Creating AI art from a bad drawing · 3:19am Apr 30th, 2023

I posted this to a group about ai, but thought I'd post it on my blog, too...

Just thought it'd be fun to create some ai art in stable diffusion by drawing badly, and document it.

So, step 1 was that I drew this bad 512x512 picture.

I put that into img2img using ponyv4 (purplesmart), with this prompt:
Prompt: A cute happy pony standing on a hill with one leg in the air, one ear down, line art
Negative prompt: EasyNegative boring_e621
I was going with denoising around 0.4.

(The negative prompt is two embeddings that are supposed to generate better pictures.)

After generating a few pictures, I went with this:

I then cleaned it up in an art program, turning it into this:

I took that to inpainting, same prompt, and inpainted just the areas I was iffy on, making this: (0.3 denoising?)

At this point, I went to text to image, and changed the prompt to this:
A cute happy pony with orange fur and blue eyes standing on a green grassy hill with one leg in the air, one ear down, sunset, intricate detail
EasyNegative boring_e621
In controlnet, I loaded the last picture, and set it to lineart anime, and told it that the prompt was more important then the picture. That got me this, after a few tries:

One ear had disappeared, but I decided I liked it. I took that into image to image, denoising 0.6. The prompt now became:
intricate detail, highres, A cute happy pony with orange fur and blue eyes standing on a green grassy hill with one leg in the air, one ear down, sunset
EasyNegative boring_e621

And I got this, which I decided was the final picture.