Most AI video prompts fail for the same reason: they describe a mood, not a shot. When I write prompts for short product ads, I break every clip into five slots and fill them in order.
1. Subject and action
One subject, one verb. "A ceramic coffee mug rotates slowly on a wooden table" works. "A cozy morning vibe with coffee" does not, because the model has to guess what moves.
2. Camera
Name the move and the lens: slow dolly in, 35mm, eye level. If you skip this, most models default to a gentle push-in, which makes every clip look the same.
3. Light and time
"Soft window light from the left, early morning" gives consistent shadows across several generations, which matters when you cut clips together.
4. Reference image
For ads, a reference frame is worth more than ten adjectives. I generate or pick one still of the product, then use it as the first frame so colors and logo placement stay stable. Tools such as the AI video generator online on Kling 4.0 accept a reference image directly, which cuts down on re-rolls.
5. Duration and format
State the target: 5 seconds, 9:16, no text on screen. Adding captions later in an editor is more reliable than asking the model to render text.
Example
A ceramic coffee mug rotates slowly on a wooden table, slow dolly in, 35mm, eye level, soft window light from the left, early morning, 5 seconds, vertical 9:16, no on-screen text.
Checklist before you hit generate
- One subject, one action
- Camera move and lens named
- Light direction and time of day
- Reference image attached
- Length and aspect ratio stated
Keeping the same five slots for every clip in a campaign also makes it easy to change one variable at a time and see what actually improved the result.
Top comments (0)