01 The Core Formula
Kling 3.0 understands structured direction: who, what they do, how the camera moves, where, in what tone. This chapter breaks a good prompt into five reusable parts, then demonstrates long-take storytelling with the official flagship sample — get this skeleton right and every later chapter is just an extension of it.
1.1 The Basic Prompt Formula
Everything starts with one sentence: subject plus motion are the two required parts; camera, environment, and style stack on as needed. Play the clip in your head first, then write it down following the formula. Close with a constraint line — "keep it free of subtitles", "do not generate watermarks" — it measurably cuts the failure rate.
[Subject] + [Motion] + optional [Camera] [Environment] [Style]
Subject
The star of the frame — a person, animal, or product. The more specific, the better.
Motion
What the subject does, with body parts and pace spelled out.
Camera
Framing and camera movement: close-up, push-in, orbit — one move per shot.
Environment
Setting and mood: place, time of day, lighting, weather.
Style
Art direction and image quality: palette, texture, cinematic feel.
A Corner Café in the Early Morning
Prompt
The outdoor terrace of a corner café in early morning, sunlight slanting across the cobblestones. A woman in a beige trench coat sits in a wicker chair and opens a book, tucking a loose strand of hair behind her ear. A waiter crosses the background carrying a tray. The camera slowly pushes in to the pages in her hands. Quiet, cozy mood, warm golden tones, rich detail. Keep the video free of subtitles; do not generate any watermark.
Two sentences for subject + motion, one each for camera and environment, then a constraint line to close. Suggested: 5s, 1080p, 16:9.
1.2 Camera Language
Camera terms are Kling 3.0's strength: push in, pull back, pan, tracking, orbit, and dive all execute accurately. One rule of discipline — a single camera move per shot; stacking push, orbit, and pan together just buys you shaky footage.
Orbiting a Cliffside Lighthouse
Prompt
A red-and-white lighthouse stands on a coastal cliff at dusk, its lamp just switching on as waves crash against the rock face in white spray. The camera slowly orbits the lighthouse one full circle, rising from sea level to a top-down view, while the setting sun dyes the clouds orange and violet. Cinematic quality, layered lighting, smooth and stable footage. Keep the video free of subtitles; do not generate any watermark.
"The camera slowly orbits … one full circle" is the steadiest orbit phrasing — add one line about the change in height. Suggested: 8 seconds, enough time for the orbit to complete.
1.3 Second-Level Beats for Long Takes
Kling 3.0's official user guide writes long takes as second-level beats: at the 4th second the camera accelerates, at the 8th it pushes to a medium shot, at the 12th music and motion peak — pin every turn of action and camera to the timeline and the model performs to the beat. The example below is the official flagship sample verbatim. Note how it differs from numbered-shot storyboarding: second-level beats suit one unbroken take; shot numbers suit multi-angle cutting.
A 15-Second Starlight Run in One Take (Official Sample)
Prompt
Opening with an ultra-wide-angle medium-long shot tracking horizontally, the stabilizer moves low to the ground, with a highly contrasting romantic cinematic tone of cold blue night and silvery white starry sky, exuding a strong poetic realism and classical epic temperament. The protagonist is a young woman in a dark green long dress, running with all her might on the garden lawn illuminated by moonlight; her skirt billows in the wind forming surging dynamic curves, she clutches a small white flower in her right hand and lifts the hem of her dress with her left, breathing rapidly yet with a firm gaze. At the 4th second, the camera accelerates forward with her, and multiple men and women in old-era ball gowns break into the frame one after another from the left and right sides in the background, running alongside her—some try to approach, some turn back to shout, yet none truly touch her, implying pursuit and escape. At the 8th second, the camera gradually zooms in to a medium shot, pans to track forward in front of the protagonist and lifts slightly; she glances back briefly at a young male character behind her, their gazes meet for a split second, emotions erupt mid-run, and the woman and man join hands to run together. At the 12th second, the music and movement reach a climax; the camera moves forward close to her side face and fluttering hair, she releases the white flower and tosses it into the air, the flower drifting down in slow motion as the crowd behind brushes past it. In the final 3 seconds, the camera keeps moving forward, the woman and man break through the crowd and dash toward the starry sky at the end of the garden, their figures gradually taking over the center of the frame. The overall atmosphere is fiery, romantic and resolute, a burst of narrative about fate, choice, and freedom.
Requires 1 first-frame image (the garden / the woman setting off); native audio is pre-enabled. "At the Nth second + a turn in action or camera" is the official long-take beat notation — this is the flagship sample from kling.ai's user guide, verbatim.
02 Native Audio
Sound effects and dialogue generation are Kling 3.0's headline feature: rain, wind, footsteps, and spoken lines are generated together with the picture, automatically in sync — no post dubbing needed. Rarer still are the languages and dialects — English, Spanish, Cantonese all get performed with tone and attitude. The writing is just as direct: like the picture, the more specific you are about sound, the more precise the result. All examples in this chapter have native audio pre-enabled.
2.1 Ambient Sound Effects
Write sound effects as "name it + layer it": list each sound you want, then state which is foreground and which is distant. If rain is the star, write it first, and let far-off thunder pad the background — leave a scene's sound blank and the model improvises, not always the mood you wanted.
Night Rain on a Forest Cabin
Prompt
At night, the porch of a wooden cabin in the forest, warm yellow light glowing from the windows. Heavy rain drums densely on the eaves and leaves, and runoff streams off the porch edge in beaded strings. Static camera framing just one corner of the porch and the curtain of rain. Audio: up close, dense raindrops hammering the wooden roof and the tick of drops hitting puddles; in the distance, an occasional low rumble of thunder — the overall mood quiet and healing. Keep the video free of subtitles.
Layer the audio — near main sound → detail sounds → distant backdrop — and close with a mood word. Suggested: 5 seconds, native audio pre-enabled.
2.2 Character Dialogue
The official dialogue notation is "character (tone cue): line" — name each speaker first, put the tone in parentheses, and follow with the exact words. In multi-character scenes, name all four speakers one by one and the model assigns voices and lip-syncs automatically. The shorter the line, the steadier the sync.
A Family of Four Watching TV (Official Sample)
Prompt
Home setting with a faint hum of the living room air conditioner in the background for a realistic daily vibe. Mom (softly, in a surprised tone): Wow, I didn't expect this plot at all. Dad (in a low voice, agreeing, in a calm tone): Yeah, it's totally unexpected. Never thought that would happen. Boy (in an excited tone): It's the best twist ever! Girl (nodding along, in an enthusiastic tone): I can't believe they did that!
The official four-character dialogue sample verbatim: one short line per person with a parenthesized tone cue, and the background hum of the air conditioner sets the everyday scene first. Suggested: 10 seconds, native audio pre-enabled.
2.3 Multilingual and Dialect Dialogue
This is where Kling 3.0 stands apart: lines can be generated in Spanish, Cantonese, and more — with accent and attitude specifiable. The notation: before the line, note in parentheses "in Spanish" / "in Cantonese" plus a tone description, and keep the line itself in the target language verbatim. Both examples below are official samples, verbatim.
Asking Directions in Spanish in Madrid (Official Sample)
Prompt
Sunlight fills the old streets of Madrid. In front of a street-side bakery, a Chinese female tourist and a male tourist wearing a gray hoodie walk toward the shop clerk, both wearing polite smiles. Female tourist (speaking slightly slowly, with an awkward accent, in Spanish): Disculpe, ¿dónde está la plaza mayor? A white-haired Spanish shop clerk (turning slightly and pointing forward, with a light and cheerful tone, in Spanish): Por allí, a dos calles. Muy cerca. The female tourist nods to express her thanks. The male tourist also nods in agreement and says (in Spanish): Muchas gracias. The shop clerk smiles and nods in response. The two tourists then turn and walk in the indicated direction.
"Speaking slightly slowly, with an awkward accent" is characterization through delivery — accent is part of the performance. Suggested: 10 seconds, native audio pre-enabled.
Cantonese Banter in a High-Rise Office (Official Sample)
Prompt
In a high-rise office building, the man leaned back, wearing a tired, disdainful expression, and said in Cantonese: 「其实……我真系唔系好 buy 你呢个 logic 啰。成个 proposal 根本 align 唔到我哋个 core value。你个 flow 咁乱，点样去 convince 个 client 呀？不如你返去 re-think 下个 angle，听朝早我要见到个 final version。」
The Cantonese line stays untranslated, and the mixed-in English work terms (logic / proposal / align) are exactly how Hong Kong Cantonese sounds — dialect authenticity lives in details like these. Suggested: 10 seconds, native audio pre-enabled.
03 Image to Video
One first-frame image plus one motion description, and Kling 3.0 brings the picture to life. One important rule: the output aspect ratio automatically follows the first-frame image — a vertical image yields a vertical video, a square yields square. Crop to your target ratio before uploading; it's easier than fighting it in the prompt. Your prompt's focus shifts from describing the picture to describing motion and sound, because the picture already exists.
3.1 Bringing a Photo to Life
An image-to-video prompt covers only three things: what moves, how it moves, and whether the camera moves. Don't force in things that aren't in the image — instead, "wake up" what's already there: fluttering hair, drifting clouds, a blink and a smile are all high-hit-rate small motions.
The Person in an Old Photo Blinks and Smiles
Prompt
The person in the photo slowly blinks, the corners of her mouth rising into a gentle smile as her gaze settles toward the camera, loose strands of hair stirred by a light breeze. The camera stays fixed; lighting and color grade carry over from the photo's original texture. Subtle, natural motion. Keep the video free of subtitles.
Requires 1 first-frame image. Describe only the motion, never re-describe the scene — "carry over the photo's original texture" holds the style. The output ratio automatically follows your uploaded image.
3.2 Product Showcase
For e-commerce product clips, write three things clearly: how the camera moves, which product details get close-ups, and what the voiceover says. Add "product appearance stays consistent with the image" to lock the details — Kling 3.0 even preserves native text on the bottle, as the official sample below shows in full.
A Perfume Ad in a Parisian Apartment (Official Sample)
Prompt
By the window of a Parisian apartment, with soft French piano BGM in the background, the gilded afternoon sunlight filters through the shutters onto the perfume bottle, casting dappled light and shadow. The camera pans slowly in from the scattered rose petals, shifting focus to the faceted cut of the Kling perfume bottle. Voiceover (lazy French female voice, British accent, slow pace): Bathe in the golden hour. The camera orbits the perfume bottle in slow motion, capturing the play of light and shadow on the golden lettering and bottle body. Voiceover: Kling, a whisper of Parisian elegance. The camera pulls back and freezes on the complete scene—the Kling perfume bottle standing on a velvet pedestal, with Parisian buildings faintly visible outside the window. Voiceover: Wrap yourself in luxury with every breath.
Requires 1 product image of the bottle; native audio is pre-enabled. The voiceover carries a timbre cue — "lazy French female voice, British accent, slow pace" — and the golden lettering on the bottle stays legible through the orbit. The official i2v ad sample, verbatim.
04 Element References
Element references (r2v) lock character and element consistency with 2–4 images: feed it different angles and outfits of the same character and the model builds a complete identity file that survives any scene change. Reference your images in the prompt as "Image 1", "Image 2" — the platform handles the citation syntax automatically; you don't need any special symbols. Hard rule: 2–4 reference images — fewer than 2 cannot be submitted.
4.1 The Element Reference Formula
A reference prompt is a three-part structure: bring the assets on stage with "Image N", describe what happens in the new scene, then close with one line of "stay consistent". Upload order is the numbering order — the first image is Image 1, the second is Image 2.
[Elements: Image 1, Image 2…] + [Scene description] + keep [element] consistent
Reference Elements
2–4 images, cited in the prompt as "Image N"; upload order is the numbering order.
Scene Description
What happens in the new scene: action, environment, camera.
Consistency
One line of "stay consistent with the reference images" reinforces the appearance lock.
A Brand Mascot's Daily Life
Prompt
The little blue whale mascot from Image 1 and Image 2 appears in a bright bakery. It stands on tiptoe at the counter, eyeing a tray of freshly baked croissants with curiosity, its tail swaying happily. The camera slowly pushes in from a medium shot to its face. Keep the mascot identical to the reference images. Warm, wholesome feel, 3D animation quality. Keep the video free of subtitles.
Requires 2–4 reference images (fewer than 2 cannot be submitted); multiple angles of the same character give the best results, and the platform handles the citation syntax.
4.2 Combining Multiple Elements
Each reference image gets one job: Image 1 for the lead, Image 2 for the co-star, Image 3 for the prop. Call out each role by number and the model won't cross them — especially when people and props are mixed, spell out what each image is for.
A Couple's Travel Clip
Prompt
The man from Image 1 and the woman from Image 2 walk side by side down an avenue of ginkgo trees in autumn, golden leaves drifting on the wind. She reaches out to catch a leaf and turns to show him with a laugh; he raises his camera and captures the moment. The camera tracks them smoothly from the side. Both people's faces, hairstyles, and outfits stay identical to the reference images throughout — no face-swapping, no distortion. Warm tones, natural and vivid. Keep the video free of subtitles.
Requires 2–4 reference images. In two-person scenes, state "Image 1 = who, Image 2 = who" up front, and close with the global lock — "no face-swapping, no distortion" — to prevent drift. Suggested: 8 seconds.
05 Long-Duration Storytelling
Kling 3.0 generates any whole-second duration from 3 to 15 seconds. Longer is not automatically better: 5 seconds suits quick iteration, 8–10 seconds holds a storyboarded narrative, and 12+ seconds belongs to "a continuous take that plays out in full" — unbroken action, complete dialogue scenes, the things short durations can't fit.
5.1 One Complete Continuous Take
The key to a continuous long take is queuing actions in time order: what happens first, what happens next, where it lands at the end — lay it out like a mini screenplay. The official sample below also schedules the emotional arc on the timeline — from despair to reunion, a whole scene played out in 15 seconds.
A Dinosaur Reunion in the Plaster Tower (Official Sample)
Prompt
This is a 15-second cinematic long take, a single unbroken shot with no edited transitions. The scene is set inside a tower of plaster statues dappled with light and shadow, surrounded by towering white plaster sculptures, evoking an air of mystery and oppression. The shot opens with the protagonist skidding to a halt at the center of the scene after a frantic run, chest heaving, expression dazed and helpless, fear glinting in their eyes. The camera orbits the protagonist in a smooth 360-degree pan. As the camera rotates, the protagonist glances anxiously around and shouts: "Alex! Alex where are you! Are you here?" A cute dinosaur cry then echoes in the background, and the camera pushes in over the protagonist's shoulder to their back— a small to medium-sized, adorable baby dinosaur steps out from behind a plaster pillar, letting out a sweet chirp. Startled by the sound, the protagonist whips around; catching sight of the dinosaur, they burst into tears instantly and rush forward without hesitation to clasp it tightly in their arms. The dinosaur nestles obediently against them. Sobbing, the protagonist strokes the dinosaur gently and trembles: "I found you! Thank God, I'm so scared!" The overall lighting and shadow boast a cinematic texture, with the emotion shifting from despair to an overwhelmingly touching reunion.
Requires 1 first-frame image (the plaster tower / the protagonist); native audio is pre-enabled. Open with the contract — "a 15-second long take, a single unbroken shot with no edited transitions" — then queue the actions and the emotional arc in time order. The official long-take sample, verbatim.