Learn the five-part formula
Every strong prompt carries the same five parts. Example: "A woman in a red coat, walking slowly toward camera, slow dolly-in, rain-lit city street at night, moody and cinematic."
Name the camera move explicitly
Pan, dolly-in, orbit, static, handheld drift. Vague prompts default to generic motion, and generic motion is what makes a clip look AI-made.
Separate subject from action
Describe what the subject looks like first, then what it is doing, so the model does not conflate the two into one muddled instruction.
Anchor lighting and time of day
"Golden hour", "overcast", "neon-lit night" — this moves the mood more than almost any other single word choice.
Callout
"Golden-hour backlight, 35mm, soft haze" reads as direction. "Nice lighting" reads as noise.
Keep one idea per generation
A prompt trying to do five things usually does all five poorly. Split a complex scene into separate shots and assemble them afterwards.
Iterate, don't restart
Adjust and regenerate rather than rewriting from zero, and reach for the Enhance control when a student is stuck.
Callout
Common mistake: writing prompts like search queries ("cool video of a car"). Reframe every prompt as a line from a shooting script.
Next up: Style control & export
Continue