How to write AI video prompts that work

Start with one clear shot

AI video models work best when a prompt describes one continuous shot. If you ask for three scenes in one sentence, the result usually blends them. Generate each shot separately, then put them together on the timeline.

Use a five-part structure

Subject: what is on screen. Action: what moves. Camera: how the view moves. Light: where the light comes from and its quality. Mood: the overall feeling, such as premium, playful or calm.

Example: "A glass perfume bottle on wet black stone. Water droplets slide down the glass. Slow orbit around the bottle. Dramatic rim light. Premium commercial."

Name the camera movement

Words like push-in, orbit, pan, tilt and handheld give clearer results than "cinematic" alone. Pick one movement. Combining several often produces unstable motion.

Draft cheaply, finish well

Test your idea on a fast model at a low resolution. When the framing and motion work, run the same prompt on a higher-quality model. In Fantygen the credit cost is on the button, so you can compare before you spend.

Use a first frame for control

When the exact product or composition matters, generate or upload an image first, then use it as the first frame. The model then animates your image instead of inventing a new one.