R-18G Stable DIffusion Repository

This is a repositry of all Loras and models for Stable Diffusion, more specifically SD 1.5, SDXL and PonyXL, both anime and realistic.

This is document is meant for people already familiar with Stable Diffusion. If you are not do not be afriad, it's easy to learn. Start here: https://stable-diffusion-art.com/beginners-guide/

I lost most of the creators of the models. Some were anonymous, other got their pixiv account banned, some might be in a chinese prison. But if your model is in this list than: Thank you.

Two other guys that made cool LoRAs but ask for money to use them. If anyone has access to their LoRA please share and I will add them to repositry.
https://frotz.gumroad.com/ and https://www.patreon.com/user?u=93157198

P.S. If you download the safetensors from catbox don't forget to change the name according to the model description.

SD 1.5

Best used for Anime, mistakes in realistic images are too annoying.

Base Models

Realistic

Realistic Vision

_

Anime

H ardcore Hentai
No idea if best but i'm stuck with it

_

LoRA

Bible Black Style

Bible Black Style
Copy the style of that famouse hentai, but it was also trained on the bllod and gore from that anime, does blood better then all otehr anime lora, i use it with weights between 0.3 and 0.6

-

Pole Stand

Pole Stand (One Bar Prison)
Good for a pole partially implaing the girl.

-

R18GAmputeeLimbless_v15:1.0

link1 and link2
Not very flexible, weight 1, trigger words limbless , weight need to be increased above 1 soemtimes,

-

Amputee v5

link1 and link2
Better amputee, weights 0.6-1, trgger words amputee, guro, bandage, injury, bondage typical prompt <lora:Amp_V5:0.6> (amputee:1.7),( bandage:1.3) ( white background:1.3)

-

amputee000

link1 and link2
Amputee, with freshly cut stump having a meat texture. I get very noisy results (pic above), maybe the model or vae i use are wrong. weight 1-1.2, trigger word amputee111 typical prompt <lora:amputee000:1>, (amputee111:1.2) quad amputee, amputation guro and add arms and legs in negatives.

-

amputee_v0-03 horror anime scene

link1 and link2
General guro lora. Best guro results, crisp image but hard to prompt, I had to use negative lora ( bad-picture-chill-75v, badhandv4 ) to make it look good. weight 1-1.2, trigger word guro , it also knows the following words amputee, object insertion, anal object insertion, ahegao, guro, horror, headless typical prompt CHARaCTER LORA, <lora:amputee_v0-03:1.1> , (amputee:1.2), (guro:1.1), blood , anal object insertion

-

*Dissection corpse *

link1 and link2
Produces open abdomens. Is a bit more anatomically accurate in depicting organs than amputee_v0-03 lora but also if used alone has a weird 2D-ishness to the open belly area. However at lower strength (~0.5) it improves details in conjunction with the "amputee_v0-03.safetensors" above.
it was trained on an extremely small sample, imho the effect is negligible, see iamge above.
Weight: 0.5, trigger words: dissection_corpse, guro, to use with amputee_v0-03

-

Whip Marks

link1 and link2
Adds whip marks, trigger words whip marks typical prompt <lora:whipMarks:0.3> whip marks
textured skin, realistic dull skin noise, visible skin detail, skin fuzz, dry skin,

-

Lashed

link1 link2
Affects the shape and position of the model too much, use with low weights, trigger word lashed, typical prompt <lora:Lashed_v3:0.3> lashed

-

beforeafter-v4

link1 link2
Typicial prompt <lora:beforeafter-v4:0.75> , beforeafter, pov, multiple views, impaledmeat
Major improvement (or regression): Removed the three new prompts in V2V3: keephead to keep the head, decap to cut off the head, and upsidedown to hang the body upside down. Switched to kotosmix training
limblessbody: Generate a female body with normal skin color
hangmeat: generates cooked hanging meat
impaledmeat: generates grilled skewered meat
multiple views: Multiple views, when the canvas of the banner is enlarged, a comic strip in which the characters are stripped and nude beheaded and skewered sequentially from left to right will be generated.
Works better with kotosmix

-

strangulation

link1 link2
Never managed to make it work properly, typical prompt <lora:strangulation-31:1> (Anime. Masterpiece, Best Quality, High Quality, Highres, Absurdres:1.3), (outdoors:1.2), (building:1.2), 1girl, (subject girl description), (full body), strangulation, garrote, noose, tongue out, rolling eyes, shaded face, saliva, tears, crying, wet, sweat, shiny, arms_behind_back, bound, restrained, suspension, stationary_restraints,

-

Lying Dead

link1 link2
Made for anime, works so-so IRL, typical prompt <lora:lying-000042g:0.65> dead_lying, expressionless, corpse , crying with eyes open, blood, ww1era,khaki uniform,equipment layout, bag, pouch

-

Jugulation

link1 link2
Throat slit from behind, very rigid and a bit overtrained, but usefull to create a base-image that can be refined via img2img.
Weight 1, trigger words: jugulation, death, 1boy, 1girl, hetero, grabbing, holding knife, bleeding
typical prompt * <lora:lora_jugulation_v1.0:1.0> jugulation, death, 1boy, 1girl, hetero, grabbing, holding knife, bleeding*

-

Cannibal Feast

link1 link2
Anime only, typical prompt <lora:re-eat:0.75>, food,cry,apple,fruit,object_insertion,chicken_(food),plate, dinner, drinking glass, wine, meat, embarrassed, crying with eyes open, shy, full-face blush

Decapitation

link1 link2
Works horribly IRL and so so in anime, Decent when combined with ControlNet
typical prompt <lora:Decapitation:0.8>, mulitple_girls, 6+girls, (body_pile:1.4), (decapitation:1.1), (severed_head:1.1), neck_cutting, neck_spray, (head_flying:1.2), sword, nude, blood, death, guro, (bad_end:1.1)

-

Decapitated head

link1 and link2
As in the show case does not works often. Two pics, one with head, one decapitated. works badly, typical prompt <lora:ghead:0.75>, severed head, blood,guro, death, ryona,trembling, kneeling,seiza,spread legs, lying, on back, defeat

-

Head fuck

link1 and link2
Anime Head fuck Lora, pretty bad, not felxible at all, only use is with ControlNet

-

Severed head 1

link1 and link2
Never managed to make it work properly, maybe I should try ControlNet, typical prompt * queen sexy head, smiling, (severed head, head on polearm, ) blood*

-

Head in Jar

civitai link
Works both in anime and in realsitic models

-

Girl in a package

link1 and link2
Not flexible, but does well what it needs to do with different aniem characters, Typical prompt <lora:Pkg_girl:0.7> 1girl, spread legs, bound, bdsm, box, in container, in box

-

Ear Fuck

link1 and link2
Made for anime works badly, I managed to make it look nice only with ControlNet, Can make horror results with realistic models as well, typical prompt <lora:earfuck:0.9> earfuck, ear penetration

-

Head destroy

link1 and link2
Can generate very graphic destroyed heads with exploded/missing parts of the skull and brainmatter. Can be used txt2img but best results when inpainting existing images.
Mask the upper head of your subject with or without the eyes. Mask additional space above and around (so there's enough space for some splatter) then set to inpaint mask only, set masked content to original, leave denoising strenght at 0.75 and cfg between 5 and 7.5. Only use the 4 triggerwords as prompt and leave lora strength at 1.
Weight 1, typical prompt <lora:head_destroy:1.0> head destroy, brain, blood, headshot

-

WrongHole

Original link and mirror
Generates a hole. To use with inpainting. Generate a bloody image, then inpaint an are where you want a hole to appear
Weight 1, typical prompt <lora:WrongHole:1> WrongHole, hole, bloody, viscera, meat

Stabbed with a sword in the belly

link1 and link2
Pretty bad.
Can generate a girl beeing stabbed with a (Katana-) sword through her belly. works better with semi realistic models, appears to be not very adaptable. I was neve able to get a good result with , picture is above is my best, probably can be used well with a inapitning or with ControlNet. Weight 1, trigger word: ci_fu , typical prompt <lora:ci_fu:1> ci_fu, sword, blood, impaled,

Impalement civ

link1 and link2
Work horribly alone in txt2img, but when paired with ControlNet of an impaled image is pretty good.
weight: 1, trigger words: impaled, impalement, execution , typical prompt <lora:impalement:1> impaled, impalement, execution

-

Impaled Effigy v1

link1 and link2
Typical promt <lora:impaled-effigy-v1:1 .2> ((impampute, amputee, , metal rod, iron rod, iron rod between neck and head , blood, decapitation , head impaled:1.2 ))
【Impfull】: All limbs are intact except for the head, which is chopped off and impaled on the same pole as the body.
【impampute】: The limbs are chopped off and then the head is chopped off and impaled on the same pole as the body.
【impwelldone】: roasted complexion. The success rate is a bit nuanced.

-

NEW BBQ LORA-V3a

link1 and link2
【butchered】a female body that has just been killed, bled, and has normal skin color but has not been roasted.
【bbq, barbecue】 used to change the color of a woman's body into a roasted color. Easy to understand, right?
【Upside down:】As unstable as ever, an independent concept was created to avoid interfering with normal image generation.
I have trouble making it work properly, typical prompt masterpiece, best quality, absurdres, high resolution, extremely detailed, <lora:nbbq-v3a:0.9>,1girl, solo, wooden rack, ((restrained,from behind,from side, no hands, no foot,incision,head impaled by wooden pole,pussy impaled by wooden pole,bbq,roasting,roasted,meat juice,all fours,campfire,cannibalism,)) arms behind back,bound, pale skin, nun, vein, purple eyes, purple hair, bdsm, bondage, open_mouth,

-

Trophy in Wall

link1 and link2
human trophy, works very good in IRL when combinde with an amputee lora, typical prompt in_wall, <lora:Concept_in_wall:0.7>, The grand entrance gate is adorned with intricate carvings of mythical creatures and divine beings.

-

Hanging 1

link1 and link2
The most felxible hanging model, works both anime and realistic, example prompt * <lora:hunged_girl:1> (hunged_girl:1.4), (noose), rope, (hanged:1.2), (asphyxiation), (bad_end:1.1 ) *

-

Hanging 2

link1 and link2
Workse very well in anime, but adds its own style, not very felxible, example prompt * <lora:hang_v2-000022:1.0>, hang, hanging, noose, hemp rope around neck, apshyxia, (drool), open mouth, strangle *

-

Belly Punch

link1 and link2
Bellypunch Lora, made for anime, actually works for IRL as well, very unstable, the prompt i found working was: <lora:BellyPunchLora-10:1.0>, harapan, (2boys:1.4) , (naked:0.5) tears (begging:1.1)

-

Ryona

link1 and link2
Appears to be trained B&W manga images of girls fighting. Not very good. Weight 1, typical prompt *<lora:Ryona-000009:1>, ryona, punches other woman's stomach *

-

Cut Lines

link1
Creates butcher "cut here" lines, weight 1, trigger words: cut line, body markings, arm marker, thighs marker, neck marker
works only in anime

-

Bruised Face

link1 and link2
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.
Should work well with other Lora's. Haven't encountered any major issues as of yet.

XYZ grid was made using based64, Lora weight seems to work slightly differently on other models.
When using the leaked 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)

-

Skinless

link1
weight 1 Trigger words skinless, woman, man

SDXL

Best used for realistic models, as of January 2024 people are still searching for a definitive anime model, Animagine XL appears to be the best candidate.

Base Models

Realistic

,%20big%20smile,%20ruby%20eyes,%20pigtails%20hair,%20dark%20makeup,%20hyperdetailed%20photography,%20soft%20light,%20head%20and%20sh.jpeg)
Juggernaut XL
Best for being felxible

_

SDXXXL
Best for realistic human skin, i believe it was trained on porn _

Anime

While several SDXL anime based models exist the best one, Pony XL, is not compatible with most SDXL LoRAs and was put as a seperate entry.

LoRA

Ahegao

Ahegao (Tongue Out, Cross-eyed) SDXL
works even if the face is small

_

Breasts Clamp

,gigantic%20breasts,_lora_breastclamp_SDXL_r.jpeg)
Breast Clamp 乳枷 for SDXL alternative downlaod link1 link2
The clamp appears woobly but still works _

Legs Behind Head

Legs Behind Head SDXL
Good for cannibal stuff in the kitchen

_

Quicksand

{Quicksand XL](https://civitai.com/models/264194?modelVersionId=297872)
Unfortuantely affects the face of the person you try to put in the sand, but otherwise good. Not very flexible

_

Shibari

[SDXL] shibari, Japanese bondage
Affects the face, otherwise works well in something like wet girl in boiling water

-

Dissection

Reference images can be found here: https://files.catbox.moe/8d96cs.rar
Model link
Creates a realistic medical dissection image. Not very flexible, Weight: 0.8-1.0 trigger word: None, but some helpfull might be: dead female body, ,detailed anatomical dissection, guts
No preview picture since it's too graphic for me.

-

Amputee Realistic

Model link
Makes an quadruple amputee. Worsk with realistic models.
Weight:0.5+ , trigger word:amputee , you can choose 3 positions: stand on ground, sit, lying

-

HOGTIED

original link, backup lin
Rope Bondage: Hogtied & Gagged.
Weight:1 , trigger word:None , But it understands a lot of tags very well
Perspective: From above | from side | from angle | facing viewer -- these may also be combined
(From behind confused the model to much so it was removed from the dataset)
Setting: On floor | on ground | on table | on bed ...
Pose: arched back | head down | head back | feet crossed | feet together | knees together | knees apart feet together | elbows together | hands together elbows apart
Bondage related: (color) + gag | (color) + rope | natural rope | feet linked to hands | feet linked to shoulders | feet linked to hair | bound + (bodypart)

-

HangingExecXL-Frotz

It works very well but it is behind a paywall: https://frotz.gumroad.com/

Works better with 3D or Western drawings, also good with multiple peopl aroudn the subject.
Weight: 0.7 - 1.0 trigger words: h4ngingexecut1on

Frotz has multiple other awesome SDXL loras. I have no idea how well they work

-

Pony XL

THE best hentai based model as of february 2024.

Pony Diffusion V6 XL
Do not be fooled by the pony tag. Does amazing things. best for hentai . knows a lot of tags and positions but not artist names.

-

amputee_SDXL

link1 and link2
Made for ponyDiffusionV6XL_v6
Produces amputee with skin or machinery at the interface,
Weight: 1, trigger word: amputee

-

Quadruple amputee

link1 and link2
Made for ponyDiffusionV6XL_v6
Produces amputee with skin or machinery at the interface, more agressive than the previous amputee_SDXL
Weight: 1, trigger words: quadruple amputee

-

amputee_XL_pony

Original link , link1 and link2
Made for ponyDiffusionV6XL_v6
Produces amputee
Weight: 0.8-1, trigger words: amputee

-

sdxlponyheadless-000006

Original link , link1 and link2
Made for ponyDiffusionV6XL_v6
Makes a body without a head. Not a guro model, but by adding words like guro or inpianting can make good results.
Weight: 1, trigger words: headless

-

sdxlpony_disembodiedhead-t1-000006

Original link , link1 and link2
Made for ponyDiffusionV6XL_v6
Makes a body holding a head. or head nearby. Not a guro model, but by adding words like guro or inpianting can make good results.
Weight: 1.2-1.4 trigger words: disembodied_head, holding head

-

sdxlpony_onlyhead-v2

Original link , link 2 and link 3
Made for ponyDiffusionV6XL_v6
Makes a head without a body, very flexible, can be from front or from side, head cna be fucked with blowjob or deepthraot. Not a guro model.
Weight: 1.2-1.4 trigger words: only head

-

XL-Headtoy

link 1 and link2
Made for ponyDiffusionV6XL_v6
Makes a head. being used as an onahole.
Weight: 1.0-1.4 trigger words: headshot, blood, 1girl, penis, cum, grabbing another's hair,cry,

-

XL-Hang-board-test

link 1
Made for ponyDiffusionV6XL_v6
Hanging.
Weight: 1.0-1.4 trigger words: * peeing,nervous, jumping, dancing, white outline,full body, blush ,sad,open mouth, tongue out,shouting, surprised, shock, solo, arms behind back, from_below, rope on neck, * Do NOT use words as hang, or hanginging, it will create a bondage hanging instead.

-

XL-Beheadv2

link 1
Made for ponyDiffusionV6XL_v6
Creates a behaeding scene. I did not manage to make it work properly
Weight: 1.0-1.4 trigger words: behead,blood,

-

Nipple Penetration

link original, link 2 and link 3
Made for ponyDiffusionV6XL_v6
Nipple fucked by either dick, dildos or tentacles,
Weight: 1.0-1.4 trigger words: nipple penetration, single penetration, double penetration, a woman being penetrated by _, cum in nipples, lactation,

-

Ear Sex

link original, link 2 and link 3
Made for ponyDiffusionV6XL_v6
Ear fuck model, Has some problems, affects the proportions of the chracter but workable.
Weight: 1.0 trigger words: ear sex

-

AyaSwan Style v2

link 1 and link 2
Made for ponyDiffusionV6XL_v6
Model of the great artist AyaSwan. It was trained on a 1000 AyaSwan images the tagging was done manually.
This resulted in a more unstable model. So play with the tag words listed here and put them in negative and positive prompt for the desired effect.
Recommended to use TOGETHER with other R-18G LoRAs.
Weight: 1.0 trigger words it knows: guro, blood, blood from mouth, amputee, severed limb, severed head, beheading, headless, neck fuck, dead, mutilated, mutilated breast, reverse blowjob, brain fuck, brain, eye fuck, empty eye socket, dangling eyeball, eye removal, disembowel, viscera, uterus, impaled, impaled mouth, noose, nipple penetration, piss, cooked, guillotine, arm insertion

the v1 of this model is available link 1 and link 2
. It is cleaner but does not understand prompts.