What follows is some geeky talk about AI generated art. Please skip down to TLDR if you just want to read the conclusion...
Here are some detailed comments on my attempts at producing
AI generated art (i.e., "Tia's Army"). The first thing I tried was
downloading GUIs (AUTOMATIC1111, InvokeAI) so as to use Stable Diffusion
locally. This would have been by far the most flexible and powerful option,
allowing the use of various advanced tools and a potentially enormous selection
of fine-tuned models (you can even train your own custom models). Unfortunately,
the GPU in my PC is so old that it's not compatible with the relevant software—very
disheartening!
Anyway, I still wanted to play around with Stable Diffusion,
so I used the free functions available at dezgo.com. The big advantage of this site
is that it allows NSFW content (no filters on the words permitted in prompts).
Online services that allow NSFW content seem to be few and far between!
I drew some rough sketches, then used image-to-image to get
the AI generated pics. I produced everything apart from Tia's head using an
anime style model (Anything 5.0), but didn't like the anime-type faces it produced,
and so created Tia's face using a different model, then edited everything
together (the guy's dick, Tia's body, and Tia's face, etc.) using Gimp and
Inkscape.
TLDR: Despite the very limited tools, I still managed to
produce something I liked (at least enough to use as personal "fap" material!) Anyway, I was so
blown away by the possibilities of this technology that I've already decided to
get a newer GPU so I can run Stable Diffusion locally—and thereby harness its
real potential! Exciting stuff!
It is really impressive to see how much the AI let you do in such a short span of time. Especially given the technical limitations you are working with.
ReplyDeleteI hope that using AI will make your future projects easier, especially given that you are upgrading your hardware.
A few words of warning about AI though. For one, AI art styles can end up looking a bit "samey", but it sounds like you worked around that by using different models and good editing.
I guess what I'm trying to say is that AI are tools that can help you but you shouldn't let them just do everything without your own input. It looks like you are not running into that trap, but I wanted to warn you nonetheless.
Another weird issue is that a lot of AI have problems drawing hands and feet. For whatever reason it just drives them nuts, so keep that in mind in the future.
Either way I'm looking forward to more projects from you in the future! I hope the AI makes your life easier.
Thank you for your interesting comment!
DeleteAs impressive as it is, the current AI that generates artwork is not at the stage where it can "understand" and represent the physical relationship between multiple characters. Perhaps for that reason, the majority of the AI generated art that you see is basically of the portrait (or pinup) variety, which is relatively easy to produce.
In my case, the illustrations are there to be part of the story, so I start with fairly specific requirements about what they will depict. For that reason, the challenge will be to find a style that's attractive, but which still allows me to produce the bits that the AI can't cope with. For example, if I produce very detailed artwork that's a good way towards being photorealistic, then it becomes very onerous to produce parts or interacting parts that the AI can't generate correctly. As you pointed out, AI is poor at generating hands, because they are so complicated, and appear completely different depending on the hand posture (although the most recent AI is reportedly much better at hands than before). So, if I needed a picture of two characters shaking hands, then I might need to redraw the hands myself to get a satisfactory result, and that's much harder if the rest of the picture is in a detailed, highly realistic style.
Anyway, I'm looking forward to experimenting, and hopefully developing a workflow that enhances my artwork while still allowing me to produce the picture content that I want.