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




    Posting a few PonyXL Lora's I've made. Will post all my future Lora's here as well.

    EyeSocket PonyXL lora
    Triggers: eye_socket | score_9,source_anime
    Supplementary: blood on eye
    Weight: 1.0
    Recommended Epoch: 000040

    Download: https://mega.nz/folder/fjph1Z6Y#knlh28rEUakDtLnUrLofsQ

    I also have a pixiv (https://www.pixiv.net/en/users/32339993)
    I post AI guro/ryona content if anyone wants to check it out
    Viewing last 50 posts
    >>
    >>3345
    I sincerely appreciate the effort. The lack of training data meant all my previous attempts at training it (on non-XL models) failed wretchedly. If I could find 40 images of uteri in jars I'd send it your way or learn to train XL models myself, but this is something niche within the niche of guro so I can't find anything worthwhile.
    >>
    >>3099
    I have another request: Self-harm scars or cutting scars. If possible, a multi-concept lora that can distinguish scar locations. I sorted the images into 3 folders based on where the scars were. Images with scars in multiple places appear multiple times.

    If training multiple concepts into a single LoRA is too difficult, you can just merge them into a single folder. The 3 datasets share filenames so if you set them to overwrite each other and put them in the same folder you won't get duplicates. I tried to avoid putting too many images of Ame-chan in there to make sure it didn't become a character LoRA.

    https://files.catbox.moe/wraqcl.zip
    >>
    >>3359
    Lora turned out pretty great. I'm doing some testing now. I'll post the Lora soon, probably later today once I've compiled all the information I find.

    For future reference please don't use "sample" images if you're downloading from gel/danbooru. I know I said anything above 512x512 is fine but if there is a better version available I would prefer if you could get those instead.

    I appreciate the separate folders but please make sure none of the images are overlapping, it just makes more work for me since I have to check the same image multiple times. Its fine if its just in one of the categories, I go through all the images one by one anyways.

    I ended up merging upper arm and wrist cuts, so there's no way to distinguish creating those using txt2img.
    >>
    >>3359
    Self-harm scars PonyXL lora
    Author: Entropy (https://www.pixiv.net/en/users/32339993), Anon (Image collection)

    DOWNLOAD: https://mega.nz/folder/2mJlnIjQ#GnPxtlV-5w0Ylrq13LOmfg

    Triggers: cuts, self-harm scar, scar on arm, scar on leg
    Supplementary: injury, blood
    Weight: 1.0
    Reccomended Epoch: 15

    Notes: Creates an image with self-harm scars on the arm or legs or both. Using the blood keyword makes the cut marks "fresh" and will have blood coming from them. Without it will appear as an older wound. Always use cuts and self-harm scar, you can try without it but I haven't tested that so can't say what happens.

    scar on arm: for a cut on arm. Note that this doesn't make distinction between arm and wrist. You will get the cut marks wherever on the arm. Shouldn't be hard to clean up and get it exactly where you want (if you want it on the wrist or somewhere specific)

    scar on leg: for a cut on the legs.

    Epoch 20 is a bit fried but it kind of look nice? Has a bit of a phsyco style to it. Might be useful.
    >>
    1241841
    >>3370
    Phenomenal work.
    >>
    >>3370
    is there enough training material to possibly add scars on the stomach in a future release?
    >>
    >>3391
    Future releases are not needed. The LoRA will add scars to the stomach if you use "Bruise on Stomach." It seems to have made an association between many forms of injury and cutting scars and only knows cutting scars do not belong on the face. Specify an injury to a body part and you can get cutting scars there.
    >>
    >>3393
    As an example, the dataset had 0 examples of cutting scars on breasts but the LoRA can produce scars there.
    >>
    >>3393
    >>3394
    Wow I didn't even know it could do that. Gotta hand it to training magic. Doing shit I didn't even add.
    >>
    Dear Entrophy.

    can I request a character LoRA?

    Princess from Angel COrps

    https://files.catbox.moe/tqhv0w.zip

    I compiled all the art I found of her.

    Please she needs to coem back to this site.
    >>
    >>3564

    not Entropy obviously, but just in case it helps, there's already a Lora for 1.5 for her: https://civitai.com/models/528348/sammie ...If nothing else maybe the preview pics are good enough to add to the dataset.

    But also there are 2 TehDave-Style Loras. Just by using the triggerwords of the sammy lora with https://civitai.com/models/318858?modelVersionId=642597 I get stuff like these pics on the left. I'd say that's pretty close! :D
    >>
    >>3564

    There's also https://civitai.com/models/497156?modelVersionId=552664. Also very close I'd say.

    And in case you like the style, these were both made with a mix of 0.5 https://civitai.com/models/500876/icatcher-cartoon with 0.5 https://civitai.com/models/500858?modelVersionId=556730 and then the result of that mixxed with 0.3 https://civitai.com/models/632040/cutecore-xlb59

    (check meta for prompt)
    >>
    >>
    >>3564
    Just saw your request. I'll work on these fairly soon, just a bit busy rn so can't say exactly when they'll be done.
    >>
    >>3571
    >>3573
    >>3570

    THank you. The body and clothing transmits very good. But the face and hair are still off.
    >>
    >>3577
    decreasing the weight of both models below 0.4 and removing chubby helped
    >>
    >>3577

    >>3578

    These are both lovely! But I must admit, that I barely see the difference on the face. The hair stands out more but still, if you'd have shown me these 3 years ago and told me these where works by dave himself I would have believed it. But the same applies to this pic here... Even though the styles look very different. Guess I really don't have an eye for that sort of thing.

    >decreasing the weight of both models below 0.4 and removing chubby helped


    Thanks for that hint, I hadn't even tried mixing both Loras at lower weights by then and thanks to you I didn't have to experiment with the weights for too long. 0.3 works just fine for me. Just like with the styles I've used both your hanging lora and the hangboard lora at lower weights for these 3 pics here.
    >>
    ...this was done prior to that with just the TehdaveV2 Lora at full weight...
    >>
    ...aaand lastly this one was done with all three style Loras at 0.3.
    Yes, all three, because someone just dropped a third TehDave style Lora: https://civitai.com/models/738490/psys-pony-style-loras

    this one is clearly overtrained but it was definitely trained on angels corps girls. Although princess is mentioned only 3 times and the tag syntax was wrong. (check Lora meta and you'll see what I mean. And don't use prompt "angel corps", you'll just get angelwings. xD)

    Hope you don't mind me posting these 3 pics here. I don't want to completely spam and derail your thread, but I thought this was relevant enough and therefor warranted. (if nothing else at least these 3 pics also showcase your hanging lora. Hope that counts^^)

    I know the faces and hair are off in all three but I really have no good eyes for that. So I'm also curious which one (if any) is closest in your opinion. :)
    >>
    >>3590
    >>3588
    Damn this is hot.

    Can I ask you to do one pic of her without the iCatcher LoRA?
    >>
    >>3591

    iCatcher Lora? That's not a Lora. That mix is my base model checkpoint. :D

    But I can try another checkpoint if the whole style is what you mean. >>3577 and >>3578 were done with baseline PonyXL_V6 if I read the meta correctly. You want my to try that style?
    >>
    >>3595
    yes please.
    >>
    >>3597

    Well, I tried, but it turns out I can't really handle basic pony_v6. It behaves so differently from what I've used until now, especially weight wise. I can't wrap my head around it! None of my prompts work as intended. "Backgrounds? What's a background?" The initial gens all look like prehistoric cave paintings, scribbled by a dieing neanderthal with late stage parkinson... Even simple shit without a lora... Tried the whole night but this is the best I could come up with. The likeness and style are still miles off, I know.
    >>
    >>3603
    that..... Is actually a very important thing you said to me.

    I never tried other tings beside the default Pony.

    Myabe I should Try another one. Thanks!
    >>
    >>3564

    Princess (angel_corps) lora
    Author: Entropy (https://www.pixiv.net/en/users/32339993), Anon (Image collection)

    DOWNLOAD: https://mega.nz/folder/OygD1ToC#sW_4GTfvcnaxRgWEDwRTKw

    Triggers: princess_\(angel_corps\)
    Supplementary: freckles, long hair, brown hair, green eyes, glasses, gloves, tiara, bracelet, fingerless gloves, armlet, ponytail | slingshot swimsuit,

    Weight: 1.0
    Reccomended Epoch: 15

    Notes: Recreates the character Princess form angle corps. Works with other loras. Clothing is not baked into the main trigger for the most part, however certain clothing tags have traning on them like slingshot swimsuit.

    pic related is made using a hanging lora
    >>
    >>
    >>
    >>3617

    Sorry for the delay. Was busy with my other project. It does work with other clothing triggers. Haven't tested it extensively so can't say how well it would all work, some triggers may have data I don't know about.
    >>
    >>3564
    And just to let you know, for next time please don't include full comic pages, and images with resolution below 256. Full pages have too much non-relevant information that could make it hard to train on, and images with resolution below 256 just don't work.
    >>
    >>
    >>3617
    that is awesome!

    THank you so much!
    >>
    >>3622
    >>3621
    It's awesome

    Also I think the weight can be brought down to 0.7 without losing to much of quality
    >>
    >>2408
    I notices some guro loras on ehentai (https://e-hentai.org/g/2928445/7da13650d4/)
    maybe you can upload them on rentry?
    >>
    >>
    >>
    If it's alright to use this thread as a sort of sounding board/venting space for guro concepts that I can't yet fulfill -- in recognition of your excellent work in making so many possible -- the one thing that I would kill for (metaphorically!) would be a lora focused on broken limbs and/or joints bent in the wrong direction. Maybe other people have some tips for managing this with existing assets?
    >>
    >>3816
    I'm with you on this. although I'm prob more of a sick fuck tbh. I wanna see exposed bone, like from a forearm snapped in half. Also, would like to see ANY male focus.. Bout to lose my lunch over all these oversized tits..
    >>
    Requesting a Pokonyan or Burn at the stake LoRa
    >>
    Mate, your LoRAs were amazing.

    Any chacne you have been working on anything cool and just have not shared it yet?
    >>
    Dear Entropy,

    I have a sorted dataset for Flux, any chance you are willing to caption and train a model fro FLux?
    >>
    I took the liberty of rearraning the repository.

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

    I created section for each individual Model family subster.

    You can access and modify each subset as before with the same password.





    Now it has a section on FLUX.
    >>
    >>4008
    Thanks, I've been super busy irl lately and haven't had much time to do anything Lora wise but I have been working on a game.

    https://www.fanbox.cc/@entropy-game

    >>4007
    As with all requests I will do them if you can provide me an image set of at least 20 images . I'll probably create a new thread with my requirements for requests clearly laid out so it's always at the top.

    >>4009
    I could try? I have never worked with Flux before so can't say if I'll be able to get any results. You can dump a link to them here or to my pixiv.

    >>4061
    Thank you for this. I've been neglecting the repository for a while now. Like mentioned before I've been busy af this month and didn't have time to fix that up. Looks much nicer and more organized.
    >>
    no one's made a branding iron/hot metal/burning torture lora yet? i've been trying to get that to work with keywords in illustrious but no luck so far.
    >>
    >>698
    Solid after a little inpainting
    >>
    Are posts deleted from this thread? I can still see an image I made using on of the posted Loras in the site search but actual post seems to be gone.
    >>
    837926
    >>
    >>
    >>6576
    Holy shit! Yes please! Can we get more cum soaked slaves obediently being brain fucked like this?
    >>
    Does anyone have a good illustrious hanging lora by any chance?
    >>
    >>7098
    It really depends on base model and stuff but idk, pony one (from Enthropy, 30 epoch) works for me just fine, e.g. picrel.
    https://boards.guro.cx/ai/res/685.html#694

    [Top]

    Delete post: []