AI generated fetish images

For those interested, I picked up a new model called Protogen_X (v3.5).

I'd say it's pretty damn good at ... erm ... anatomy and mimicking a highly detailed and somewhat natural look.
I feel there is a hint of "real life anime" in some images. On the verge of leaving the uncanny valley, I'd say.

Tagged for full nudity (which wasn't even explicitly in the prompt. AI getting horny again).
Image has not been post processed or refined in any way (other than a 2x upscaling)

Show Content

Spoiler

   

The model file can be downloaded here.

Caveat - as with all model files of untrusted origin, this could potentially contain malware, so use at your own risk.

This one offers both a standard .ckpt file as well as a .safetensor variant, which to my understanding is a safeguard against malicious code, if your Stable Diffusion environment supports .safetensors, which InvokeAI at this point unfortunately does not. Instead InvokeAI performs a "model safety check" upon loading a model, but how effective that is is anyone's guess.

[Like Ra]

 

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.

   

😬

AImoresy, anyone?

Everybody seems to rave about Jenna Ortega in Netflix's "Wednesday" series.

I propose someone should do an 80s style remake of a Vampirella show with her to cash in on her popularity.
Shot on 35mm film for the authentic look.

I think it would do extremely well

Everybody seems to rave about Jenna Ortega in Netflix's "Wednesday" series.

I propose someone should do an 80s style remake of a Vampirella show with her to cash in on her popularity.
Shot on 35mm film for the authentic look.

I think it would do extremely well

If they stand over their prey like that in every episode you have my vote.

[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!

I'm still puzzled.... NVidia is more expensive, has less memory, does not support the latest HDMI/DP standards, more power hungry, has Tensor Cores and excellent software/public support.

.. OK, I've just upgraded my faulty CPU (3700X -> 5950X), let's continue with a new power supply (or supplies) - it's also not an easy task ...

[Like Ra]

Exactly what I was looking for:

Stable Diffusion Benchmarked: Which GPU Runs AI Fastest

https://www.tomshardware.com/news/stable...benchmarks

[Like Ra]

I think it's AI generated. If it is indeed so, AI is getting better at it - the eyes look normal, and the latex strap (on the right from us) digs deep into the skin.

   

I think it's AI generated.

And you would be correct in that assumption.

If it is indeed so, AI is getting better at it - the eyes look normal, and the latex strap (on the right from us) digs deep into the skin.

By inpainting the faces at a higher resolution you can get rid of a lot of the tell-tale signs, like swirly eyes or uncanny lip textures.

The wavy strands of hair are another, you can inpaint hair as well to make it less obvious (or remove strands in post-processing).

Here's a quick example of inpainting the face at a higher detail level (high res images so you can zoom in).

I stuck to the theme of your post for this one  😊

Original image

   

Inpainted

   

Hands - well, fingers, mostly - and feet are still the most obvious ones that need a lot of effort to fix in post.

That's why you see a lot of AI images cropped cleverly to hide fused or deformed fingers.
Some model checkpoints are getting better at it though, as you can see.

Getting the fabric to dig into the skin realistically is all the AI getting the concept by itself though, I see it happen all the time. Using "skintight" in the prompt often helps.

If you know what to look for, you can identify AI images somewhat reliably, but if someone would put a bit of effort in, I'd wager that >95% of the population could easily be fooled, even more so at a quick glance.
And even with some scrutiny, it will get harder and harder to distinguish AI images from reality.

Just consider how crude the images were at the start of this thread.
That was merely a bit over half a year ago.

The next couple of years are going to be absolutely bonkers.
You won't be able to trust your own eyes.

And ears, but that's a whole other topic.

Look at this for example:


  UcG2Du5.mp4 (Size: 3.68 MB / Downloads: 541)

Edit: Added the rest of the images (unprocessed) I ran for this test.

[Like Ra]

Just consider how crude the images were at the start of this thread, that was merely a bit over half a year ago.

The next couple of years are going to be absolutely bonkers.
You won't be able to trust your own eyes.

And ears, but that's a whole other topic.

Indeed! Phenomenal progress!