When an AI video comes out flat, the usual reason isn’t the generator. It’s that the prompt described a topic instead of a shot. “A girl in a garden” is a topic. The generator has to guess her age, what she’s doing, where the camera sits, what time of day it is and whether it should look like a phone video or a film. Each guess is a chance to land somewhere you didn’t want.
The fix is to answer those questions yourself. I use six parts, always in the same order, and it turns a vague idea into a shot list the generator can follow.
The six parts
1. Subject
Who or what is the clip about? Be specific enough to picture: “a woman in her sixties with silver hair in a bun” beats “a woman.” For objects, give material and size: “a small brass bell,” not “a bell.”
2. Action
One clear action. Clips are 3 or 5 seconds long, which is enough for one thing to happen well. “Turns and smiles” works. “Turns, smiles, picks up her bag and walks to the door” will look rushed or broken.
3. Setting
Where is this? Name the place and one or two details that make it real: “a narrow Jaipur lane with blue doors,” “a quiet library with green lamps.”
4. Camera
How is it filmed? Shot size (close-up, medium, wide) and movement (static, slow push-in, pan left, drone flyover, handheld). This one part does more for the “filmed, not generated” feel than anything else.
5. Light
Golden hour, overcast daylight, neon at night, soft window light, candlelight. Light sets mood instantly and gives the generator a strong visual anchor.
6. Look
The overall finish: “35mm film,” “clean commercial look,” “vintage home video,” “soft pastel animation.” Keep it to a few words.
Weak to strong: three rewrites
Here’s the formula applied to prompts people commonly type.
Example 1: a portrait
Weak: “A man drinking tea.”
Strong: “A bearded man in a grey shawl sips chai from a clay cup and looks out the window, small hill-station café, medium close-up, static camera, soft morning window light, warm film look.”
Now the generator knows his age range, clothing, the cup, where he is, how close we are and what the light is doing.
Example 2: a product
Weak: “Perfume bottle.”
Strong: “A square glass perfume bottle with a gold cap rotates slowly on wet black stone, water droplets on the glass, macro shot, slow orbit, single hard spotlight from above, luxury commercial look.”
Example 3: a landscape
Weak: “Beautiful mountains.”
Strong: “Snow peaks above a valley of prayer flags fluttering in wind, high Himalayan pass, wide drone shot drifting forward, clear blue sky and late-afternoon sun, crisp documentary look.”
Notice that none of these are long. Each is one sentence with commas. Length isn’t the goal; coverage is.
Cheat sheet
| Part | Question it answers | Useful words |
|---|---|---|
| Subject | Who or what? | age, clothing, hair, material, colour, size |
| Action | What happens? | smiles, turns, waves, pours, rotates, drifts, blinks |
| Setting | Where? | rooftop, kitchen, market lane, beach, studio backdrop |
| Camera | How is it filmed? | close-up, wide, push-in, pull-back, pan, orbit, drone, handheld |
| Light | What does it feel like? | golden hour, overcast, neon, window light, candlelight, backlit |
| Look | What finish? | 35mm film, commercial, vintage home video, anime, claymation |
Habits that improve results
- Change one part per retry. If the clip is close but the motion is wrong, only rewrite the camera. Changing everything at once means you learn nothing.
- Put the important thing first. Generators pay more attention to the start of a prompt. Lead with the subject.
- Skip the filler. “Ultra realistic, 8K, masterpiece, best quality” adds little. Describe the light and camera instead.
- Avoid text in the scene. Signs and labels usually come out garbled. Add words later as captions in the Reels Maker.
- Match the frame to the platform. Choose vertical for Reels and Shorts before you generate, not after.
Using the formula with a photo
When you start from an image in image to video, the photo already answers subject, setting, light and look. Your prompt only needs the action and the camera: “she turns her head and smiles, slow push-in.” Repeating what’s already in the picture can confuse things, so keep it short.
Building your own library
Every time a clip turns out well, save the prompt somewhere: a notes app is fine. After a few weeks you’ll have a personal set of camera moves and lighting phrases that suit your style. Mix and match them with new subjects and you’ll get good results far more often than by starting from scratch.
If you’d like ready-made starting points, the prompt library has dozens arranged by theme. Pick one, swap the subject for yours, and run it through text to video. Then try the same prompt with a different camera line and compare. That side-by-side is the fastest way to learn what each part really does.
