Kling 3.0 prompt guide

The Kling 3.0 prompt generator writes a prompt in this order; this guide explains each part. For the model, see the Kling 3.0 page.

Kling’s text-to-video guide gives the formula as subject, movement and scene, then camera, lighting and atmosphere. For 3.0, add the frame and take first, since they decide how much action fits, and the sound last, since 3.0 makes it with the picture.

1. 1Open with the frame and the length: “Widescreen 16:9, one continuous 5-second take with no cuts.”
2. 2Name the subject with the details that must not change: clothes, colors, shape.
3. 3Write the action in the present tense, from how the shot starts to how it ends, in short, simple sentences.
4. 4Add one camera move, the setting, the light and its direction, and one look.
5. 5End with the sound, and any line in quotes after the person who says it.

Kling designed 3.0 to keep characters and objects steady while the camera moves. Name one move, with its speed, straight after the action it follows. Kling’s app also has a camera control with six basic moves; in a prompt, words do that job.

One move per shot: a track that turns into an orbit smears. For dialogue and products, a locked-off camera or a slow push in keeps faces and lettering sharp.

With sound on, Kling 3.0 makes ambience, effects and speech with the picture. Its Video 3.0 user guide says giving each character their own lines lets it match line to speaker, even with three or more people. See native audio.

• Name the speaker, then the line in quotes: The owner says, “Last one today.”
• Put each person’s action before their line, and say who listens in silence.
• Keep to Kling’s five dialogue languages and name the language: says in Japanese, “どうぞ。” Kling translates a line in any other language into English.
• Ask for an accent or dialect in the description: English with a British or Indian accent, or Cantonese.
• Add a tone where it matters, and keep to two or three short lines in 10 seconds.

Kling’s separate Lip Sync feature syncs the lips of a person in a video Kling made to an uploaded or Text to Speech voice. A line in a 3.0 prompt needs neither: it is spoken in the same generation. See lip sync.

The clip’s length is fixed before Kling reads the prompt, so write as much action as the seconds hold: an action that needs longer is rushed or stops mid-move.

On DreamBench Kling 3.0 renders 9:16 for Reels, TikTok and Shorts, 16:9 for YouTube, 1:1 for feeds, 4:3, and 21:9 for a film look. In 9:16, frame one subject large.

When Kling starts from a picture, the picture already fixes the subject, the place and any text; repeating them wastes words. Kling’s image-to-video guide gives a shorter formula: subject and movement, then background and movement. See image to video.

1. 1Name the subject as the picture shows it: the woman in the red coat.
2. 2Say what it does, with a clear verb, and what moves around it: she turns to the window; her hair lifts in the breeze.
3. 3Add the camera move and the sound, and keep the motion plausible: Kling warns that a description far from the image can cause a cut.

Kling’s example: “Mona Lisa puts on sunglasses with her hand” works where “wear sunglasses” doesn’t. On DreamBench every Kling clip starts from an opening frame drawn from your references, so its prompts spend their words on motion, camera and sound.

Kling lists Start & End Frames-to-Video for 3.0: give a first and a last picture, and the prompt describes the change between them. Keep the two close in subject, light and framing, or the change can become a cut. See first and last frame.

DreamBench’s composer has no start or end slot: you add reference pictures, DreamBench draws each opening frame from them, and the prompt decides how the clip ends. DreamBench gives Kling 3.0 a last frame in one place, the logo animation: five seconds that end on your logo.

Two Kling shots agree only on what both were given. In Kling’s app, an element made from 2 to 4 pictures holds a character’s look and voice. In a prompt:

• Describe each character in the same words in every shot: “an old lighthouse keeper with a white beard and a navy wool sweater”.
• Keep clothes, props, the time of day and the direction of the light.
• Give every shot the character is in the same reference pictures.

On DreamBench each character, place and prop is drawn once as a sheet, and each shot’s opening frame from the sheets it needs. See character consistency.

Kling’s app has a Multi-Shot switch that plans cuts inside one generation of up to 15 seconds. DreamBench plans a film as several Kling generations instead, one continuous take each, so it can run past 15 seconds and any shot can be redone.

• Break the idea into moments, one action each, with the seconds it needs; a DreamBench film’s shots run 3 to 12 seconds.
• Change the framing at every cut, keep a line inside one shot, and change place only at a cut.

Write each shot with the formula and repeat the character descriptions, or describe the film on DreamBench: it writes every shot’s prompt, draws the cast, renders each shot on Kling 3.0 and cuts them together. See multi-shot video.

Paste these into Kling’s app, or use one as a DreamBench brief. Change the subject; keep the shape.

Sport: a skateboard grind

Widescreen 16:9, one continuous 5-second take with no cuts. A skateboarder in a green hoodie rolls at a concrete ledge in an empty parking lot, pops an ollie, grinds its length and lands clean. Low tracking shot. Late sun from behind. Wheels rumbling, the scrape of the grind.

Dance: a popping routine

Vertical 9:16, one continuous 10-second take with no cuts. A dancer in a red shirt stands still in a bare studio, then breaks into a popping routine: arms locking, a body wave, two spins, a freeze on the last beat. The camera orbits her slowly. Soft window light. Sneakers squeaking, a muffled beat.

Product with legible text: a honey jar

Square 1:1, one continuous 5-second take with no cuts. A honey jar with a cream label reading “WILD MEADOW HONEY” in bold black capitals stands on an oak table. A hand lifts the lid and a dipper draws out a slow golden ribbon. Slow push in, the label sharp throughout. Warm morning light from the right. The pop of the lid, honey dripping.

Dialogue: two speakers in a bookshop

Widescreen 16:9, one continuous 10-second take with no cuts. In a bookshop at closing time, an elderly owner in a gray cardigan looks up at a young woman holding a paperback and says, with a British accent: “That one’s been waiting for you.” She laughs: “Then I’d better take it home.” Locked-off two-shot. Warm lamplight, rain on the window.

A line in Chinese: the dumpling stall

Vertical 9:16, one continuous 5-second take with no cuts. At a night-market stall, a grandmother in a floral apron hands a tray of steaming dumplings to a small boy and says gently in Chinese: “小心烫，慢慢吃。” The boy nods. Locked-off medium shot. Warm bulbs, drifting steam. Market chatter, sizzling pans.

Vehicle: a motorcycle on a coast road

Cinematic 21:9, one continuous 10-second take with no cuts. A vintage silver motorcycle, its rider in a brown leather jacket, leans through a bend on a cliff-top road and accelerates toward the sea. Drone shot from behind, rising to reveal the coast. Golden hour, sea haze. The engine climbing, wind.

Animal: a fox in the snow

Widescreen 16:9, one continuous 5-second take with no cuts. A red fox trots across a snowy meadow, stops, tilts its head, then leaps and dives nose-first into the drift. Locked-off camera low in the snow, long lens. Overcast winter light. Wind, paws crunching, a muffled thump.

Food: chilli oil on noodles

Vertical 9:16, one continuous 5-second take with no cuts. Close-up on a cast-iron pan of thick noodles. A hand pours chilli oil and they sizzle; chopsticks lift a tangle toward the camera, steam rising. Slow tilt up with the chopsticks. Hard window light from the left, dark background. A loud sizzle, oil crackling.

Fashion: a coat on a cobbled street

Vertical 9:16, one continuous 10-second take with no cuts. A model in a long cobalt-blue coat walks toward the camera down a cobbled street at dusk; she stops, looks back over her shoulder, and the hem settles. The camera pulls back at walking pace, then holds. Warm shop-window light. Heels on cobbles.

Vertical social clip: a hook to camera

Vertical 9:16, one continuous 5-second take with no cuts. A young woman in a yellow raincoat films herself on a busy crossing at night and says: “Three things nobody tells you about this city.” She turns the phone to show the crowd. Handheld selfie camera. Neon on the wet street. Crowd noise.

A two-shot sequence: the lighthouse keeper

Shot 1. Widescreen 16:9, one continuous 5-second take with no cuts. An old lighthouse keeper with a white beard and a navy wool sweater climbs the last steps of a spiral staircase, lantern in hand. Wide shot, craning up with him. Cold blue dawn light. Boots on iron steps, wind.

Shot 2. Widescreen 16:9, one continuous 5-second take with no cuts. The old lighthouse keeper with a white beard and a navy wool sweater strikes a match and lights the great lamp; warm light floods his face. Locked-off close-up. Cold blue dawn turning gold. A match striking, the hiss of the lamp.

Both shots describe the keeper in the same words and keep the light; the framing changes at the cut.

Most weak prompts are short, not wrong: each gap is a choice the model makes at random.

• One prompt run twice gives two clips.
• Hands, crowds and fast physics are where errors show first.
• A real person or product can’t be known from words; it takes a picture you have the right to use.
• One generation stops at 15 seconds, and on DreamBench a prompt past 2,500 characters is trimmed.

• One prompt: the Kling 3.0 prompt generator writes it in this guide’s order, with a line lip-synced in one of Kling’s five languages. It writes the prompt only.
• A film: describe it, add up to 8 reference pictures and pick Kling 3.0. DreamBench writes every shot’s prompt, draws the opening frames and cuts the shots together; redo any shot with a note.
• Not offered: a typed prompt sent to Kling as it is. DreamBench’s Video page runs models that start from text alone; Kling 3.0 starts from a frame here.

The composer shows the price in credits before you generate; the first time, create a free account and choose a plan. For numbered shots inside one clip, see the Seedance 2.5 prompt generator; for prompts from words alone, text to video.