Clipwave
Media ForgeGenerate images & video, fastMedia StudioNewBuild images & video by chatting with AIWorkflowsAutomate pipelines at scaleClipperNewBreak a video down & cut the highlightsMarketing AgentDone-for-you UGC video ads
Explore NewMCPPricingFAQ
Log inStart free
AI Models/Kling 3.0/Prompting

How to prompt Kling 3.0

Kling 3.0 prompts read like a page of a screenplay, because the model can execute one: a single generation holds up to 15 seconds and can cut between shots you number in the prompt, with native audio underneath and fidelity built for 4K-at-60fps-class delivery. Structure beats vocabulary here. Decide the shot plan first, open each shot with its framing, give the arc explicit per-shot durations, and keep character labels identical across shots so identity survives the cuts. Artifacts go in the separate negative-prompt field, not the prompt.

Anatomy of a Kling 3.0 prompt

  1. 1

    Shot plan

    One take or a sequence? For sequences, number the shots with rough durations that sum to the clip length — "Shot 1 (5s): … Shot 2 (5s): …". For a long take, say "no cuts" explicitly.

  2. 2

    Framing per shot

    Each numbered shot opens with its own shot type — wide, macro, over-the-shoulder, Dutch angle. The first clause of a shot is its lens.

  3. 3

    Action & progression

    What changes inside each shot, with verbs, and how shots hand off — end a shot on the movement the next one picks up and the cut feels edited, not stitched.

  4. 4

    Character labels

    Fixed labels reused verbatim ("the violinist", "the candidate") plus one wardrobe anchor repeated per shot. Identity drift across cuts is the multi-shot failure mode.

  5. 5

    Sound & dialogue

    Declare continuous ambience once, then place point sounds and quoted lines per shot. One line of dialogue per shot lands cleanly; English and Chinese are native.

  6. 6

    Negative prompt (separate field)

    Recurring artifacts — blur, warped hands, distortion — belong in the dedicated field, short and targeted. Keep the main prompt purely positive.

Template

Shot 1 (Xs): [framing], [subject] [action], camera [move]. Shot 2 (Xs): [framing], [continuation]. Shot 3 (Xs): [payoff]. Continuous audio: [ambience]; [dialogue in quotes where it lands]

10 example prompts that work

Three-shot product film

t2v · 16:9 · 15s · audio on · Pro
Shot 1 (5s): macro glide across a brushed-titanium watch face, light sweeping over the crown. Shot 2 (5s): the camera pulls back to reveal the watch on a rotating obsidian pedestal, single spotlight. Shot 3 (5s): lifestyle — a man in a charcoal suit checks the watch stepping out of a car at night, neon reflections on wet asphalt. Continuous audio: deep ambient pulse; a soft tick in shot 1, city rain in shot 3.

Macro, reveal, lifestyle — the three-beat ad structure in one generation, no editor. Equal 5-second beats keep the pacing even, and audio assigned per shot makes the cuts feel deliberate.

15-second mini narrative

t2v · 16:9 · 15s · audio on · Pro
Shot 1 (5s): wide shot, a young violinist stands alone on an empty subway platform under fluorescent light, she raises her bow. Shot 2 (5s): close-up on the strings and her fingers as she plays, camera slowly circling her. Shot 3 (5s): the platform now crowded, commuters stopped and listening, she finishes and lowers the bow with a small smile. The violinist wears the same green coat in every shot. Continuous audio: her violin melody building through all three shots, reverberant station acoustics.

The wardrobe anchor ("the same green coat") is repeated because identity is what breaks across cuts. The melody declared as continuous stitches three shots into one story instead of three clips.

Single-shot long take

t2v · 16:9 · 10s · audio on · Standard
One unbroken shot, no cuts: slow dolly through a rain-soaked greenhouse at night, plants brushing past the lens edges, a single hanging bulb swinging gently ahead, moths circling the light, the camera settling on a workbench with an open notebook. Audio: rain on glass, the slow creak of the swinging bulb.

On a model trained to cut, an undeclared long take may get cut on its own — "one unbroken shot, no cuts" locks it. Foreground contact ("plants brushing the lens") shows off the fidelity headroom.

Dialogue payoff sequence

t2v · 9:16 · 10s · audio on · Pro
Shot 1 (5s): medium shot, a job candidate in a red blazer waits in a hallway and rehearses under her breath: "You've got this." Shot 2 (5s): inside the office, shot-reverse-shot — the interviewer closes the folder, pauses, and says: "When can you start?" Her face lights up. Same candidate in both shots: red blazer, hair in a bun. Ambient: quiet office hum, a clock ticking in shot 1.

One line per shot is the rhythm dialogue holds best. "Shot-reverse-shot" is understood as staging, and the described pause before the interviewer's line becomes real comic-dramatic timing.

Image-to-video from a packshot

i2v · 1:1 · 5s · Pro
The bottle stays center frame while the background transforms from studio white into a slow-motion splash of citrus slices and water arcs behind it, droplets catching the light, a gentle dolly push through the last two seconds.

The i2v rule: the image holds the product, the prompt directs only what changes. "Stays center frame" protects label fidelity while everything around it earns the motion.

End-frame guaranteed packshot

i2v + end frame · 1:1 · 5s · Pro
Golden liquid pours in slow motion into an empty glass flacon suspended in darkness, filling it exactly, the pour thinning to a final drop as the bottle settles into a centered, label-forward hero pose, rim light flaring briefly.

Upload the finished hero shot as the end frame and the clip must land on it — a guaranteed packshot you can hard-cut into a logo card. The prompt narrates the pour; the frame owns the destination.

Negative-prompt cleanup on hard subjects

t2v · 1:1 · 5s · Pro · negative: "blur, distortion, warped hands"
Macro close-up of hands pouring sparkling water over a pyramid of ice spheres in a crystal glass, bubbles streaming upward, razor-sharp refraction, studio black background, one hard backlight.

Glass, liquid and hands are where artifacts concentrate. Keep the main prompt positive and put "blur, distortion, warped hands, extra fingers" in the Negative Prompt field — short, targeted lists beat essays.

Sports POV with impact physics

t2v · 16:9 · 5s · audio on · Pro
POV from a mountain biker's chest mount bombing a forest trail, handlebars visible in frame, roots and rocks rushing under the front wheel, lens shudder on each impact, strobes of sunlight through the canopy, a low branch ducked at the last second.

Equipment framing ("chest mount, handlebars in frame") locks the perspective, and "lens shudder on each impact" binds camera motion to physics. High-frame-rate-class motion keeps the shake legible instead of smeared.

Vertical fashion editorial, uneven beats

t2v · 9:16 · 10s · audio on · Pro
Shot 1 (3s): low-angle full body, a model in a flowing crimson dress walks toward camera down a brutalist concrete corridor. Shot 2 (3s): tight profile close-up, wind lifting the fabric as she turns her head into the light. Shot 3 (4s): overhead top-down as she spins once, the dress blooming into a circle on the gray floor. Same model, same crimson dress throughout. Audio: echoing heels, fabric rustle, a low fashion-film drone.

Shot durations don't have to be equal — 3/3/4 builds an accelerating rhythm with the payoff on the longest beat. The overhead spin is the money shot, so it gets the extra second.

Architecture fly-through

t2v · 16:9 · 10s · Standard
Aerial establishing shot gliding over a cliffside modern villa at golden hour, then the camera descends and flies through the open glass wall into the living room in one continuous move — curtains breathing in the wind, ocean glittering beyond the infinity pool. No cuts.

For a fly-through, describe the interior before the camera arrives so the model builds the destination, not just the approach. "One continuous move" plus "no cuts" keeps the transition from exterior to interior seamless.

Settings that matter

  • Model tier

    Draft on Kling 3.0 Standard, finish on Pro — identical grammar and durations, so the winning prompt carries over unchanged.

  • Duration

    5, 10 or 15s. Budget roughly one shot per 5 seconds: 15s fits a three-beat sequence, 10s a two-beat or a long take, 5s a single statement.

  • Aspect ratio

    16:9, 9:16 or 1:1. Multi-shot works in all three — just restate framing per shot in the ratio's terms ("full body" reads differently in 9:16).

  • CFG Scale

    Default 0.5 is tuned. Move it one step at a time and only with a reason: up when a named instruction keeps being dropped, down when outputs feel stiff.

  • Generate Audio

    Leave on — dialogue, foley and ambience are generated with the picture. Write dialogue in English (or Chinese) for native delivery.

  • Negative Prompt

    Empty by default on Clipwave. Add short artifact lists ("blur, warped hands") for glass, liquid, hands and text-heavy scenes; keep it under a handful of terms.

  • End frame

    Optional final-frame upload alongside the start image. Use it for guaranteed packshots and exact scene-to-scene handoffs.

Do and don't

Do

  • Decide the shot plan before writing — one beat per shot, durations that sum to the clip length.
  • Open every shot with its framing; the first clause of each shot is its lens.
  • Repeat identity anchors verbatim across shots — "the same green coat" is cheap insurance.
  • Declare "no cuts" explicitly when you want one continuous take.
  • Keep dialogue to one line per shot, written in English or Chinese for native delivery.
  • Draft on Standard, finish on Pro with the identical prompt.

Don't

  • Don't cram shots — below roughly three seconds each, cuts start eating the action.
  • Don't use pronouns across shot boundaries; "she" in shot 3 may not be the woman from shot 1.
  • Don't write negatives into the main prompt ("no blur") — the Negative Prompt field exists and works.
  • Don't re-describe the uploaded image in i2v — direct the change, not the contents.
  • Don't default to 15s — pacing is tighter at 10s unless the story genuinely earns three beats.

Advanced techniques

Engineering the cut

Multi-shot output is only as good as the continuity you write. The model cuts where you number, but whether the result feels edited depends on what crosses the cut: keep location and light coherent (or state the change), restate the character with the same label and wardrobe anchor, and declare ambience once as continuous so the soundtrack bridges the shots.

The strongest cuts are motivated by motion: end a shot on an action the next shot picks up — she raises the bow / close-up as she plays. Handoffs like that read as editing choices rather than model behavior.

Prompting for the fidelity headroom

Kling 3.0's output is built for large screens and fast motion, and that headroom is only used if the prompt feeds it detail worth rendering. Micro-texture words — fabric weave, condensation, brushed metal, dust motes — pay off visibly here, where lesser models would blur them away. The same goes for motion: fast action stays legible, so you can ask for impacts, whip-fast turns and handheld shudder that other models would smear.

CFG and the negative field as a tuning loop

Iterate one dial at a time, on Standard. Start from CFG 0.5 and an empty negative field. If a named instruction is dropped across two runs, raise CFG one step. If an artifact recurs, add exactly one term to the negative field and rerun. When the take is right, re-render on Pro with the same settings — changing multiple dials per run tells you nothing about which one worked.

Which variant to use

  • Kling 3.0 Pro

    Finals — maximum fidelity, native audio, the full 5/10/15s range. The version clients see.

  • Kling 3.0 Standard

    Iteration — same prompt grammar, durations and controls at a lower render cost. Find the take here, finish on Pro.

Frequently asked

How many shots fit in one generation?

Budget one shot per five seconds: three beats in a 15-second clip is the sweet spot. The model can execute more, but each shot below about three seconds loses its action to the cut.

Kling 3.0 Pro or Standard?

Same grammar, same durations, different fidelity. Iterate on Standard until the shot plan works, then re-render the identical prompt on Pro for the final. Never prompt-engineer on Pro.

What languages does dialogue support?

English and Chinese are native; lines in other languages get auto-translated into English before delivery. For ads, write the dialogue in English and keep it to one line per shot.

Does Kling 3.0 support negative prompts?

Yes — a dedicated Negative Prompt field, empty by default on Clipwave. Use short artifact lists ("blur, distortion, warped hands") rather than long essays, and only for problems you actually see.

How is Kling 3.0 different from Kling 2.6?

Three things: 15-second takes (2.6 stops at 10s), multi-shot cutting inside one generation, and the negative-prompt field — on top of higher, 4K/60fps-class output fidelity. 2.6 Pro remains the quick i2v workhorse, and motion transfer stays on the 2.6 Motion Control models.

Can I control the exact last frame?

Yes — upload an end frame next to your start image and the clip is guaranteed to land on it. Use it for packshots, state changes, and chaining generations into longer sequences.

Try Kling 3.0 in Media Forge →About Kling 3.0All models
Clipwave

Tell me what you need and grab a coffee. I'll have it ready before you finish the cup.

Product

Marketing AgentWorkflow StudioMedia ForgeTemplatesAI Models

Open in app

Workflow StudioMedia ForgeProjectsAvatars

Company

ContactPrivacy PolicyTerms of ServiceLegal Notice

© 2026 Clipwave

support@clipwave.io