What is the best prompt structure for Kling 4.0?

Label each reference, set the scene in concrete detail, write the action as time ranges, direct the camera, then add the stereo bed, effects, and music.

How long can a Kling 4.0 prompt be?

Kling’s guide lists up to 8,000 tokens. A one-line idea still works; the model expands it. A fully specified prompt is followed more closely.

How do I use images and videos as references?

In Omni Reference, number them in upload order and assign a role: appearance, action, camera, or composition. The cap is 15 assets, including at most 5 videos and 30 seconds of video in total.

How do keyframes differ from a start image?

A start image is the first frame. Keyframes let you place up to 10 images along the timeline. They are a separate entry and cannot be combined with Omni Reference.

Which languages can Kling 4.0 speak?

Kling lists Chinese, English, Japanese, Korean, Spanish, Portuguese, German, French, and Hindi, with accents and dialects called out in the user guide. Put the language in the voice label.

Is Kling 4.0 available in Krea Video?

Not yet. Krea has not published a selector name, duration, resolution, or price for Kling 4. These prompts follow Kling’s guide, and the clips on this page are earlier Kling generations.

Kling 4.0 prompt guide

How to write prompts for Kling 4.0, Kuaishou’s video model for clips of 3 to 30 seconds with stereo sound, up to 10 keyframes, and prompts of up to 8,000 tokens. Kling 4.0 is not in Krea Video yet. Every clip on this page was made with an earlier Kling model on Krea. The prompts follow Kling’s 14 October 2026 guide.

Get notified See examples

Last updated: 2026-10-19

Prompt formula

The Kling 4.0 prompt formula

Asset roles 01

Say what each upload controls: @Image1 preserves the face, @Video1 sets the camera path.

Scene 02

Place, time, light, and texture in concrete words. Skip “premium” and “vintage”.

Timed beats 03

Write 0–Xs, then the next range, through the end. One action per range.

Camera 04

Shot size and behavior: locked, push in, orbit, tracking, long take that never cuts.

Stereo audio 05

Ambience first, then sound effects on their beat, then music with an entrance. Say “no subtitles” or “no background music” when you need them gone.

Examples

Kling 4.0 prompt examples

Nine prompts, one per use case, written from Kling’s 14 October 2026 guide. The videos were made with earlier Kling models on Krea. Copy a prompt now. Kling 4 is not selectable on Krea yet.

Marketing Kling 4.0 · Earlier Kling model · 15s · 21:9 · stereo · text only Earlier Kling model

Water bottle ad at sunrise

Timed beats, a 21:9 frame, and music that enters on the last beat.

Prompt

21:9. Stereo wind in the pines, birds far off. No subtitles. No logo.

0–4s: Macro of condensation on a matte sage-green steel bottle. Locked camera.
4–10s: Pull back. A hiker in a rust-orange fleece drinks on a rock ledge above a misty valley at sunrise.
10–15s: Acoustic guitar enters as the hiker lowers the bottle. The guitar swells as the sun clears the ridge.

Get notified

Product Kling 4.0 · Earlier Kling model · start image · locked camera · no music Earlier Kling model

Product lid lift from a still

@Image1 is not described again. The prompt only says what the hand does.

Prompt

@Image1 is the product. Preserve the box, the label, and the grey table. Do not redesign them.

Locked overhead camera. A hand enters from the right, slides the box a few centimetres toward center, and lifts the lid. Soft cardboard scrape. The lid taps the table. Quiet room tone. No background music.

Get notified

Film Kling 4.0 · Earlier Kling model · dialogue labels · stereo rain Earlier Kling model

Goodbye on a rainy platform

Character labels with voice tone, and an action before each line.

Prompt

A night train platform in light rain. Stereo: rain on the metal roof, a humming departure board, a distant horn. No subtitles.

Character A: a young woman in a yellow raincoat holding a paper ticket. Character B: an older man in a grey wool coat with a small suitcase.

The woman holds out the ticket.
[Character A: Young Woman in yellow raincoat, steady quiet voice]: "You can still stay."
Immediately, the man looks down at the suitcase.
[Character B: Older Man in grey coat, tired warm voice]: "I have stayed too long already."

Medium two-shot, locked camera, slow push-in on the last line. A low cello note enters.

Get notified

Games Kling 4.0 · Earlier Kling model · long take · no cuts Earlier Kling model

Fox knight long take

The camera is fixed behind the character and the prompt says it never cuts.

Prompt

Stylised hand-painted 3D, soft dusk light. No subtitles.

0–8s: A small fox knight in dented silver armor and a red scarf runs along a crumbling stone bridge over a misty canyon.
8–12s: The knight leaps a gap and rolls onto a grassy ledge.

Long take: a third-person camera stays level behind the knight at shoulder height and never cuts. Armor clinks. Wind howls. Heroic strings rise as the knight lands.

Get notified

Photography Kling 4.0 · Earlier Kling model · 9:16 · sound bed first Earlier Kling model

Rain on a café window

The ambient bed opens the prompt. Each effect is its own sentence.

Prompt

Rain tapping on a café window. Muffled street traffic, stereo. No background music.

0–5s: Macro of raindrops on the glass. City lights sit as soft bokeh. One drop gathers others and runs down the pane. Locked camera.
5–8s: Rack focus to a red umbrella passing outside.
A cup is set on a saucer. Ceramic clinks softly.

Get notified

Architecture Kling 4.0 · Earlier Kling model · 2 of 10 keyframes Earlier Kling model

Canal house across two keyframes

Two of the ten keyframe slots pin the start and the end. The prompt is the motion between them.

Prompt

Keyframe 1 is the still of the canal house. Preserve the facade, the laundry line, and the green shutter.
Keyframe 2, at the end: the same house, shutter open, two pigeons on the ledge.

Between them: a slow push-in, a light breeze moving the laundry, the shutter swinging open. Wind and fabric. Distant church bells. Pigeons coo. No subtitles.

Get notified

E-commerce Kling 4.0 · Earlier Kling model · on-screen text held Earlier Kling model

Sweater folded, letters kept

The start image owns the word PICNIC. The prompt says not to redraw it.

Prompt

@Image1 is the sweater flat lay. Preserve the knit, the word PICNIC, and the background. Do not redraw the letters.

Locked overhead camera. Two hands fold the sweater: sleeves cross, then the body folds into a square with PICNIC still readable on top. Soft fabric rustle. Quiet room tone. No background music.

Get notified

Fashion Kling 4.0 · Earlier Kling model · timed multi-shot · 9:16 Earlier Kling model

Three-beat corduroy lookbook

One description of the suit is repeated in every time range so the person stays put.

Prompt

9:16. Pale-pink seamless studio. The same woman in a mustard corduroy suit and white sneakers in every beat. No subtitles.

0–3s: Full shot, locked camera. She walks to center. Sneakers squeak.
3–6s: Close-up. She adjusts the jacket cuff. Ribbed corduroy fills the frame. A shutter clicks.
6–10s: Medium shot, slow orbit. She turns to camera and smiles. Soft electronic beat enters in stereo.

Get notified

Personal Kling 4.0 · Earlier Kling model · Hindi and Indian English Earlier Kling model

Hindi and Indian English in a kitchen

Each voice label names the language. Hindi is one of the languages Kling added for 4.0.

Prompt

A small kitchen at night. A pot simmers. Soft radio, stereo. No subtitles.

Character A: a grandmother with silver hair in a blue apron. Character B: her grandson in a green hoodie.

The grandmother lifts the lid.
[Character A: Grandmother, warm teasing voice, Hindi]: "Bhook lagi, beta?"
Immediately, the grandson leans into the steam.
[Character B: Grandson, laughing, Indian English]: "Since noon, dadi."

She laughs and hands him a spoon. Medium two-shot, locked camera.

Get notified

How Kling 4.0 reads a prompt

Kling’s guide says to describe what you want to see, not what to avoid. “Side light outlines the bottle, the background is soft, the surface shows a fine matte grain” beats “premium look.” Negative lines raise errors, except two: “no subtitles” and “no background music,” because the model may add both.

A short idea is enough for a first pass. Kling 4.0 expands a simple prompt with blocking and camera. When the prompt is already specific, follow it instead of inventing. Prompts can run to 8,000 tokens.

Name characters and key objects once, then keep those words.
Put the action on a clock: 0–4s, 4–10s, 10s–end.
State who speaks, in which language, and with what tone.

Label every reference

Omni Reference takes up to 15 assets: 10 images, 5 videos totaling 30 seconds, 7 elements (up to 3 of them video), and a voice. Files are numbered in upload order: @Image1, @Video1. If you only pass the id, the model guesses the role.

Omni Reference and the Keyframes entry cannot be combined in one task. Keyframes are images only, up to 10, and they pin visual states on the timeline.

Prompt

@Image1 defines the actor. Preserve the face, hair, and coat.
@Video1 is the camera path only. Do not copy its location or wardrobe.

Write the clip as timed beats

Kling lists a single generation from 3 to 30 seconds, in 1-second steps, default 10 seconds. Multi-shot pacing is on by default for longer clips, so write the sequence instead of one frozen moment.

Keep one framing and one camera behavior inside each range. Repeat the character description when the shot changes, or the person can drift.

Prompt

0–3s: Full shot, locked camera. She walks to center.
3–6s: Close-up. The same woman adjusts the jacket cuff.

Pin moments with keyframes

Use keyframes when the important frames already exist: a start still and a later still, up to 10. The prompt describes the motion between them. Do not re-describe the pictures.

Keep an end frame close in time to the start. For a large jump, such as day to night, Kling’s guide says to make two clips.

Prompt

Keyframe 1 is the canal house. Preserve the facade.
Keyframe 2, at the end: the same house, shutter open.

Direct the stereo soundtrack

Kling 4.0 generates two-channel stereo with the picture. Open with the ambient bed, then give each effect its own sentence on its beat. Give music an entrance and a change.

Krea does not offer a Kling 4 audio switch yet. Kling’s guide treats sound cues in the prompt as part of the generation, and “no background music” is how you keep score out.

Prompt

Rain tapping on a café window. Muffled street traffic, stereo.
A cup is set on a saucer. Ceramic clinks softly.

Use the new languages and accents

Kling lists speech and on-screen text in Chinese, English, Japanese, Korean, Spanish, Portuguese, German, French, and Hindi, plus further accents. German, French, and Hindi are new relative to Kling 3.0. Dialects in the guide include Beijing, Taiwanese, Northeastern, Sichuan, Cantonese, and American, British, and Indian English.

Write the line in its language and put the accent in the voice label. For text already in a start image, tell the model to preserve the letters instead of redrawing them.

Prompt

[Character A: Grandmother, warm teasing voice, Hindi]: "Bhook lagi, beta?"
[Character B: Grandson, laughing, Indian English]: "Since noon, dadi."

Describe the change, not the still

With a start image, Kling keeps identity, layout, and text. Describing the picture again can make it drift. Use the prompt for what moves: the subject, the camera, and the room tone.

Video references share one quota with video elements and edit clips: up to 5 videos and 30 seconds combined. Say which video is action, which is camera, and which details must not carry over.

Prompt

@Image1 is the product. Preserve the box and the label.
A hand slides the box to center and lifts the lid.

Settings Kling published

From Kling’s 14 October 2026 guide: 720p, 1080p, and 4K; aspects 21:9, 16:9, 1:1, 9:16, and Auto; duration 3–30 seconds. 10-bit HDR at 1080p and 4K, and extension out to 2 minutes, are listed as coming soon, not part of that rollout. A faster Kling 4.0 Flash mode was announced; its exact limits are not in that spec table.

Krea has not published Kling 4 durations, resolutions, or a price, because Kling 4 is not in Krea Video. Do not treat the Kling guide’s maximum as a Krea setting.