There's limitations to what can be achieved with Stable Diffusion. Randomness is unavoidable due to the nature of the process by which the image is formed which is difficult to explain. The computer is looking at random noise until it finds what you tell it to find.

Minor features are also often missed, many times there is no bellybutton, no skin texture, hardly ever is there linea nigra. A lot of times the "bellybutton" is a finger. Many partial hands appear on the bellies. 

Negative Prompts are very useful for lessening the creation of deformed freaks, currently to use them requires running your own implementation that supports the feature. Either on your own hardware or a cloud computer. But hardware is expensive and cloud computer is not private. 

Settings also matter. CFG Scale causes problems if too low or too high. Scheduler can make the difference also, some seem better for making photography, others are often better for creativity, for paintings and artwork. 

Faces can be repaired by feeding the ugly monster face back into the machine with beautyfying prompts, then pasted back into the image. Upscaling can also be performed in this manner or via separate AI processes. There are also specialized face repair AI and integrated upscaling in some implementations. Eventually there will be options to redraw the parts that is not as desired.

If you try to produce beautiful artwork rather than blatant pornography you will get better results. But it is all fun! In the hands of the skilled and patient artist it will be a magnificent tool. img2img and infill show great promise for morph work. And eventually someone will train a model to specialize in creating pregnancy content. 

   
   
   
   
   
Imgur Link to Video of Permutations of the four girls

Festival Girls

                                                   

                           

Elements
               

there are implementations of stable diffusion that let you disable the nsfw filter.

the img2img feature of stable diffusion lets you use prompts to modify images and apply a mask to define which parts will and won't be altered.

   

   
   
   

Have a read through this guide, it's not specific to NMKD but a lot of it applies. There are other guides but practise will be your teacher. There are no magic settings that create perfection every time, it is always random.

                   

   
   
   
   
   
   
   

When I have my own setup I may share prompts and other parameters and methods used if people are interested. But for now I am using an online service and I am paranoid about accounts being identified