30 AI Video Prompt Templates: One Set Each for Camera, Lighting, Motion and Duration
A prompt only needs four things
A video prompt that actually works has a fixed structure:
`` subject + action + camera + lighting + aspect ratio and duration + negative ``
Write "who is doing what" first, then "how it is shot," and only then "what you do not want." Do not reverse that order — plenty of people open with "cinematic, 8K, cyberpunk," the model never gets subject information, and the output inevitably drifts.
The 30 templates below are split by module and can be combined across modules. For instance, "medium shot, locked-off camera" + "early-morning side backlight" + "slow turn" + "5 seconds, 16:9" is already a complete prompt.
Module one: camera (8 templates)
| # | Template fragment | Best for |
|---|---|---|
| 1 | Medium shot, locked-off camera, subject centered | Talking head, product demos, the most reliable opener |
| 2 | Close shot, slow push-in | Emphasizing an emotional shift |
| 3 | Extreme close-up, macro, shallow depth of field | Material, texture, hand detail |
| 4 | Low-angle shot looking up | Making a character feel powerful |
| 5 | High-angle wide shot, wide lens | Establishing environment and spatial relationships |
| 6 | Side tracking shot, subject moving at a steady pace | Walking, running, commuting shots |
| 7 | Orbiting camera, 90 degrees around the subject | Showing dimensionality |
| 8 | Slight handheld shake | Documentary feel, vlog texture |
Put only one camera move in a 5-second clip. Ask for a push-in and an orbit at once and the model averages the two moves, leaving neither one convincing.
Module two: lighting (6 templates)
| # | Template fragment | Best for |
|---|---|---|
| 9 | Early-morning side backlight, warm rim light | People, outdoors, openings |
| 10 | Harsh midday sun, short shadows | High contrast, documentary |
| 11 | Golden-hour warmth, orange-gold sky | Emotional closings, nostalgia |
| 12 | Overcast soft light, no hard shadows | Beauty, product detail |
| 13 | Neon night, teal-and-magenta contrast | City, cyberpunk, nightlife |
| 14 | Indoor window light, side light on half the face | Interviews, monologue |
Lighting is the easiest anchor for multi-shot consistency: keep the same lighting sentence across a set of shots and the grade will not jump.
Module three: motion (6 templates)
| # | Template fragment | Best for |
|---|---|---|
| 15 | Slow turn, gaze moving from over-the-shoulder to camera | Character entrances |
| 16 | Walking toward camera, natural gait | Street photography, commuting |
| 17 | Reaching for an object just outside frame | Product interaction, plot setup |
| 18 | Wind moving hair and the hem of a coat | Adding life to a frame |
| 19 | Turning to look back, expression shifting from calm to a smile | Emotional turns |
| 20 | Ripples spreading on water, subject still | Insert shots, transitions |
One action per 5 seconds. Ten seconds can hold two (action A then action B), and 15 seconds at most three beats. Ask for more and the model returns a blurred in-between state.
Module four: duration and aspect ratio (5 templates)
| # | Template fragment | Notes |
|---|---|---|
| 21 | 5 seconds, 16:9, single action | Cheapest on credits, good for style tests |
| 22 | 10 seconds, 16:9, action A to action B | Use when you need one turn |
| 23 | 15 seconds, 16:9, three beats: establish, develop, resolve | The smallest unit of complete narrative |
| 24 | 5 seconds, 9:16 vertical, subject centered slightly high | TikTok, Instagram, Facebook |
| 25 | 15 seconds, 9:16 vertical, person talking to camera | Digital humans, educational content |
Cost comparison (each generation deducts a fixed number of credits): 480P at 5 / 10 / 15 seconds is 5 / 10 / 14 credits; 720P is 10 / 19 / 28 credits. Test at 480P for 5 seconds, then switch to 720P for 15 seconds once locked — that is the most direct way to save money.
Module five: stability and negatives (5 templates)
| # | Template fragment | Notes |
|---|---|---|
| 26 | Subject keeps a stable share of the frame | Prevents characters from growing and shrinking |
| 27 | Lighting direction stays consistent throughout, no jumps | Prevents color temperature shifts between shots |
| 28 | Negative: no text, subtitles, watermarks, logos | Text is the first thing to dissolve into gibberish |
| 29 | Negative: no extra fingers, distorted limbs, two heads | Essential for any human subject |
| 30 | Reference image: keep facial features and clothing consistent | Essential for image-to-video |
One complete example
Combine 1, 9, 15 and 21:
`` A woman in a dark grey trench coat stands in the middle of an empty street, turning slowly, her gaze moving from over her shoulder to the camera. Medium shot, locked-off camera, subject centered. Early-morning side backlight, warm rim light. 5 seconds, 16:9. Negative: no text, no watermarks, no extra fingers. ``
How to iterate without burning credits
- Change one variable at a time. Adjust lighting this round and camera next round, otherwise you cannot localize what broke.
- Run 3 clips at 480P for 5 seconds first (15 credits), pick a direction, then go to 720P.
- Log your prompts. The exact sentence behind a finished clip is the starting point for the next one.
- Reuse the lighting sentence across a shot set so the grade does not jump when you assemble multiple shots.
The 40 credits you get at signup are enough for 8 test clips at 480P for 5 seconds (40 credits); if you want to lock a cut directly, it covers one 720P 15-second clip (28 credits) plus two 480P 5-second clips (10 credits). Testing first and locking second is cheaper than going hi-res in one shot.
FAQ
Q: Are prompts better in English or in another language? A: Prompting in the language you are most precise in works well for subject and action; technical camera and lighting terms (dolly in, rim light) are often sturdier in English. Mixing the two is common practice and there is no need to force uniformity.
Q: Why does adding "cinematic, 8K, hyperrealistic" make things worse? A: Those words add noise without adding information. Replace them with a concrete lighting direction, camera position and focal length and the result becomes far more controllable.
Q: Do image-to-video generations need a prompt? A: Yes, but the emphasis shifts from "what it looks like" to "how it moves." The reference image already fixes the look, so the prompt should describe motion, camera movement and lighting changes, plus template 30 to lock character consistency.
Q: How long can one prompt stay usable? A: Results shift when the model updates. Record "prompt + model version + date" for every finished clip and rerun one comparison clip after a version change.
Next steps
- See credit tiers for video: Pricing
- Work out costs clearly: How to calculate AI video cost
- Break down a storyboard: How to write an AI video storyboard
- Dimensions to compare when choosing tools: How to choose an AI video tool
Want to try it? Create a free account