AI generated fetish images

[Like Ra]

Someone has taken the time to start training a bondage and fetish focused model, and it gives some pretty good results (sometimes). I've attached a few images I generated without a lot of testing.

Looks VERY promising! Where did you get the model? Please, post more!

Not sure if I can post magnet links here?

You can, absolutely.

Someone has taken the time to start training a bondage and fetish focused model, and it gives some pretty good results (sometimes). I've attached a few images I generated without a lot of testing.

Looks VERY promising! Where did you get the model? Please, post more!

Not sure if I can post magnet links here?

You can, absolutely.

I found it on the Unstable Diffusion discord server. One of the community members posted the magnet link (magnet:?xt=urn:btih:8563df35bf5baf34e5914ce062d41714a5739ac4&dn=bondage-v10-14k) after training it themselves. Just drop it into your UI of choice like any other SD model.

They posted some previews along with the model and they are very promising seen as the model hasnt been trained on a massive dataset and its just one guy. Specifically this chair one:
   
(inpainting was used to fix the face)
and this bitgag one:
   
(no inpainting was used on the face)

Ive attached a few other ones I generated of some more positions/gear (the last three are using a similar prompt to the preview chair image above)

[Like Ra]

   

I like the rendering of this sheer, shiny nylon suit!

This looks like it could be a lot of fun, especially with a little more effort put in via inpainting and masking.
Should be interesting to merge this model with others.

Promising 😋

inpainting and masking

Took an image (also AI generated, in case you're wondering 😁 ) and used masking around the mouth area to apply a ballgag (and some drool)

No further post processing in an external program, just some upscaling/shrinking on the third one for some extra sharpness and detail.

Not perfect, but serviceable.

Kinda impressed that the AI picks up on the red light source to the right in the original image (the images above have been cropped, the original is larger) and adds a reflection and some ambient occlusion to the right side of the ballgag and to the piece of strap, even though the bounding box for processing was just around the face.

I'm just going to leave these here ...

Poor Mina ...

           


It's pretty easy to spot where the training material is coming from.
Adding "redhead" will almost always give you a dead ringer for Sarah Brooke (but no Eve Ellis so far, what a pity)

   

[Like Ra]

Adding "redhead" will almost always give you a dead ringer for Sarah Brooke (but no Eve Ellis so far, what a pity)

   

That's Emily Marilyn IMO https://www.likera.com/blog/wp/archives/3198

[Like Ra]

and used masking around the mouth area

How did you do that?

and used masking around the mouth area

How did you do that?

I'm still using InvokeAI, which was a bit slow out of the gate, but has made a lot of progress with its UI and quality of life changes. Highly recommended. It also comes with a one-click installer basically, which lowers the entry bar even more (the main basic prerequisite is a Python install if I remember correctly, but the installation guide is concise and comprehensive).

Starting with v2.4, they introduced their "unified canvas" as part of the WebUI. Check their Youtube channel for an introduction of its main features, tools and terminologies, can't explain all of them here.

One of which is masking. Using the mask brush, you just paint over the area you want to replace.
So you pick an image, upload it to the unified canvas, switch to the appropriate model (in this case, the Bondage model), set the bounding box to the area you want to edit, switch to the mask tool and mark the inpainting area.
Put in a text prompt, "red ballgag with leather straps, drool" and hit Invoke. Everything "masked" will be replaced, the rest will not be touched.

You will need to fiddle with the Image strength slider - the lower, the more it will try to preserve the original image data underneath, if you go to high, the seams will become very obvious and the new parts will not blend in nicely.
Somewhere between .4 and .6 is the sweet spot usually.

If the results are unsatisfactory, you can also switch to the base layer mode and paint the gag in very broad strokes with the paint tool to jumpstart the AI on what should go where.

Next time I do an edit, I will try to remember to make a screen recording of the process.

 

That's Emily Marilyn IMO https://www.likera.com/blog/wp/archives/3198

I can see the resemblance to Emily Marilyn, but my money is on Sarah Brooke.

[Like Ra]

I'm still using InvokeAI

InvokeAI is supported across Linux, Windows and macOS. Linux users can use either an Nvidia-based card (with CUDA support) or an AMD card (using the ROCm driver).

Oh, AMD is supported! That's interesting!