AI generated porn

[chess8]

Yes I saw some of those but I was not in mood to create huggingface account when I saw that 5700xt is not much faster than cpu :)
But I will give it a shot.

You don't need a huggingface account for that. AUTOMATIC1111 provides an alternate download link for the model in the readme. Onnx is indeed slower. I didn't looked much into it but perhaps this might help https://www.reddit.com/r/Amd/comments/y ... t/iukzipx/

[edger477]

Hmm I will check it out... I have 5700xt, which might not be any faster than running it on my Ryzen 5800x. I don't want to give money to NVidia, I might just wait to see how much perf/w the new Radeons give and if they have better tensors equivalent then bother with setting this up, and run in WSL on laptop till then (I have laptop with 1070 MaxQ with 8GB vram).

[chess8]

Hmm I will check it out... I have 5700xt, which might not be any faster than running it on my Ryzen 5800x. I don't want to give money to NVidia, I might just wait to see how much perf/w the new Radeons give and if they have better tensors equivalent then bother with setting this up, and run in WSL on laptop till then (I have laptop with 1070 MaxQ with 8GB vram).

DirectML, Vulkan (But that probably will never work beyond just images), OneAPI are all interesting ones. Unfortunately none of them can replace CUDA. Only ROCm is good enough to replace it for our use case, if only if AMD actually cared about their cards enough and added support for it on Windows. AMD kind of neglected that part of their user base, since most people don't want to buy a GPU just to try out ML, and not saying its easy, but NVIDIA managed to get CUDA on all their cards somehow, so its definitely possible for AMD too, if the pressure to do it is high enough, which it sadly isn't.

[Kapierik]

Thanks to hevelius i realised the importance of a thorougly trained model for the right purpose. Before, i was quite frustrated with the creation of pictures with a photorealistic style. Thats no wonder as a lot of companies take a sharp stance against NSFW AI art, stable diffusion was kicked from Kickstarter (pun intended) for trying to finance a (not exclusive but nonetheless) NSFW model. Stable Diffusion is now subsequently removing NSFW from their 2.0 version upwards.
With photorealistic style i mean it looking like real, or close to (uncanny valley effect im looking at you). Dedicated Hentai models were there from the beginning, but i missed that in the meantime some models emerged with a more realistic style, im currently catching up on that front.
So after playing around with new models a bit the last days i can show some of my results, let me know what you think of that style and if you consider that sufficient. Fingers are still the end boss of AI so dont wonder when they are missing, i spare you those monstrosities.

[shinini]

That looks nice! And absolutely realistic! Quite vanilla though ...

Quite sure that AI generated porn will be the future.
Being able to exactly mix your fetishes the way you want and get the result you've fantasized about sounds like paradise

However, we are not there yet.

One of my main fetishes is Tied Girls Kissing (you guessed it: viewtopic.php?t=13821). Recently, I tried to generate some pictures with Stable Diffusion, putting something in like "Two young women wearing pink bikinis having their hands tied behind their backs kissing each other". To my astonishment, the problem was not the kissing. Stable Diffusion had no problem/rules there obviously. Neither was the bikini. The problem was the bondage. Whatever keyword I've tried ("bound", "tied", "cuffed", "in bondage") - it never worked. Their wrists were always free ...

[Shattered]

The day when Eos has a built in 'generate a model' button for new teases is a day I look forward to Being able to type the exact image I want rather than scour the internet is a dream...that may come to pass by 2025 or so at this rate!

[Kapierik]

That looks nice! And absolutely realistic! Quite vanilla though ...

Quite sure that AI generated porn will be the future.
Being able to exactly mix your fetishes the way you want and get the result you've fantasized about sounds like paradise

However, we are not there yet.

One of my main fetishes is Tied Girls Kissing (you guessed it: viewtopic.php?t=13821). Recently, I tried to generate some pictures with Stable Diffusion, putting something in like "Two young women wearing pink bikinis having their hands tied behind their backs kissing each other". To my astonishment, the problem was not the kissing. Stable Diffusion had no problem/rules there obviously. Neither was the bikini. The problem was the bondage. Whatever keyword I've tried ("bound", "tied", "cuffed", "in bondage") - it never worked. Their wrists were always free ...

Something like this?

[shinini]

Not quite, unfortunately.

Their hands are not tied together (which is necessary to tingle main brain at its sweet spot )

With which engine/model did you generate it?

[Kapierik]

Not quite, unfortunately.

Their hands are not tied together (which is necessary to tingle main brain at its sweet spot )

With which engine/model did you generate it?

For a realistic western style the big three are Deliberate, URPM and RealisticVision.If you can put up with an asian style ChilloutMix, ChikMix and RealDosMix are also to consider. I used URPM, also tried it with Deliberate (attachment3) which gives a similiar result and with ChilloutMix (attachment1+2).

[bakerboy]

Any AI generated video ?? Maybe with a babe like this one?