Kling VIDEO 4.0
Kling VIDEO 4.0 is a unified video model that generates 1080p clips of up to 30 seconds from text, images, reference videos, and reusable elements. It follows up to 10 keyframes per clip, including first, middle, and last frames, and combines reference images, reference videos, and elements in a single request to hold subjects, style, and motion. It also edits an existing video from a prompt, which suits long narrative shots, storyboard-driven sequences, and reference-driven production work.
Prompting
How to write prompts for Kling VIDEO 4.0, from one-line ideas to structured briefs with timed beats, camera direction, and an audio section that sets what you hear.
Introduction
Kling VIDEO 4.0 turns a text prompt into a clip of up to 30 seconds with native sound. The sound comes out of the same request as the picture, so the prompt directs both. Short and long prompts both work: a one-line idea gets expanded into shots and camera moves, and a detailed brief is followed as written.
Overall settings: a 10-second trail running shoe commercial on a rocky mountain ridge at sunrise, natural color, crisp low morning light, light handheld energy. 0–3 s: ground-level close-up as a runner's orange and gray trail shoe lands on wet rock, throwing up small droplets, the mesh upper flexing. 3–7 s: the camera tracks alongside the runner, a woman in her thirties in a black running vest, as she climbs the ridge with steady breathing, loose strands of hair moving in the wind. 7–10 s: wide shot from behind as she reaches the summit and slows to a walk against a pale orange sky. Audio: shoes crunching on gravel, steady breathing, light wind, no dialogue, no background music, no subtitles.
One prompt produced it, with three beats scheduled by the second and an audio section at the end. Each part of that brief has its own section below.
Request shape
Text-to-video takes a positivePrompt, an output size in width and height, and a duration from 3 to 30 seconds in whole seconds. Sizes come in fixed pairs, four aspect ratios at each of two resolutions:
- 720p:
1280 × 720,720 × 1280,940 × 940, and the ultrawide1460 × 624 - 1080p:
1920 × 1080,1080 × 1920,1440 × 1440, and the ultrawide2200 × 940
import { createClient } from '@runware/sdk'
const client = await createClient({ apiKey: process.env.RUNWARE_API_KEY })
await client.connect()
const [result] = await client.run({
model: 'klingai:kling-video@4.0',
positivePrompt: 'Overall settings: a 10-second trail running shoe commercial on a rocky mountain ridge at sunrise, natural color, crisp low morning light. 0–3 s: ground-level close-up as a runner\'s trail shoe lands on wet rock. 3–7 s: the camera tracks alongside the runner as she climbs the ridge. 7–10 s: wide shot from behind as she reaches the summit. Audio: shoes crunching on gravel, steady breathing, light wind, no background music, no subtitles.',
width: 1920,
height: 1080,
duration: 10
})import asyncio
import os
from runware import Runware
async def main():
async with Runware(api_key=os.environ["RUNWARE_API_KEY"]) as client:
results = await client.run({
"model": "klingai:kling-video@4.0",
"positivePrompt": "Overall settings: a 10-second trail running shoe commercial on a rocky mountain ridge at sunrise, natural color, crisp low morning light. 0–3 s: ground-level close-up as a runner's trail shoe lands on wet rock. 3–7 s: the camera tracks alongside the runner as she climbs the ridge. 7–10 s: wide shot from behind as she reaches the summit. Audio: shoes crunching on gravel, steady breathing, light wind, no background music, no subtitles.",
"width": 1920,
"height": 1080,
"duration": 10
})
asyncio.run(main())curl https://api.runware.ai/v1 \
-H "Authorization: Bearer $RUNWARE_API_KEY" \
-H "Content-Type: application/json" \
-d '[
{
"taskType": "videoInference",
"taskUUID": "6f1e2d3c-4b5a-4968-8776-5a4b3c2d1e0f",
"model": "klingai:kling-video@4.0",
"positivePrompt": "Overall settings: a 10-second trail running shoe commercial on a rocky mountain ridge at sunrise, natural color, crisp low morning light. 0–3 s: ground-level close-up as a runner's trail shoe lands on wet rock. 3–7 s: the camera tracks alongside the runner as she climbs the ridge. 7–10 s: wide shot from behind as she reaches the summit. Audio: shoes crunching on gravel, steady breathing, light wind, no background music, no subtitles.",
"width": 1920,
"height": 1080,
"duration": 10
}
]'runware run klingai:kling-video@4.0 \
positivePrompt="Overall settings: a 10-second trail running shoe commercial on a rocky mountain ridge at sunrise, natural color, crisp low morning light. 0–3 s: ground-level close-up as a runner's trail shoe lands on wet rock. 3–7 s: the camera tracks alongside the runner as she climbs the ridge. 7–10 s: wide shot from behind as she reaches the summit. Audio: shoes crunching on gravel, steady breathing, light wind, no background music, no subtitles." \
width=1920 \
height=1080 \
duration=10{
"taskType": "videoInference",
"taskUUID": "6f1e2d3c-4b5a-4968-8776-5a4b3c2d1e0f",
"model": "klingai:kling-video@4.0",
"positivePrompt": "Overall settings: a 10-second trail running shoe commercial on a rocky mountain ridge at sunrise, natural color, crisp low morning light. 0–3 s: ground-level close-up as a runner's trail shoe lands on wet rock. 3–7 s: the camera tracks alongside the runner as she climbs the ridge. 7–10 s: wide shot from behind as she reaches the summit. Audio: shoes crunching on gravel, steady breathing, light wind, no background music, no subtitles.",
"width": 1920,
"height": 1080,
"duration": 10
}Response
[
{
"taskType": "videoInference",
"taskUUID": "6f1e2d3c-4b5a-4968-8776-5a4b3c2d1e0f",
"videoUUID": "0a9b8c7d-6e5f-4a3b-9c2d-1e0f2a3b4c5d",
"videoURL": "https://vm.runware.ai/video/os/a14d18/ws/2/vi/0a9b8c7d-6e5f-4a3b-9c2d-1e0f2a3b4c5d.mp4"
}
]The aspect ratio is a creative choice, the resolution a delivery one. Iterate on the prompt at 720p, where a second costs less, and render the final take at 1080p. There is no seed, so each render is a new take rather than a sharper copy of the draft.
Starting from one line
A short prompt is a valid brief. Give Kling 4.0 a subject and an action, and it fills in the shots and the sound on its own:
A food delivery rider weaves through evening city traffic.
This is the fastest way to explore an idea. The trade-off is control: every decision you leave out is one the model makes for you. Once you know what you want, write it down, because a clear description is followed as written.
Concrete details over adjectives
Words like "premium", "luxurious" or "cinematic" carry no visual information. The model has to guess what you mean, and the guess changes from run to run. Describe what the camera would see instead: the light source, the surface materials, the background, and what stays in focus. The same product, prompted both ways:
A premium, luxurious, cinematic shot of a glass serum bottle in a bathroom. High-end feel.
A frosted glass serum bottle with a black dropper cap stands on a pale stone bathroom counter. Soft side backlight from a window traces a bright line along the edge of the bottle, the background is a blurred white-tiled bathroom with a green plant, and fine matte texture is visible on the cap. A single drop of serum slides slowly down the glass. Slow push-in at counter height. Audio: quiet room tone and one faint drip, no music, no subtitles.
A useful habit is to translate every adjective into the thing that would produce it. In the second prompt, "premium" became a side backlight tracing a bright line along the edge of the bottle:
| Instead of | Describe |
|---|---|
| "premium look" | the light direction, the surface finish, what is in and out of focus |
| "cozy home" | the materials, the light source, the objects on the counter |
| "energetic" | the camera move, the pace of the action, the sound |
| "vintage" | the wear on the furniture, the color of the walls, the type of lamp |
A structure that scales
Longer briefs stay readable when they follow the same order every time. Kling recommends overall settings, then timed events, then audio. The settings fix what stays constant, the timed beats say what happens when, and the audio section describes the track:
Overall settings: a bright living room with light oak floors and a large window, natural daylight, realistic home workout video, fixed camera on a tripod at hip height. 0–3 s: a man in his late twenties in a gray t-shirt and black shorts walks in from the left and stops in the middle of a blue yoga mat, feet shoulder-width apart. 3–7 s: he does two bodyweight squats at a steady pace, arms extended forward, heels flat. 7–10 s: he stands up, exhales, and gives the camera a short nod. Audio: soft footsteps on the mat, controlled breathing, quiet room tone, no music, no subtitles.
Timestamps give every action a slot, so the squats happen in the middle of the clip instead of whenever the model decides. Size each action to its slot: two squats fit in four seconds, three slow ones do not. When the request also carries reference images or videos, their roles go at the very top of this structure, as covered in Reference images.
Directing the camera
Camera language works the same way as visual language: name the move, the height, the speed, and what the shot ends on. "Tracks alongside at handlebar height, then pulls ahead and pans to her face" is a plan the model can execute. "Dynamic camera" is not.
A commuter cyclist in a yellow rain jacket rides along a protected city bike lane on a wet morning. The camera tracks alongside her at handlebar height, matching her speed, then slowly pulls ahead and pans to frame her face as she checks traffic over her shoulder. Reflections of storefronts slide across the wet asphalt. Audio: tires hissing on wet pavement, one ring of a bicycle bell, distant traffic, no music, no subtitles.
Push-ins, pull-outs, pans, tracking shots and orbits all work. Tie each move to something happening in the frame, like the pan landing on her face as she looks back.
Ultrawide framing
The 2200 × 940 and 1460 × 624 sizes produce an ultrawide frame close to 21:9, the shape of a cinema screen. It suits landscapes and subjects that travel sideways through a wide space:
An ultrawide cinematic shot of a dark green electric SUV driving along a coastal highway cut into the cliffs at golden hour. The camera holds a wide side view from a parallel vehicle, keeping the car small in the frame against the ocean and the long curve of the road ahead. Audio: tire noise, wind, distant surf, no music, no subtitles.
Write the composition for the shape. Keep the subject small and let the width do the work: the road and the coastline ahead of the car are what make the shot read as ultrawide.
Sound and the two exclusions
Every clip comes back with sound, and there is no parameter to turn it off. The audio follows the prompt, so end each brief with what the track should contain: dialogue, ambience, effects, music.
A high school basketball player in a white practice jersey shoots three free throws in an empty gym under fluorescent lights. Medium shot from the baseline, steady camera. Audio: the ball bouncing on the hardwood, sneaker squeaks, the swish of the net, the hum of the lights. SFX only, no music, no voices, no subtitles.
Kling's guidance is to describe what you want rather than what you don't, because negative instructions add room for error. Two exclusions are the exception: the model can add burned-in subtitles and a background music bed on its own, so write "no subtitles" and "no background music" whenever you don't want them.
Rewrite every other negative as a positive. "No people in the background" works better as "an empty street behind her".
Tips
-
Start short to explore, then write it out. A one-line prompt shows you what the model does by default. A full brief makes the result repeatable.
-
Swap adjectives for things the camera can see. Light direction and surface materials produce the look that words like "premium" only gesture at.
-
Order long prompts the same way every time. Settings, then timed beats, then audio keeps a long brief easy to edit.
-
Put actions on the clock. Seconds like
3–7 splace each action where you want it in the clip. -
Give the camera a plan. A move, a height, a speed, and the shot it ends on.
-
Exclude subtitles and music explicitly. They are the two things the model adds unasked, and the only negatives worth writing.