hangs

    Btw, what is "unacceptable content" on DeviantArt? My an several of my friends accounts suspended without any prior warnings.
    >>
    >>
    >>
    [lady knights 15-06-23 originals.zip](https://pixeldrain.com/u/QjYefq6C)
    >>
    How do I prompt hangings please
    >>
    File: 1.png
    (3.71 MB, 1536x1536)
    3889114
    >>992
    execution_on_gallows - main activation tag
    Trained tags:
    gallows_pillar_presence gallows_pillar_absence
    gallows_beam_presence gallows_beam_absence
    gallows_frame_presence gallows_frame_absence
    big_gallows_platform_presence big_gallows_platform_absence
    small_gallows_platform_presence small_gallows_platform_absence
    gallows_dropzone_above_presence gallows_dropzone_above_absence
    gallows_dropzone_below_presence gallows_dropzone_below_absence
    gallows_dropzone_front_presence gallows_dropzone_front_absence
    https://pixeldrain.com/u/8Lx8dSNN
    execution_on_gallows-[OriginMix][CS2][32-16-16-1][4E-4][725x1]-E-05.safetensors
    https://pixeldrain.com/u/zVkiGSNa
    execution_on_gallows-[OriginMix][CS2][32-16-16-1][4E-4][725x1]-E-04.safetensors
    https://pixeldrain.com/u/fdacTRnq
    execution_on_gallows-[OriginMix][CS2][32-16-16-1][4E-4][725x1]-E-03.safetensors
    >>
    File: 3.png
    (1.85 MB, 1536x1024)
    1942663
    >>
    Can you do it as a prompt or what can I get this result on Prodia?
    >>
    >>4750614 how do you make such guro pictures in stable diffusion (what model or lore is it just that I often see AI generated guro and don’t know how to make it myself)
    >>
    >>
    >>
    >>
    >>995
    Go to my pixiv, download originals from link in post description, look to metadata.
    >>
    >>996
    Lora posted above. Also use any good anime model from CivitAI.
    >>
    >>
    >>
    >>
    >>
    >>
    >>
    >>
    >>
    >>
    >>
    >>
    I can’t figure out how to prompt this are you able to do the same with Prodia.com cause there u can get real people
    >>
    >>1013
    I dont know what is "prodia".
    This is StableDiffusion.
    >>
    >>
    >>
    The download have been deleted. Any chance of a reupload?
    >>
    >>1017

    https://pixeldrain.com/l/UZUFKS4k

    Here are all released gallows loras, different training setting and models, but activation tags all the same and listed above.
    >>
    New LORA release:headdisplayv1

    https://drive.google.com/file/d/1HdgPTnWwvQ9EUSMOi6IWgvhHfojCOcMD/view?usp=sharing

    Because every time I post a picture, some people will say "I want to see the head", but I am getting too lazy to make img2img, so I decided to train a LORA that can complete the task at one time.
    >>
    This LORA is very similar to the previous before-after LORA. It is a before-after composition of the same character in different states.

    LORA code name:
    <lora:headdisplayv1:1>
    0.8~1.2 weight.
    CFG:7~12
    The base model is trained using kotosmix_v10. In theory, it can correspond to all types of body types.

    Start keywords:
    multiple views,
    headdisplay,
    displayshelf,
    wooden Display Shelf,

    According to the situation, you can add words such as "blood, guro, death, ryona, trembling, kneeling, seiza, "etc. to enhance the effect or determine the action posture of the character.
    >>
    In addition to 1girl, solo, which can show a single person's before and after beheading pictures, you can also try to use multiple girls and a horizontally long canvas to generate multiple people's collective beheading for public display.
    However, this effect is not stable, and it seems that I can only generate about three people at most, and I can't control it stably if there are more.
    >>
    This LORA also includes some unstable concept words. In the future, training materials may be added or separated. At present, it is difficult to produce pictures effectively, but you can try to challenge them.
    >>

    flyhead: The head is cut off from the body, and no head display shelf is generated.
    >>
    ghead: The head rests on the ground, usually by the feet.
    >>
    bhbehind: beheaded from behind
    >>
    impaledhead: a head or headless body impaled on a pike or rod
    >>
    >>986
    Am I doing it right senpai?

    I used sciencemix7 as the base model (it does good lolis) and your execution_on_gallows-[SD1.5][CS2][32-16-16-1][4E-4][725x1]
    I'm not sure if that's the right LoRa to use for a stable diffusion-based model?

    Prompt:

    ((hanged)), ((hanged to death)), ((noose)), bondage, crying, ((urine)), (cute blonde child in lingerie hanged to death), (((little girl))), nsfw, topless, flat_chest, lolicon, (beautiful), ((cute)), ((arms_behind_back)) (masterpiece), ((young)), ((child)), princess, lace stockings, execution, gallows, bound_ankles, (guro), asphyxiation, hanging, corpse, rope, (((noose around neck))), tongue_out, (((execution_on_gallows))) <lyco:execution_on_gallows-[SD1.5][CS2][32-16-16-1][4E-4][725x1]-E-18:1.0>
    Negative prompt: (low quality), sketch, (worst quality), disfigured, ugly

    Then I upscaled.
    >>
    4684730
    >>1027
    [SD1.5] version trained on stock SD1.5 and is better for semi-realistic, for anime better take NAI or OriginMix version.

    SRDA - last released one seems to work more or less good both on anime and realistic
    >>
    2809318
    >>1027
    SD is not AI and will not understand more then half of you prompt, here are more appropriate prompt example:

    (anime, masterpiece, best quality, high quality, highres, absurdres, intricate details),

    girl, small breast, tshirt, thighhighs, lace, panties, cameltoe,
    scenery, outdoors, nature, landscape,

    arms behind back, bound, restrained, (noose:1.3), (rope:1.3), pillar, beam,
    <lora:execution_on_gallows-[SRDA][CS1][16-8-8-1][2E-4][725x1]-E-15:1>,
    (execution_on_gallows:1.3), (gallows_pillar_presense:1.3), (gallows_beam_presense:1.3), (gallows_frame_presense:1.3),

    negative prompt: (worst quality:1.5)

    Recommended steps: 40, sampler: dpm++ 2m sde karras, cfg scale: 8, denoising strength: 0.3
    >>
    This >>1029 image is not generated with SRDA version, just some tags, here are example from actual SRDA lora version.

    SRDA states for Semi Realistic Digital Art.
    >>
    >>1029

    Where did you get the Lora <lora:execution_on_gallows-[SRDA][CS1][16-8-8-1][2E-4][725x1]-E-15:1> ?

    I know only those posted here >>993

    But your maybe is newer
    >>
    >>1032

    wow! thanks!

    I didn't expect so many files, are they all different in trigger words or they are all the releases of the same file?
    >>
    Would you consider making a shooting lora? I'd be willing to pay for the service
    >>
    >>1033
    Yes, mostly same.

    >>1034
    Already done, check my pixiv.
    >>
    Pretty new to Lora's & stuff, this is my first lora. Not explicitly guro but Ryona is close enough i hope. Made a Bellypunch Lora. Results are a bit on the gacha side so I think I need to do more research. Trigger is "harapan". Having (1boy:1.5) or some high modifier for the boy seems to make generation more consistent. Also seems to push towards exposed stomachs, I typically use a weight of 0.7~1.0. Will goto a default "shocked" facial expression if none is given in the prompt.

    https://drive.google.com/file/d/1-hCtolpEPWMJ8TjROJU_Wg7YsIV9Zooi/view
    >>
    >>1036
    2boys seems to generate more consistent punches
    >>
    Bondage Dreams SD model release
    Set of bondage and torture tuned custom models in different styles.
    Default modern, simplistic, retro-alike anime, and two semi-realistic, may be ever can be brought to realistic using prompt keywords and loras/embeds but overwhelming part of training data is anime, so expect bias.

    Huge-breasted NSFW sample grids:
    Darkness from Konosuba - https://pixeldrain.com/u/GSXMdHdm
    Gaping, large insertion, weight 1.0 - https://pixeldrain.com/u/VfLdo37e
    Wooden horse (no loras, 1.0 weight) - https://pixeldrain.com/u/GpXRpssd

    Good compatibility with bondage loras, but overburned loras will become ever more overburned when overlaps with already existed same-concept model data.

    IMPORTANT: I know everyone now hyping over SDXL, but SDXL have almost ZERO capabilities in NSFW, and still need to be hardly finetuned. This models still SD1-based and capable up to 768x768 before upscale.

    Download safetensors from huggingface:
    https://huggingface.co/Aleteian/bondage_dreams/tree/main
    >>
    >>1038
    Haven't tried yet, but have a question: can you make it a LoHa, to be used with other models?
    >>
    >>1039
    There already plenty or loras with such data, from me and not.
    >>
    >>986
    >>1038
    Hello!

    If you're still monitoring this thread: Thank you for making and working on the LORAs!!! Your results are very useful for Quick Pose Ideas for sketching for fun! I cannot use AI myself on my computer, so Im looking at images generated by you and other people and they are very helpful!

    Hope to see more of your images and I hope more people make and share their AI generated images like these too!
    >>
    >>988
    Drawn from this AI image, it's one of my favourites!

    Also currently DeviantArt will probably ban any guro and snuff content, even without NSFW. The current Terms of Service forbid sexual violence fetishes. All my favourite artists got deleted.
    >>
    >>1036
    >Bellypunch Lora

    Wow. This works Very well. Top 3 LORAs ever published on this board.

    You made my day anon.
    >>
    >>1036
    >>1037
    How could I have missed it 2 months ago. We need a sticky thread that collects all useful LORAs
    >>
    >>1044
    >>1043
    >>1036
    The trick is using many negative embeddings to correct the errors.
    >>
    >>1045
    It even combines well with character LORAs!!

    Thanks anon
    >>
    >>1038
    >>1036

    How big was the dataset you guys used for trianing?
    >>
    >>1047
    400-800 images usually
    >>
    >>1048
    thanks
    >>
    >>1049
    I even managed to make it work with IRL models. The results require a lot of cleanup.

    but something like:

    <lora:BellyPunchLora-10:1.3>, (harapan:1.3), (2boys:1.2) , (large breasts:1.2), full body, (sexy:1.1), gasping , (open mouth:0.5), (pain:0.3), tears, surprised, naked, (boxing gloves), punching, <lora:BoxingWomenV3:0.15>,
    (Russian), model, Braids, ruined,
    sharp focus focus, (boxing bag:1.0), (Motion lines), Punching bag, [dripping sweat],
    Negative prompt: bare hand, closeup, old, fat, ugly, CGI, (poorly drawn hands:1.1), poorly drawn face, bad art, low res missing fingers, extra finger, long finger, ((blurry)), (low quality, worst quality:1.3), 3DCG, clone, mulitple nipples, child, multiple girls, cartoon, comic, anime, drawing, monochrome, (badhandv4), (negative_hand-neg), bad-picture-chill-75v, ad_prompt_version2-neg, easynegative, ng_deepnegative_v1_75t, verybadimagenegative_v1.3 , mask
    Steps: 44, Sampler: DPM++ 2M Karras, CFG scale: 8, Seed: 516621542, Size: 960x1264, Model hash: ef76aa2332, Model: realisticVisionV51_v51VAE, Denoising strength: 0.37, Lora hashes: "BellyPunchLora-10: 9a68f183a72a, BoxingWomenV3: 0f77469b6595", Version: v1.6.0
    >>
    >>1027

    Sciencemix? I can't find that model online. do you have a link by any chance?
    >>
    >>1047
    I made the bellypunch lora, I used 100 images (200 if you count flipped version). I just followed a Lora guide I found on civitAi. I've seemed to have lost the exact link but I'll post it again if I find it.
    >>
    >>1036

    Bruised Face Lora

    Trigger: n_bruisedface
    Optional (Supplementary): tears, one eye closed, crying, nosebleed

    Trained only on girls so not sure how well it will transfer over to male characters if anyone is interested in doing that. Attached XYZ grid was using based64, Lora weight seems to work slightly differently on other models. When using the NAI model I found higher weights like 2 or 2.5 running into problems (I prefer using 1.7 on NAI, and 2.2 on Based64)

    Also figured out how XYZ plots work, and that you're supposed to test out the different epochs to see which one is best.

    https://drive.google.com/drive/folders/1I-3KoWECfh_U6LZsPzDefr4VDz7x-5Ah?usp=sharing
    >>
    >>1053
    Should work well with other Loras' as well. Haven't encountered any major issues as of yet.
    >>
    >>1054
    Kind of works with the bellypunch lora as well.
    >>
    Does anyone know where I can find a good resource for guro loras, specifically masculine ones? No shade against those who like women, but it's all I seem to find whenever I look for amputation, hanging, decapitation, etc.
    >>
    >>1056 bump somehow juggernaut or sdxl dont do guro well
    >>
    Well anon made ac omiplation of the guro lora he foudn aroudn here

    https://rentry.org/R-18G_Stable_Diffusion

    SOme more needed to ba dded yet,

    A ryona lora came out on civitai but was removed
    >>
    OP's pixiv is gone. Does anyone know what happened or where they might have moved to?

    Also https://www.pixiv.net/en/users/16453490 is gone (they also made a lot of relevant loras) same questions here.

    [Top]

    Delete post: []
    First
    [1]
    Last