AI generated fetish images

I thought I would just send an image or two through the new flux2klein4bFp8 model using a text prompt on an existing image and was amazed.

The input images were very low quality grey scale bondage images that I downloaded from this site in the magazine area.  

The images were upscaled, colorized and given a lot more details than the originals had.  The only bad thing is that the model does not know what a cloth gag is, but I am sure a lora could be trained with bdsm themed images to fix that.

This is much better than any previous ai model worked at this task before.

I thought I would just send an image or two through the new flux2klein4bFp8 model using a text prompt on an existing image and was amazed.

The input images were very low quality grey scale bondage images that I downloaded from this site in the magazine area.  

The images were upscaled, colorized and given a lot more details than the originals had.  The only bad thing is that the model does not know what a cloth gag is, but I am sure a lora could be trained with bdsm themed images to fix that.

This is much better than any previous ai model worked at this task before.

Using this to convert a bunch of old files. Other than the tendency to put heads on backwards it is doing an amazing job stilll.

[Like Ra]

new flux2klein4bFp8 model

Can we have more details, please? 😇 E.g. about the GUI, prompts, LORAs?

Very nice results. Can't blame that one with upper body backwards, because that drawing wasn't very detailed and with very unusual position.

This is all with no loras yet!!! 

This is just using latest comfy ui with flux2klein4bFp8 workflow.

These are the directions I used.  I had to update to latest version of everything.

https://www.patreon.com/posts/flux-2-kle...dium=email

Then I used this link: https://www.patreon.com/file?h=149436927&m=605206778 to download the text image editing workflow.  

I did have to bypass the  Sage patch node and model patch torch settings to make it work.
 
Then I loaded the images into the input directory and went through each image in turn, running it with just the generic "hyper realistic photograph, enhanced, " prompt.

Going though a second pass with colorized black and white images to see if that helps informing the ai on the images it is struggling with and will try to customize the prompt to fix errors.

Trying out the 9b version of the same model.   Don't have a setup for this, just downloaded the 9b models and bigger qwen text model and got it working.  Used 3 loras that are named in the image in comfy ui. 

The flux 2 klein 9b model is better than the 4b version for quality, text and number of loras being made for it. 

I was able to keep adding layers to the prompt and it just followed each new thing I asked of it. 

Show Content

A Bambi image

       

[Like Ra]

A Bambi image

These images are perfect for at least two threads:

https://www.likera.com/forum/mybb/Thread...minization
https://www.likera.com/forum/mybb/Thread...-by-Gemini

Remotely controlled and monitoring devices on the neck, arms and penis for forced feminization.

[Like Ra]

Trying out the 9b version of the same model. 

Could you please drop the links here? I do not see 9b mentioned on SOTAI Patreon (I might have missed them, though...)

[Like Ra]

This thread has reached 100k (100,488) views!

Trying out the 9b version of the same model. 

Could you please drop the links here? I do not see 9b mentioned on SOTAI Patreon (I might have missed them, though...)

I just kind of found them on civitai and put them in the right place.

If you drag the image into comfy ui it shoulld give the model name to look for, put them in the same comfy ui directories the 4b models went into. 

I had to try a bunch of stuff, but the image file should have a workflow in it.