Oops. The little link doesn’t work.
But here is that Picture in the third hanging thread.

https://boards.guro.cx/g/koko.php?res=100654

Experimented combining the original and the AI filtered version.

Interesting. I should try another line drawing for you to combine.
Maybe I’ll filter my favourite “Hanging Thief”
through an A.I. filter.

Original.

how do you generate these?

for non-guro /f/ stuff: https://boards.guro.cx/f/koko.php?res=62906

How do i get these ai generated things?

stable diffusion + photoshop pretty gud

I’m gonna be honest.
This is what I was afraid to see.
It feels a little too realistic to post here.
Most if not every post here is animated/3D renders.
So I encouraged that you post something more, let say “drawn”.
If you catch my drift.
Until next time.

Original + A.I. Art combined

What does it draw if you put “child girl shot to death” in the box?

Which sites do you recommend to try?
All the ones I tried censored the violence

We should discuss how to train AI, guro is complicated, it will be probably be the last to barrier.
This what Novel AI gives me with a guro tag

>>30566
Another one

>>30562
I just downloaded it and run it on my PC.

Go to 4chan, they have threads explaining what to do.

Then come back here

i humbly suggest to leave real life shit out of this, go to gorechan and make a thread there if this is your fetish

>>30561
I'm getting awy different reuslts

Soem interestign reuslts can be obtained with fire.

>>30584
breasts on a tray also works

useful prompts
Roasted pig
sweating
breasts pressed

What would be incredibly useful is for people to share prompts and tags that AI would recognize.

For example, I ran "roasted pig" and "roasted like a pig" and from there understood what kind of results better suits what I want. Remember that the order of the words separated by comma matters. How? No idea from the articles I have read.

tried different approach

I... cut my right nipple and prepared some sashimi, hope you enjoy!

My right breast healed after you ate it. Do you want to cut more?

Darling I already cut of 2 of my finger. DO I need to keep going?

You want it? You want to see me stab myself? Cut my nipple off? Or do a seppuku?

If you want I can cut myself open from cunt to stomach. Gut myself like a fish.
Would you like me to husband?

So you wawnt me to cut off both of my nipples as your wedding gift?

Sure! Oh god. I'm actually enjoy the pain. Are you gonna add the nipples I just cut to your collection?

it can generate missing eyes. How to replicate this?

I started doing captions

please tell me what do you think

butcher and meat-related prompts give nice results

what program do you use to generate these images ?

>>30626
How do ou add wounds? I'm failing.

>>30627
Stable Diffusion.
Novel Ai model.

Guro is not supported apaprently

>>30629
at least I was unable to do hanging and a head separated from the body.

Probably will need to create a dataset to train.

>>30627
what about you? What do you use? what have you discovered so far?

Had some success with roasting pictures. More incoming

The coloring tool in NovelAi is our friend in this case. As well as keywords with whatever pose you want and the keywords mentioned above.

>>30615
wait, did you generate those with the basic stable diffusion 1.4 model, or other more specialized models?

>>30637
I used NOVEL AI

for pic related those prompts


slave, multiple girls, (butcher shop), (suspended meat), chains, (meat hooks), (masterpiece_1.2), slender, medium breasts,


Negative
lowres, text, error, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry

>>30636
amazing. I have not yet played with img2img.

What keywords?
I don't see much mention above besides the word roasting

looks like there are a wide variety of models here, maybe we can train our own?
https://rentry.org/sdmodels

>>30640
we should.
We need to creat a good dataset of guro images.

With proper tags etc.

I have tons of images but I need to go through all fo them and added tags.

Also I have no idea how to tag amges properly

>>30639

girl, nude, medium breasts, roasting pig, sweating, dark brown skin, fire, steaming body, oven, empty eyes, dead eyes

A lot of the work is done by the color tool, I must stress.

More experimenting to be done today.

>>30656
If you want t beautify an image add tags,
(masterpiece:1.2), slender, medium breasts, nsfw, perfect face, pixiv, (extremely detailed)

and as negative

lowres, text, error, extra digit, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry


but it tends to create less creative results.

Also, dead hookers thrown into the garbage are hard to do.

Pre-Snuff Execution (AiArt)

Snuff Headshot Execution (Ai-Art)
"Here's my attempt in getting the Ai, with some Photoshopping, to get a proper snuff Ai Art :3"

1.draw a rough sketch
2.load to Mr. AI
3.add and modify the output image

Training hypernetworks is now a thing in new stable diffusion versions. I have enough vram to do it if someone could send over a zip of images by the same artist, or I'll probably find some eventually anyway

>>30669
yessss

the top guro artist would be
ayaswan and Beaver

Beaver is hard since he mostly draws comics.

Ayaswan should be good.

>>30648
What settings are you using to generate these?
I am using novel AI locally via stable diffusion but can't get anything close to these.

Is AI able to create pictures where child or children are getting killed? Or is that a taboo for AI?

>>30676
Looks like png data isn't deleted on this board so you can just put them in the PNG info and find out

trying out various things, not being very scientific about it. posting some of the more successful stuff to see if anyone can glean something from it

roughly speaking, i've been interrogating various existing pics with deepbooru and then running that as a prompt either in img2img with the original image as input or in txt2img. so far for me waifu diffusion 1.3 gives way better results than the novelai leak, but i've gotten good and bad out of both.

haven't messed with guro inpainting any yet, nor have i bothered trying to shop any of these to be closer to what i want, but i think that's probably going to be a pretty significant part of the process. unless somebody's willing to train up a whole .ckpt off of guro i'm not sure we'll ever get to the same level of convenience the normals enjoy

that cloth in the upper right is almost like a gloved fist if you squint. not too many more decent ones

the anatomy on this one is totally fucked and not in a guro way, but i really like the face on this one.

using both interrogations, CLIP and Deepbooru, seems to produce more consistent results.

>>30689
While NovelAi can't do actual open wounds it is pretty good at emulating or faking the look of guro with the right tags.

This here is one of my creations that is a food example,

May the owner of a toaster that has yet to receive the new iNvidia RTX, request some inflations&burst bellies ?

>>30710
not one of my higher priority fetishes so i doubt i could do it from scratch. but if you want to sketch/block some shots for img2img i'm down to run some tests a couple hours from now. ive used koikatsu for that before and the results were promising, but even mspaint can work.

tried a bunch of img2img cycles aimed at getting faces to bruise how i like them, im basically confident i can get adequate results in that arena now but it takes time and work.

did some initial exploration into debreasting as well, but that hasnt panned out yet.

>>30714
>img2img
Do you happen to have one that does NOT require a log-in?

Getting passable guro content out of NovelAI, at least, is very difficult in my experience, likely in large part due to how little there is of it on Danbooru, and I feel what success I have had to date is more a result of lucking out with the seed rather than actually engineering a good prompt. I'll keep trying, but I think a better training set is going to be necessary before we can consistently get half-decent results out of it.

When the team of sexy adventurers get party wiped.

>>30761
not sure what you mean by this. i host the AUTOMATIC1111 repo locally. if you're the person i was replying to before, what I was offering was to experiment with a base image to try to get quality inflation and popping. I'm not wanting to start from scratch on that since I'm still trying to solve some of my main fetishes, but if I have a base to work from I don't mind putting some time on it.

>>30781
Sorry to confuse. I'm new to AI and I can only practice patience until my new RTX arrives. How many images do you need?

>>30783
i think i can work with that one, ran out of time today but i'll put it at the top of the list tomorrow. haven't had much success with open wounds yet but i have gotten it to simulate disembowlment by putting intestines over the belly area. expecting to try a similar strategy in this case but with an emphasis on motion lines and keywords like that.

>>30784
Thank you.

>>30775 can you please make more this character
With beheading and dismemberment
Which AI site do you use for this?

>>30785
ok im not totally satisfied with this result but i learned some stuff so I'll give you a brief rundown for when you get your own setup.

went with breasts instead of belly for a couple reasons, i think belly will mostly be easier because you can put guts in front of it when you do img2img, and the AI knows more about intestines than it knows about the fatty tissue in the breasts.

the AI is decent at fixing shoddy coloring, so you can get expansion pretty easily by taking your first result from the AI into your image editor and just lassoing+resizing the desired area, then using a blending brush to push the skin color over the transparency.

i got lazy at the end so the breast pop isnt too well aligned with the breasts, so you can definitely expect better results than this if you're willing to put in the work. I think you could draw up a generic 2- or 3-color starburst that you could use to speed up the workflow, just pasting it over the expansion area and recoloring as desired.

photobashing guro from your collection into the image may not be super ethical if you're going to be trying to make Fanbox money or whatever, but that can also be a quick way to get shit done if you're good with selectors.

>>30785
>>30789

for reference this was the initial image i made to work from. i didn't end up using the img2img you provided, i forgot. but inflation might end up being a pretty good fit for AI art because of how often there's multiple versions of an image or multiple stages of inflation in one image to show progression. if you wanna get advanced you could probably use content like that to train an embedding or hypernetwork on inflation. there's also a syntax to have the prompt itself change partway through generation, but i forget how it's done.

>>30789
>>30790
Thank you so much for considering my request. While this is nowhere near what was envisioned, scrace reference material is to blame. Will you try red veiny abdomens?

merged berrymix earlier, got some decent results

does anyone know how to get results like these?

>>30551
Question for admin (or OP), I guess a new thread for "realistic" AI prompts is needed here, but which board?

>>31436
at the moment we are evaluating the implications for AI art, limits and their actual intrinsic value (as in if they should be considered spam or not)