I was struggling to get good groups and tried to micro manage everything. But then I decided to simple let stable diffusion into the drivers seat. This means the results vary widely, but with batches big enough there is always something usable.
This is actually my favorite, but the fingers are a little too messed up and I didn’t redo them.
Prompt: full body shots of group of different young women of multiple races, (naked), topless, nude, smiling. in a forest clearing, bright sunlight. (photorealistic), (raw, 8k:1.2), (textured skin:1.2), (imperfect skin:1.1)
See exif data for everything.
how do you avoid bad anatomy? just generating enough iterations until you find one that looks normal enough?
Dunno, I used a 2.1 model and the images just came out like this. I threw away something like 80% though, for various reasons (bad anatomy, bad eyes and missing variety).
After playing more with the prompt, it definitely helps that they are standing next to each other and that there’s little overlap, that’s when bad anatomy happens a lot.