kling-3-prompting

Overview

Kling 3.0 is a unified multimodal video model. It understands cinematic direction, not keyword lists. Write prompts like a director — describe what the audience sees, hears, and feels over time.

Core shift: Description → Direction. Think "direct a scene" not "describe an image."

Interactive Builder Workflow

When invoked, guide the user through these steps using AskUserQuestion:

digraph builder {
  "1. Generation mode?" [shape=diamond];
  "Text-to-Video" [shape=box];
  "Image-to-Video" [shape=box];
  "Multi-Shot Sequence" [shape=box];
  "Keyframe Transition" [shape=box];
  "2. Gather scene details" [shape=box];
  "3. Assemble prompt" [shape=box];
  "4. Present & refine" [shape=box];

  "1. Generation mode?" -> "Text-to-Video";
  "1. Generation mode?" -> "Image-to-Video";
  "1. Generation mode?" -> "Multi-Shot Sequence";
  "1. Generation mode?" -> "Keyframe Transition";
  "Text-to-Video" -> "2. Gather scene details";
  "Image-to-Video" -> "2. Gather scene details";
  "Multi-Shot Sequence" -> "2. Gather scene details";
  "Keyframe Transition" -> "2. Gather scene details";
  "2. Gather scene details" -> "3. Assemble prompt";
  "3. Assemble prompt" -> "4. Present & refine";
}

Step 1: Determine Generation Mode

Ask the user which mode:

• Text-to-Video — prompt from scratch
• Image-to-Video — animate a reference image
• Multi-Shot Sequence — 2-6 shot storyboard (up to 15s)
• Keyframe Transition — start frame → end frame with interpolated motion

Step 2: Gather Scene Details

Ask about each element (adapt questions to mode):

Element	Question	Why it matters
Subject	Who/what is the focus? Specific appearance details?	Anchors consistency — define distinguishing traits early
Action	What happens? Describe the timeline (first → then → finally)	Kling 3.0 excels at sequential action over 15s arcs
Environment	Where? Be specific (not "a street" but "narrow Tokyo alley, steam from grates")	Grounds the scene physically
Camera	Shot type and movement? (See camera reference below)	Cinematic language produces far better results
Lighting	What light sources? Name them specifically	"Flickering neon" beats "dramatic lighting"
Mood/Emotion	What should the audience feel?	Drives color grade, pacing, music
Audio	Dialogue? Ambient sound? Music?	Kling 3.0 generates native audio + lip-sync
Duration	How long? (3-15s)	Longer = describe progression over time
Aspect Ratio	16:9 / 9:16 / 1:1 / 21:9?	16:9 cinematic, 9:16 social, 21:9 ultra-wide

Image-to-Video: Focus on how the scene evolves from the image — movement, camera motion, environmental change. The model preserves identity/layout from the source.

Keyframes: Ask for start and end frame descriptions. Frames should match in color, style, and lighting. Prompt sparingly — Kling infers motion well.

Multi-Shot: Define each shot separately with its own framing, subject, action, and duration. Label shots explicitly.

Step 3: Assemble the Prompt

Use the Master Formula:

[Scene/Environment] + [Subject & Appearance] + [Action Timeline] + [Camera Movement] + [Audio & Atmosphere] + [Technical Specs]

Writing rules:

• Use cinematic motion verbs: dolly push, whip-pan, crash zoom, rack focus, tracking shot — NOT "moves" or "goes"
• Name real light sources: neon signs, candlelight, golden hour, LED panels — NOT "dramatic lighting"
• Include texture for credibility: grain, lens flares, condensation, fabric sheen, smoke, sweat
• Describe temporal flow: beginning → middle → end
• Keep to 1-3 rich sentences per shot (specificity > length)
• For dialogue: use character labels, assign voice tone/emotion, use transitional words ("Immediately," "Pause")

Step 4: Present & Refine

Present the assembled prompt. Ask if they want to:

• Adjust any element
• Add a negative prompt
• Generate variations (different duration, different camera, different mood)

Quick Reference

Camera Movements

Movement	Effect	Example phrase
Dolly push-in	Builds intimacy/tension	"slow dolly push-in toward her face"
Dolly zoom	Vertigo/dramatic reveal	"dolly zoom creating disorienting depth shift"
Tracking shot	Follows subject laterally	"camera tracks alongside as she walks"
Whip-pan	Energy/surprise	"whip-pan to reveal the door"
Crash zoom	Shock/emphasis	"sudden crash zoom on the object"
Rack focus	Shift attention	"rack focus from foreground hand to background figure"
Handheld/shoulder-cam	Raw/documentary feel	"handheld shoulder-cam with subtle sway"
Static tripod	Composed/observational	"locked-off static tripod, wide shot"
FPV drone	High-energy immersion	"dynamic FPV drone shot chasing through corridor"
Low-angle tracking	Heroic/imposing	"low-angle tracking shot, subject towers above"
Truck left/right	Lateral reveal	"camera trucks right revealing the cityscape"
Tilt up/down	Vertical reveal	"slow tilt up from boots to face"

Lens & Film Stock

Phrase	Effect
"Shot on 35mm film"	Warm grain, organic texture
"Macro 85mm lens"	Tight detail, shallow depth of field
"Wide-angle steadicam"	Smooth, immersive, spatial
"Handheld camcorder"	Raw VHS energy, nostalgic
"Anamorphic lens flare"	Cinematic horizontal streaks

Lighting

Use specific sources, not adjectives:

• "Golden hour sun cutting through dusty warehouse windows"
• "Flickering neon casting magenta/cyan across wet pavement"
• "Single bare bulb swinging, casting moving shadows"
• "Cool blue LED panels reflecting off glass surfaces"
• "Candlelight warming skin tones, deep shadows beyond"

Color & Grade

• "Desaturated teal grade, crushed blacks"
• "Amber nightclub strobe cutting through smoke"
• "Cool blue haze filling the corridor"
• "Magenta neon reflecting off wet asphalt"
• "Overexposed highlights, blown-out whites"

Multi-Character Dialogue

Rule	Do	Don't
Name characters	[Character A: Silver-haired CEO]	[Man] says...
Anchor to action	Agent slams table. [Agent, angrily]: "Where is it?"	Just dialogue without visual action
Assign voice tone	[CEO, deep authoritative gravelly voice]	Generic "says"
Control timing	"Immediately," "Pause," "After a beat"	Back-to-back dialogue without transitions

Multi-Shot Structure

Shot 1 (0-5s): [Wide establishing shot description]
Shot 2 (5-10s): [Medium/close-up with action progression]
Shot 3 (10-15s): [Resolution/reaction with camera payoff]

Atmosphere: [Overall mood, color grade]
Audio: [Sound design, music, dialogue]

Label every shot. Assign durations. Describe framing + subject + motion per shot.

Start & End Frame Tips

• Frames should match in color palette, style, and lighting
• Identical start/end frames = seamless loop
• Prompt sparingly — Kling infers motion between frames well
• Simple camera directions: zoom in/out, pan left/right, tilt up/down
• 5s for dynamic transitions, 10s for complex transformations
• Start frame aspect ratio drives the whole clip

Negative Prompts

Use to prevent common AI defaults:

smiling, laughing, cartoonish, bright saturated colors, low resolution,
morphing, blurry text, disfigured hands, extra fingers, static pose,
frozen expression, stock photo aesthetic

Customize based on scene — remove items that conflict with your intent.

Weak → Strong

Element	Weak	Strong
Camera	"Camera follows person"	"Handheld shoulder-cam drifts behind subject with subtle sway"
Subject	"A woman walking"	"Woman in red dress, heels clicking wet cobblestone"
Environment	"In a city"	"Narrow Tokyo alley, steam from grates, glowing vending machines"
Lighting	"Dramatic lighting"	"Flickering neon casting magenta/cyan across wet pavement"
Texture	"It looks realistic"	"Rain beading on leather jacket, condensation on glass, visible breath"
Motion	"She walks away"	"She turns slowly, hair catches light, disappears around corner"

Common Mistakes

Mistake	Fix
Keyword lists instead of scene direction	Write like directing a shot: subject + action + camera + environment
Vague motion ("moves," "goes")	Use cinematic verbs: dolly, track, whip-pan, crash zoom
Generic lighting ("dramatic")	Name the source: neon, candle, golden hour, LED panel
Overlong prompts	1-3 rich sentences per shot; specificity > length
No temporal progression	Describe beginning → middle → end of the shot
Mismatched keyframes	Match color, lighting, and style between start/end frames
Unattributed dialogue	Label every speaker with name, tone, and emotion
Cramming multi-shot into one paragraph	Separate and label each shot with duration