all / g / s / fur / art / ai / 3dcg / f / rp / req / p2p / dis / lit / kaki chat
[Settings] [Home] [Catalog] [Search] [Thread list] [Stats] [Reports] [Watcher] [PMs] [Admin]
[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)
    File: 00027-2216965335.png
    (1.46 MB, 1024x1280) ▶
    1534751
    No.4264 [Report] AS
    ▶
    Lora request rules got buried in my previous thread so making a new thread with them clearly at the top.

    At the time of creating this thread I only make PonyXL lora's so the following will mainly be for PonyXL. But since the "best" checkpoints are always changing make sure to check through the thread if I'm doing anything else if. Should be easy enough by just Ctrl+F searching my name and quickly skimming through the thread.

    I want to accept as many Lora requests as possible but to make the best use of my time and yours I request that you provide the following when you make a Lora request.

    1. At least 20 images high quality images that clearly show the concept/character/clothing you want in your Lora (quality > quantity)
    a. By high quality I don't mean high resolution. Anything above 512x512 is good enough (That said higher res is always better so avoid downloading samples if possible, only one dimension needs to be above 512. So something like 512x368 is fine)
    b. Shows what you want trained clearly. If many images partially obstructed it may not train correctly
    c. Avoid full comic pages, and things that may have a lot of detail. If its a part of a comic page, just crop it out and send me the panel. I crop out a few myself but if you send something like 50% of your images as comic pages I may decline the request.
    d. Images should be of a anime/cartoon style, 3D renderings, etc. Lets avoid realistic images whenever possible (They're also guro-chan rules)

    2. Upload all of your images onto a easy to download file service. Any is fine.
    a. Catbox seems like the most popular file upload method but anything is fine so long as I can get them.
    b. I would recommend zipping it with a password if you're going to use Catbox. (They typically don't allow anything super hardcore guro so just to make sure its contents are hidden)
    c. And obviously if you use a password make sure to share it here

    Where you get the images doesn't matter but I would recommend gelbooru or pixiv. When using gelbooru make sure you download the full image and not the sample you typically see by default. If you need any help gathering images or don't know what tags to use, just ask here and me or other users can share tags/pointers for getting them.

    There is no need for you to rename, tag or sort the images yourself. Although appreciated I will be doing that part myself.

    Check out the other stuff I'm doing:
    Pixiv: https://www.pixiv.net/en/users/32339993
    Fanbox: https://www.fanbox.cc/@entropy-game
    Viewing last 50 posts
    >>
    No.4936 [Report]
    ▶
    File: xyz_grid-0002-51363527.jpg
    (1.92 MB, 4800x2841) ▶
    2015721
    >>4935
    I would recommend raising the weights on the trigger words a bit. Here's an XYZ graph to show how the two main types of triggers interact. I think raising the weight of the needle on <Bodypart> triggers give better results. Raising the weight on just needle too high will start placing needles on all parts of the body.
    >>
    No.4945 [Report]
    ▶
    have any future plan to make ilXL Lora ?
    >>
    No.4958 [Report]
    ▶
    >>4945
    if you are interested in il models, maybe you shoule check this https://civitai.com/models/904011/punishmentdiffusionxl-05
    I have tested it, and it can do hanging,Garrote,drowning and roasting. I believe there are more, but due to the restriction of civitai, the creator didnt prove more prompts, as i checked that impalement is kind of working, but i dont know what prompt can make it work properly
    >>
    No.5180 [Report]
    ▶
    File: 00088-241230,Tanya Degurechaff Youjo Se(…).png
    (3.29 MB, 1536x2304) ▶
    3452114
    thanks for all the loras
    the "zunko" amputee loras works especially well.
    here's a dump of most of my stuff so far:
    https://exhentai.org/g/3178724/7226f3dbd8/
    some older stuff
    https://exhentai.org/g/3143593/1003385af1/
    >>
    No.5189 [Report]
    ▶
    File: ComfyUI_temp_kimao_00011_.png
    (1.76 MB, 1216x832) ▶
    1845704
    >>4929
    Made an illustrius lora. Activator is acid_digestion. Add other guro tags for best results. Works decent for what I tried. This is a 3 day DL so hopefully it gets re posted somewhere else!

    https://litter.catbox.moe/c6oat0.safetensors
    >>
    No.5190 [Report]
    ▶
    File: 115584532_p0_master1200.jpg
    (547 KB, 1200x877) ▶
    560714
    Could someone make or at least suggest a lora for neck snapping/broken? I've scrubbed every topic here and the concept just doesn't seem to exist.
    >>
    No.5191 [Report]
    ▶
    File: 00056-467876994.png
    (2.41 MB, 1344x1808) ▶
    2526897
    >>5190
    Ican get somewhat close at super low resolutions, but upresing always just turns it vanilla. I cant even get a head turned sideways.
    >>
    No.5193 [Report]
    ▶
    File: 00061-1099287480.png
    (2.34 MB, 1344x1808) ▶
    2457041
    >>5191
    With a lot of cherry picking and re renders I can almost aproximate it just by mixing other random Loras but the effect could be a lot better.
    >>
    No.5198 [Report]
    ▶
    File: 00039-3873542494.png
    (1.24 MB, 1024x1280) ▶
    1302326
    >>5190
    Whoops I forgot to post the neck snapping Lora someone requested me to make a few weeks back. I made both an illustrious and Pony version of this one. It creates a sort of "twist" effect on the neck, and it's not great at creating a sideways neck. I would recommend you use photoshop or something to manually put the neck sideways than apply this lora. neck twist can sometimes work with txt2img like pic related. It will create a neck that's facing backwards.

    DOWNLOAD: https://mega.nz/folder/m3wlxDQQ#2NMoAyU_231Tx2kNbOLP-A

    Neck Snap - lora
    Author: Entropy (https://www.pixiv.net/en/users/32339993), Anon (Image collection)

    Triggers: neck snap, neck twist
    Supplementary: guro, blood, asphyxiation, back view (if using neck twist)
    Weight: 1.0+ (~1.5 ish for Pony)
    Recommended Epoch: 20

    Notes: Another primarily img2img Lora. txt2img may not give ideal results.
    - Neck twist creates (**tries to create) a neck twisted backwards.
    - Neck snap is (supposed to be) a neck snapped to the side.
    Both create a "twisted" effect on the neck. Should work well with img2img/ControlNet
    >>
    No.5199 [Report]
    ▶
    >>4929
    >>4924
    Sorry I completely missed this. But it seems like someone else already made one?

    >>5189
    Link is already dead. Did someone manage to download this?
    If no one has the Lora I'll try making one myself.
    >>
    No.5200 [Report]
    ▶
    >>5189
    Link is dead for me too. Maybe try Mega because Catbox seems to be down right now. From that image alone it already looks better than my attempts.
    >>
    No.5201 [Report]
    ▶
    >>5198
    Hah, thanks, that would have saved me a lot of time. Would a sideways neck effect require its own lora? I'm not really sure what the process of training is like.
    >>
    No.5202 [Report]
    ▶
    >>5199
    It's fine. I realized that my dataset might not be the most consistent or diverse, since there aren't many artists who illustrate the concept and the few who do have vastly different takes on it.
    >>
    No.5203 [Report]
    ▶
    File: 00000-255470390.png
    (2.56 MB, 1536x1920) ▶
    2683124
    >>5198
    Having good luck so far, it combines well with the creepy head tilt lora

    https://civitai.com/models/864631/normal-and-creepy-head-tilt-slider-request
    >>
    No.5204 [Report]
    ▶
    >>5203
    Getting higher resolutions is almost impossible, their faces just start to bend like rubber.
    >>
    No.5207 [Report]
    ▶
    >>5199
    Hey all, I gave in and made a mega. Let me know how it works!
    Activation tag is acid_digestion
    https://mega.nz/file/iJIDTRKY#wx26DugxXpJ-mIpwc6B6Ax2cVqR_4cRcM5-zGhte2Is
    >>
    No.5210 [Report]
    ▶
    >>5198
    OMG!! It's so HOT! This reminds me a death scene from a movie called "Army of the Dead". If you know, you know.
    >>
    No.5237 [Report]
    ▶
    >>5190

    This neck snap Lora for Pony was uploaded to civit sometime in august and nuked within hours: https://files.catbox.moe/bo4tti.rar ...pw: gurochan

    Main triggers that were listed: neck_snap, head tilt, eyes rolled up. Check meta for more. I didn't have much success with this one but also didn't try very much. maybe you'll get more out of it.

    >>5191
    >>5204

    are you using High-Res-Fix and/or Adetailer? If so maybe you might want to try if multidiffusion is better at this.
    >>
    No.5260 [Report]
    ▶
    File: 1719501863363.jpg
    (892 KB, 4004x3395) ▶
    913527
    how specific do you think loras need to be? I'm wondering if it would be possible to make just a general torture guro lora trained on a huge variety of tortures? one use case would be for when you want to generate a guro torture image, but want to allow the ai some randomness on the specific torture thats being applied? or when you just want to make it clear to the ai that this is a guro torture scene, it should be bloody, and the character should be reacting as if in horrible pain?

    if that's not possible I'm also just interested in specific torture loras. pulling of fingernails, slicing off pieces of flesh, lingchi, flaying, roasting over coals/fire, hot metal(we did branding but what about just more red hot metal in general?), boiling oil, pliers, cats claw, pear of anguish, vaginal insertions covered in spikes, torture with medical instruments, tasers, crucifixion, stretching on the rack, etc. interested in anything were the person is experience extreme pain or fear.

    maybe even just a lora trained on pained horrified facial expressions
    alone could be good. often times the facial expression doesn't seem extreme enough in my generations.


    do you think any of those could be particularly easy to do or work particularly well?

    I could start collecting and categorizing images and come back with a large amount later.
    >>
    No.5264 [Report]
    ▶
    File: 8.png
    (2.14 MB, 1920x1275) ▶
    2245070
    >>
    No.5265 [Report]
    ▶
    File: 10.png
    (1.63 MB, 1291x1920) ▶
    1709825
    >>
    No.5267 [Report]
    ▶
    can you make a general or at least a throat slit lora for NoobAI new model? loras for noob need to be made for it due to how it handles light, but it seems to be the most advanced local model yet
    >>
    No.5307 [Report]
    ▶
    File: 00069-00004-250109,1girl upside down lo(…).png
    (3.19 MB, 1536x2304) ▶
    3347112
    >>5207
    not quite my thing but pretty good results
    >>
    No.5308 [Report]
    ▶
    File: 00071-00007-250109,1girl limbs dissolvi(…).png
    (4.31 MB, 1536x2304) ▶
    4520379
    >>
    No.5339 [Report]
    ▶
    >>5308
    This is stunning!
    >>
    No.5411 [Report]
    ▶
    >>5198
    >neck snap
    will you make an Il version of your not so fresh corpse lora?
    >>
    No.5412 [Report]
    ▶
    >>5260
    What is the source for this? Saucenao isn't bringing up anything and I'm not sure where else to look.
    >>
    No.5430 [Report]
    ▶
    File: 1719501851371.jpg
    (1.68 MB, 2893x3850) ▶
    1758892
    >>5412
    im not sure, but unfotunately I think it might have been from this artist https://x.com/natunisinu?t=TrQXGKiO1MtWyZcH8yaJ6w&s=09 who has privated and deleted all their guro content off pixiv. I could be wrong though.
    >>
    No.5510 [Report]
    ▶
    >>4266
    will you make illustrious version of those loras, as well as hanging and half-rotten corpse,arrow loras?
    >>
    No.5842 [Report]
    ▶
    I'm interested, I would love a "Vagina Dentata" LoRA, but I think my data set is too few pictures and maybe too low quality. What should I do?
    >>
    No.6025 [Report]
    ▶
    File: 127146919_p0.jpg
    (2.28 MB, 2480x3508) ▶
    2389337
    I'm starting to get interested in training loras myself.
    I think I've got enough vram to do it.
    looked up tutorials online.
    do you have any tips that might not be in the online tutorials? especially relating specifically to making guro loras?

    interested in making one's for general torture. needles. bleeding whip marks. burning with fire or hot metal. flaying alive. rack stretching. and maybe ones for horrified and pained facial expressions.
    >>
    No.6248 [Report]
    ▶
    >>5842
    Number 1 is check out pixiv or gelbooru (or any booru really) and try to get as many images as possible.

    Step 2 if you still don't have enough images you can try creating them yourself without a Lora. It'll be a lot more difficult but with creative use of inpainting, photoshop (or whatever art tool you prefer) you can create the necessary images to create an initial model.

    We can then use those to create a lora, then use that Lora to create more training images, which we can use to further improve the Lora. Let me know if you're interested in doing something like this
    >>
    No.6249 [Report]
    ▶
    >>6025
    I don't have many "guro specific tips". But I can share what worked for me in general.

    I would separate them out each into their own lora, you can try to put them all into their own lora but for beginners like you and me it'll save a lot of headaches if you just make each of them their own. A lot of triggers can conflict with each other or be affected by pre-existing keywords in the base model. Having multiple concepts in your Lora can make it really difficult to figure out which concept is causing issues.

    If you have enough sample data creating your own unique keyword can save you a lot of time. I'm currently working on an improved impaled Lora with another guro-chan user and the one thing we found is that using pre-exisiting triggers can be difficult with guro. Since guro is typically not a part of popular base models we constantly ran into issues with base version of the prompt messing up our gens. At the same time not having enough training data can make it hard to train a new trigger from scratch.
    >>
    No.6294 [Report]
    ▶
    >>5180
    What is this amputee lora?
    >>
    No.6296 [Report]
    ▶
    >>6294

    This one: >>698

    But there is only the 40 epoch version remaining in that mega link, the image in question used the 50 epoch version: https://files.catbox.moe/gkvv2a.rar
    >>
    No.6300 [Report]
    ▶
    >>6248
    I had not considered creating dentata images with existing tools before i saw your post. Yes, I'll try this. When I have 15 images, how should I get in touch with you?
    >>
    No.6301 [Report]
    ▶
    >>6296
    what password?
    >>
    No.6303 [Report]
    ▶
    >>6296
    >>6301

    Ah, shit, sry! Forgot that of course... >.< Pw is:

    gurochan
    >>
    No.6342 [Report]
    ▶
    >>6303
    thank u sir
    >>
    No.6608 [Report]
    ▶
    made a collection of "screaming in agony" images.
    maybe could be used to make a lora for agonized screaming facial expressions?

    this is part 1, of 6. catbox has a file size limit and is slow, I'll add the other parts when they finish upload.

    password for all parts is "guro"
    https://files.catbox.moe/o7m1dy.rar
    >>
    No.8495 [Report]
    ▶
    >>5180
    where download this lora?
    >>
    No.8608 [Report]
    ▶
    Looking for a basic throat slit lora. Not bloody but like the body is ready to be cooked
    >>
    No.8615 [Report]
    ▶
    Looking for a lora that those stump fucking. Especially the neck/throat
    >>
    No.8638 [Report]
    ▶
    File: 00452-2598348382.png
    (1.37 MB, 1024x1024) ▶
    1432965
    >>6608
    hey my guy not to shoot you down or anything but I took a look at this file and most of these can be attained through prompting without the need of a lora. Something like "wide-eyed, pain face, constricted pupils, foaming at the mouth, drooling, screaming". You can play with other tags too but they'll get you pretty consistent results.
    >>
    No.9012 [Report]
    ▶
    Anyone has prompting tips, or lora suggestions, to make guro eye stuff? As in eye gouging, pulling eye from the socket, pushing finger or fingers inside the eye socket, anything eye trauma related
    >>
    No.13876 [Report]
    ▶
    https://uploaditonline.com/download/wLzUnZRTeKa

    Acid Digestion Lora for IllustriousXL remade.
    Trained again with the images provided.

    Example prompt (for Forge UI)

    pink background
    high quality photo front, laying, safe_pos, score_9, score_8_up, score_8, (front), (highres), (solo), visible face,
    of one medium build woman, large breasts, european, 25 years old, worried, beautiful face, (blue) eyes, (looking at viewer), open mouth, (brown hair:1.4), medium hair, (visible feet), (full body portrait:1.8), (muscles), nude, barefoot, wet body, (ripped white shirt), (red nail), (pubic hair), ripped clothes, ((guro)), gore, acid_digestion, wounds, blood, stomach interior, (ripped skin:1.2), (exposed flesh:1.2), ((disemboweled:1.2)), (intestines), ((green liquid:1.4)), ((bones:1.2))

    Negative
    (((extra fingers:1.6))), gilf, old, grandma, ((bangs)), child, underage, extra limbs, deformed hands, deformed feet, (((hand with 4 fingers:2))), (((hand with 6 fingers:2))), 7 fingers, (((foot with 4 fingers:2))), (((foot with 6 fingers:2))), ugly, ((childface)), ((chinese:)),, (low res), ((squint eyes)), ((deformed hand)), ((extra limb)), ((third arm)), deformed, ((missing finger))
    >>
    No.13877 [Report]
    ▶
    Trained with:
    illustriousXLPersonalMerge_v30Noob10based
    https://huggingface.co/Toc/toc/blob/main/models/illustriousXLPersonalMerge_v30Noob10based.safetensors
    >>
    No.14921 [Report]
    ▶
    File: 61060348_p27_master1200.jpg
    (285 KB, 800x800) ▶
    292757
    Been experimenting with trying to replicate something like this for a while, albeit focused on specific genre niches, but it's been difficult to get it to work correctly. I'm wondering if it's a problem with either my training data or my tagging...

    Specifically, I've been trying to come up with a LoRA that will do a heart that is either midway through bursting apart or is already exploded, showing the heart interior or as a limp chunk of flesh. It's been *maddeningly* difficult to get the training to work for that though, so I'm hoping all of you would have better luck making it work...

    Old image provided as a little context for what I'm after, don't remember who originally made it.
    >>
    No.15074 [Report]
    ▶
    >>4264
    Dang. High quality abdominal inflation is plenty but exploding bellies bursting gorily are rare.
    >>
    No.24392 [Report]
    ▶
    anyone can share a link to debreast LORA for illustr? Thanks!
    [Top]
    Delete post: []