Chapter 16 · Kling: 3.0, and 4.0 as announced

Yasser Ashour

2 hours ago

37 min read

Part 3 · Motion

Kling is the motion camera you reach for when the move is the shot: a dolly, a track, a crane, a push on a face or a label, or one small action from a frame you have already signed. This chapter teaches Kling 3.0 as it runs today through the Higgsfield connector: what it reads, how to write to it, the settings, the checks and the limits. It ends with six prompts at the standard. Kling 4.0, announced on 28 Sep 2026 for a full release in October, has only its Flash member out, in early access for a limited group, and nothing of it on the connector; it is recorded beside 3.0 without changing a single 3.0 fact.

In this chapter

What Kling 3.0 is today, its two siblings, and Kling 4.0 as announced
What the connector lets you give it: a start frame, an end frame, a prompt and a sound switch
The prompt in Kling's order: the camera first when the move is the shot, one measured magnitude for each moving system, an endpoint, protection by name
Six prompts at the standard: the Mastorna crane in corrected form, a beauty push-in, a pour, a car reveal, a scene with a person, and a talking plate
Settings, checks, failures and fixes, limits, prices and rights, and what to re-check the day 4.0 ships

Before you start. Chapter 13 chooses the route. Chapter 14 teaches the seven elements of a moving shot and how to size and measure a move; this chapter is how to say them to Kling. Chapter 22 teaches the speaking shot, and Chapter 24 the lip-sync pass that the talking plate in 16.5 is made for.

16.1 What it is

Kling is the video model family from Kuaishou's Kling AI. The generation in use is Kling VIDEO 3.0 (the maker's guide is dated 6 Feb 2026), with a native 4K mode added on 23 Apr 2026, and a faster Turbo tier and stronger video editing in the Omni variant added on 17 Jun 2026 (all checked 30 Sep 2026 on Kling's own release notes and guides). Kling 4.0 was announced on 28 Sep 2026 and is covered at the end of this section. Facts about the Higgsfield connector below are its contract as last read on 28 Sep 2026; read it again on the day you spend.

What it is for. Kling takes one signed still and moves it. It is the first choice when the camera is the shot (a dolly, a track, a crane, a small push whose size matters), for a product or hand insert, for lettering that must survive a gentle move, for a quiet performance with no words, and for the wordless plate that the exact-take speaking route needs. Its start frame is its default operation: one approved keyframe, one action, one clip.

Reach for another tool when:

The shot needs	Use	Where
More than one liquid or cloth system, a crowd near the camera, a chain of cause and effect	Seedance 2.5	Chapter 17
A recurring face or product with no fixed frame, or a reference video that lends a camera or a performance	Seedance 2.5 references	Chapter 17
Footage extended or edited by a supplied clip	Seedance 2.5; for one scoped change in an accepted clip, Kling's Omni Edit (16.3)	Chapters 17, 16
A very cheap candidate	Wan 3.0, after a test	Chapter 18
A mouth fitted to an exact recorded take	a Kling plate, then Sync Lipsync 3	16.5, Chapter 24
Egyptian or Saudi dialogue	an actor or ElevenLabs, never Kling	Chapters 21, 23
The keyframe itself	a still tool	Chapters 11, 12

How it is reached. On the Higgsfield connector, three ids: kling3_0 (VIDEO 3.0, the working model), kling3_0_turbo (Turbo) and kling_video_edit (Omni Edit). Name the model in every request: the connector's own tool text steers some requests to other models. The connector exposes a start frame, an end frame, a prompt and a few settings, and nothing else (16.2, 16.6). Kling's own site and app, and its developer API, offer more: Elements, voice binding, storyboarded multi-shot, camera controls. Those are marked "on Kling's own site" wherever they appear, because a shot planned on them cannot be run from the connector. The connector also lists an older Kling generation; this book does not use it.

The two siblings. Turbo is a separate model, not a mode of VIDEO 3.0. Kling introduced it on 17 Jun 2026 as the efficient option with quick generation, and on the connector it takes a start frame only, has a resolution setting in place of modes, and has no sound switch (16.6). Omni Edit changes one thing inside a clip you have already accepted: it takes a source clip and, optionally, picture references. Its prompt and limits are in 16.3 and 16.9.

Kling 4.0, announced

Kling's release note of 28 Sep 2026 announces Kling 4.0 and says it will launch officially in October. The same note says the first member of the family, Kling 4.0 Flash, became available on 28 September to a limited group of users for early access; Kling's post of the same day on X says Flash is live for Ultra Yearly subscribers. Kling's own pages give no price. As of 30 Sep 2026 the connector's model list, last read on 28 Sep, has no 4.0, and no Higgsfield page found that day announces it. Kling 4.0 has not launched: the full release is due in October, only Flash is out, for a limited group, and neither is on the connector. Everything in this book about running Kling is Kling 3.0.

What Kling states, in its own release note (official, read 30 Sep 2026):

Kling 4.0	Kling 4.0 Flash
Operations	text to video; image to video; first and last frames; multi-keyframes; Omni Reference; video editing	text to video; image to video (first frame); Omni Reference
Clip length	3 to 30 seconds	3 to 20 seconds
Resolution	720p, 1080p, 4K	720p
Dynamic range	10-bit HDR at 1080p and 4K, marked as coming soon	8-bit SDR
Aspect ratios	21:9, 16:9, 1:1, 9:16, auto	not listed
Keyframes	up to 10 images	not listed
References	up to 15 items in all: up to 10 images, up to 5 videos (30 seconds combined), up to 7 subjects (3 of them from video), voice references for audio	Omni Reference, limits not listed
Prompt	up to 8,000 tokens	not listed
Audio	two-channel stereo; dialogue in Chinese, English, Japanese, Korean, Spanish, Portuguese, German, French and more, with regional accents	not listed
Editing and extension	editing of up to 5 clips (3 to 30 seconds each, 30 seconds combined); extension up to 2 minutes, marked as coming soon	not listed
Price	not published	not published

Arabic is on neither list, and Kling's note names no Arabic dialect. Its note says 4.0 generates legible text in the languages it lists, again without Arabic.

Third-party reading. Bloomberg and Briefs Finance reported the announcement on 28 Sep 2026: clips doubled to 30 seconds, more than a dozen input references, up to 10 keyframes, a lighter version already open to annual subscribers, the full model due in October. That agrees with Kling's note and adds no spec. Pages that publish other figures (a 1080p ceiling, 60-second extension steps, 120-second clips) cite no Kling source, and are not Kling's.

How to hold it. The table is what Kling says it will offer, in Kling's words on Kling's date. How 4.0 reads a prompt, whether the connector will expose its keyframes and references, what it costs, and whether 3.0's habits carry over are not known. Nothing in 16.3 to 16.9 is taught for 4.0. What 4.0 would change for a director, and what to re-check the day it is listed, is in 16.10.

16.2 How it reads what you give it

The connector gives Kling four inputs and three settings (last read 28 Sep 2026). Each input owns one thing, and nothing is labelled, ordered or bound: there is one picture per slot.

Input	What it owns	What it must not carry
Start frame	Everything that exists: the faces, the product, the room, the light, the lettering and the instant the clip begins in. It becomes frame 0	The finished action; a wrong wordmark or a stray hand, which the clip will keep
End frame	The landing. It costs nothing extra	Any difference from the start frame beyond the change
Prompt	What changes: the move, the action, the pace, the endpoint and what is protected	The frame's contents again; Arabic dialogue; lens numbers
Sound switch	Kling's own soundtrack. Off returns a file with no audio stream; on returns AAC stereo at 44.1 kHz	Anything you can direct beyond what the prompt says (16.3)

The three settings, mode, duration and aspect, are in 16.6.

The start frame is a strong reference, not a pixel lock. On the one standard clip compared (26 Sep 2026), frame 0 was the keyframe scaled to the output size, with nothing re-framed and no prop re-rendered. Kling scales the still to its own raster: a 2752 × 1536 still came back at 1284 × 716, so small print and thin lines soften, and the still is made for that size (Chapter 9). What is wrong in the still is in the clip. Kling's own guide says the same about the prompt: with a frame supplied, the scene is given, and the words need only the subjects and their intended movement.

Because nothing is bound, the start frame is the whole identity. Three consequences follow.

Every shot needs its own approved keyframe. Continuity on Kling is a chain of them, each made from the masters and the last frame you accepted (Chapter 10, 14.14).
A beat between two people needs one relationship keyframe. For a handover, or a look that passes from one face to another, put both faces in the one start frame, at the size they will play, with the object in its state before contact and the eyelines set (Chapter 9 prints one, Example 9.2). Kling cannot be given a second face separately; a second person who appears in a clip from a one-face frame is invented.
Name each figure by its role in the frame, "the dealer", "the man at the desk", never by a character name. There is nothing for a name to bind to.

An end frame turns the clip into a journey between two pictures. Make the two frames relatives: the same subject, scale family, light logic and aspect, ideally one made from the other by a one-change edit (Chapter 10). Kling's guide to start and end frames advises the same, because large differences may make the model switch shots. The prompt then describes the travel, not the two pictures (14.8). On the one pair checked (26 Sep 2026) the last frames matched the end frame, and mid-path a box rotated about a third of a turn before it straightened to land: judge the middle, not only the ends (16.7). A landing on an end frame with no start frame is described by Kling's and Higgsfield's guides; whether the connector accepts it is not yet known, and a price check with only an end frame attached costs nothing. The same free check answers whether kling3_0 accepts a prompt with no frame at all (Chapter 13).

Sound is an input too, and it is mostly closed. The connector takes no audio input, so no cast voice can be given to Kling. From text it speaks Chinese, English, Japanese, Korean and Spanish; anything else it translates into English (Kling, checked 30 Sep 2026). Never Arabic dialogue: Egyptian and Saudi speech comes from an actor or ElevenLabs (Chapters 21 and 23). A clip whose sound the edit will build is made with the switch off.

What the connector does not take: references, Elements, voice binding, storyboard or multi-shot fields, a negative prompt, a seed, a frame rate, a resolution field, a language field and camera values. Each job also stores guidance 0.5, prompt enhancement off and multi-shot off, none of which you can set. A shot planned on any of these cannot be run from the connector.

On Kling's own site the picture is wider. These are marked "on Kling's own site" wherever they appear in this book:

Feature	What Kling says (checked 30 Sep 2026)	On the connector
Elements	A character or object held from two to four pictures or a short video, usable beside a start frame	Not exposed
Voice binding	A voice built from five to thirty seconds of one speaker (Omni)	Not exposed
Custom multi-shot	Each shot its own description and duration inside three to fifteen seconds	Not exposed; stored off
Motion Control 3.0	A reference video drives a character (released 4 Mar 2026)	Higgsfield's wrapper names it as its engine; untested and in no route
Camera controls	Six basic movements and four master shots, with the extent set by "displacement parameters" (guide listed 24 Nov 2025)	Not exposed; the extent can be written only as a magnitude in the prompt

16.3 The prompt, in this tool's dialect

Chapter 14 teaches the seven elements of a moving shot and how to size and measure a move. This section is how to say them to Kling. A finished-grade Kling prompt is one flowing paragraph of prose, as a drafting expectation 120 to 220 words (the maker gives no count), and never labelled fields: shot labels, cast keys and a negative field belong to Kling's multi-shot surface, which the connector does not expose.

The order Kling wants

When the camera move is the shot, open on the camera: the verb, the rig with its weight, the destination, one magnitude and a speed profile that ends. When the shot is an action or a performance, open on who does what and declare the camera in the next sentence: "The camera is locked; everything that moves, moves inside the frame." Kling's own formulas open on the subject (its image-to-video guide of 24 Nov 2025 and its motion-prompt article of 3 Jul 2026), while the long showpiece in its 3.0 guide opens on the camera. No Kling page says the first words weigh more, so the order is a craft rule, set by what the shot is about. The rest follows in the order of Chapter 14: the action, the world's life from systems the frame contains, protection by name after the action, the endpoint as a composition, and the closing line.

A fixed camera is legal when it is declared as design and all the motion is given to the world. Something visible moves at once. The words "from frame one" and hedge words ("barely perceptible") stay out. Opening every prompt with "Start on the first frame" was harmless and optional on the connector (26 Sep 2026), and is no substitute for a declaration.

Measured moves

Give each moving system one approximate magnitude, its duration and a speed profile that ends, with the visible result beside it: "advancing approximately fifteen centimeters over five seconds, a settling push". Write the numbers as words, as the Mastorna prompts do. Three things stay out: stacked metrics on one move (a distance, an angle and a speed together), coordinates, and contradictory spatial demands, such as a close push and a full-body view at once, which are the documented trigger for warping. Lens and camera-body numbers never belong in the prompt. Bold moves stay in words: "whip pan from the older man to the younger one".

Kling's own guide uses in-take markers ("at the 4th second") as its grammar. Whether the connector passes them is not yet known, so use them as pacing ("over the first two seconds", "by the third second"), never to schedule a fast action.

Size the number from the subject's distance (14.5). With D the distance the still implies and d the travel, the subject grows by D ÷ (D − d). The sizes finished Kling prompts use:

The shot	Subject distance	Ask	The subject grows by
Push on a face or a hairline	1 m	approximately 15 to 20 cm over 5 s	× 1.18 to × 1.25
Push on a product or a label	0.6 m	approximately 10 cm over 5 s	× 1.20
Push on a figure in a room	3 m	approximately 30 cm over 5 s	× 1.11
Push on a car or a facade	6 m	approximately 40 cm over 5 s	× 1.07
Crane, or a track past a near object	6 m or more	approximately 60 cm to 1 m	almost nothing: the value is parallax

At the foot of the table the enlargement is too small to read, and the move earns its keep by shifting near things against far ones. Whether Kling honours a stated distance, and how closely, is not yet known (16.10); 16.7 measures it clip by clip.

Camera words

Kling reads plain film terms and gives them the physics of the rig you name. Push in and pull back are its most obedient verbs. A lateral track holds best at the pace of a subject. A crane rise or descent floats unless its mechanism is named, once and first. Arcs and orbits are the highest-warp family: keep them slow and simple, or split them into two shots. "Zoom" reads as a lens effect, not a move. One dominant intention to a shot: a compound device with one purpose counts as one, but "slow push-in then pan left" is two shots. Kling's camera-control guide (read 30 Sep 2026) belongs to its app and does not say it applies to 3.0; on the connector the extent of a move is a magnitude in the prompt and nothing else. A director's name stays out of a Kling prompt, as a style or as a technique; only Seedance 2.5 motion prompts may use one, as a technique token (17.3).

What the frame owns, and what to leave out

The start frame owns what exists, so the prompt names the mover, says what moves and protects by name. It does not describe the room, the clothes or the light again. Do not name what is off frame: a named thing may be given a body. A laptop "beside the camera" rose into the frame as a lid, and a passer-by's shadow summoned the whole passer-by (26 Sep 2026). Describe the light instead: "the cool grey light comes back onto his near cheek."

Kling has no negative field, and two prose exclusions were ignored on the connector ("does not tip or turn"; "no one enters"). Say what the thing does instead: "its printed face stays square to camera for the whole slide." Five is the ceiling; work to three or fewer, each earned. With the sound switch off, close on "No captions, no subtitles, no on-screen text." With it on for a talking plate the text line alone closes the prompt, because the pass replaces the audio; Kling documents no sound half of the line. Where a sign or other lettering in the frame must survive the move, close on "No overlaid text, subtitles, logos, or interface graphics." Kuaishou states that 3.0 keeps the signage of an uploaded image consistent, including under gentle camera movement; read that as "must be checked at glyph level". Protect the lettering by name ("the painted letters keep their exact shapes, thicknesses and joins"), and put the correct label in the keyframe or composite it from the artwork. For a hand on a product, give the path, the grasp and where the hand rests, and size a small move in a unit the eye can check: "approximately a finger's width down along the hair over two seconds, and stops."

A cut inside a clip

Multi-shot is stored off on the connector, but a cut can be written as a sentence. One clip asked for a cut "at the middle of the shot" and landed near the middle (26 Sep 2026, one clip, no verdict). The second framing has no picture behind it, so write it in full, as in text to video. A cut written this way bakes the edit into the take; the default stays one shot, one clip (Chapter 13).

Performance

Kling acts what a body does. Stillness, a blink, a head turn, lifting a cup and a breath are reliable. Micro-expressions, precise eyelines, hands on small objects and turns past about sixty degrees are fragile (other directors report this). Give the performance a task and one visible event in the order it happens (Chapter 15), keep a turn small, and split anything past three-quarter into two shots cut on the look.

Speech and the talking plate

Kling can speak a line from text with its sound on, and that is a test, not a shipped voice (Chapter 21). The wordless plate for the exact-take route (Chapter 22) asks the face to talk quietly for about as long as the line: "talks quietly toward the lens for about three and a half seconds". Never give the plate the words. It would perform its own line, and outside its five languages it would turn them into English. Never tell it to keep its mouth rigidly closed: on 12 Sep 2026 a quietly talking plate passed the lip-sync pass with sound on, and a still, closed-mouth plate failed. Keep the business before the talk small, protect the mouth by name, keep the camera locked or slow, and end on a landing. Time the plate by measurement after it returns (22.4): a time written into a Kling prompt is an ask, not a control.

Omni Edit and Turbo

Omni Edit changes one thing inside a clip you have accepted. Write plain sentences about "the supplied clip" and "the image", never Seedance's "Edit Video1" grammar. There is no mask or region field, so the prompt is the only fence: "Change only [the one thing] to [its new state]", what the new thing inherits (grip, swing, the moment it stops, size, what it passes behind), that none of the original remains, and a keep-list that includes the framing and the timing. Attach an image only for a look you cannot describe, and say what to take from it and what not. Two changes in one sentence are two edits and no fence. The clip's own sound is dropped when a video goes in, so keep the original aside and lay it back in the edit.

Turbo takes the same grammar in a shorter prompt: the camera or a locked frame, one action with its stop, the protected things, the endpoint, the closing line. It has no end frame and no sound switch, and it returned a stereo track on four of four clips (12 Sep 2026) although each prompt asked for silence in words. Silence is a setting, and words do not reach it.

16.4 Templates

Fill every bracket from the shot record (Chapter 8) and delete any bracket you cannot fill; a template left half full is the commonest way to send a hedge to the model. Every template ends on the closing line, and the sound switch is set in 16.6, not in the words.

Written for this book · template; not run · finished grade, camera-led, from a start frame

Template — the camera is the shot

[Camera verb] on [rig, with its weight] toward [destination], approximately [one distance or angle] over [duration], [a speed profile that ends: "a settling push that eases to rest"]. [The one committed action, through its beginning, middle and arrival.] [The world's life: systems the frame contains, each sized where the eye can check it.] Keep [the named things that must not change] exactly as they are[, and the period: nothing modern enters the frame]. The move settles as [the composition you can cut on], held to the end. [Up to three earned exclusions.] No captions, no subtitles, no on-screen text.

Written for this book · template; not run · finished grade, action-led, camera locked

Template — the action is the shot

[Who] [does the one action, through its contact and its stop, in the order it happens]. The camera is locked; everything that moves, moves inside the frame. [What the action causes in the room: one approximate magnitude for the consequence; a light change with its cause.] Keep [the named things that must not change] exactly as they are, one [prop] throughout. Everything comes to rest on [the composition: where each hand and the eyes rest] and holds to the end. No captions, no subtitles, no on-screen text.

Written for this book · template; not run · test grade, to prove that a still moves

Template — the test clip

[Camera verb] toward [destination], approximately [one distance] over [duration], settling. [The one thing that moves in the frame.] Keep [the one thing that must not change]. The move settles on [the composition], held to the end. No captions, no subtitles, no on-screen text.

A test clip proves a route for one std attempt; it is never the model of a clip you deliver (14.4).

Written for this book · template; not run · start and end frames

Template — the landing is the point

Over the first [two] seconds [the object] [moves how] toward [where], about [one size in body units], and by the [third] second it stops [against what] with a small settle; [its orientation, held for the whole path: "its printed face stays square to camera the whole way"]. The camera is locked. Keep [the named things that must not change]. Only after it has stopped [the release]. The move lands on the last frame and holds to the end. No captions, no subtitles, no on-screen text.

Written for this book · template; not run · the wordless plate for the exact-take route, sound on

Template — the talking plate

The camera is [locked / pushing in approximately [n] centimeters over five seconds]; everything that moves, moves inside the frame. In the first half second [one small breath or settle]. Then, for about [n] seconds, [she / he] talks quietly toward the lens, the jaw and lips working through an easy rhythm, the head making a small nod with each phrase. [One or two systems of the world's life.] Keep the lips, teeth and jaw clear and true throughout, both corners of the mouth visible, with nothing crossing the mouth, and keep the face exactly as it is. As [she / he] finishes, [the landing: the lips close, the eyes stay on the lens], and the frame holds there to the end. No captions, no subtitles, no on-screen text.

Set the plate to std, five seconds, sound on. The pass discards the plate's own sound and lays the take, so plate sound follows what has passed: Kling 3.0 on, Seedance 2.5 off, Wan 3.0 off (24.3). Keep the padded take 0.3 to 0.5 s shorter than the 5.04 s file (22.4).

Written for this book · template; not run · Omni Edit, one scoped change in an accepted clip

Template — the one change

Edit the supplied clip. Change only [the one thing] to [the new state, in full unless the image carries it]. Keep its [shape, motion, occlusion by the actor], and let the change affect [nearby light or reflection] consistently. Preserve [the people, their movements, the other objects], the framing, the timing and the light. One [object] throughout; none of the original remains. The clip ends as it did.
No overlaid text, subtitles, logos, or interface graphics.

16.5 Examples at the standard

Six prompts follow. Each runs from an approved start frame, is complete against the seven elements of Chapter 14, and gives one approximate magnitude to each moving system whose size matters. The first is a Mastorna shot in corrected form; the next four are a beauty push, a food and drink pour, an automotive reveal and a telecom scene with a person; the last is the talking plate. None has been run. The start frame of each is described in its label only so you can see what the words leave out.

From Mastorna, corrected · film trailer, shot 2.1, the crane down into the piazza · Kling 3.0, the approved still as start frame, 5 s, std, 16:9, sound off · corrected for this book; not run

Example 16.1 — the crane down into the piazza

Crane down slowly from the elevated position on a studio crane arm, heavy and steady, with the slight lateral sway of a boom reaching its lower limit, descending approximately sixty centimeters over five seconds, a settling descent that eases to rest. Rain falls steadily in visible diagonal streaks through the emergency vehicle headlight beams, ambulance beacons rotate in slow orange arcs casting sweeping light across the wet cobblestones, medics in white coats cross the square as small moving figures, the cobblestone reflections shimmer and ripple with the continuous rain impact, and reflected headlight streaks crawl slowly across the wet stone toward the camera. Keep the cathedral exactly as framed, its two spires and dark facade unchanged, and keep the ambulances and the fire truck period-correct, with nothing modern entering the frame. The descent settles with the cathedral still holding the vertical center of the frame and the beacons still turning at its base, the medics small against the wet stone. No captions, no subtitles, no on-screen text.

The rig is the declaration and has one mechanism: a studio crane arm, with its physics (heavy and steady, the slight lateral sway of a boom at its lower limit), one magnitude, a duration and a speed profile that ends.
The world's life is five systems the frame contains (rain, beacons, medics, reflections, headlight streaks). The cathedral is protected by name and not described, and the period is locked in the same sentence.
The endpoint is a composition, the cathedral holding the vertical centre with the beacons still turning, and the closing line is the text line because the sound switch is off.
Check on return: the two spires at 20, 40, 60 and 80 per cent, the period of the vehicles, and the last frame against the endpoint.

Written for this book · beauty and personal care · Kling 3.0 pro, start frame (the signed profile of a woman at an open window, hair down, subject 1 m from the lens), 5 s, 16:9, sound off · finished grade · not run

Example 16.2 — a push-in on the hair

Push in on a heavy studio dolly toward her profile, advancing approximately twenty centimeters over five seconds, a settling push that eases to rest, the dolly's weight reading as a slight lag at the start and a damped arrival at the end. Over the first three seconds she turns her head approximately ten degrees toward the window, and the long lengths of her hair follow a half-beat behind, swinging a hand's width and settling back onto her shoulder, each strand catching the window light in turn. The sheer curtain beside the window lifts approximately a hand's width in the draught from the open sash and falls back, and dust turns slowly in the shaft of light behind her. Keep her face, the parting of her hair and the fine flyaways at her temple exactly as they are, the hair one unbroken mass at the roots, with no strand crossing her eyes. The push settles with her cheek and the sweep of her hair filling the right of the frame and the window light rimming its edge, held to the end. No captions, no subtitles, no on-screen text.

The camera opens the prompt because the move is the shot. At the shot record's one metre, twenty centimeters enlarges her by a quarter (× 1.25, 14.5), the size for a face.
Three systems carry a size each (the camera, the head, the curtain) and the hair follows the head, so its cause is visible. The dust keeps a pace word, because its size does not matter.
Protection names what hair invites: a parting that shifts, a strand across the eyes, flyaways that vanish. Pro is chosen for strand detail.
The endpoint is a composition with a light on it, so the cut can land on the rim-lit edge.

Written for this book · food and drink · Kling 3.0 std, start frame (the signed bowl of ful with the jug already tilted and the first thread falling), 5 s, 16:9, sound off · finished grade · not run

Example 16.3 — the oil goes on the ful

The oil keeps pouring: the jug tilts approximately eight degrees further over the first two seconds, the golden thread thickening as it falls a hand's height into the bowl, then rights itself over the next two seconds, the thread thinning and cutting clean at about four seconds. Where it lands, a gold pool spreads across the ful approximately five centimeters and stops. The camera is locked at table height; every movement is inside the frame. Steam rises from the bowl in slow curls, leans once to the left as the oil lands, and thins by the last second. Keep the jug's shape and glaze, the bowl's rim, the copper spoon and the scatter of parsley exactly as they are, one thread of oil throughout and unbroken until it cuts, and the hand on the handle with its five fingers separate. The pour ends with the jug level above the bowl, the gold pool spread and settled, the steam thinning above it, held to the end. No captions, no subtitles, no on-screen text.

One liquid moves. The thread and its pool count as one system, and each stage has its size and time: the tilt, the cut at about four seconds, the pool. The steam is the only other system, and the camera is locked as design.
The pour has a beginning, a middle and a stop: it thickens, rights, cuts clean and settles, so the model has somewhere to end.
Protection names what a pour breaks: the jug, the rim, the single thread, the fingers on the handle.
The sound switch is off because the edit builds the pour's sound (Chapter 25), so the prompt carries only the text line.

Written for this book · automotive and luxury · Kling 3.0 pro, start frame (the signed rooftop frame at dusk: a stone pillar in the left foreground, the saloon beyond it), 5 s, 16:9, sound off · finished grade · not run

Example 16.4 — a track that reveals the saloon

Track right on a heavy dolly along rails at a slow walking pace, travelling approximately one meter over five seconds, a rolling start with a slight lag and a firm, damped arrival at the end, so that the stone pillar in the foreground slides left out of the frame and the saloon is revealed behind it. As the camera passes, the last orange of the sky slides along the saloon's shoulder line as a single unbroken highlight, the pillar's shadow crosses the bonnet and leaves it, and the far lights of the city behind the parapet flicker. Keep the saloon's badge, grille bars and wheel spokes exactly as they are, the body lines unbroken as the pillar's edge passes them, one saloon throughout with its colour unchanged, and nothing else entering the frame. The track settles with the saloon in three-quarter view, the badge on the vertical centre line of the frame and the pillar gone, held to the end. No overlaid text, subtitles, logos, or interface graphics.

The reveal is parallax. A metre of track slides the near pillar out of the frame while the far saloon shifts only a little, which is why this move is larger than a push on the same car (16.3).
The world's life is three systems the frame contains, and the highlight and the shadow move because the camera does: reflections stay attached to the bodywork.
The badge and the grille are lettering and mark, so protection names them, the pillar's edge is told to leave the body lines unbroken, and the closing line is the overlay exclusion. Pro is chosen for the spokes and the badge.
The endpoint is a composition on an axis, the badge on the centre line, so the shot cuts to the packshot.

Written for this book · telecom and tech · Kling 3.0 std, start frame (the signed medium close-up of a young woman on a sea-front balcony at night, phone in her hand), 5 s, 16:9, sound off · finished grade · not run

Example 16.5 — the message, and the look up

The camera is locked; everything that moves, moves inside the frame. She reads the message on her phone, eyes on the screen, until the lights of the far ferry crawling across the horizon behind her pass her shoulder. Then, over about two seconds, she raises her eyes from the screen to the sea, her chin lifting approximately ten degrees as she breathes out, and her hand lowers the phone approximately a hand's width so that the cool light of its screen falls away from her cheek while the warm light of the string lamps along the rail stays. The lamps sway approximately a hand's width in the sea air and settle. Keep her face, the phone held loosely at her chest and the fingers of both hands separate exactly as they are, one phone throughout. The shot settles on her face lifted toward the sea, the phone low at her chest, the lamps steady and the ferry's lights small behind her shoulder, held to the end. No captions, no subtitles, no on-screen text.

One visible event, with a cause the audience can see: the ferry's lights pass her shoulder, and only then does she look up (Chapter 15). The sequence is written in the order it happens.
The acting is in Kling's reliable list: a head lift, a breath, a lowered phone. No micro-expression and no precise eyeline is asked for.
The light change has a cause and a place: the screen's cool light leaves her cheek because the phone drops, and the warm lamps stay.
The camera is declared locked as design, the frame carries the phone and its screen, and the prompt describes neither.

Written for this book · food and drink · Kling 3.0 std, start frame (the signed speaking frame of a baker at her shop counter, lips barely parted), 5 s, 16:9, sound on · the wordless plate for Sync Lipsync 3 on the exact-take route · timed to a stated take of 4.6 s: 0.6 s head, 3.6 s of speech, 0.4 s tail; the figures belong to your take · finished grade · not run

Example 16.6 — the plate for a testimonial

The camera is locked on the medium close-up, level with her eyes; everything that moves, moves inside the frame. In the first half second she draws one small breath through her nose and settles her shoulders. Then, for about three and a half seconds, she talks quietly toward the lens, her jaw and lips working through an easy, unhurried rhythm, her head making a small nod with each phrase. Her hands stay below the frame. Steam from the fresh loaves behind her drifts up and to the left, and the shop's warm light holds steady. Keep her lips, teeth and jaw clear and true throughout, both corners of the mouth visible, with nothing crossing the mouth and nothing held near it, and keep her face exactly as it is. As she finishes, her lips close and her eyes stay on the lens, and the frame holds there to the end. No captions, no subtitles, no on-screen text.

The plate speaks without words. The prompt asks for quiet talking and carries no line, so the pass has a speaking face to redraw and the model has no words to perform or turn into English. The line itself is an actor's or ElevenLabs take, locked before the plate is made.
The plate is timed to the take. A small breath in the first half second, about three and a half seconds of talking under a 3.6 s speech, then a landing. The sum 0.6 + 3.6 + 0.4 is 4.6 s, 0.44 s shorter than the 5.04 s file. The measured talk onset, not the prompt, sets the final head padding.
The mouth is protected by name, the hands stay below the frame and the camera is locked, because body motion competes with the sync.
Sound is on because that is the configuration that passed (24.3); the pass replaces the audio, so the prompt has no sound line.

16.6 Settings

Settings are chosen in the request, never written in the prompt. The connector's contract was last read on 28 Sep 2026; read it again on the day you spend, and name the model in every request, because the connector's tool text steers some requests to other models.

Setting	Values	What to use, and why
Model	kling3_0 (VIDEO 3.0) · kling3_0_turbo · kling_video_edit	kling3_0 for the working shot. Turbo is a speed tier, not a budget tier (16.9). Omni Edit only for one scoped change in an accepted clip
Mode	std · pro · 4k, default std	std to find the shot and for most work. Pro (1912 × 1080, about 1.2 times std) for a hero shot delivered at 1080p, or where fine detail decides: hair, skin, lettering. 4k only after a probe (below)
Duration	3 to 15 whole seconds, default 5, no auto	Five seconds for a single action. Six to eight when the landing needs time to settle. Split a long action into two clips rather than stretch one: 10 s has run once and 15 s never
Aspect	16:9 · 9:16 · 1:1, no default shown	Set it every time, and make the keyframe at the delivery aspect first (14.15). 4:5, 4:3 and 21:9 do not exist: reframe in the edit. Only 16:9 has been judged
Sound	on · off, default on	Off for every clip whose sound you design. On for a talking plate (24.3), and for a test of Kling's own speech
Start frame	one image	The signed keyframe, at the instant the clip begins in (14.3)
End frame	one image	Only when the landing is the point. It costs nothing extra
Preset decline	declined_preset_id	When a preset suggestion answers a job instead of running it, resubmit with the decline. Never accept the preset

Defaults that bite. Sound is on unless you turn it off, and a generated mix cannot be taken apart: an unset switch is a decision you did not make. The aspect has no default, so an unset one is a refused or a surprising job. The job record can disagree with your log: the plate that passed on 12 Sep 2026 was logged "sound off" everywhere and ran with sound on at 8.75 credits, so write down what the job record says.

The 4k mode is not yet a delivery mode. Higgsfield describes it only as "4K resolution". Kling launched native 4K for the 3.0 series on 23 Apr 2026, and Higgsfield's post of the same day calls its Kling 3.0 4K native and not upscaled, but no 4k file has been probed. Until one measures 3840 × 2160, write "4k mode, provenance unverified" in the delivery notes, and get 4K from a judged upscale after picture lock (Chapter 30).

Set by the job, not by you: guidance 0.5, prompt enhancement off, multi-shot off.

The route in one line: kling3_0 · std · sound off (on for a plate) · 5 s · 16:9 · a start frame, plus an end frame only when the landing is the point → probe → the director's eye.

Turbo (kling3_0_turbo) takes a start frame only. It has a resolution setting, 720p or 1080p, in place of modes, three to fifteen seconds, the same three aspects and no sound switch. Omni Edit (kling_video_edit) takes a source clip and optional picture references, has std, pro or 4k modes with pro as the default, and no aspect list: the source sets it. Run the free cost check with the clip attached, and choose the mode after it.

16.7 Checks before you sign

Check the returned file, not the job record. Work at 100 % and at 200 %, frame by frame where the shot depends on it, and write what you found in the run record.

Before you sign a Kling clip

Probe the file for raster, frame rate, frame count, length and audio stream. Expect std 1284 × 716 (the job record says 1280 × 720), pro 1912 × 1080, 24 frames a second, 121 frames for five seconds (5.04 s) and 241 for ten. Sound off gives no audio stream; sound on gives AAC stereo at 44.1 kHz. Turbo's "1080p" in 9:16 was 1076 × 1928, with a stereo track on all four clips (12 Sep 2026). Quote the probed figures, never the label.
Frame 0 against the keyframe, at 200 %: face, hands, label, lettering, the count of props. It is the keyframe scaled, so small print softens, and what was wrong in the still is still wrong.
The first second. Something visible moves at once; there is no dead second before the photograph wakes.
The move, measured (below).
The middle, at 20, 40, 60 and 80 per cent: the protected things (the spires, the badge, the label, the fingers), one prop where you asked for one, and nothing that entered from the frame edge. For a start and end pair, the path between the frames.
Lettering at full crop in the first, middle and last frames. Arabic must be exact letter for letter (Chapter 10). A letter that drifts twice goes to the compositing layer.
The last frame against the endpoint you wrote, and the hold. The hold after the action is your handle; nothing exists beyond the ends of the file.
Period and place. Nothing modern in a period frame; the light is still the frame's light.
Sound. Absent when the switch was off. When it was on, listen: a plate's sound is replaced by the pass, and a spoken line in a clip is a test.
A talking plate, watched as a performance before it goes to the pass, with its talk window measured (22.4).

Measuring a push. A measured move is a request, and the returned clip is the answer. Take the first and last frames at the file's own size, pick one rigid feature at the subject's distance that does not move, measure its height in each, and work out the travel as in 14.5. On a std file the frame is 716 pixels high, not 720; the ratio is unaffected. Then measure a second feature well behind the subject. If it grew by the same factor, Kling zoomed instead of moving, and the rig sentence needs its weight or its mechanism. Write the distance asked and the distance measured in the run record. A few clips answer the question that no maker answers: whether Kling honours a stated distance.

16.8 Failures and fixes

The general failures of a moving shot (a prop that never moves, an overshoot, a dead first second, melting fingers, a duplicated prop, an object from the edge of the frame, a floating camera, drift) and their fixes are in the table of 14.16, and they apply on Kling unchanged. These are Kling's own.

You see	Why	Smallest fix
The camera zooms and does not move	The far feature grew as much as the near one	Name the mechanism and the destination in the rig sentence; measure again
A detail flips (a bangle's wrist, a shirt's colour)	The prompt contradicts the frame	Correct the prompt; the frame does not always win
An exclusion is ignored ("does not tip or turn")	There is no negative field, and prose negatives are weak	State the positive state held all the way
A face warps on a turn	A turn past about sixty degrees	A small turn, or two shots cut on the look
One action is missing and the rest is good	The model skipped one event	Keep the take; generate the missing event alone from the same start frame and composite it (14.13)
A mouth opens uninvited on a silent plate, or speech comes back in English	A listening line invites speech; dialogue outside the five languages	Name the mouth ("lips together and relaxed"); never ask for dialogue; sound off if the plate is silent
A preset suggestion instead of a job	The platform intercepted the request	Resubmit with the decline field; no charge for the interception

The retry ladder is the one in 14.16, smallest lever first. Change one variable per attempt; on Kling the fifth rung moves the magnitude in, or writes the move as timed phases. A std five-second attempt costs 6.25 credits, about thirty cents. After two failures of the same kind, change the input, not the adjectives.

16.9 Limits, prices and rights

Limits

Limit	Value
Clip length	3 to 15 whole seconds. Five seconds has run often, ten once, fifteen never. Plan takes of five to eight seconds
Inputs	One start frame, one end frame, a prompt and a sound switch. No references, no audio, no video, except Omni Edit's source clip
Modes and aspects	std, pro, 4k; 16:9, 9:16, 1:1
Speech from text	Chinese, English, Japanese, Korean and Spanish; other languages are translated into English (Kling, checked 30 Sep 2026)
Returned files	std 1284 × 716, pro 1912 × 1080, 24 frames a second, 121 frames for 5 s and 241 for 10 s (probes of 12 to 26 Sep 2026). The 4k mode has never been probed

Omni Edit. Kling's Omni guide (checked 30 Sep 2026) gives a source clip of three to ten seconds, at most 200 MB and 2K; up to four images or elements when a video goes in, seven when none does; output at 720p or 1080p; and native audio only when no video goes in. Kling's release note of 17 Jun 2026 says Omni video editing now takes input and output up to 15 seconds and 4K. The guide still shows the old limits, so treat the note as later, and let the free cost check with your clip attached say what the connector accepts. Until it does, trim the source to ten seconds.

Prices

Connector prices are in credits, about five US cents each. They are planning figures with their dates: ask for the cost on the day, which the platform quotes at no charge.

Configuration	Credits	Date and basis
std · 5 s · sound off · start frame	6.25	checked 11 Sep; charged 12, 26 and 27 Sep 2026
std · 5 s · sound off · start and end frame	6.25	checked and charged 26 Sep
std · 10 s · sound off	12.5	checked and charged 26 Sep
std · 5 s · sound on	8.75	charged 12 Sep; checked 27 Sep
pro · 5 s · sound off	7.5	quotes of 17, 20 and 21 Sep
4k · 5 s · sound off	30	quote of 21 Sep, nothing attached
Turbo · 5 s · 720p, 16:9 / 1080p, 9:16	7.5 / 10	quoted 21 Sep / checked 12 Sep
A std plate, sound on, plus a Sync Lipsync 3 pass	about 26.9 (8.75 + 18.15)	charged 12 Sep
A std plate, sound off, plus the pass	about 24.4 (6.25 + 18.15)	a sound-off talking plate has not been run
Omni Edit; any 15 s job; sound on at pro or 4k	not known	never priced

Turbo is a speed tier, not a budget tier: it quoted dearer than std on every quote, and its speed has never been timed.

On Kling's own site and app, priced in Kling's own credits per second of output (checked 30 Sep 2026), VIDEO 3.0 with native audio costs 12 at 1080p and 9 at 720p, and without audio 8 and 6; voice control adds 2. Omni with a video input (audio off only) costs 16 at 1080p and 12 at 720p; without a video it costs 12 and 9 with audio, 8 and 6 without. Native 4K is 30 credits a second (release note, 23 Apr 2026). This is a different currency from Higgsfield's, and the ratios differ: use it to compare tiers on Kling's site, never to convert.

Rights and terms

Three sets of terms touch a Kling clip made through Higgsfield.

Kling's Terms of Service (21 Apr 2026). You own the content as far as the law allows. Users without a membership must label output with Kling's brand. Commercial use of output needs Kling's written permission. Kling holds a non-exclusive licence to what you put in and what comes out, including to train and improve its models. Output about individuals must not be used for decisions with a material impact.
Kling's Terms of Paid Service (21 Apr 2026). A paid member's commercial use is not restricted, except for products that compete with Kling, and members hold the right to remove the watermark.
Higgsfield's terms (updated 26 Jul 2026, read 30 Sep). It claims no ownership of inputs or outputs and does not restrict commercial use of outputs. Its help pages (August 2026) allow ads, broadcast and paid media on any paid plan, with no attribution. Each model provider's acceptable-use policy also binds you, and the stricter rule wins. Keep auto-publish off.

Which set governs a clip made through the connector is not settled (16.10). Meanwhile work on a paid plan; treat a start frame you upload as licensed to Kling and the platform, and do not upload an unreleased client packshot or a real person's likeness without the consents and clearances of Chapter 34; look for a mark on the first file of a job and record that you looked; and put model, tier, date and plan in the run record.

What may ship. A Kling clip may go into a client final when it was made on a paid plan, the rights in its start frame are cleared, the file has passed 16.7 and the ACCEPT gate, and the mark check is done. No Kling voice ships (Chapter 21). Kling certifies no small legal text: legal lines are set in the edit.

16.10 Version notes

Where 3.0 stands

Kling's own release history (checked 30 Sep 2026) dates the steps that matter to a director.

Date	What Kling records
31 Jan 2026	Kling 3.0 fully rolled out (its user guide is dated 6 Feb)
4 Mar 2026	Motion Control 3.0
23 Apr 2026	Native 4K for the 3.0 series
17 Jun 2026	Turbo live; Omni video editing extended to 15 seconds and 4K
28 Sep 2026	Kling 4.0 announced; Flash in early access

Every fact in 16.2 to 16.9 belongs to Kling 3.0. Nothing in the announcement changes one of them, and 3.0 stays the route until 4.0 is on the connector and has been tested.

What 4.0 would change for a director

This is Kling's announcement read for its effect on a shot, not a teaching of 4.0. Every row waits on what the connector exposes.

Kling 3.0 on the connector	Kling 4.0, as announced (28 Sep 2026)	What it would change
Clip length	3 to 15 seconds	3 to 30 (Flash 3 to 20)	One take could be longer. Keep planning five to eight seconds until a longer clip has been judged; the late part of a long take drifts on 3.0
Keyframes	a start and an end frame	up to ten images	A path with waypoints in one clip. The rule that frames are relatives, and the check at 20, 40, 60 and 80 per cent, would apply to every waypoint
References	none	up to 15 items: images, videos, subjects, voice references	The rule that the start frame is the only carrier of identity, and the relationship keyframe, may no longer be needed. Chapter 13's two routes may then meet in one tool
Resolution and range	720p and 1080p; 4k unprobed; 16:9, 9:16, 1:1	720p, 1080p, 4K; 21:9 and auto; 10-bit HDR marked as coming	Wide origination without a reframe. HDR reaches the conform and the grade (Chapter 30). Probe every raster
Sound and language	five dialogue languages; stereo	eight named languages "and more", regional accents; stereo	Arabic is named nowhere. The rule that Egyptian and Saudi dialogue comes from an actor or ElevenLabs stands until a native ear has judged an audition
Editing and extension	Omni Edit on one clip	editing of up to five clips; extension up to two minutes, marked as coming	Not yet in reach; re-read when listed
Price	known (16.9)	not published	The cost line of the spending packet changes for every job

The day 4.0 is listed

Re-checking Kling on release

Read the connector's model list and contract. Note the id, the modes, the durations, the aspects, and which of keyframes, references and voice are exposed. Run the free cost check.
Name the model in every request, and keep 3.0 named where a job has already been approved on it.
Re-run one approved keyframe with its finished 3.0 prompt unchanged, and compare: frame 0 against the keyframe, the move measured (16.7), the hold, the sound.
Probe the first files for raster, frame rate, frame count, audio and bit depth, and write what you measured.
Audition Arabic only as a test, with a native ear, and ship no Kling voice until it has passed.
Read Kling's and Higgsfield's terms again.
Update 16.6 and 16.9, and leave the 3.0 text as it is. Do not cut 3.0 and 4.0 clips together in one sequence without grading them together.

Not yet known

These matter because each changes what a director does. Until they are settled, the chapter teaches the cautious default.

Does kling3_0 accept a prompt with no frame, or an end frame alone? A price check with nothing, or only an end frame, attached costs nothing.
Does Kling honour a stated distance, and how closely? No maker documents it and no measurement is on record; 16.7 answers it clip by clip.
What is the 4k mode? Native or upscaled, and its raster and frame rate.
Do fifteen-second clips hold? Never run.
Does the connector pass Kling's in-take markers? A route fact settled by one run.
Which talking plates pass? A sound-off plate, a pro plate and a plate over five seconds have not been run.
What does Omni Edit accept and return on the connector, and what does it cost?
Are 9:16 and 1:1 clips good? Three 9:16 clips ran without a verdict; 1:1 never ran.
Do reverse motion and the composite repair work on Kling? Neither has been run.
Which terms govern a clip made through the connector?
Kling 4.0: when the connector lists it, which keyframes and references it exposes, what it costs, how it handles Arabic, and whether 3.0 stays listed.

What to remember

Kling 3.0 is the camera you reach for when the move is the shot: a signed start frame, one action, one clip. Kling 4.0 was announced on 28 Sep 2026 for a full release in October; only Flash is out, in early access for a limited group, and neither is on the connector, so write for 3.0.
The connector gives a start frame, an end frame, a prompt and a sound switch. The start frame is the whole identity, so every shot needs its own keyframe, and a beat between two people needs one relationship keyframe.
Open on the camera when the move is the shot, otherwise on who does what with the camera locked. Give each moving system one approximate magnitude, a duration and a profile that ends, and end on a composition and a hold.
Do not re-describe the frame, do not name what is off it, and say the positive state, because prose negatives are ignored. Three or fewer exclusions (five is the ceiling), then the closing line.
Set the sound explicitly: off for every clip whose sound you design, on for a talking plate. Never Arabic dialogue from Kling.
Probe every file, measure the move, judge the middle and the lettering, and quote the probed raster, never the label.
Kling's terms and Higgsfield's both apply, and which governs a connector clip is not settled. Check the price on the day.

Contents of The AI Commercial Director's Handbook

Next: Chapter 17 · Seedance 2.5