Disagree that curation and prompts adds artistry (dense intent reflected in the output) to AI generations.
"Curation" in AI can only surface the curator's local maxima among a tiny and arbitrary grab-bag of seed integers they checked among the space of 2^64 options; it's statistically skewed 99% towards the model's whims rather than anyone's unique intent or taste.
Prompt crafting is likewise terribly low fidelity since it's a constant battle with the model's idiosyncratic interpretation of the text, plus arbitrary perturbations that aren't actually correlated with the writer's supposed intent. And lord spare me the "high quality high resolution ultra detailed photorealistic trending on artstation" type prompts that amount to a zero-intent plea for "more gooder". And when pursuing artistry, using artist names / LORAs are a meta-abandonment of personal direction, abdicating artistic control and responsibility to a model's idea of another artist's idea of what should be done.
Fancier workflows generally only multiply this prompt-and-curate process across regions/iterations, so can't add much because they're multiplying a tiny fraction by a fixed factor.
I agree with you on the idea of prompts and seeds leaving much to be desired. So that's why I think more sophisticated steering is necessary.
The models' latent space is extremely powerful, but you get hamstrung into the text encoders whims when you do things through a prompt interface. In particular, you've hit exactly an issue I have with current LLMs in general in that they are locked into wors and concepts that others have defined (labelings of points in the latent space).
Wishy washy thinking: I'd be nice if there were some sort of Turing complete lambda calculus sort of way to prompt these models instead. Where you can define new terms, create expressions, and loops and recursion or something.
It would sort of be like how SVGs are "intent complete" and undeniably art, but instead of vector graphics, it is an SVG like model prompt.
"Curation" in AI can only surface the curator's local maxima among a tiny and arbitrary grab-bag of seed integers they checked among the space of 2^64 options; it's statistically skewed 99% towards the model's whims rather than anyone's unique intent or taste.
Prompt crafting is likewise terribly low fidelity since it's a constant battle with the model's idiosyncratic interpretation of the text, plus arbitrary perturbations that aren't actually correlated with the writer's supposed intent. And lord spare me the "high quality high resolution ultra detailed photorealistic trending on artstation" type prompts that amount to a zero-intent plea for "more gooder". And when pursuing artistry, using artist names / LORAs are a meta-abandonment of personal direction, abdicating artistic control and responsibility to a model's idea of another artist's idea of what should be done.
Fancier workflows generally only multiply this prompt-and-curate process across regions/iterations, so can't add much because they're multiplying a tiny fraction by a fixed factor.