Kling 4.0 Prompt Guide: Shots, Camera, Sound and References

Kling 4.0 works best with plain, concrete sentences: what we see, how the camera moves and what we hear. Sound is always generated with the picture, so a good prompt describes the audio too. This guide follows the structure in Kuaishou's Kling 4.0 user guide and adds what is specific to this site.

Start with one sentence

Kling 4.0 expands short prompts on its own. "An old man ranting about AI" is a real example from Kling's guide, and it produces a full scene with staging and camera moves. Start there when you're exploring.

When you know what you want, add only the details that change the result: camera, timing, sound, and what must stay the same. A long list of adjectives doesn't help.

The structure for longer prompts

Kling's guide suggests four blocks. Use the ones you need:

[References]   @Image1 is ___; keep its ___. Follow the ___ of @Video1.
[Overall]      Setting and texture ___; visual style ___; camera feel ___.
[Sequence]     0–X s: ___ (picture + action + emotion)
               X–Y s: ___ (camera move + extra detail)
               Y s–end: ___ (closing shot)
[Audio]        Dialogue or voice-over + ambience + music + effects

Describe things, not qualities

Abstract words make the model guess. Replace them with what you would see:

Avoid negative instructions in general; they raise the chance of mistakes. Kling's guide names two that are worth writing, because the model may add them by default: no subtitles and no background music.

Camera moves that Kling 4.0 handles well

Name the move and what it follows:

• Push in / pull out — "slow push-in on her face as she reads the letter"
• Tracking shot — "the camera tracks beside the cyclist at wheel height"
• Orbit — "the camera orbits the dancer once, keeping her centred"
• Handheld — "handheld, slight natural shake, follows them through the crowd"
• One take — "one continuous shot, no cuts, no angle changes"

Kling 4.0 is multi-shot by default: it splits longer prompts into shots by itself. If you want a single take, say so, as in the last line above.

Sound and dialogue

Every Kling 4.0 clip has two-channel stereo sound. Write the audio as its own part of the prompt:

• Dialogue in quotes, in the language it should be spoken in. Kling 4.0 speaks Chinese, English, Japanese, Korean, Spanish, Portuguese, German, French, Hindi and more, with accents such as American, British or Indian English.
• Who speaks and how. Age, tone, speed and emotion: "a woman in her thirties, warm and unhurried, slightly amused".
• Ambience and effects. "Rain on the window, distant traffic, a kettle clicking off."
• Music, or none. "Soft piano under the dialogue", or "no background music".

Example:

A barista leans over the counter and smiles. She says in English, warm and
relaxed: "Your usual? It's on the house today." Handheld medium shot, morning
window light. Café ambience, cups clinking, no background music.

Referencing your files: @Image1, @Video1, @Audio1

In Reference mode, uploads are numbered in order: @Image1, @Image2, @Video1, @Audio1. Type @ in the prompt box to insert one. Then give each file a job; without one, the model infers what it's for and results get less predictable.

@Image1 defines the scene; keep its layout, furniture and lighting.
@Image2 is the main character; keep her face, hairstyle and clothes.
Follow the camera path and pacing of @Video1.
She walks to the window, stops and looks back. The voice in @Audio1 says:
"I thought you'd never come."

What works on this site:

• Images (up to 10): characters, products, scenes, style, storyboards.
• Video clips (up to 5, 3–30 s each, 30 s in total): motion, camera moves, pacing.
• Audio (up to 3 clips of 5–30 s): a speaker's voice. Audio needs at least one image or clip. Music and effects can't be uploaded; describe them instead.

When a reference should give only one thing, say what not to take: "use only the camera move from @Video1, not its colours or people."

First and last frame

In Image mode you upload the frames directly, so the prompt only has to say what happens. A first frame is required; a last frame is optional.

The raw iron bar is heated, hammered and shaped into a sword on the anvil,
sparks with every strike, ending on the finished blade on the table.
One continuous side tracking shot. Hammer clangs, fire roar, no music.

The video takes the shape of your first image. Kling's own app can also place keyframes in the middle of the timeline; this site offers first and last frame.

Pacing a 30-second clip

Kling 4.0 makes 3 to 30 seconds in one generation. A long clip needs beats, not a longer description of one moment. Kling's own advice is to break the 30 seconds into story beats:

0–5 s: wide shot of a quiet harbour at dawn, a fishing boat ties up.
5–12 s: medium shot, an old fisherman lifts a crate onto the pier.
12–22 s: his granddaughter runs in from the left and helps him; they laugh.
22–30 s: slow pull-back over the harbour as the sun clears the hills.
Audio: gulls, rope creaking, waves; she says in Spanish: "¡Abuelo, déjame a mí!"
Soft guitar from 22 s.

Tips:

• One main action per beat. Three or four beats fit comfortably in 30 seconds.
• Use rough timings ("at 6 s she turns") when the order matters.
• Draft short, then go long. Test the idea at 5 seconds and 720P, then render the full length at 1080P. Price is per second, so drafts are cheap.

Text on screen

Kling 4.0 can render legible text, logos and emoji. Write the exact words in quotes and say where they go: "the sign above the door reads 'OPEN 24H' in red neon".

Quick checklist

1. One sentence that says who, what and where.
2. A camera move, named.
3. The light, described as you'd see it.
4. Sound: spoken lines in quotes, ambience, music or "no background music".
5. For references: a job for every @Image1, @Video1, @Audio1.
6. For long clips: beats with rough timings.

Our box takes up to 5,000 characters, which is plenty; most good prompts are a few lines. Open the generator and try one of the examples above.