Well, at least real artists haven't been replaced yet. They can still have work touching up mistakes the AI make, such as drawing hands.
Ah, the “AI art vs “real” art” discussion is amusing to me. I see so many people lamenting AI taking away jobs from real artists, yet I’m not sure how many real artists are even out there. A lot of modern artists have given up learning figure drawing, perspective, and properly rendering light and shadow, and give us stuff that’s just stripes or paint splatters on canvas. Yeah, AI should take their jobs.
Otherwise, though, for the people who aren’t just writing a prompt, hitting the button, and calling it a day, I think AI is going to be more of a tool to work with that can help replace the tedious work.
Watching the process of SD, I can see how it moves from blocking out basic shapes to building up details, much like a real artist would. And I can see a lot of places where I would want to do more to guide that process, doing things like painting the rough image myself, editing the output and running it through again, building 3D models to better match what I want to see from the characters, photo-bashing scene elements to show what I want the output to look like. At that point, the AI is really just doing the time-consuming, tedious details.
And, yeah…hands. Until AI actually learns proper anatomy, I think artists are going to still be spending a lot of time fixing things like hands.
You're still in 2022, my dude. Head on over to /hdg/ on /h/ to get up to date with LoRAs, controlnet, and the like.
While I personally don't think the game needs any AI CG, here's hoping it works the way you want.
Well, I started with TI because it was built-in to Automatic1111 and I liked the idea of not tearing through a ton of HD space every time I wanted to train a face, outfit, hair style, etc. Plus, most of my experiments have been on trying to preserve characters who were created by the AI in the first place, meaning there shouldn’t be any external data needed in order to do it…
But, yeah, it’s been hit-and-miss and very time-consuming. I’ve tried a couple of LoRAs, but I really need to find some good guides that are written for dummies that explain what all the settings are.
And, yes, I’ve been playing a lot with ControlNet as well. Did some previous experiments to see if I could get the same character with different breast sizes via drawing with Scribble or painting on top of depth maps. Got some interesting results.
There’s really just so much to explore that it’s overwhelming. I’ve been at it for months now, and I’ve got thousands of images from different experiments. I‘ve got a blog and Patreon/SubscribeStar post written up about some of what I’ve worked on, but I need to go through the hassle of organizing all of the images and actually
making the post to show people more of what I’ve been up to. (Not to mention a lot of playing around with ChatGPT for coding and planning help…)
