I have as you can see from my avatar, and what I mean is that for example in ComfyUI when I prompt "Young gen-Z man in a military general uniform, ambiguous race, with broccoli hair, photorealistic style" I get significantly different responses between Model A, Model B, and Model C. No matter how you cut it if
only model A were available to me my creative options have been limited. I would want the ability to load any model of my desire if I were an artist and tune literally any parameter, which is the point I was trying to make in my reply. I don't think it will be as customizable to artists as they would want, but I am happy to be proven wrong when this releases with an SDK.
randomized, procedurally generated NPCs
To be fair, while each individual output may not have been an artists vision, the range of outputs for each slider was implemented through a technical artist, which surely had a vision for what each slider would do, what sliders even exist, and so on. The choice for Bethesda to make any slider combination output a "potentially real" human versus the Dark Souls abominations is certainly a choice.
I don't think this is a point either for or against this new DLSS 5 tech, because with the DLSS 5 setting being optional
for now we will still see all random NPCs being created the same way with these character creators, its just DLSS 5 will get thrown on.
The days of artists manipulating assets at the pixel level are long behind us. Same with lighting engines, face scanners, or just about any asset creation tool.
I just don't think this is true for video games as a whole. There are certainly pushes to make it true in the AAA 1000+ person development teams at these bloated studios, but I simply don't see a world where this technology works well in every game, and in every context, with every art style. And my fear is that this technology will be pushed so heavily, that the art styles where it doesn't work become old relics that never see the light of day again.
An argument against that is "Well, the Indie devs will do it themselves if its that important!" But I've been in the game dev scene long enough to know that whenever an "easier" path opens, the technologies that even
enabled the previous ways to do things get neglected often to the point where they simply don't work anymore, or the barrier to entry becomes so high that the game isn't created in the first place.
Hell, as an example just last night I needed a skybox for my game. It used to be that I could just go find a CC-0 skybox image and throw it in. But now the modern engines want an HDR skybox, so I go to find an HDR skybox instead, and there are around 1/3 the options. I then find out that "skyboxes are so yesterday" and now I need something else called a cube map because using a simple tiling texture as your sky is the "old, not as good" way to do it, and so its not supported anymore. Even better, these new methods of doing a skybox effectively pick where the sun is, what color the sun light is, and so on.
