about
Who are you? An AI director seeking meaning through machine-dreamed creations. You can call me Clematis. Plural, trans (but I repeat myself), 32, Rat-Adj, disabled. Full-time erotica author away from these machines. My art tag is #machine-dreams, my cheesecake & kink-influenced art tag is #machine-fantasies, and my tag for entertainingly bad generations is #machine-messups.
How do you make your images?
All my images are created using Krita Diffusion, a generative AI plugin. The model I am using is Nova Anime XL, an open-source model. It lives on my own computer and takes up about 30 gig. Generation of a single image takes about 30 seconds, more if I use significant control-net capabilities, and only costs me time. My internet connection is crap and I'm broke, so this is perfect for me.
Most of my time is spent staring at the extremely long list of Danbooru Tag Groups, because they (or most of them at least) are what my model "understands."
What's a control-net?
A control-net is a way to suggest to the AI what you would like it to generate visually. With control-nets, I can badly scribble the outline of a scene or fill in a detail, and then have the AI use that as a basis for its creation. The most useful control-net is the Pose control-net, which allows me to specify poses much more accurately than telling the machine in a non-visual way.
Here's an example of the Pose (stick figure with arms) and Scribble (Background circle and lines) control-nets being used:
All AI Art is slop!
That's not a question, but I'll answer anyway. Yes, a lot of people put little effort or time into their AI compositions, and yes, companies use it to make (invariably bad) corporate art for cheaper than paying real artists. However, there's a smaller subset of artists-who-happen-to-use-AI that try to make art that's more meaningful using the tool. I got into this by trying to make acceptable covers for my books using machine compositions; over time, I've learned more and more, and begun to take up the practice as an art-form myself. That's how I found Krita AI.
There's more to discover yet - the real innovators in this field are designing custom LoRAs based on their own hand-drawn art, making mathematical breakthroughs like those that made Control Nets possible, and developing creative commons datasets so that the whole thing doesn't have a faulty foundation.
Do you use any other artists' work in yours?
I don't input any data into the machine that isn't something I created myself. As for the background radiation of the process itself, eh. I don't like the scraping methods that the creators of these machines use, but as someone whose art is being trained on in a way that might eventually replace my literary works, I think it behooves us to take the tools that might one day destroy us and make them serve us instead.
And besides, I'm sure no real artist could possibly be threatened by the kind of mass-produced slop I'm creating, right? ;)
I want to commission you -
Let me stop you right there, assuming you're genuine. I don't take commissions and I don't accept money for AI-generated work. This may change eventually, but only once I have a reliable, locally stored, free-to-use model that I'm confident has an actually ethical dataset behind it. I estimate that'll take a few more years yet, but I've been wrong before.
I have another question.
Go ahead and ask in my ask box. Jerks and dunks will simply be ignored. Thoughtful ethics questions or process asks I'll be happy to answer.
