AI generated images

[Condemned One]

Thank you so much, these are so great! Love picture #1 most. More nails, skin and hair, please! Can't wait for it! Would you also try crucified beauties being burned?

Will certainly try. Been trying to achieve nails for some time. Have tried multiple checkpoint merges but honestly think I’ll need to train my own LoRA to achieve it without impacting my aesthetic. Lots of work involved.

I’ll get there eventually but for now may have to rely on some of the fantastic manip artists here to help me out. My skills just aren’t there yet. Someday!

 

[Condemned One]

Thank you so much, these are so great! Love picture #1 most. More nails, skin and hair, please! Can't wait for it! Would you also try crucified beauties being burned?

Oh, and yes(ish) on the fire. More to come!

 

[fallenmystic]

Will certainly try. Been trying to achieve nails for some time. Have tried multiple checkpoint merges but honestly think I’ll need to train my own LoRA to achieve it without impacting my aesthetic. Lots of work involved.

I’ll get there eventually but for now may have to rely on some of the fantastic manip artists here to help me out. My skills just aren’t there yet. Someday!

I have a suggestion for all who love crux and AI.

I believe the current generation SDXL models can generate a crux scene without even requiring much inpainting or using ControNet images. The only missing puzzle is the hands, how they should face forward, and the wrists pierced with nails.

As Condemned One pointed out, the best way to deal with the problem is to train a dedicated Lora. The good news is that it's not even difficult. You only need 20 or more high-resolution (1024p) images of nailed hands and someone to run a training program over them.

Granted, a training program typically has a daunting number of options. But there are many guides to help you choose decent settings. And it can be done for free or $1-2 at most on a cloud instance.

As we have quite a few people showing interest in AI-generated images now, I believe it won't be too difficult a task to open up a thread to collect source images. And if you can find just one person willing to spend 2-3 hours & $1-2, everyone will be able to generate proper crux images with nailed hands easily.

I feel reluctant to be directly involved in the effort since crux isn't my thing. But I've trained a few Loras, so I can guide those willing to try if needed.

 

[Condemned One]

I have a suggestion for all who love crux and AI.

I believe the current generation SDXL models can generate a crux scene without even requiring much inpainting or using ControNet images. The only missing puzzle is the hands, how they should face forward, and the wrists pierced with nails.

As Condemned One pointed out, the best way to deal with the problem is to train a dedicated Lora. The good news is that it's not even difficult. You only need 20 or more high-resolution (1024p) images of nailed hands and someone to run a training program over them.

Granted, a training program typically has a daunting number of options. But there are many guides to help you choose decent settings. And it can be done for free or $1-2 at most on a cloud instance.

As we have quite a few people showing interest in AI-generated images now, I believe it won't be too difficult a task to open up a thread to collect source images. And if you can find just one person willing to spend 2-3 hours & $1-2, everyone will be able to generate proper crux images with nailed hands easily.

I feel reluctant to be directly involved in the effort since crux isn't my thing. But I've trained a few Loras, so I can guide those willing to try if needed.

Would love some pointers to get started! I’m know the basics but still very new to all this.

 

[fallenmystic]

Would love some pointers to get started! I’m know the basics but still very new to all this.

I haven't used it myself yet, but I suspect the best & most straightforward method is using the new online trainer Civitai provides.

There are also a few Google Colab notebooks you can use for free, like this one, for example.

The method I used is running Kohya's GUI on an instance I created on RunPod.

As you can see, many options are available, and things are always in flux as the field moves rapidly.

So, I'd suggest trying the Civitai trainer first, and if it doesn't work well, find the latest Lora training guide on Civitai (there's an article/guide section) and follow its steps.

 

[Hornet1ba]

I haven't used it myself yet, but I suspect the best & most straightforward method is using the new online trainer Civitai provides.

There are also a few Google Colab notebooks you can use for free, like this one, for example.

The method I used is running Kohya's GUI on an instance I created on RunPod.

As you can see, many options are available, and things are always in flux as the field moves rapidly.

So, I'd suggest trying the Civitai trainer first, and if it doesn't work well, find the latest Lora training guide on Civitai (there's an article/guide section) and follow its steps.

Thanks!
I will probably try the Civitai trainer with my own 3d renders or others, should i try one Lora for the hands and another for the nailed feet or combined both of the limbs ?

 

[fallenmystic]

Thanks!
I will probably try the Civitai trainer with my own 3d renders or others, should i try one Lora for the hands and another for the nailed feet or combined both of the limbs ?

I can't be sure without trying myself. But my gut feeling tells me that making one general crux Lora would be feasible. At worst, we can always split the same training dataset to create more specific Loras if it doesn't work.

As for the source images, I think it'd be best to mix both 3D renders and photos (e.g. screen captures from films or TV shows, preferably upscaled using AI) so that the Lora doesn't mistake a particular style (3D or photorealistic) is crucial to the concept of "crucifixion".

Note that you need to specify only such circumstantial traits in the captions, so you shouldn't include terms like "cross", "nail", or "blood" because they are essential parts of what it means to be crucified, while things like "3D render" or "photo" better be included.

 

[Condemned One]

Continuing to try various prompts… with varying results

 

[lbwbl_bats]

these are wonderful! Please continue.

 

[Laura1949]

I find of interest that AI produces images of nude female bodies at once toned and fleshy.

View attachment 1401439

I love these.

 

[Peter_the_giant]

Merry Christmas to you all.

 

[Mikerytel]

Merry Christmas to you all.

View attachment 1408543

Ms clause after santa found a new woman

 

[Zara]

After Santa finally gets off work the real party can begin in North Pole village.

 

[Yupar]

Continuing to try various prompts… with varying results

The last two are done pretty well

 

[Yupar]

...

very nice set expecially 3 4 5

 

[torturer4]

With the exception of crucification there is not enough tortures depicted

 

[Peter_the_giant]

With the exception of crucification there is not enough tortures depicted

It's simply because there is a lack of images of that kind to train AI tools on, plus some other fundamental restrictions of the tech. Like it's not good at separating multiple characters. They all bleed into each other.

so we do what we can, fuelled by our horniness and creativity

 

[Condemned One]

Peter is so right. It can be difficult to just get
the AI to generate a person doing what you want. It takes a lot of tweaking and negative prompts to produce a useful image. Thankfully you can continue to learn and train your model so it’s getting better.

Here are a few funny examples of an AI artists struggle. The AI has no concept of reality or physics. So frustrating when it’s so close but so far from what you actually want to produce.

 

[LukeBlanc]

Peter is so right. It can be difficult to just get
the AI to generate a person doing what you want. It takes a lot of tweaking and negative prompts to produce a useful image. Thankfully you can continue to learn and train your model so it’s getting better.
Here are a few funny examples of an AI artists struggle. The AI has no concept of reality or physics. So frustrating when it’s so close but so far from what you actually want to produce.

Add stiches in the middle of the second one, you get the gruesome result of a particularly devious executioner, who unleashed his dark fantasies on a pair of willing (if I believe the grin on the lower one's face) sisters.

 

[Peter_the_giant]

Here's a fun little one I did for someone who asked nicely

From a group of slutty friends having fun:



To a group of enslaved sluts:



What do you think should happen to them next?