# How to Write AI Video Prompts for Veo, Sora and Kling

> The five-part structure for AI video prompts: subject, action, camera, light and sound. With copy-paste prompts for Veo 3.1, Sora 2 and Kling and fixes for messy clips.

Canonical: https://promptitin.com/blog/veo-sora-video-prompts
Last updated: 2026-11-18

A video prompt is a shot description, not a story. Veo, Sora and Kling make one short clip at a time, and they follow a prompt best when it reads like a single line from a shot list.

## The five parts

| Part | Example |
| --- | --- |
| Subject | A lone white lighthouse on a black rock |
| Action | Huge waves crash and spray into the air |
| Camera | Slow dolly push-in over 8 seconds |
| Light | Cold blue moonlight, flashes of lightning |
| Sound | Roaring waves, heavy rain, distant thunder |

Give the camera move a speed or duration. "Slow push-in over 8 seconds" works; "cinematic camera" doesn't tell the model anything.

**Weak prompt**

```
A cool cinematic video of a lighthouse in a storm with lots of camera movement.
```

**Better prompt**

```
Wide shot of a lone white lighthouse on a black rock at night in a violent storm. Waves crash and spray into the air, the beam sweeps across the frame. Camera: slow dolly push-in over 8 seconds. Cold blue moonlight with flashes of lightning. Sound: roaring waves, heavy rain, distant thunder.
```

## Prompts to copy

### Product reveal (Veo 3.1)

```
A glass perfume bottle with pale pink liquid on a round mirror plinth against a soft beige backdrop. Camera: slow 360-degree orbit around the bottle over 8 seconds. A soft light moves so highlights glide over the glass. Tiny dust sparkles in the air. Sound: soft ambient pad. Vertical 9:16.
```

### Talking to camera (Veo 3.1 or Sora 2)

```
A friendly man in his late 20s with a short beard sits at a café table and talks to the camera with natural hand gestures. He says: "Here are three AI tools I use every single day." Café blurred behind, soft window light. Camera: static selfie-style framing at eye level. Sound: his voice clear and close, soft café ambience. Vertical 9:16.
```

Put spoken lines in quotes after "says:". Veo 3.1 and Sora 2 lip-sync them.

> Browse [cinematic](https://promptitin.com/video-prompts/cinematic), [product](https://promptitin.com/video-prompts/product-video), [Reels and TikTok](https://promptitin.com/video-prompts/social-video) and [nature](https://promptitin.com/video-prompts/nature) video prompts, or describe your own clip in the [Video Prompt Generator](https://promptitin.com/video-prompt-generator).

## Which model for what

- **Veo 3.1:** cinematic shots with sound and dialogue. See [Veo 3.1 prompts](https://promptitin.com/models/veo-3-1).
- **Sora 2:** realistic motion, POV and handheld social clips. See [Sora 2 prompts](https://promptitin.com/models/sora-2).
- **Kling:** people moving, dance, and turning a still image into a clip. See [Kling prompts](https://promptitin.com/models/kling).

## Fixing a messy clip

- Too much happening: cut to one subject and one action.
- Camera jumps around: name one move only.
- Wrong product or face: start from your own image as the first frame.
- Bad audio: add a "Sound:" line that lists sounds in the order they happen.

## FAQ

### How long should a video prompt be?

Around 50 to 120 words. Enough to cover subject, action, camera, light and sound, but only one shot.

### Can I make a longer video?

Make several short clips, one prompt each, and edit them together. Keep the subject description identical between prompts for consistency.

### Do these prompts work in Runway and Seedance?

Yes. The five-part structure works in any text-to-video model; only Veo and Sora generate the sound.

## Related

- [Nano Banana Photo Prompts: How to Edit Your Photos in Gemini and Keep Your Face](https://promptitin.com/blog/nano-banana-photo-prompts)
- [Best AI Tools for Content Creators in 2026](https://promptitin.com/blog/best-ai-tools-for-content-creators)
- [How to Write Better AI Image Prompts (Complete Guide for 2026)](https://promptitin.com/blog/how-to-write-ai-image-prompts)
