# Kling 3.0 prompt guide: camera moves, multi-shot scenes, dialogue and consistent characters

By SeedRouter · Published 2026-10-06 · Updated 2026-10-06

**A good Kling 3.0 prompt names five things in plain language: the subject, what it visibly does, where it is, how the camera frames and moves, and the light and mood.** Kling's own prompt guide puts it the same way: strong prompts "begin with clear human writing: subject, action, setting, camera language, lighting, and mood".<sup><a href="#ref-2">\[2]</a></sup> Everything else in this guide, multi-shot scenes, dialogue and consistent characters, builds on that structure.

## What should a Kling 3.0 prompt contain?

Kling's prompt guide lists the parts of a clear prompt and what each one does:<sup><a href="#ref-2">\[2]</a></sup>

| Part           | What it sets                         | Example                                                           |
| -------------- | ------------------------------------ | ----------------------------------------------------------------- |
| Subject        | The main character, product or place | A woman in a striped shirt; a perfume bottle on a velvet pedestal |
| Action         | Visible movement                     | Walks toward the camera, swirls juice in a glass                  |
| Scene          | Location and surrounding detail      | Outdoor terrace, an old street in Madrid                          |
| Camera         | Framing, angle and movement          | Close-up, low angle, slow push-in, tracking shot                  |
| Light and mood | The atmosphere                       | Golden hour, cold blue night, soft haze                           |

Write the subject concretely. The same guide suggests replacing a vague word such as "magic" with "swirling blue energy particles with an ethereal glow", so the model has something to draw.<sup><a href="#ref-2">\[2]</a></sup> Describe motion the viewer can see, such as smoke drifting upward or a runner leaning forward, rather than feelings or intentions.

A prompt from our own tests, with all five parts in one sentence:

> A street violinist plays under a stone arch at night while light rain falls, slow push-in from across the square, warm lamplight on wet cobblestones, the violin melody echoing with soft rain. No text, no logos.

<video preload="none" poster="https://static.seedrouter.ai/media/landing/models/kling-3-0/v1/sample-3.webp" src="https://static.seedrouter.ai/media/landing/models/kling-3-0/v1/sample-3.mp4" width="1920" height="1080" style="{ width: '100%', height: 'auto' }" />

*Kling 3.0, 1080p, 5 seconds, with sound.*

## Which camera directions does Kling 3.0 follow?

Name the camera the way a director would. Kling's guide recommends plain framing and movement terms and gives these uses:<sup><a href="#ref-2">\[2]</a></sup>

| Direction     | Write it as                                              | Good for                          |
| ------------- | -------------------------------------------------------- | --------------------------------- |
| Close-up      | Close-up on the speaker's face                           | Dialogue, emotion, product detail |
| Wide shot     | Wide shot showing the terrace and both characters        | Setting and scale                 |
| Push-in       | The camera slowly pushes toward the subject              | Emphasis, tension, a reveal       |
| Pan or tilt   | The camera pans across the room, or tilts up to the sign | Revealing space or height         |
| Tracking shot | The camera follows beside the runner                     | Action and continuity             |

Add pace when it matters: "slow" for calm or suspense, "steady" for readable action, "quick" for energy.<sup><a href="#ref-2">\[2]</a></sup> Framing terms such as extreme close-up, medium close-up, full body and establishing wide shot tell the model how far the viewer is from the subject.<sup><a href="#ref-2">\[2]</a></sup>

## How do you write a multi-shot scene?

Kling 3.0 can cut one clip into several shots. Kling describes two ways: let the model plan the cuts from a single prompt, or define each shot and its length yourself, which Kling calls Custom Multi-Shot and says the model will "strictly follow".<sup><a href="#ref-1">\[1]</a></sup> Kling's own examples write each shot as one line of framing plus action: "Shot 1, low-angle rear wide shot, tracking behind the rider… Shot 2, low-angle side close-up, a detailed shot of the motorcycle wheel…"<sup><a href="#ref-1">\[1]</a></sup>

Through the API the custom version is `multi_shots: true` with a `multi_prompt` list, one entry per shot with its own prompt and duration. Our test asked for two 3-second shots:

```json
{
  "model": "kling-3-0",
  "mode": "pro",
  "sound": true,
  "multi_shots": true,
  "multi_prompt": [
    {"prompt": "Wide shot of a small open kitchen, a chef tosses vegetables in a wok, flames rising, warm light. No text, no logos.", "duration": 3},
    {"prompt": "Close-up of the wok, vegetables flipping through the flames, oil sizzling, steam drifting. No text, no logos.", "duration": 3}
  ]
}
```

<video preload="none" poster="https://static.seedrouter.ai/media/landing/models/kling-3-0/v1/sample-4.webp" src="https://static.seedrouter.ai/media/landing/models/kling-3-0/v1/sample-4.mp4" width="1920" height="1080" style="{ width: '100%', height: 'auto' }" />

*Kling 3.0, 1080p, two shots of 3 seconds, with sound.*

Give every shot its own framing, and keep the subject described the same way in each so it reads as the same person or object. The shot lengths must add up to 3–15 seconds.

## How do you prompt dialogue and sound?

Write the line in the prompt and attach it to the character who says it. Kling's guide shows the pattern "Character (tone): line", for example "Mom (softly, in a surprised tone): Wow, I didn't expect this plot at all", and says Kling 3.0 matches each line to the right speaker even with three or more characters in the frame.<sup><a href="#ref-1">\[1]</a></sup>

* **Languages.** Dialogue works in Chinese, English, Japanese, Korean and Spanish, and characters can switch languages within one clip. Kling says a line in any other language is translated into English.<sup><a href="#ref-1">\[1]</a></sup>
* **Accents and dialects.** Tag them next to the line, for example "speaks in English with an Indian accent"; Kling lists American, British and Indian English accents and several Chinese dialects, including Cantonese and Sichuanese.<sup><a href="#ref-1">\[1]</a></sup>
* **Ambience.** Name the background sound too: Kling's example opens with "a faint hum of the living room air conditioner in the background".<sup><a href="#ref-1">\[1]</a></sup>

Sound is generated only when you ask for it: set `sound` to `true` in the request.

## How do you keep a character or product consistent?

Use an element. In Kling 3.0 an element is a subject built from 2–4 reference images; once bound, Kling says the subject stays "clear and stable without shifting or disappearing" through zooms, pans and tilts.<sup><a href="#ref-1">\[1]</a></sup> In the API it goes in `kling_elements` with a `name`, a short `description` and the image URLs, and the prompt mentions it by name:

```json
{
  "prompt": "@hero walks out of the elevator, takes off her sunglasses and nods to a colleague, steady tracking shot",
  "kling_elements": [
    {
      "name": "hero",
      "description": "a professional woman in a camel coat with a black commuter bag",
      "element_input_urls": ["https://example.com/hero-front.png", "https://example.com/hero-side.png"]
    }
  ]
}
```

Use a front view as the first image and add side or three-quarter views; the description should name what must not change, such as clothing or a logo on a product.

## How do you control timing in a 15-second clip?

Mark the beats with times. Kling's own 15-second example moves its action on the clock: "At the 4th second, the camera accelerates forward with her… At the 8th second, the camera gradually zooms in to a medium shot… At the 12th second, the music and movement reach a climax… In the final 3 seconds…"<sup><a href="#ref-1">\[1]</a></sup> For a single continuous shot, say so explicitly, as Kling's other example does: "a single unbroken shot with no edited transitions".<sup><a href="#ref-1">\[1]</a></sup> Clip length is any whole number from 3 to 15 seconds, so match the duration to how many beats you wrote.

## What should you leave out?

* **Vague quality words on their own.** "Cinematic" or "masterpiece" without a camera move or light gives the model little to act on.<sup><a href="#ref-2">\[2]</a></sup>
* **Unwanted text and marks.** Kling 3.0 renders lettering well, including signs and logos it is asked for.<sup><a href="#ref-1">\[1]</a></sup> When you do not want any, end the prompt with "no text, no logos", as all our test prompts do.
* **Several cuts in one sentence.** If a scene needs specific cuts, use multi-shot with one entry per shot rather than chaining "then the camera cuts to…" in a single prompt.

## Kling 3.0 prompt questions

### How long can a Kling 3.0 prompt be?

Up to 2,500 characters on SeedRouter, and up to 500 characters per shot in a multi-shot request. Most strong prompts are one to four sentences.

### Does Kling 3.0 support negative prompts?

SeedRouter's Kling 3.0 request has no separate negative prompt field. Write what to avoid into the prompt itself, for example "no text, no logos". Kling's current API also folds this into the prompt: its documentation says "the prompt can include positive and negative descriptions".<sup><a href="#ref-3">\[3]</a></sup>

### What are good camera movement prompts for Kling 3.0?

Slow push-in, tracking shot, pan, tilt and low-angle reveal are the moves Kling's own guide uses, written as plain sentences such as "the camera slowly pushes toward the subject".<sup><a href="#ref-2">\[2]</a></sup>

### Where can I test a prompt?

The [Kling 3.0 playground](https://seedrouter.ai/models/kling-3-0) runs the same request the API takes, and the [Kling 3.0 API guide](https://seedrouter.ai/blog/kling-3-0-api) shows how to send it from code.

## References

1. <span id="ref-1" />Kling AI. *Kling VIDEO 3.0 Model User Guide*. February 6, 2026. Retrieved October 6, 2026 from [kling.ai](https://kling.ai/quickstart/klingai-video-3-model-user-guide).
2. <span id="ref-2" />Kling AI. *Kling AI Prompt Guide: The Secret to Cinematic Video Prompts*. August 7, 2026. Retrieved October 6, 2026 from [kling.ai](https://kling.ai/blog/kling-ai-prompt-guide).
3. <span id="ref-3" />Kling AI. *Kling 3.0: Text to Video* (API reference). Retrieved October 6, 2026 from [kling.ai](https://kling.ai/document-api/api/video/3-0-omni/text-to-video).
