How to write prompts that produce usable video
Subject, action, setting, light, lens, motion. A six-part structure that consistently beats long descriptive paragraphs.
Most weak generations come from prompts that describe a feeling instead of a shot. The model does not know what 'inspiring' looks like; it does know what 'low-angle shot, golden hour backlight, slow push-in' looks like.
Use six slots in order: subject, action, setting, lighting, lens or framing, and camera motion. 'A young woman in a linen apron (subject) rolling out dough (action) in a sunlit kitchen (setting), warm morning light through a lattice window (lighting), 50mm close-up (lens), slow dolly right (motion).'
Keep it under about 60 words. Past that, models start dropping the earliest clauses, and you lose control of the subject to gain adjectives you did not need.
Generate at 480p while you iterate on wording, then re-run the winning prompt at 1080p. That is 5 credits per test instead of 20, so you can afford four attempts before you commit.