Yep, the software has changed some different images are being generated for the same input seed and parameters, something I wasn't aware of which kind of ticks me off, it was so difficult to get these images generated in the first place.
You raise some very good comparisons and I agree with them. I am unfortunately a lousy musician and conductor, seeing as it took me longer to generate the content than coding it
Well, you are a coder. Toolmaking is a thing that makes us human. In game makeing this is apparant with procedural level design. Why design a level, when you can design rules for levels and let the computer do what compuers do: compute the level.
Once those AI thingies understand procedurally the inner workings of humans and other stuff depicted in images, you can train them differently and not just imitate other art's "hash values". They heavily play the simulacrum game of splottering random stuff that will look, like it is the thing related to the prompt. Boosted to 11 with mechanically tenacity.
It might be possible to do this without "training". Just look at character creation of modern AAA games. One can press the random button there. And with rigging you can do random poses of those models. That character you have there could have been rendered by one of those engines. They are also good with lighting effects. The accessories are another matter.
But if we only train them with pictures of hands, they will see many pictures of hands that are interlocked with other hands and "learn" this, literally not realizing, nor learning, how a 2d projection of a 3d human is supposed to look. What baffles me, is that the negative prompt of disfigured and such work. If it "recognizes" the disfigurement, why did it put it in, in the first place.
Actual artists work with those small wood puppets you can bend to a pose to have reference material, to put it into context.
Anyways, it is a gloryfied random number generator (with imprecise numbers). You can command that one AI to show you the steps by putting out a picture with each iteration of the same starting seed. And it is not some constant kind of evolution towards a fine tuning in direction of the prompt. It takes leaps and bounds all around what it considers fitting to the prompt. You can actually make the image "worse" by having too many iterations. It was fun to play around with the thing, but its limitations became apparant quickly. Or maybe the end of my skillz became apparant. Creating the same character twice is such a limitation. Unless you prompt for some celebrity. Maybe if you have art of your own and teach it, this is MC, make that image look like it is MC weilding an oversized sword in a lightning storm.
