• Sign up or login, and you'll have full access to opportunities of forum.

My AI torture art

Go to CruxDreams.com
Nice pic but I notice when AI makes bondage pics that usually chains or ropes connected to cuffs lead in a direction that will not properly secure the victim. If AI can't fix this, maybe a program like microsoft paint and somebody that knows what to do wit that can do magic.
 
Really well-made images! One can see that you put a lot of effort into making them, like paying attention to the facial expressions, or cleaning up artifacts by postwork. I also like the electricity 'specil effects'. You might want to try the prompt "motion blur' to get these dildo things REALLY to work more...

I am also using the pony model a lot. What I haven't quite igured out are the booru tags, like the list here. Some seem to work, others are totally ignored.
 
thecuriousone said:
Nice pic but I notice when AI makes bondage pics that usually chains or ropes connected to cuffs lead in a direction that will not properly secure the victim. If AI can't fix this, maybe a program like microsoft paint and somebody that knows what to do wit that can do magic.
Click to expand...
Hah well, at least the cuffs are connected to something in this case! But I feel that she's pretty well secured with that massive energy beam hitting her from both sides... :devil:

These sorts of details are usually fixable in photoshop, given enough time, but in this case I used up all of my photoshop stamina trying to repair her face and body, which were seriously messed up in the raw version.
 
jucundus said:
Really well-made images! One can see that you put a lot of effort into making them, like paying attention to the facial expressions, or cleaning up artifacts by postwork. I also like the electricity 'specil effects'. You might want to try the prompt "motion blur' to get these dildo things REALLY to work more...

I am also using the pony model a lot. What I haven't quite igured out are the booru tags, like the list here. Some seem to work, others are totally ignored.
Click to expand...
Thanks jucundus! Your art (both AI and 3D) is amazing! With the pony-derived models I've definitely found that some of them are much better than others at certain themes. Like some of them will have no problems doing a plausible electrocution scene, whereas others will struggle with the same prompt. Still haven't figured out why this is. For now I keep hopping between models trying to find a good match for each scene....
 
2amjockey said:
Ah well, like I said I am a beginner myself and am still figuring things out! But I can tell you what little I have learned:

- If you want to generate torture scenes, you need to use a model that has been trained on such scenes! Most of the AI generators out there either barely understand what a vagina is, or have been trained only on mainstream porn and definitely have never seen anything like sexual torture.

- The best option right now (that I know of) is Pony Diffusion and its derivatives, a Stable Diffusion base model trained on basically the entirety of the boorus (sites like danbooru, etc.). These sites have a decade+ of carefully labelled hentai drawings of almost every possible kink, including many we like here. And since the tagging system on these sites is so good, it's easy to know what keywords to use in a prompt.

- Since the pony models were trained on drawings, to get more realistic results you need to use a derivative that has also been trained on real photos, and/or LORAs (mix-in mini models) that understand realistic nudity. This is the part I am still figuring out -- how to refine the raw pony-generated images into something closer to the style I want. Not quite there yet...

- Be prepared to spend an hour+ on each image in photoshop fixing glaring flaws...
Click to expand...
Stable diffusion,Pony diffusion,Lora's, Booru's, cleaning up in Photoshop. Sounds like AI is a whole discipline of education and work. Wish I only spent an hour cleaning up a page ,LOL. Your results are a major step forward in AI torture images from what I've gleaned from the internet. I particularly like the lighting realism of the cunt coil. I could have great fun learning this sort of thing but the curve to get to what I want to see would really distract me from a very busy schedule. One day. One thing I'm still not sold on with AI, other then fingers and toes and body parts morphing into solid objects is the lack of real wound presentation. Whipping, stumps , burns etc. Lots of times its just a superficial splash of blood. Though I'm completely jealous of skin, sweat and shine presentation. Keep up the great work. PS. What sort of computer might I need to get started and do you touch up with a tablet and stylus (draw another words).
 
zerns said:
Stable diffusion,Pony diffusion,Lora's, Booru's, cleaning up in Photoshop. Sounds like AI is a whole discipline of education and work. Wish I only spent an hour cleaning up a page ,LOL. Your results are a major step forward in AI torture images from what I've gleaned from the internet. I particularly like the lighting realism of the cunt coil. I could have great fun learning this sort of thing but the curve to get to what I want to see would really distract me from a very busy schedule. One day. One thing I'm still not sold on with AI, other then fingers and toes and body parts morphing into solid objects is the lack of real wound presentation. Whipping, stumps , burns etc. Lots of times its just a superficial splash of blood. Though I'm completely jealous of skin, sweat and shine presentation. Keep up the great work. PS. What sort of computer might I need to get started and do you touch up with a tablet and stylus (draw another words).
Click to expand...
By the way, Mr. Zerns, where did you learn to give such realism to injuries? They look like medical-grade illustrations.
 
zerns said:
Stable diffusion,Pony diffusion,Lora's, Booru's, cleaning up in Photoshop. Sounds like AI is a whole discipline of education and work. Wish I only spent an hour cleaning up a page ,LOL. Your results are a major step forward in AI torture images from what I've gleaned from the internet. I particularly like the lighting realism of the cunt coil. I could have great fun learning this sort of thing but the curve to get to what I want to see would really distract me from a very busy schedule. One day. One thing I'm still not sold on with AI, other then fingers and toes and body parts morphing into solid objects is the lack of real wound presentation. Whipping, stumps , burns etc. Lots of times its just a superficial splash of blood. Though I'm completely jealous of skin, sweat and shine presentation. Keep up the great work. PS. What sort of computer might I need to get started and do you touch up with a tablet and stylus (draw another words).
Click to expand...
Some of the models can do convincing wounds...I've made a few images featuring complete cunt removal that are not too bad (but which I probably can't post here).

Anyway, it's not hard to get started and play around wth it. For total freedom you'd want a local installation of Stable Diffusion and a beefy GPU, but it's not necessary. There are good free services out there like tensor.art, which I recommend. A free account there allows about 100 generations per day and exposes all of the features you need. It has a good selection of models, including many pony models, and lets you upload your own as well when you get more advanced. And no censorship that I can detect, which is important for the sort of images we're talking about!

Basically you just select "text2img" if you want to create a brand new image from a prompt, "img2img" if you want to start with an existing image, and "inpaint" if you want to change only part of an image, Then select your model -- for torture as I said I recommend the pony models that have also been trained on real images, which are easy to find. If you want to mix in a certain character or kind of girl, you can also add a LORA for her as well to supplement the main model. Then craft your positive and negative prompts using either descriptive sentences or keywords that you know the model was trained on (for example, tags from danbooru), or the keywords for your LORAs. The docs for the model will have suggestions. Some models will do well with certain themes and understand more or less what you're after, while others won't. Lots of trial and error. There are also some important settings you need to play around with like: "CFG Scale" (controls how closely the model obeys your prompt), Sampling Steps (more is usually better), VAE (some models need a custom one), and Denoising Strength (when doing img2img, controls how close to stay to the original).

Usually you eventually end up with an image or a couple of images that have approximately the elements you want. Then you maybe take the screaming face from one, the bleeding pussy from another, and photoshop them all together, or use img2img or inpainting to refine individual elements using AI. Then, unfortunately, you do usually need to go into photoshop and manually redraw little details on a tablet, remove deformities, smooth out the transitions between different combined elements, etc., until it doesn't look completely ridiculous. It can be a lot of work, but not nearly as much as drawing something from scratch. But of course you don't have 100% control over every detail, which can be frustrating when you want something very specific....
 
2amjockey said:
Hah well, at least the cuffs are connected to something in this case! But I feel that she's pretty well secured with that massive energy beam hitting her from both sides... :devil:

These sorts of details are usually fixable in photoshop, given enough time, but in this case I used up all of my photoshop stamina trying to repair her face and body, which were seriously messed up in the raw version.
Click to expand...
In addition, the AI's always have the habit of not displaying the number of fingers or toes correctly. Often the characters have 6 fingers or toes and with others it is possible that they worked in a sawmill and only have 3 or 4 fingers. I think that you have to edit them with Photoshop then.
 
2amjockey said:
Some of the models can do convincing wounds...I've made a few images featuring complete cunt removal that are not too bad (but which I probably can't post here).
Click to expand...
Would love to see such artwork that is too violent to be posted on this forum. Do you upload it anywhere else, like Visions of Darkness or Dark Fetish Net?
 
2amjockey said:
Then, unfortunately, you do usually need to go into photoshop and manually redraw little details on a tablet, remove deformities, smooth out the transitions between different combined elements, etc., until it doesn't look completely ridiculous. It can be a lot of work, but not nearly as much as drawing something from scratch. But of course you don't have 100% control over every detail, which can be frustrating when you want something very specific....
Click to expand...
Hugin&Munin said:
In addition, the AI's always have the habit of not displaying the number of fingers or toes correctly. Often the characters have 6 fingers or toes and with others it is possible that they worked in a sawmill and only have 3 or 4 fingers. I think that you have to edit them with Photoshop then.
Click to expand...
You rarely need to use an image editor like Photoshop to fix deformities or add details as an after work. Most of the time, inpainting with, or without control net images would suffice. It just takes a bit of getting used to before you can use them proficiently. The same goes for combining different elements, although more serious compositions are best be done with regional prompting.

In fact, manual drawing is best done before, rather than after an AI process. It can be a great way to add details, if used that way. :)

(Those brand marks in my profile image were made in such a manner. It’d be very difficult to hand paint them to look as realistic. But you can draw them first, then use AI to enhance the result, which would be much easier.)
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom