5 prompt-writing tips that make AI video generation actually work

1. Describe the shot, not just the subject. "A woman in a park" is vague. "Close-up tracking shot of a woman walking through a sunlit park, shallow depth of field" gives the model something to actually compose around.

2. Name the camera movement if you want one. Static shot, slow pan, handheld, drone pull-back — models respond well to explicit camera language, and it's the single biggest lever for making a clip feel cinematic instead of flat.

3. Keep one clear action per generation. Prompts that try to pack in three separate events tend to produce muddled results. Split multi-beat ideas into separate generations instead.

4. Set the lighting and mood explicitly. "Soft natural light", "neon-lit night scene", "golden hour" — these two or three words do more for the final look than almost anything else in the prompt.

5. If you're using the avatar mode, keep the reference photo simple: one person, face clearly visible, decent lighting, no sunglasses. The cleaner the input photo, the more natural the lip-sync and motion.