A cute cat boy maid, waiting in bed and excited to see you.

This was an experiment in generating a base image with a computer, and then painting over flaws or weird aesthetic choices by hand. In the end I think I ended up redrawing a significant portion of the image, probably over 1/3.

I think this was still much faster than doing the whole thing by hand, and I suspect most people wouldn't be able to tell the difference between this and a fully hand-drawn image.

There's definitely an aspect of the creative process that is less satisfying this way. I feel like the computer came up with (or depending on your perspective, stole, primarily from seth-iova who the LoRA is trained on) the character design, the colors, the outfit, the pose, the composition... In the end this doesn't feel like something I made. I feel like I got someone else's idea, and then I did a bunch of mindless polish work to bring it up to par.

As egotistical as it sounds, I normally love looking at the art I make. It makes me feel accomplished and proud. But the sense of "I made that" pride is not there when I look at this piece.

I also definitely would not be comfortable doing this for commission work. I really do feel like I'm basically stealing and frankenstein editing somebody else's work, especially since it's so close to one specific artist's style. I think I even recognize specific seth-iova pieces that were blended into portions of this image.

Even if I didn't have any ethical quandaries, I also don't think you could use this process for a client with a very specific idea. I found the computer was very poor at following any sort of detailed instructions, you basically have to feed it a set of tags and accept whatever comes out of it. You cannot tell the computer, for example, "Keep this pose, but move the camera to a side view". You must instead change the "three-quarter view" tag to a "side view" tag and pray everything else stays the same. For anything without a tag, or with insufficient numbers of images that have been tagged ("Make his skirt outline his butt", "keep his tail under the skirt instead of coming out of a tailhole in the skirt"), you are at the mercy of random chance.

I had people suggest using inpainting and img2img workflows to apply these sorts of details, but I did not have much success. The computer would too often randomly ignore my changes, or decide it wanted to change other portions of the piece. It was much less frustrating, and took less time, to give up on the computer and just do it myself.

Because of these issues, I suspect trying to use it to generate a base image for a very detailed brief would be more frustrating and take longer than just drawing it yourself.

Overall, I don't think I will use this process going forward. To me, this is unacceptable for commission work, and defeats the purpose of personal work.

I am curious to see if anyone can tell what parts of the image are hand-drawn, and what parts are generated. If you feel like speculating on that, feel free to leave a comment!