Kling 4.0 Prompting Guide: Write 30 Seconds, Not 5

A five-second prompt is a sentence. A thirty-second prompt is a script. Most people will try the new model by pasting in the same one-line prompt they used last year, get a long clip where the interesting thing happens in the first four seconds, and decide the extra length is padding.

It isn't. It's a different job. Kling 4.0 generates native clips from 3 to 30 seconds, takes up to 10 keyframes and up to 15 references in one task, and accepts prompts of up to 8,000 tokens. Those numbers only pay off if you write for them. This Kling 4.0 prompting guide covers how: the structure Kling itself recommends, how to plan beats and keyframes, how to give references jobs, and how to write dialogue and sound. Kling says the full model launches this October, and it's coming to each::labs. Where Kling has published guidance, we say so. Where we're extrapolating from Kling 3.0 habits, we say that too.

What Kling Has Confirmed About Prompting Kling 4.0

Start with the official material, because there's more of it than usual for a model that isn't fully out. Kling's Kling 4.0 announcement gives a prompt formula in plain words: Subject + Action + Setting + Camera + Lighting + Pacing + Sound. For dialogue, it asks you to describe the character's expression, movement and tone, so the words connect to the performance.

It also lays out a workflow. Clarify the purpose first (an ad, a short narrative, UGC, a product demo, a brand film). Break the 30 seconds into story beats. Assign a specific role to each reference: character identity, product details, scene layout, visual style or motion. Write the prompt as shot direction. Then review the whole sequence for continuity rather than judging single frames.

The capability side is just as specific. Kling 4.0 supports up to 10 images, up to 5 videos (30 seconds combined) and up to 7 subjects, within a total of 15 reference assets. Kling lists 4K output, a 21:9 ultrawide format, two-channel stereo audio with improved lip-sync, and on-screen text in Chinese, English, Japanese, Korean, Spanish, Portuguese, German, French, Hindi and more. One thing to keep in mind if you're testing now: the early-access build is Kling 4.0 Flash, which Kling's feature page lists at up to 20 seconds and 720p, so judge prompts there with that ceiling in mind.

Use Kling's Formula as a Checklist, Not a Template

Subject, Action, Setting, Camera, Lighting, Pacing, Sound. Seven slots. The mistake is filling every one with an adjective. The formula is a checklist for what the model needs to know, and some slots deserve a sentence while others need two words.

Pacing is the slot people skip, and on a 30-second clip it's the one that matters most. Kling put it in the formula for a reason. "Slow build, quick cut to the reveal, hold on the final frame" tells the model how time should feel. Without it, a long clip tends to spread its action evenly, which is how you get 30 seconds that feel like 10 seconds stretched.

Subject: a ceramicist in her fifties, grey linen apron, clay-stained hands. Action: she centers clay on a wheel, pulls up a tall vase, then stops the wheel and smiles at it. Setting: a small sunlit studio with shelves of glazed pots. Camera: starts on a close-up of her hands, slowly pulls back to a medium shot. Lighting: soft afternoon light from a side window, warm dust in the air. Pacing: unhurried and meditative, the final smile held for three seconds. Sound: the hum of the wheel, wet clay, a quiet radio in the background.

You don't need the labels in a real prompt. Writing them once, though, shows you which slot you forgot.

Plan 30 Seconds in Beats Before You Write a Word

Kling's second workflow step is the one that changes your results most: break the 30 seconds into story beats. A clip that long needs a beginning, a turn and an end. If you can't say what happens at second 20, neither can the model.

Kling's own keyframe guide shows the pattern with timestamps in the prompt, segment by segment, such as 0 to 5 seconds, then 5 to 12 seconds. That's the most useful habit to bring over. Write in time order, one segment per beat, and keep each segment to what changes in it.

0 to 6s: a rainy city street at night, a cyclist in a yellow jacket waits at a red light, close-up on her face, breath visible. 6 to 15s: the light turns green, she pushes off, tracking shot alongside her through traffic, reflections streaking on wet asphalt. 15 to 24s: she turns into a narrow alley, the camera rises above her, rooftops and neon. 24 to 30s: she stops at a door, takes off her helmet and looks up at the camera. Pacing: tense, then fast, then a held quiet ending. Sound: rain, tires on water, distant traffic, a low synth pulse that drops out on the final beat.

Today on each::labs you can rehearse this exact structure with Kling 3.0 Pro text-to-video and its multi-shot mode. A beat sheet that works as a set of shots will translate directly into one longer generation later.

Keyframes Mark Moments, the Prompt Moves Between Them

Kling puts the division of labor in two sentences: "Keyframes show the important moments. The prompt explains how the video moves between them." Kling 4.0 accepts up to 10 keyframes, and Kling is clear you don't need all of them. Use one only when it defines an important visual change or story beat.

The official example layout for a 30-second product video is a good default: establish the character and setting at 0 seconds, introduce the product at 6, move into product use at 12, shift the scene or camera angle at 20, land the hero shot at 28. Five keyframes, not ten. Each one is a state worth pinning.

Then write the prompt along the same timeline and don't describe what the images already show. If keyframe 3 is the product on the counter, the prompt for that segment should say how the hand reaches it, how the camera moves and what we hear. Kling's guide names the mistakes directly: too many keyframes, changing too much inside a single keyframe, inconsistent references across inputs, and prompts that contradict the keyframe order. The fastest way to break a long clip is a prompt that says the product appears at second 10 when its keyframe sits at second 6.

If you want to practice now, Kling 3.0 Pro image-to-video on each::labs takes a start and end frame, which is the two-keyframe version of the same discipline.

Give Every Reference One Job

Fifteen reference slots is a lot of rope. Kling's instruction is to assign each reference a role: identity, product, scene layout, style or motion. Its keyframe guide adds the sentence that ties it together: "Keyframes define when something should change. Omni Reference helps define what should stay consistent."

So name the roles in your prompt. Kling hasn't published an exact syntax for referring to inputs, so treat the wording below as our suggestion, not an official format:

Use the first two images for the presenter's face and outfit. Use the third image for the bottle, keep its label exactly as shown. Use the video reference only for the handheld camera rhythm, not for the setting. The presenter stands in a bright bathroom, picks up the bottle at 4 seconds and turns it toward camera.

Fewer, aligned references beat many conflicting ones. Keep the lighting the same across your character images, show one subject in each image, and use clean frontal shots. Our guide to keeping a character consistent across AI video shots covers the reference habits that carry straight over, and you can test multi-reference prompts today on Kling O3 Pro reference-to-video.

Direct the Camera in Order

Kling says keyframes and prompts can drive push-ins, pull-backs, pans, tracking shots, orbits, cuts and transitions, and that large camera moves and fast motion are improved in Kling 4.0. It doesn't document a separate camera-control parameter, so camera direction lives in your words.

Two rules make that work. Use real camera vocabulary ("slow push in", "orbit left around the subject", "locked-off wide shot", "handheld, slight natural shake") instead of mood words. And describe moves in the order they happen, attached to the beat they belong to. "Starts on a close-up of the hands, tilts up to her face, then pulls back to a wide as she stands" is a choreography. "Dynamic cinematic camera" is a hope. Our AI video camera movement prompts guide has the vocabulary.

Write Dialogue Like a Performance Note

Stereo audio and better lip-sync make dialogue one of the most interesting parts of Kling 4.0, and Kling's advice is specific: describe expression, movement and tone alongside the words. A line on its own gets read. A line with direction gets performed.

A two-shot in a small kitchen at night, warm overhead light. The older man sets down his mug, looks at his daughter and says quietly, almost smiling, "You were right about the job." She laughs once, surprised, and looks away. Sound: fridge hum, rain on the window, no music.

Kling lists dialogue in multiple languages and accents, and on-screen text in the languages above. If a clip needs a line in Spanish or Hindi, write the line in that language and say who speaks it. And always fill the Sound slot. The model generates audio either way, so naming ambience, effects and whether there's music is free direction.

Kling 4.0 Prompt Examples You Can Adapt

These are our examples, built on Kling's formula and beat structure. Adapt the content, keep the shape.

A 15-second product reveal for social:

0 to 5s: a matte black wireless speaker on a concrete plinth, dark studio, a thin line of light sweeps across it. 5 to 11s: slow orbit right, the speaker's ring lights pulse to a deep bass note, fine dust in the beam. 11 to 15s: camera settles front-on, the speaker centered, hold. Vertical 9:16. Pacing: slow and confident. Sound: a single bass drop at 5s, low ambient hum, silence on the final second.

A 30-second brand moment in ultrawide:

21:9. 0 to 8s: dawn over a fishing harbor, wide locked-off shot, boats rocking gently. 8 to 18s: tracking shot along the pier following a young fisher carrying a crate, gulls overhead. 18 to 26s: close-up as she opens the crate and smiles at the catch. 26 to 30s: pull back to a wide as the sun clears the horizon. Pacing: patient, building to the sunrise. Sound: water lapping, gulls, wooden planks creaking, soft strings entering at 18s.

A UGC-style testimonial:

Handheld selfie framing, 9:16, 12 seconds. A woman in her twenties sits in her car in a parking lot, daylight through the windshield. She holds up a small tube of sunscreen and says, relaxed and a little amused, "I've reapplied this three times today and my face still isn't greasy." She rubs a bit onto the back of her hand and shows it to camera. Sound: muffled traffic, the click of the cap.

When Kling 4.0 lands on each::labs, these run through the same API you use for every other text-to-video model. If you chain steps (a product still, then video, then upscale), build that chain now as an each::labs flow and swap in the new model on launch day.

Frequently Asked Questions

What's the best prompt structure for Kling 4.0?

Kling's own formula: Subject + Action + Setting + Camera + Lighting + Pacing + Sound. For clips longer than a few seconds, add timestamps and write one segment per story beat, in order. Pacing and Sound are the slots most people forget, and on long clips they matter most.

How should I use keyframes in Kling 4.0 prompts?

Pin only the moments where the visual state really changes, and let the prompt describe how the video moves between them. Kling supports up to 10 keyframes but its example 30-second product video uses five. Never let the prompt contradict the keyframe order.

Can I use Kling 4.0 prompt examples on Kling 3.0?

Mostly, yes. The formula, camera language and dialogue direction carry over. Shorten the timeline to 15 seconds or split it with multi-shot on Kling 3.0 Pro, and drop references beyond what Kling 3.0 accepts. It's a good way to rehearse before launch.

When can I run Kling 4.0 on each::labs?

Kling says the full model launches in October 2026, with no exact day published and no API announced yet. It's coming to each::labs, and you'll be able to run it through the same API as the Kling 3.0 models already live there.