seeking AI art generation advice

[cisco47]

I have been experimenting with AI art. I put up a couple of GIMP stories on this website here: https://ralphus.net/thegimpforum/viewto ... =16&t=1014 and here: https://ralphus.net/thegimpforum/viewto ... =16&t=1146 Hope some of you find entertainment in these.

I used perchance.org and tensor.art . Both put up decent bondage scenes with enough coaxing and editing. One lacks ability to do any GIMP under bondage in the same frame as her torturer and the other lacks the ability to illustrate male genitals and any sexual intercourse. Anyone know a AI art generator that does it all? I am no artist but am learning.

[cisco47]

hkaa.jpeg (27.97 KiB) Viewed 5082 times

cd (2).jpg (15.65 KiB) Viewed 5082 times

[cisco47]

Here above is a sample of Perchance.org and of Tensor.art. neither does decent horizontal whip wounds and I had to manually draw it into some of my illustrations.

[cisco47]

In addition, I have failed so far to prompt either websites to do a decent rack scene or rape scene.

[cisco47]

One big problem is when submitting a prompt, the result is often goofy. I found out recently that the technique is to continue to submit the prompt, then when a satisfactory image is created, to save the 'seed' which can be made part of the prompt. That way, the prompt can be tweaked with a decent image.

[torman2011]

I was in the same boat as you. I tried using Stable Diffusion on my PC to create GIMP images using a number of models from opensource market places. None of them can do a descent scene or draw wounds. I finally had to settle for using inpaint for creating facial expressions. How I work is I find an image with the kind of scene I want to create. I then use inpaint to create the facial expression and then photoshop for the peril bit. You can check out some of my works at - viewtopic.php?f=3&t=993

[Praetor3d]

I use 3d modeling to create stills or seeds for AI to work off of. For instance, if I'm making an animation using AI, I'll render the first and last frames so that AI can be guided to interpolate in between. Then I use Framepack or Hunyuan via ComfyUI on a local PC to allow for NSFW material. You typically have to use Loras to get certain explicit actions to work correctly.

Tbh, I find it's faster for me to 3d render something than to constantly babysit an AI to get it to do what I want. Typically, I'll use it for undressing/stripping and facial animations, but other types of actions are just too much work imo where things are. But the good news is the technology and models are evolving at such a fast pace, even for nsfw models.

[cisco47]

I see the difference in only the past couple of months. for example, playing with perchance.org, it seems to have learned to do multiple people in a scenario. I doubt that my prompting descriptions have gotten much better, but the software seems to be catching up with fewer goofy results.

[cisco47]

In making a series of images using text-to-image generation (I use the perchance AI website), I end up with random different faces in each image. In trying to get the same GIMP's face in each image, I find that lots of specific description raises the probability of getting the same basic face. For example: '(blonde, blue eyes, straight shoulder length hair ...)' I learned through trial and error that simply naming a well known celebrity also yields consistency. With Perchance, more often than not, the face does not really look like the targeted celebrity girl. But I get consistency, which, in my case, is the objective.

[yyy02]

Hello folks, would you mind posting any further AI info to: viewtopic.php?f=3&t=1247

There's so much good information that we wouldn't want to get lost among diverging threads. Thanks!