It takes a huge factor more to train the AI model than it does to run the AI model to make images from it. The amount of computing power to run the AI to generate images is on the desktop level. The amount of compute to train it kind of depends on how big the model is. But the ones which are working very well and produce the good results are huge and therefore only able to be trained on big clusters of servers like the sort that Amazon and Google have.

The 6 digit hands, text, and many of the other issues that the current AI models have will be fixed in no time. There will be other tells but most of the obvious ones will be gone by the next year or two.

The soul thing you see is because the AI is not intelligent and is reproducing the style of other peoples work. Though not exact copies, it is extracting the average essence of all other peoples work and producing its own from it. Its like generic pop music and other generic forms of art. It will have creativity but never have true unique exceptional brilliance. Its good for mundane generic art for which there is a whole lot needed in the world. But don't expect anything truly novel and ground breaking from it (in the art / pictures area) from AI.

The reason we have human brilliance is that each person is its own trained (A)I based on their unique experience in the world. When we have so many of these trained models that it makes up millions of them then we might expect some of them to be different, unique and have something surprising to add into the worlds creativity.

Computer AI models need to have feelings, experiences, taste, smell and all sorts of other non image input put into them before it can create new picture art from them. Right now all that happens is that millions of pictures are fed in. Like a baby growing up inside a white featureless box with no other experiences except a continuous stream of images flashed before it. Your not going to get any human wisdom or soul from it.