All-purpose AI video prompt with motion design principles

The #1 mistake in video prompting: describing a photograph instead of a MOTION SEQUENCE. Video generators need to know what CHANGES — in subject, camera, and light — across the duration.

'[CAMERA MOVE + SPEED] of [SUBJECT] [ONE CONTINUOUS ACTION WITH START AND END STATE], [SETTING], [TIME OF DAY], [LIGHTING + HOW IT CHANGES], [MOOD], [FILM LOOK], [DURATION]s'

Example: 'Slow dolly-in of a lone astronaut walking toward a habitat dome across red dunes, footprints forming behind them, a distant dust storm building on the horizon, golden hour transitioning to blue twilight, warm light cooling as shadows lengthen, solitary and awe-struck, anamorphic 35mm with subtle lens flare, 5s'

Notice the difference from a photo prompt: the dust storm is BUILDING, the light is TRANSITIONING, the astronaut is walking TOWARD something. Every element implies motion through time.

How different generators handle motion:
• Sora — handles complex multi-element motion well; you can be more ambitious with camera + subject + environment all moving
• Runway Gen-3 — best with a single clear camera move; lead with the camera direction
• Veo — strong with natural/physical motion and audio; describe sounds for richer output
• Seedance — rewards explicit 'smooth motion, consistent lighting' stability tags; keep physics grounded

Describing lighting for video (not photo):
Static: 'golden hour light' ← this is a photo prompt
Dynamic: 'golden hour light with long shadows sweeping left as the sun drops' ← this is a video prompt
Always ask: what is the light DOING during these 5 seconds? Flickering? Shifting color temperature? Clouds passing over?

Duration strategy:
• 3-4s — one clean action, one camera move. Highest quality.
• 5-6s — room for a slow build or reveal. Sweet spot for most generators.
• 8-10s — only if the action is simple and continuous. More time = more chance of artifacts.

Rule: ONE continuous action + ONE camera move per clip. If you need a cut, generate two clips. Describe every motion explicitly — 'the camera slowly rises' not 'aerial view.' Vague prompts get vague motion.