Resource dump: a structured prompt template that ports across image AND video models
Resource flair. After a lot of testing I converged on a slot-based template that works as a starting point almost everywhere. Fill the slots, delete empty ones, tune per model.
[SUBJECT] + [subject's key attribute] + [ACTION/motion] + [ENVIRONMENT] + [single LIGHT source] + [LENS/framing] + [MOOD]
The discipline isn't the template, it's the CONSTRAINTS it enforces: one subject, one action, one light, one mood. The slots stop you from stacking five lights and three moods, which is what actually kills most prompts.
It's deliberately boring. Boring and reliable beats clever and inconsistent. Riff on it, don't worship it.