Just to be clear the GUI doesnt mean its weaker. You can find very specific models to use and get very well done renders. I have seen stable diffusion users draw stick figures and turn them into full fledged anime poses.
After a while getting good results starts needing inpainting.
automatic1111 webui is amazing, so many features and gadgets that its overwhelming at first, second, and third times. But its amazing, I use it to clean up images, enlarge them, or merge models together and see what the mix gets me. Im slowly learning all the bells and whistles, it will probably be my top pick by the end of the week.
I used SD at work for a gallery of directors of a foundation that's celebrating its centenary. The author is having a hard time finding pictures: sometimes she sends a scan of a 1930s photo of some guys visiting an exposition, taken from afar, saying "I think the president is the one in the third line with a bowler hat."
I forced faces out of newspaper prints 9 dots wide and 13 dots tall (OK, that one required a lot of work and editing in Photoshop).
Photo-retouching the face of an aged lady that's making a photo book for the press ("Well, this photo is good, but can't you remove some wrinkles?" yes, and that webbed throat, stooping checks and swallowing eyelids?) used to take hours. Now it's a 5 minutes work.
It's seriously scary at times.
I came to the forums looking to see if anyone was training a model on DAZ models. Havent tried training a model myself but I know a few people that might give it go if they can get a hold of enough raw images to process.
As far as I can tell, user trained models are usually built on a very limited number of images, as low as just a few dozens, and hyper-specialized.
They then get mixed in order to have some flexibility in the prompts.
I don't see the advantage in training specifically with Daz models, as style tags allow to turn the result into various kinds of 3D renders. Real porn seems a better choice to me.
