Doubt I’m the first to think of this but I discovered an interesting process to create artwork with ai that I though I’d jot down. The key idea is instead of simply trying to generate the image you want, you go though a process of generate -> edit -> generate -> edit. Lets follow along with how I came to the image above to get a better idea what I’m talking about.
The first thing I tried was generate what I wanted directly from a prompt.
1girl, solo, cute, fox ears, fox tail, red hair, long hair, blue bikini, beach, key animation, dynamic shot, soft lighting
The results are promising, but none of them are quite right. In theory I could edit one of them to fix it’s issues, but they are complicated enough that that would be very difficult for me to do. So lets step back, what if I created something I could edit instead?
1girl, solo, cute, fox ears, fox tail, red hair, long hair, blue bikini, beach, simple sketch, clean lines
Ok, still not quite right but the style is simple enough that I could work with it. The first thing I did was break up the foreground from the background.
Lets start with the background first, I removed the extra tails and filled in those ares with simple colors.
I then fed this back though to get something a bit nicer
beach, simple sketch, clean lines
Cleaned it up a bit
Ok, clouds would be nice, lets generate that
and paste them in
Ok now for the foreground. I did want some blue eyes so I generated those as well.
1girl, solo, cute, fox ears, fox tail, red hair, long hair, blue bikini, gorgeous blue eyes, beach, simple sketch, clean lines
As for the rest of the cleanup, it was quite a bit of work but not too bad. The hand was the the hardest. Neither me nor stable diffusion seem to be good at drawing hands 😂.
Finally I ran this though again to up the quality
1girl, solo, cute, fox ears, fox tail, red hair, long hair, blue bikini, gorgeous blue eyes, beach, high detail, soft light
Sigh, it messed up the hand again, time to fix.
And there we go! All and all I’m pretty impressed with this process. I was able to work around stable diffusion’s limitations and get exactly what I want with my limited art skills.