PINNED: The prompt-engineering starter guide — read this before asking 'why doesn't my prompt work'
Welcome. This sub is about the CRAFT, not any single model. Before posting, internalize these:
Models parse differently. A weighting syntax that works in one image model is literally plain text in another. There is no universal prompt language. Always say which model.
Natural-language models ≠ tag models. Newer models (most video models, modern image models) want sentences. Old-school tag-salad ('masterpiece, 8k, best quality, trending') is often noise or harmful. Know which era your model is from.
Negatives behave wildly differently per model. Some have a real negative field, some fake it, some ignore it. Don't assume.
Test one variable at a time. If you change five words and it improves, you learned nothing. Change one. This is a science sub, act like it.
Flairs explained in the comments. Now go bend some models.