>>2561
>>2562
>>2563
>>2564
>>2565
>>2566
>>2567
>>2568
>>2569
>>2570
>>2571
Thank you so much! They look really good!
>not delighted

Comment too long, view post No.2576 to see the full comment.

>>2571
And an image or two made by an excellent friend of mine.

>>2576
My absolute pleasure!

>>2577
>>2578
Again, thank you! It's been only a year since the first inflated bellies looked better than an attached ball.

>>2633
My absolute pleasure, and I look forward to trying my hand at other requests! :)

Real stuff not allowed

>>2280
NAI Furry V3 fairs very well with vibe transfer.

bisected, black hair, hair over one eye, not furry, humanoid, arm support, gore, intestines, looking at viewer, solo, backlighting, nipples,

>>2359
Whoa.I was trying very hard to make the intestine come out of the character's stomach with Anime V3, but it looks like Furry V3 with not furry+ humanoid is doing a much better job!

>>2359
How do you get the gut spill without bisection?

>>2423
The only influence the bisected tag has is a minor effect on the gore. There's too little in the e621 training data for it to produce actual bisection consistently. Vibe transferring a source pic featuring guts spilled out will likely help tremendously.

>>2394
Thanks for the advice.
When I tagged the images I also commonly tagged 'guro' on all the images, and 'bisected' on transverse plane cutting and sagittal plane cutting images (and only a few images with coronal plane cutting, since I think just the face and breasts cut away doesn't count as bisection.). Also most of the images had common tags like 'blood'.
Do you think that having common tags in the training set also caused the bleeding?

>>2395
It's possible. The inability to use something similar to a search engine's quotation marks is frustrating.
It's also worth noting that while bleeding is a thing it appears to happen at lower influence than the full tag. Maybe in this case because there's no other familiar thing in the noised composition it's guessing "that looks like this other plane cutting tag" even though it's kind of mentioned in the prompt, so it switches to using those other ones too.
I don't exactly know how models work, but the denoising algorithm is handled by the sampler. So maybe it's also worth checking other samplers, maybe the bleed is handled differently.

I don't know what the best approach to a novel concept lora is, I can only theorize.
Common sense would say because there isn't a base model to take advantage of tags such as "guro", then the tag is pointless, unless you assume this would be used in conjunction with the ayaswan lora.

i was collecting a dateset to try training it.

if yo uever wish to retrain it I cna send my dataset to you. (it will probably overlap 90%)

>Should I really make three seperate LoRA for this?
yes

Really into the nude girl(s) in a forest theme! Love how the first two seem more annoyed than anything else about being cut in half!

>>2299
so you made your own realistic guro model?

I saw a guy opening a website like a year ago with very nice realistic guro content, was that you?

I only remember Samus torn to pieces in the preview.

>>2300
No I use a program called Remix, for $50 a month it's unlimited uncensored generation AND free commercial use. It's the best program I've found. I can't code for SHIT so I had to find someone to do it for me. I've generated over 8k images in the past 3 weeks

>>2301
$50/y actually excuse me

>>2301
is it local, or you run it on their servers?

>>2314
I run it on theirs, haven't figured out local yet but once I do it'll be local.

>>1066
yea some of them got deleted off aco xd

woww thank you for that guide! all of the above posts are just using the 'ghead' lora. goto prompt for that lora is '10_ghead, ghead, 1girl, photo of a of a (naked:1.2) (headless:1.2) woman, (guro, decapitated, severed head), blood_splatter, blood_from_mouth, short dyed hair, <lora:ghead:.5> (any extra loras/characters here)' its mainly using triggers from the lora training information.

>>1066
what settings for controlnet do you recommend for these? not sure which control type to use

>>1068
for controlnet I foudn that the depth is my favorite.
4chan advise more advanced stuff, or combining multiple, but imho the depth is th ebest. with strength of around 0.6

So I have head the following tags work well together: pregnant, hose, anal insertion. How would you generate an exploding belly? If it's not possible I'll take bloody from the mouth.

I used https://rentry.org/R-18G_Stable_Diffusion guide, and it's good! I've been making some good stuff.

Would love it if OP share his details; models and such

>>2054
But of course!

My first prompt for her — Perchance [dot] org — was very simply: “Tinkerbell, fairy, Disney, Peter Pan, pretty, blonde, wings, pixie, green dress, high heels, choker”

… then i started getting violent; my final prompt was “Tinkerbell, fairy, Disney, Peter Pan, pretty, blonde, fairy wings, pixie, green dress, high heels, choker, bondage, prisoner, public humiliation, girl wearing dog leash, dog collar choking girl, dog chain, bloody, cage, dog collar, dog leash, girl wearing dog leash, slutty, shibari, bondage, exhibitionist, choker, bloody limbs, bloody stumps, blood splatter, blood splatters, raw meat, serving knife, bloody cunt, bloody tits, knives, bleeding, bloody skin, bleeding, guro, gore, Tinkerbell, fairy, Disney, Peter Pan, pretty, blonde, fairy wings, pixie, green dress, high heels, choker, Dolcett, lovely, blood, blood spatter, big nipples, used, abused, tattoos, body writing, human toilet, rape victim, cum, cum slut, cuts, self harm, dog collar choking girl, dog chain, bloody, dog collar, dog leash, girl wearing dog leash, bloody tits, sliced breasts, bloody nipples, cut nipples, cut breasts, terrified, scared, hurt, dripping blood, bloody cunt, spread legs, Tinkerbell, fairy, Disney, Peter Pan, pretty, blonde, fairy wings, pixie, green dress, high heels, choker, pregnant”

>>1204

Damn. The new Pony SDXL model is amazing.

incredible work, love the one w/girl serving up her tits on a tray with a knife through the breast... more like that?

>>1998
I'm happy someone ressurected this thread.

I also love knife in the breast pics, but that was done with photoshop. not with AI

>>1907
>Whats some nice place to host albums (anonymously, perhaps) and easily browse it like in phones things with free scrolling idk

I would suggest SadPanda but its not anonymous. You can disown libraries at a later date so that they're not tied to you forever. I use Ehviewer for mobile use.

>I never use anything but sd1.0 and so on

When you say you only use Sd1.0 and so on, does that mean you haven't used SDXL/PonyXL? PonyXL is currently the best for anime so if your PC can handle that I suggest switching.

>Also why is AI so hardly produce poses like getting eaten or exploding, or perhaps beheading

That's because training for those topics are relative small in base models. You can train Lora's to specifically train those types of images if that's what you want. I'm going to assume you're new so as a reminder don't post realistic AI images, they will get removed. If you don't have a beefy enough PC to train Lora's (I can with a gtx3060 12gb, but I think I'm near the minimum) you can gather training data and ask someone to train them for you from this board.
I'm pretty sure a beheading lora is currently in the works by members of this board. Haven't heard anything about being eaten or exploding tho.

>>1908
what sadpanda what ehviwer? i mean like e-hentai dot org or exhentai but, uh, anonymous and free to use, no login maybe idk

>ponyxl
...yea i only use those from sdxl site or stable diffussion dot org, so on
i dont use one on pc nor do i have one for thisin particular.

what's my other alternative.

i can never get decapitation on sdxl or anything on hugging space or sd1.0

the body is always attached

vore and such also
why?

The tags in that page are surely correct, you can use them of course. But what's more important is you understand the concepts you want to train. And I suggest you to train a Lora, not a full model. Looking forward to your results.

>>1780
If you want to tag images for AI training you should consider using wd14 to tag images for you. It'll tag most of them correctly and you can simply look through to make sure they are correct/make changes.

For Automatic1111: https://github.com/toriato/stable-diffusion-webui-wd14-tagger
For ComfyUI: https://github.com/pythongosssss/ComfyUI-WD14-Tagger

If you are deadset of tagging them yourself using gelbooru would be a good idea. I'm pretty sure ponyXL is trained using mostly gelbooru tags. If you're looking to train a ponyXL model it would be a good idea to look there. https://gelbooru.com/index.php https://gelbooru.com/index.php?page=tags&s=list

thanks!

Will try to follow. I wan tto make artist LoRA and they are usually a mix of all guro concepts

>>1796
When you tag images it's important you tag everything, even the non-guro aspects of the image. wd-14 will do that for you which is why I recommend you use it. If you take tags out it will bake them into your trigger word. So for example if you're training on a character, by removing clothing tags the clothing becomes a part of the character's trigger word and will always generate with that keyword. The trigger word you choose by the way should ideally be something the model is already trained on so that it gets a head start on training. Not everything works with it however. Hanging for example is notorious for being a trigger word you cant train hanging on since on ponyXL, hanging will become a bdsm tied up image of the person being strung up. In such a case either create your own trigger word or use another adjacent trigger word that wont conflict. ex, noose, rope_around_neck etc. etc. When in doubt just check gelbooru.

Good luck can't wait to see what you create.

>>1780
>>1801

Not sure if it will apply to what you're trying to do but I figure I should at least mention it.

If you are planning on training a PonyXL lora you only need about 20 images per concept. More is not always better. Quality > Quantity. You will get much better results if each image clearly shows what you are trying to replicate. Pruning images that aren't as useful is also a part of Lora training. Having more is better but make sure they are all quality images.

Can you please provide the whole promts you used? I appreciate the effort but I can't get it to work, like at all, character's head just won't leave their body. I'm using base Pony.

I can't get it work either (I'm using Pony btw). I'm a bit confused, does OP use SDXL or SDPony? Cause there isn't really a lora that's "XL-Pony", two checkpoints have totally different database, many loras that are trained on XL can't be used on Pony at all. Also I found this lora will make everything kinda anime style.

score_9,score_8_up,score_7_up,<lora:nbbqXLpony-v2b-000002:1:lbw=1,1,1,1,1,1,1,1,0,0,0,0>,masterpiece,best quality,absurdres,high resolution,extremely detailed,zPDXL2,indoors,2boys,multiple boys,1girl,on wooden floor,ass,back,doggystyle,sex,from side,(penis into neck),(gangbang:1.1),(group sex:1.1),(headless female body),(bhdsx:1.2),pubic hair,blood,blood on clothes,blood splatter,breasts,cum,dark-skinned male,(((neck sex))),hetero,injury,clothed female nude male,sex,rape,testicles,uncensored,(beheaded female),bleeding,blood,blood_splatter,bleeding,outdoors,<lora:genshin_pony_v3:1:lbw=1,0,0,0.5,1,1,0,0,1,1,1,1>,pussy juice,headless female body,eula_\(genshin_impact\),large_breasts,blue_necktie,eula_\(genshin_impact\),sidelocks,high-waist_shorts,black_gloves,long_sleeves,spoken heart,

It seems that this training is still not good enough. I will probably update another version later, and everyone will try it again.