AI Musings #1 (Patreon)

I was going to release Multiverse Elf today, but I got sidetracked with some experiments in AI, which mostly came about because it takes less concentration and I've had a killer headache all day despite many drugs.

I generated these images by first creating a quick character in Daz, rendering her out in 1:1 aspect ratios, and then using those images as the image prompt in NMKD Stable Diffusion. These are presented as my rendered images followed by the AI images generated using that render.

All of the AI images used the same text prompt of: huge muscular female bodybuilder flexing at the beach

I have found that, when using an image as part of the prompt, it's best to describe the starting image fairly well as a basis.

Obviously some of these are better than others, and some are a fairly radical departure from the starting images. This is due to altering the Init Image Strength (Influence).

For the numbers:

The Init Image Strength on the more outlandish ones is around 0.350, while most of the rest are at 0.700.

The Detail (Steps) is 500, the Creativeness (Guidance Scale) is 17.5.

The Sampler is k_heun.

The Resolution is 512x512 with 4x Upscaling and Face Restoration set to 1 in the Post-Processing Settings.