Here is my attempt at a workflow with DAZ and Stable Diffusion. 

I used several of my own blender renders to train an embedding. Then I wanted to combine the embedding and a Genesis 8 custom character that I created,who resembles the embedding, essentially using Genesis 8 as an avatar. That way I can pose her, alter the lighting, change her clothing and hair, render in iray, and then use imgtoimg to add the embeddeding. It could also mimic her facial expressions.

Before I've used iray renderings plus imgtoimg without any embedding involved and with a low denoising value, but it was hit-and-miss. Even though I was using the same seed and prompt, changing the lighting, the clothing, or the camera angle would cause imgtoimg to change the face of my character. Additionally, the aspect ratio itself changed the face too. 

Given that I'm using low denoising values and my G8 character already resembles the embedding (albeit not exactly, but real people look differently depending on the lighting), it ought to have worked.

If I had the same seed each time, it should to have producted the same face. Okay, no! SD doesn't know that this is the same person (or that that it is a person at all). When I tried to increase the denoising value, it initially merely combined the two faces, but it eneded up with different combination for each picture,but and it also made some caricatures, blurry images and nightmare fuel. Also, if I adjust the lighting or aspect ratio, I would get a different person. 

I made several attempts to make it work. It was really hit-and-miss. I was able to make it work in the end and keep the lighting, pose, clothing and hair of the G8 while also producing a person who is reasonably consistent and somewhat resembles my character. The process as a whole was interesting and somewhat irritating. It is still far from ideal, but I like promise of what this could do someday.

Even with lower denoising levels, SD managed to mess up the hands and other little things. 

The embedding is not very consistent in and of itself. The images I've shared are the ones that are more reliable. Even when used in SD with just a prompt it will change the contour of her face, the color of her eyes, and the shape of her eyebrows. Sometimes it would dramatically age her or turn her into a caricature.

I used to overlook all things in the beginning. Now looking back at it, I see plenty of nightmare fuel.