suomynonA:
The biggest problem is that stuff like that is going to be out of the hands of the masses for quite some time until AI training gets a bit cheaper. StableDiffusion was the one to break the mold and make it available to the masses at a considerable cost to themselves from my understanding. Though some pretty amazing things can happen once the gestalt gets their hands on things and have the enthusiasm to back it as we've been seeing. Again, things are moving VERY fast in this sector, and it's a bit of a crap-shoot as to where things will land once the proverbial dust settles. Either way, we're in the middle of a pretty major development technology-wise.

But, pulling things back a bit, these are going to be raging discussions throughout the interwebs in public forums much larger than our little corner. I think it would behoove us to rather discuss our findings about what works and what doesn't work in regards to getting decent artwork of our particular persuasion whilst also sharing our cherry-picked generations to others.

Using Automatics1111's webui version of SD, you can use any of the pictures I've uploaded here to recreate them and see what settings and prompts I used by plopping them onto the PNG info tab. (Keeping in mind the model hashes, also the Feather-chans were img2img so no dice there, but you'll be able to still see the settings.)

Negative prompts make a huge difference and can usually get decent to good artwork out of even the simplest prompts. My most common one is as follows:

bad anatomy, extra legs, extra arms, poorly drawn hands, poorly drawn feet, disfigured, out of frame, tiling, bad art, deformed, mutated

A side effect is that it tends to hide hands and feet rather than just making them decent... Oh, adding 'breasts' to the negative will also often win the battle of the bulge, but you'll need to add 'nipples' to the regular prompt. Something I've forgotten to do in some of my examples...



Tried making a decent set of classic halloween monsters, but man for the life of me I can't get a mummy wrapped one that works.

Gushes:
What I'm using is based on StableFusion, and runs on a person's own equipment. Costs nothing to generate unlimited images. More and more of these will become available to the public, free. Including ones that generate posable 3D models, Movies, novellas, Short films, and eventually feature films and video games. All within the next 10 years.

suomynonA:
So documentation on Waifu 1.3 was released. It contains some very good information that should help with prompts: hxxps://gist.github.com/harubaru/f727cedacae336d1f7877c4bbe2196e1