>>106502694

I prefer tags recognizing abstract concepts (happy, sad, confused) and object-position relationships (left hand grabbing X, right hand touching her cheek)

Prose prompting It's way too subjective.

People making and fine tuning these models.
For example: If you feed the AI a whole paragraph and it messes up, how is the creator supposed to know what part of the prompt caused the error? They can't "fix" the model for "wistful gazing." It's not a clear, linear problem they can solve. You're just giving them noise.


For example pic related, I have two Jalters, how can I modify my prompt to have one Jalter?