How to Write AI Video Prompts That Look Real (With Examples)
Last updated: October 20, 2026
A good AI video prompt describes one shot the way a director would: who or what is in frame, what they do, where it happens, how the camera moves, what the light looks like, and the overall style. Prompts that follow this order consistently produce more realistic video than long lists of adjectives.
Below is a formula you can reuse, the most common mistakes, and real examples you can copy.
The six-part prompt formula
Write your prompt in this order:
1. Subject — who or what the shot is about, with two or three concrete details.
2. Action — one clear thing that happens.
3. Setting — where and when.
4. Camera — shot size and movement.
5. Light — time of day, light source, mood.
6. Style — the look you want, such as documentary, commercial or film noir.
Example: "A young chef with rolled-up sleeves flips vegetables in a pan, flames rising, in a small busy restaurant kitchen at night. Medium shot, slow push-in. Warm practical lights and steam. Cinematic, shallow depth of field."
Describe one shot, not a whole story
The most common mistake is asking for too much: "A man wakes up, makes coffee, drives to work and gives a presentation." AI video models generate one continuous shot of a few seconds. If you need a story, write one prompt per shot and edit the clips together.
Use camera language
Models understand basic film terms. Useful ones:
- Shot size: extreme close-up, close-up, medium shot, wide shot, aerial shot.
- Movement: static, slow push-in, pull-back, pan left, tracking shot, handheld, orbit around the subject, drone glide.
- Lens feel: shallow depth of field, wide-angle, telephoto compression.
One camera move per shot works best. "Slow push-in" is far more reliable than "push in, then orbit, then crane up".
Light is what makes it look real
Realism comes mostly from lighting. Name the light source and the time of day: "golden hour sunlight from the left", "overcast soft daylight", "neon signs reflecting on wet pavement", "single window light". Vague words like "beautiful lighting" add very little.
Be specific about people
For realistic people, describe age range, clothing and one or two visible details instead of general beauty words. "A woman in her 30s in a grey wool coat, light freckles, hair tied back" beats "a stunning woman". Give the person one simple action with their hands visible only if needed — hands are still a weak point for many models.
Before and after examples
Weak: "A beautiful car driving, cinematic, 4K, ultra realistic."
Strong: "A dark green vintage sports car drives along a winding coastal road at sunset. Low tracking shot from the side, slow motion. Warm golden light, sea spray in the air. Commercial car-ad style."
Weak: "A cozy coffee video."
Strong: "Close-up of milk being poured into a cup of espresso, forming a heart in the foam, on a wooden café counter in the morning. Static camera, shallow depth of field. Soft window light. Calm, minimal style."
Weak: "Fantasy castle."
Strong: "Wide aerial shot of a stone castle on a cliff above the sea, waves crashing below, low clouds drifting past the towers. Slow drone glide toward the castle. Cold blue dawn light. Epic fantasy film style."
Image-to-video prompts are different
When you start from an image, the model already knows the subject and setting. Describe only what should move and how the camera behaves: "The woman turns her head toward the camera and smiles; slow push-in; hair moves gently in the wind." Repeating the whole image description can confuse the model.
Prompts for dialogue and sound
On models with native audio, put spoken lines in quotation marks and describe the sound: "She says, 'We open at nine tomorrow.' Quiet café ambience, cups clinking." Keep lines short — one or two sentences per clip.
Common mistakes to avoid
- Stacking quality words ("8K, masterpiece, ultra HD") — they rarely help.
- Contradictions, such as "night scene, bright sunlight".
- Several actions in one shot.
- Readable text, logos or signs — models still struggle to render text correctly.
- Real people's names — most platforms block them, and results are unreliable.
How long should an AI video prompt be?
Usually 40 to 80 words. Long enough to cover subject, action, setting, camera, light and style, short enough that every word matters.
Why does my AI video look fake?
The most common causes are vague lighting, too much happening in one shot, and over-the-top adjectives. Name a real light source, keep one action, and pick a grounded style such as "documentary" or "commercial".
Should I use the same prompt on different models?
Yes. The same prompt often looks quite different on different models. Testing one prompt on two models is the fastest way to find the best result. Verocine lets you run your prompt on several leading models from one studio and one credit balance.
Ready to try it?
Start creating