AI video prompts: 20 copy-ready examples and the script format behind them

A good AI video prompt reads like a line from a shot list: who or what is in frame, what happens, what the camera does, and what the light and sound are like. The 20 examples below are written that way and can be pasted as they are. The second half of the guide covers the format for longer videos, where a single line is not enough: timed beats, a script the engine reads one moment at a time. If you want the reasoning behind prompt structure first, start with how to make an AI video from text; for animating a still, see image to video prompts.

What makes a good AI video prompt?

A good AI video prompt names five things in one or two sentences: the subject, the action, the camera, the setting and light, and the sound. It describes one moment clearly, not a story, and it describes what should be seen and heard rather than how it should make you feel.

The shape that works on most engines:

[Shot size and camera move], [subject] [action], [setting], [light and time of day], [style or lens], [sound].

"Beautiful, cinematic, epic" adds nothing the engine can act on. "Low angle, slow push in, golden hour backlight, shallow depth of field" does.

AI video prompts for ads and product shots

1. Macro close-up, slow orbit around a matte black wristwatch on a marble slab, soft window light, water droplets on the glass, quiet ticking.
2. Top-down shot, hands unbox a pastel sneaker from tissue paper on a wooden table, morning light, paper rustle.
3. Slow dolly past a row of skincare bottles on a bathroom shelf, steam in the air, warm practical light, soft drips.
4. A cold soda can slides into frame on a wet bar counter, condensation, neon reflections, a can hiss and fizz.

AI video prompts for B-roll

1. Wide aerial of a coastal road at sunrise, a single car driving north, long shadows, wind and distant waves.
2. Handheld walk through a busy night market, lanterns overhead, steam from food stalls, crowd murmur.
3. Static wide, fog rolling over a pine forest at dawn, birdsong.
4. Close-up, coffee poured into a white cup in slow motion, dark background, the pour audible.

See the AI B-roll generator guide for building a library of shots like these.

AI video prompts for music videos

1. Low angle, a singer in a red jacket on a rooftop at dusk, wind in her hair, the city lit behind her, camera slowly rising.
2. Dolly along a tiled corridor lit by flickering fluorescent tubes, a dancer moving toward camera, 1990s film grain.
3. Close-up, fingers on the strings of an acoustic guitar, warm lamp light, shallow focus.
4. Slow-motion confetti falling over a crowd with raised hands, stage lights sweeping, crowd cheering.

AI video prompts for openers and intros

1. Push in on a glowing neon sign that reads OPEN in a rainy alley, reflections on wet asphalt, rain on metal.
2. Title card style: ink spreading in water in slow motion, deep blue on white, a low hum.
3. The camera flies through a doorway from a dark hallway into a sunlit room full of plants.
4. Static shot, a vintage radio on a windowsill, dust in a beam of light, the radio crackling to life.

AI video prompts for short films

1. Over-the-shoulder, a girl watches a train leave the platform, snow falling, cold blue light, a distant whistle.
2. Wide, a lone astronaut walks across red dunes toward a small rover, 2.39:1 framing, wind.
3. Close-up, an old man's hands wind a pocket watch at a kitchen table, candlelight, the clock ticking.
4. Tracking shot beside a cyclist on a country road at golden hour, wheat fields, gravel under tyres.

The prompt enhancer turns a rough idea into this format if you would rather start from one line.

How do you prompt a longer AI video from a script?

A single sentence describes a moment. A 20- or 30-second video needs a sequence of moments, and a paragraph of prose leaves the engine to guess the order and timing. The fix is timed beats: one line per beat, each with a time range.

0-5s: wide, a lighthouse on a cliff at dusk, waves below, gulls
5-12s: the camera climbs the lighthouse wall toward the lamp room
12-20s: inside, the keeper lights the lamp; warm light floods the room
20-26s: the beam swings out over the dark sea
26-30s: pull back to the wide, the beam turning, the first stars out

PrismPoster's Premium engine reads beats like these as directions inside one continuous take of up to 30 seconds, so the camera moves from beat to beat without a cut. On the Standard engine, which renders 4, 6, 8, 10-second shots, turn each beat into its own shot and assemble them on the timeline. The shot script builder on the AI video generator page writes this format for you, checks the total against the take limit, and shows the price at 480p, 720p and 1080p.

Rules for beat scripts:

• One action per beat. Split "she lights the lamp and looks out and smiles" into beats.
• Give beats room. A camera move needs four seconds or more to read; two-second beats blur together.
• Describe transitions as movement. Inside one take, "cut to" means nothing. Write "the camera turns toward" instead. Our guide to the oner explains why.
• Put sound in the beat where it happens. "The keeper lights the lamp; a match strikes" places the sound on the right beat.

How do you fix a prompt that is not working?

Change one thing at a time, render short and cheap, then scale up. A four-second draft at 480p costs 72 credits on Premium, so finding the right wording there before rendering 30 seconds at 1080p saves most of the budget. The most common failures and their fixes are in nine AI video prompt mistakes.

Frequently Asked Questions

What is a good prompt for AI video?

Name the shot size and camera move, the subject and action, the setting and light, the style, and the sound, in one or two sentences describing a single moment.

How long should an AI video prompt be?

One or two sentences for a single shot. For a longer take, write one line per timed beat rather than one long paragraph.

Can I make an AI video from a script?

Yes. Write the script as timed beats, such as "0-5s: ..." and "5-12s: ...". On PrismPoster's Premium engine the beats render as one continuous take of up to 30 seconds; on Standard, each beat becomes a shot you assemble on the timeline.

Should I describe sound in a video prompt?

Yes, if the engine renders sound. Both PrismPoster engines render sound with the picture, and naming it ("rain on metal", "crowd cheering") steers it.

Do AI video prompts work the same on every generator?

The structure transfers well between engines; exact behaviour does not. Test a prompt with a short, low-resolution draft before a long render.