Kling 4.0 prompt guide

This guide covers what 4.0 changes. The basics carry over from the Kling 3.0 prompt guide; for the model itself, see the Kling 4.0 page.

Kling 4.0 doubles one generation and decides its own cuts, so a prompt becomes a timeline. The numbers are from Kling’s Kling 4.0 and Kling 3.0 user guides.

Kling’s guide gives one structure: the role of each picture you upload, the overall settings, the sequence of events with times, then the audio. Its rule is to describe what can be seen and to drop abstract words: “premium look” becomes side backlight outlining the edges and fine matte grain.

1. 1Open with the frame and length: “Vertical 9:16, one 15-second take.”
2. 2Write the overall settings as things to see: the place, the light, one look, the camera.
3. 3List the events in order, each with its seconds: “0–5 s: … 5–15 s: …”, one action per block.
4. 4End with the sound: each line in quotes after its speaker, then ambience and effects.
5. 5Close with “No subtitles, no background music” unless you want either.

A short prompt works too: Kling says 4.0 expands a simple idea and follows a clear one as written.

The length is fixed before Kling reads the prompt, so give each beat its seconds. Blocks that meet end to end, “0–8 s”, “8–20 s”, “20–30 s”, keep the order clear.

Allow a spoken line about 2 words a second, with a pause between speakers. DreamBench takes 2,500 characters, so spend them on the timeline, not on adjectives.

Kling 4.0 splits a clip into shots by itself: its Smart Storyboard is on by default, with no switch. That suits a story; a product turn can cut too soon.

• One shot: write “one continuous take, no cuts, fades or dissolves”, as Kling’s own example does, and keep one camera move.
• Several shots: number them with times and say how they join: “Shot 2 (8–15 s), hard cut, close-up …”.
• At each cut, change the framing; keep the light and the clothes.

Kling says 4.0 handles tracking shots, orbits, push-ins and pull-backs that follow fast action. Name the move in plain words, with a speed, inside its timed block.

In a long take, give each block one move: track for the run, orbit for the stop. For lines and labels, hold still or push in slowly.

Kling 4.0 makes stereo sound with every clip: speech, ambience, effects and, unless told not to, music. Kling says to write each line in its own language, inside quotation marks. See native audio.

• Name the speaker and how they sound, then the line: The fisherman (gruff, in Portuguese): “Hoje o mar está calmo.”
• Ask for a dialect or an accent Kling lists: Cantonese, Sichuanese, Beijing or Taiwanese Mandarin; American, British or Indian English.
• One speaker at a time, as Kling’s own example asks: the others listen in silence.
• Indonesian isn’t one of the 9 languages, so an Indonesian line may not come out as written.
• Kling renders on-screen text in the same languages, and burns in subtitles if you describe their size and place.

For a score, describe it in the sound block: “a slow cello line under the dialogue”. Kling’s app takes a voice profile as audio, not music. See lip sync.

In Kling’s app, the Keyframes entry fixes pictures at points on the timeline: a first frame, a last frame, both, or middle frames, up to 10. The prompt then describes what happens between them. See first and last frame.

1. 1Say which picture is which: “Use @Image1 as the first frame and @Image2 as the last frame.”
2. 2For middle frames, give each one a time block: “4–8 s | Keyframe 2: she reaches the door.”
3. 3Describe only what moves and how the camera gets from one picture to the next; the pictures already show the subject and the place.

The Keyframes entry takes no other references and can’t be combined with Omni Reference. See image to video.

Kling’s Omni Reference takes up to 15 references, numbered by upload order: @Image1, @Image2, @Video1. Give each a role, or the model guesses.

• @Image1 defines the kitchen; keep its layout and light. @Image2 is the chef; keep her face, hair and apron.
• Without references, describe each character in the same words every time they appear.

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

One Kling 4.0 generation can hold a 30-second scene, with Kling choosing the cuts. A film can also be several generations joined at hard cuts, each planned, and redone, by itself.

In Film Studio, DreamBench plans the scenes, draws the cast and has Kling 4.0 animate each one, in clips of 3 to 12 seconds. For one long take, use the Video page. See multi-shot video.

Each follows Kling’s order and fits in 2,500 characters. Change the subject; keep the shape.

A 30-second one-take chase

Widescreen 16:9, one continuous 30-second take, no cuts. A night market in Bangkok, steam and neon, wet tarmac, handheld at running pace. 0–10 s: a boy in a blue school shirt snatches a mango and runs. 10–20 s: the vendor, a woman in a straw hat, chases him past food carts; the camera tracks beside her. 20–30 s: she catches his collar, he holds out coins, and she laughs. Sizzling woks, flip-flops on wet ground, crowd noise. No subtitles, no background music.

Parkour on the rooftops

Cinematic 21:9, one 15-second take. Concrete rooftops at golden hour, long shadows, a realistic sports look; the camera tracks low beside the runner. 0–5 s: a traceur in a gray hoodie sprints and vaults an air vent. 5–10 s: he leaps a gap and lands in a roll. 10–15 s: he stands at the edge, breathing hard, as the camera orbits him once. Sneakers slapping concrete, wind, distant traffic. No subtitles, no background music.

Two languages at a station (German and English)

Widescreen 16:9, one 15-second take. A Munich station platform in the morning, a red regional train, soft overcast light; locked-off two-shot. 0–6 s: a conductor with a gray mustache checks a ticket and says (briskly, in German): “Der Zug nach Salzburg fährt von Gleis drei.” 6–11 s: the young backpacker frowns and asks (in English, with an Indian accent): “Platform three?” 11–15 s: the conductor points, smiling. Station announcements, rolling suitcases. No subtitles, no background music.

A line in Hindi: the tea stall

Vertical 9:16, one 10-second take. A roadside tea stall in Delhi at dawn, a brass kettle on a gas flame, warm light from one bulb; slow push in. 0–5 s: the tea seller pours chai from height into a small glass, a long amber stream. 5–10 s: he hands it to a taxi driver and says (warmly, in Hindi): “पहली चाय, मेरी तरफ़ से।” Pouring tea, the flame’s hiss, horns far off. No subtitles, no background music.

Product with a bilingual label

Square 1:1, one continuous 10-second take, no cuts. A bottle of yuzu soda on a stone counter, its label reading “柚子ソーダ” above “YUZU SODA” in white capitals; side backlight, condensation beads. 0–5 s: a hand twists the cap off and fizz rises. 5–10 s: the bottle turns slowly, the label sharp throughout. The cap’s hiss, bubbles, a glass set down. No subtitles, no background music.

Pixel-art animation

Widescreen 16:9, one 12-second take. 16-bit pixel-art animation, not photorealistic: a side-scrolling forest of chunky pixel trees, a purple dusk sky. 0–6 s: a small pixel fox in a red scarf runs right, hops two stumps and grabs a glowing acorn. 6–12 s: the acorn bursts into a four-frame sparkle and the fox does a happy spin. Chiptune-style jumps and a coin chime as effects. No subtitles, no background music.

Three shots with hard cuts

Widescreen 16:9, 15 seconds, three shots joined by hard cuts. A boxing gym at night, one hanging bulb, a gritty look. Shot 1 (0–5 s): close-up, a fighter in black gloves wraps her hands, jaw set. Shot 2 (5–10 s), hard cut: medium shot, she hits the heavy bag in fast combinations. Shot 3 (10–15 s), hard cut: wide shot, her coach in a gray tracksuit says (quietly, in English, with a British accent): “Again.” Gloves thudding, chains creaking. No subtitles, no background music.

First and last frame (Kling’s Keyframes entry)

Use @Image1 as the first frame (a block of clay on a wheel) and @Image2 as the last frame (a finished blue vase). 15 seconds, realistic documentary look, one continuous side shot, no cuts. 0–7 s: wet hands center the clay and draw it up into a tall cylinder. 7–15 s: the potter shapes the neck, and the clay turns into the glazed vase of @Image2. The wheel humming, wet clay. No subtitles, no background music.

Omni Reference with two pictures

@Image1 defines the café; keep its layout, plants and window light. @Image2 is the barista; keep her face, short black hair and green apron. Vertical 9:16, one 10-second take, handheld. 0–5 s: she pours a leaf pattern into a flat white. 5–10 s: she slides it across the counter and says (cheerfully, in English): “Your usual.” Milk steaming, cups clinking. No subtitles, no background music.

A clip with a score

Cinematic 21:9, one continuous 20-second take, no cuts. A lone sailing boat on a gray North Sea swell, low cloud, cold light; drone shot circling slowly, then rising. 0–10 s: a sailor in a yellow jacket hauls in the sail as spray breaks over the bow. 10–20 s: the cloud parts and sun falls on the water. Wind, rigging, waves; a slow string score that swells with the sun. No subtitles.

The last prompt asks for music, so it drops “no background music”; the others keep both exclusions.

Most weak prompts lack times, or use words a camera can’t film.

• One prompt run twice gives two clips.
• One generation stops at 30 seconds; Kling lists its extension to 2 minutes as coming soon.
• A real person or product can’t be known from words; it takes a picture you have the right to use.
• On DreamBench, Kling 4.0 always makes sound, and a Video page prompt past 2,500 characters is refused.

• One clip: paste your prompt into the Video page and pick Kling 4.0, 4 to 30 seconds, 16:9, 9:16, 1:1 or 21:9, 720p or 1080p. It goes to Kling as written.
• A film: describe it, add up to 8 reference pictures and pick Kling 4.0. DreamBench writes each scene’s prompt, makes its opening frame and joins the clips; redo any scene with a note.
• Not offered with Kling 4.0: keyframes, Omni Reference, voice profiles, video editing or extension. Kling 4.0 Flash shows as Soon.

The Kling prompt generator writes Kling 3.0 prompts; its order holds for 4.0, so add the times. The AI video prompt generator writes for any model. The price in credits shows before you generate; the first time, create a free account and choose a plan. See also text to video.