woman detailed

There's a lot of talk, a lot of friction online currently about AI generated art. A lot of it comes from a profit motive perspective, but some is about the quality of art (setting aside its provenance).

Some people view the amount of effort involved in creating a piece of art as contributing to the worthiness of the art.

I don't think like that. I think the aesthetics of art should be evaluated independently of such things.

Part of the effort involved in creating a piece of art using Stable Diffusion is crafting a prompt.

I didn't do that with any of these images. I didn't even use a negative prompt. It's just "woman detailed", run through a few checkpoints.

These are the ones I liked. Honestly, most of them came out well.

For the ones I did like, I boosted the steps from 20 to 30, set hires fix to 2x using 4x-UltraSharp with .4 denoising, and let adetailer work on their face. I've got tiled diffusion and tiled vae running as well, though I don't think that should impact the final image overmuch.

With one of the images, when boosting the steps (and mind you, this is DPM++ 2M Karras) I got a completely different image. Instead of a woman standing, I got a portrait. After some XYZ, I found that the image wasn't fully baked until 35 steps, and there was a small range around 20 where it was a completely different image (not just the same basic thing with some of the details changed).