ManyMotions
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
MCPPricingFAQ
Log inStart for $1
AI Models/Kling/Prompting

How to prompt Kling AI

Kling is a family, and good prompting starts with casting: Kling 3.0 for cinematic takes, 2.6 Pro for dependable image-to-video with native audio, Motion Control to transfer movement from a reference video onto your character, O3 Pro for long takes, AI Avatar to make a portrait speak. What every member shares is a director's contract: open with the framing, state the subject's movement explicitly — Kling animates what you describe and only what you describe — and phrase camera work relative to the subject, "the camera follows her", not "the camera moves right".

Anatomy of a Kling prompt

  1. 1

    Framing first

    Open with the shot type — "profile shot", "macro close-up", "POV". Kling reads the first clause as the lens choice; burying it mid-prompt weakens it.

  2. 2

    Explicit subject motion

    Say exactly what the subject does, with verbs. An undirected subject barely moves — Kling never assumes motion you didn't write.

  3. 3

    Camera relative to subject

    Anchor the camera to the action: follows, orbits, pulls back from, leads. Subject-relative phrasing keeps the move coherent when the subject travels.

  4. 4

    Temporal progression

    How the scene evolves from first second to last — "the light brightens as she reaches the counter". Progression words are what separate a clip from a moving photo.

  5. 5

    Physical detail

    Impacts, materials, reactions: "dust plume erupting", "suspension visibly working". Kling's smooth motion looks best when it has physics to render.

  6. 6

    Audio (on audio models)

    On 2.6 Pro, 3.0 and O3 Pro, quoted lines are spoken and ambience is generated. English and Chinese are native; other languages get auto-translated.

Template

[shot type], [subject] [explicit movement], camera [movement relative to subject], [how the scene evolves], [materials and physical reactions], [audio if enabled]

10 example prompts that work

Product ad push-and-orbit (image-to-video)

i2v · Kling 2.6 Pro · 16:9 · 5s
Slow dolly push toward the serum bottle, then the camera gently orbits fifteen degrees right, revealing the texture of the frosted glass and the embossed logo, studio light gradually brightening, shallow depth of field keeping the label crisp.

The proven ad move on Kling: push in, orbit slightly, let the light do a slow reveal. The prompt describes motion FROM the image, never what the image contains — the start frame is the anchor for identity, layout and label text.

Car shot with equipment language

t2v · Kling 3.0 Pro · 16:9 · 5s · audio on
Suction cup car mount on the passenger door of a rally car, camera locked to the body as it drifts around a gravel corner, dust plume erupting behind, suspension visibly working, background whipping past in motion blur while the car stays razor sharp.

Kling understands real film equipment — naming the rig produces the physics of that rig: car crisp, world blurred. "Suspension visibly working" is the physical-detail layer that sells speed.

Motion Control: choreography transfer

Kling 2.6 Motion Control Pro · 9:16 · 10s · Character From: image
The woman from the character image performs the dance from the reference video on a rooftop at sunset, full body in frame, city skyline behind her, warm backlight with lens flare, her dress flowing with each spin.

Motion Control reads movement from your reference video and identity from your character image — the prompt only sets who and where. Clean full-body reference footage with a stable camera transfers best; the model recreates the choreography, not the reference's background.

Rack focus attention shift

t2v · Kling 3.0 Standard · 16:9 · 5s
Medium shot over a wooden cafe table: a steaming espresso cup in sharp focus in the foreground, a woman reading blurred behind it. Rack focus from the cup to her face as she looks up and smiles at someone off-screen, morning window light, dust motes drifting.

Name both focus planes and the trigger ("as she looks up") — Kling times the focus pull to the action instead of drifting it randomly. The cheapest way to add narrative to a static scene.

Pull back to reveal

t2v · Kling 2.6 Pro · 16:9 · 10s · audio on
Macro close-up of hands kneading dough, flour dust in the air. The camera slowly pulls back and cranes up to reveal a full bakery kitchen at dawn — three bakers working, trays of croissants, warm tungsten light — in one continuous move.

A reveal only works if the destination is described: give the wide shot its content ("three bakers, trays of croissants") or the model invents an empty room. "One continuous move" guards against an unwanted cut.

Dialogue with native audio

t2v · Kling 3.0 Pro · 16:9 · 10s · audio on
Two-shot at a kitchen counter: a chef in a denim apron holds up a jar of chili oil and says: "Three ingredients. That's the whole secret." His friend leans in, takes the jar and answers: "You're lying." Natural handheld feel, warm evening light. Ambient: sizzling pan, low music.

Attribute each quoted line to a distinct, described speaker so the voices land on the right faces. Write dialogue in English (or Chinese) — those are native; other languages get auto-translated.

Subtle parallax cinemagraph (image-to-video)

i2v · Kling 2.6 Pro · 1:1 · 5s
Subtle parallax: the camera drifts a few centimeters right, foreground plant leaves shifting slightly against the window, steam rising from the mug, rain streaking down the glass outside. Everything else stays still.

Minimal motion verbs plus one drifting camera gives a living-photo effect with full fidelity to the source. Kling's motion smoothness shows at low intensity — no jitter or warping on the near-static regions.

Crash zoom hook

t2v · Kling 3.0 Pro · 9:16 · 5s · audio on
Static wide of a skater alone on a rooftop, then a crash zoom into a tight close-up of his grin as he drops in, Dutch angle, harsh midday sun, the board's wheels hitting concrete with a visible jolt.

Crash zoom plus Dutch angle is a first-second thumb-stopper, and Kling executes both as named techniques. "A visible jolt" gives the cut-in a physical beat to land on.

O3 Pro long take

t2v · Kling O3 Pro · 16:9 · 15s · audio on
One unbroken steadicam shot: the camera follows a florist from the back room of her shop, through the doorway, weaving between flower buckets to the front counter, where she hands a wrapped bouquet to a customer, morning light growing brighter toward the storefront.

O3 Pro goes up to 15 seconds, which fits a genuine oner. Subject-relative phrasing ("the camera follows") holds the follow steady through the whole path; the light progression marks time across the take.

End-frame product transition

i2v + end frame · Kling 3.0 Pro · 1:1 · 5s
The perfume bottle rises slowly out of black water, droplets sliding off the glass, rim light igniting the amber liquid as it settles into a centered, label-forward hero pose, the water surface calming to a mirror.

Upload your finished hero shot as the end frame — the clip is guaranteed to land on the packshot, ready to hard-cut into a logo card. Kling interpolates a smooth path from the start image to it.

Settings that matter

  • Model

    Cast before writing: 3.0 Pro for finals, 3.0 Standard and 2.6 Pro for iteration and i2v ads, Motion Control for movement transfer, O3 Pro for 15s takes, AI Avatar v2 for talking portraits.

  • Duration

    5s or 10s on 2.6 and Motion Control; 5, 10 or 15s on 3.0 and O3 Pro. One clear action beat per 5 seconds.

  • Aspect ratio

    16:9, 9:16 or 1:1 across the family. Match your framing language to it — "full body in frame" wants 9:16.

  • CFG Scale

    Leave at the 0.5 default — it is tuned. Nudge up one step only when a key instruction keeps getting ignored, down for looser interpretation.

  • Generate Audio

    Native on 2.6 Pro, 3.0 and O3 Pro. Leave on and direct it; turn off when you'll score the clip yourself or for silent loops.

  • Character From (Motion Control)

    Keeps orientation and framing anchored to your character image ("image") or to the reference clip ("video"). Start with "image" when your character photo is well framed.

  • End frame

    Optional final-frame image on 2.6 Pro, 3.0 and O3 Pro. Use it to guarantee packshots and build invisible transitions between clips.

Do and don't

Do

  • Open every prompt with the framing — shot type is the first thing Kling reads.
  • Describe subject movement explicitly with verbs; an undirected subject barely moves.
  • Phrase camera work relative to the subject: "the camera follows her" beats "the camera moves right".
  • Name real equipment when you want its look — car mounts, steadicam, shoulder cam.
  • Give characters stable labels ("the chef", "the skater") and reuse them; pronouns drift.
  • Pick the family member before writing — a Motion Control prompt and a 3.0 prompt do different jobs.

Don't

  • Don't write catalog descriptions — a list of visual attributes with no motion produces a near-still.
  • Don't describe the uploaded image's content in i2v — describe only what moves and where the camera goes.
  • Don't expect native dialogue outside English and Chinese; other languages are auto-translated.
  • Don't stack more action than the clip holds — one clear beat per 5 seconds.
  • Don't feed Motion Control messy reference footage — shaky, cropped or multi-person clips transfer badly.

Advanced techniques

Motion Control: what transfers and what doesn't

Motion Control splits the job three ways: the reference video contributes movement, choreography and timing; the character image contributes identity and wardrobe; the prompt contributes scene, lighting and mood. Nothing else crosses over — the reference's background, outfit and camera are discarded.

  • Best references: single performer, full body in frame, stable camera, clean background
  • The clip's energy is the reference's energy — pick footage with the exact tempo you want
  • Character From = "image" keeps your character's framing; "video" lets the reference's framing win
  • Pro for finals, Standard for testing which reference clip transfers best

Making a portrait speak with AI Avatar v2

AI Avatar v2 is the no-prompt member of the family: upload a portrait and an audio file in Lip-Sync Studio and it animates the face to speak the audio, blinks and micro-expressions included. Pick frontal portraits with the mouth clearly visible and neutral expression — the animation has room to move. Clip length follows the audio, so script first, record or generate the voiceover, then animate.

Character continuity across clips

Separate generations don't share identity, but the family gives you three levers. Reuse the same start image for every clip of a character — i2v anchors identity harder than any description. Chain scenes by making one clip's final frame the next clip's start frame (the end-frame upload makes that frame exact). And keep the character's label and wardrobe wording identical in every prompt, so what the image anchors the text doesn't contradict.

Which variant to use

  • Kling 3.0 Pro

    Cinematic finals with native audio, up to 15s, multi-shot capable — see the dedicated Kling 3.0 guide.

  • Kling 3.0 Standard

    Same grammar and durations as 3.0 Pro at a lower render cost — drafts and volume work.

  • Kling 2.6 Pro

    The dependable i2v workhorse with native audio — product ads and cinemagraphs at 5-10s.

  • Kling 2.6 Motion Control Pro

    Transfer movement from a reference video onto your character image — dance, sports, gesture work.

  • Kling 2.6 Motion Control Standard

    Same transfer at draft quality — test which reference clip works before the Pro pass.

  • Kling O3 Pro

    Omni model for long takes — up to 15 seconds with audio, strongest on continuous single-shot scenes.

  • Kling AI Avatar v2

    Portrait + audio file → talking head, driven from Lip-Sync Studio. No prompt writing needed.

Frequently asked

Which Kling model should I pick?

By task: Kling 3.0 Pro for cinematic finals and multi-shot, 3.0 Standard or 2.6 Pro for iteration and image-to-video ads, Motion Control to copy movement from a real video, O3 Pro for 15-second continuous takes, AI Avatar v2 to make a portrait speak.

Why is my subject barely moving?

Because the prompt didn't tell it to move. Kling renders described motion only — replace attribute lists with verbs: "she turns, lifts the cup, looks out the window" instead of "a woman with a cup by the window".

Does Kling generate audio and dialogue?

Yes on 2.6 Pro, 3.0 and O3 Pro with Generate Audio on: quoted lines are spoken and lip-synced, ambience is generated from your description. English and Chinese are native; other languages get auto-translated, so write ad dialogue in English.

How does Motion Control actually work?

You provide two inputs: a character image (who) and a reference video (how they move). The model re-performs the reference's movement with your character in the scene your prompt describes. The reference's background and outfit are ignored.

What does CFG Scale do?

It sets how strictly the model follows your prompt, from 0 to 1. The 0.5 default is tuned for Kling; raise it slightly only if a key instruction keeps being ignored, lower it for more creative interpretation.

Can I keep the same character across multiple clips?

Reuse the same start image for every clip — image-to-video anchors identity far better than text. For sequences, chain clips by making one clip's last frame the next one's start frame, and keep wardrobe wording identical in every prompt.

Try Kling in Media Forge →About KlingAll models
ManyMotions

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 NoticeCancel subscriptionWithdraw from contract

© 2026 ManyMotions

support@manymotions.com