AI Tutorials for spanking content

Welcome to SpankingForum!
Feel free to login or register using the links below.
Please send a message in the shoutbox below for password reset assistance.
Register
Sort by newest Sort by date
xaodj said:
You should be able to use it with CPU, it's slower but still works.
Click to expand...
I am trying this as "CPU only" since AMD doesn't work on Windows, but it is insanely slow to the point I am sure I've set up something wrong. Any tips or help?
 
What speed are you getting more or less?
Also what CPU do you have?
 
xaodj said:
What speed are you getting more or less?
Also what CPU do you have?
Click to expand...
Speed is so low I'm not sure how you'd measure it. Maybe fifteen-twenty minutes for a single sentence of text? Like I said, maybe that's normal but it seemed unusable to me at that speed.

CPU is a Ryzen 7 7840HS.
 
Sorry if it took me so long to reply, here is a script that I use with cpu.
It takes me about about a minute to generate a 20 sec clip with a 5ghz i5 CPU.

Edit: I forgot to add the repo I used:
You must be registered to view links and images

I have made it so that it has 2 identical sections (text set 1 and text set 2), so that you can alternate between generating with 2 characters if you need to.

It has both TTS with reference voice and voice conversion (optional).

Use reference voice for voice cloning and target voice for voice conversion.

In the top part, you can specify a subfolder name you want it to use/create, otherwise it will default to date and time (it will save the folders and files in the "saved" folder).

This means you could generate using an angry reference voice to get the emotion, then use target voice to apply the speaker voice you actually want.

Let me know if it works!

Here is the script:
You must be registered to view links and images

Here are the pip dependencies if you need to install them, just run these 3 commands:
pip install torch torchvision torchaudio --index-url
You must be registered to view links and images

pip install gradio soundfile numpy librosa
pip install git+
You must be registered to view links and images



Also here is a screenshot of the interface, for reference:
You must be registered to view images
 
Last edited:
I'll put in my 2 cents here and say that I've played around with Flux as well as QWEN and other "home-brewed" AIs, and I'll just say that any of us F/M lovers are in for a LOT of disappointment. Trying to generate a decent F/M image is virtually impossible right now due to the lack of good Loras.
 
After the free AI I was using got repetitive I decided to take the plunge and have installed Invoke to see how it goes. Images only, I'm not interested in creating videos (yet). I don't have a dedicated GPU, so I'll have to see how it copes. Was it my imagination or did someone post a link to an 'model' to install that was good for spanking imagery (not the cartoon, illustration one)? I can't seem to find it.

I was inspired by using Grok, wow did that produce some good (non pornographic, sadly) stuff.
 
Xenon said:
To add to this, here's another cool way you can get started (esp. helpful if you don't have a separate GPU in your machine) using Google Colab.

Setup is pretty easy (and free) – you should be able to generate your first AI image within 5 minutes!

I added some context about the tools being used in the spoiler below if you are interested; otherwise, feel free to dive in –

Google Colab
Google offers a product called "Google Colab," which is a free service that let's you write and run python scripts using Google's computers instead of your own. Everything is done through the web browser, so you don't need to install anything on your computer. Colab has recently expanded its capabilities so that users can now access Google's GPUs, which has opened the door to some cool AI applications.

Fooocus
Fooocus is an AI image generator with a user interface based on stable diffusion. A more detailed description of the implementation of Fooocus can be found via their
You must be registered to view links and images
.

Of note, I chose Fooocus for this tutorial because it's super easy to get up and running, though it's worth noting that the models are a bit out of date at this point (last repository update was 10 months ago). I would encourage anyone who is interested to do some more digging – there are a bunch AI image generators that leverage Google Colab out there! Feel free to share any better / more recent notebooks you come across 🙂

Setting Up AI Image Generation via Google Colab

  1. Open the
    You must be registered to view links and images
    (note: login to your Google account if prompted)

  2. [Optional] At the top of the page, click "Copy to Drive" – this just creates a copy of the notebook to your account so you can easily access it later on

  3. To start the notebook, click the "Run All" button (next to "Copy to Drive")
    You must be registered to view images


  4. As the code runs, you should start to see text output on the page – you can ignore all of this (it's just showing you the steps it is taking to install the AI image generator on Google's computers)

  5. After a minute or two, you should see two lines in the log similar to this:

    Running on local URL: http://127.0.0.1:7865
    Running on public URL: https://fa7e224dd3030e0412.gradio.live

    This means that the image generator has been installed and is ready to use! Click on the second link ("public URL") to open your personal AI image generator in a new tab.

  6. Once you have the image generator open, you can quickly get started by entering a prompt and pressing "Generate." However, I would encourage you to select the "Advanced" checkbox at the bottom, which will give you a lot more control over what the image will look like (including image size, number of image generated at once, etc.). Notably, you can also play around with different styles to make the images look more/less realistic
A few other notes worth mentioning:
  • There shouldn't be any limits to the app, so feel free to play around with it as much as you want!
  • Based on my initial testing, images take around 30 seconds to generate
  • It's worth noting that Google Colab notebooks can run for a maximum of 12 hours, so you will need to restart the app once that time has elapsed. Since this is using the free version of Google Colab, there's also a chance that you may not be able to run the notebook if Google is experiencing a heavy load from other users (I've never actually had this happen to me, but it is outlined in Google's terms and conditions).

App UI Screenshot (w/ a few examples of image output)


You must be registered to view images


Enjoy!
Click to expand...
I'm getting this error below when I tried this and I don't really understand what I've got to do to rectify it. If there is a more up to date one let me know.

I tried 'free' Grok just to try it out on women I found attractive in different outfits and it was fantastic. The minute paid for a subscription it turned into a (fairly) slow, very elderly relative who doesn't quite understand what you're saying and gives you the wrong thing.
 
Can anyone help? I'm enjoying using AI to create "spanking pictures". I have 2 problems:

1. Getting the view from the rear of a model. What words should I use?
2. How to create the actual spanking? Any recommended simple site to use?

Hope you can help
 
westpier said:
I'm getting this error below when I tried this and I don't really understand what I've got to do to rectify it. If there is a more up to date one let me know.

I tried 'free' Grok just to try it out on women I found attractive in different outfits and it was fantastic. The minute paid for a subscription it turned into a (fairly) slow, very elderly relative who doesn't quite understand what you're saying and gives you the wrong thing.
Click to expand...
I haven't used colab in a very long time, so I honestly have no idea how to fix this.
An alternative could de to use runpod, it's a service that allows you to rent a pc and pay by the minute (a good pc would cost 40-80 cents per hour).
They have templates for most ai image generation softwares.
At the end of this video the guy gives instructions to use Zimage on runpod:
You must be registered to view links and images
 
SaabMan said:
Can anyone help? I'm enjoying using AI to create "spanking pictures". I have 2 problems:

1. Getting the view from the rear of a model. What words should I use?
2. How to create the actual spanking? Any recommended simple site to use?

Hope you can help
Click to expand...

I use something like this:
Indoor scene, raw unedited amateur candid shot featuring a woman. The woman is looking back at viewer, after a spanking. She is crying, and she is wearing cotton panties. The background is a living room. She is facing a wall, and the photo is a head-to-toes full body shot. The skin on her bottom appears to be reddened from a recent spanking.


"she is looking back at viewer" or "looking over her shoulder" usually does the trick.
 
xaodj said:
I use something like this:
Indoor scene, raw unedited amateur candid shot featuring a woman. The woman is looking back at viewer, after a spanking. She is crying, and she is wearing cotton panties. The background is a living room. She is facing a wall, and the photo is a head-to-toes full body shot. The skin on her bottom appears to be reddened from a recent spanking.


"she is looking back at viewer" or "looking over her shoulder" usually does the trick.
Click to expand...
Many thanks I will try that
 
SaabMan said:
Still no joy. I can't get the aI to understand a spanking over a desk, knee, etc. Any suggestion for words to use?
Click to expand...
What AI generator are you using?

For Perchance there is no point even using the word 'spanking' it doesn't understand it.
 
SaabMan said:
Still no joy. I can't get the aI to understand a spanking over a desk, knee, etc. Any suggestion for words to use?
Click to expand...
I meant for a corner time style image after the spanking.
If you want an actual 2 people spanking pose image, you need to use an AI model that knows the concept.

I have only made them locally on my pc, with open source models.
The best IMO is Qwen Image with a spanking lora.
 
  • Like
Reactions: westpier
xaodj said:
I meant for a corner time style image after the spanking.
If you want an actual 2 people spanking pose image, you need to use an AI model that knows the concept.

I have only made them locally on my pc, with open source models.
The best IMO is Qwen Image with a spanking lora.
Click to expand...

Would you be willing to share which base model and lora you've had the best results with? I'm starting to experiment in Invoke.
 
I mainly use Qwen Image, but I've also had good results with Zimage.
As for the Lora, the one I've had best results with was made by Shagrath for Qwen Image, but I don't know if he shared it yet.
The other one is this one I made for Zimage, using Shagrath's dataset (i have included the lora and the comfyui workflow I use):
You must be registered to view links and images


Here are some images I generated with Zimage:

You must be registered to view images
You must be registered to view images
You must be registered to view images
You must be registered to view images
You must be registered to view images
 
Last edited:
  • Like
Reactions: Lorenzopaddle and westpier
xaodj said:
I mainly use Qwen Image, but I've also had good results with Zimage.
As for the Lora, the one I've had best results with was made by Shagrath for Qwen Image, but I don't know if he shared it yet.
The other one is this one I made for Zimage, using Shagrath's dataset (i have included the lora and the comfyui workflow I use):
You must be registered to view links and images


Here are some images I generated with Zimage:

You must be registered to view images
You must be registered to view images
You must be registered to view images
You must be registered to view images
You must be registered to view images
Click to expand...

Thanks for sharing the lora!

I'm using Invoke AI, not ComfyUI, so I can't use the workflow file directly, but I can see the prompt in it as well as the strength of your lora (1.2).

It looks like there's another Z-Image lora in there that I don't have (zimage\\spk_shagrath_zimagebase_000025000.safetensors), but I can't tell if that's necessarily being used in the workflow or if it's just an option that you select when you run it.

I tried the prompt in your workflow with this Z-Image Base checkpoint model (
You must be registered to view links and images
), thinking that the NSFW checkpoint would work well for this, but I'm getting some very strange/poor results.
 
samhel55 said:
Thanks for sharing the lora!

I'm using Invoke AI, not ComfyUI, so I can't use the workflow file directly, but I can see the prompt in it as well as the strength of your lora (1.2).

It looks like there's another Z-Image lora in there that I don't have (zimage\\spk_shagrath_zimagebase_000025000.safetensors), but I can't tell if that's necessarily being used in the workflow or if it's just an option that you select when you run it.

I tried the prompt in your workflow with this Z-Image Base checkpoint model (
You must be registered to view links and images
), thinking that the NSFW checkpoint would work well for this, but I'm getting some very strange/poor results.
Click to expand...
Sorry about that, the lora is actually the one I uploaded.
I had it previously named like that because I trained it on Shagrath's dataset.
Now I renamed it to differentiate between the the 2.

You should try with non finetuned checkpoints first in my opinion.
 
xaodj said:
Sorry about that, the lora is actually the one I uploaded.
I had it previously named like that because I trained it on Shagrath's dataset.
Now I renamed it to differentiate between the the 2.

You should try with non finetuned checkpoints first in my opinion.
Click to expand...

Thanks for the tip! I got it working with a vanilla Z-Image Turbo checkpoint. Here are a few that I thought turned out reasonably well:

You must be registered to view images
You must be registered to view images
You must be registered to view images


It would be helpful to know more details about the Lora such as which images it was trained on and which keywords were used to label them.

In particular, I found myself wanting to change the angle a bit to see more of a behind, full view of the ass, rather than the side profile view that you see in the hermione images, but I couldn't get it to do that no matter what I tried. It seems like the Lora was trained on a very specific pose and viewpoint, unless I'm missing something.

Occasionally I'd get a fully behind ass viewpoint but the image would be completely distorted and bizarre, and I also noticed that it wouldn't draw detailed exposed genitals even though I prompted specifically for it.

I tried a completely different model/checkpoint (
You must be registered to view links and images
) with
You must be registered to view links and images
and was able to easily get the more exposed angle and anatomy I wanted, but with not great faces and some other details I didn't like. Fortunately, with Invoke AI, it was pretty easy to use the canvas and inpaint tool to modify those.

Here's before inpainting:

You must be registered to view images


And here's after:

You must be registered to view images



I'm finding this very addicting to play around with!

Curious if any others on here have also tried.
 
  • Like
Reactions: westpier and cirdan
You must log in or register to reply here.