SEEDANCE AI VIDEO PLAYBOOK


Seedance Prompt System to generate insanely good AI videos and curate expert-level prompts. 

Subscribe to metricsmule on 🎯 YouTube

The Seedance 2.0 Playbook

ONE MODEL.
EVERY PLAY YOU NEED.

Eight focused breakdowns of what Seedance 2.0 does that other models can't — Omni-Reference tokens, timestamp scripting, GLOBAL LOCK, cinematic vocabulary, scene breaks, and audio sync. Every play, in one place.

3reference token types
12camera moves mapped
3audio sync modes
shots from one lock

01 · Inside the model

Seedance 2.0 Unfair Advantages

Seedance 2.0 was architected for cinematic coherence, not just per-frame beauty. Most models hallucinate motion — Seedance 2.0 choreographs it. Here's where it legitimately outperforms the field.

Exclusive
🔗

Omni-Reference tokens

Inject @image, @video, and @audio references in a single prompt. Character, motion, and sound composited from separate files — while other models accept just one input.

Seedance
Others
Exclusive

Timestamp shot scripting

Write camera events at exact timestamps inside one generation. [0:00] wide → [0:03] push in → [0:06] rack focus. No other consumer model exposes in-prompt timeline control.

Seedance
Others
Exclusive
🔒

GLOBAL LOCK system

Persistent identity lock across shots. Character face, lighting rig, and color grade stay consistent — no drift between generations. Your reference becomes a contract, not a suggestion.

Seedance
Others
Superior
🎬

Cinematic vocabulary

Anamorphic bokeh, Steadicam follow, rack focus, lens breathing — trained on real cinematography. These terms change output meaningfully, not just as keywords but as compositional logic.

Seedance
Others
Unique
✂️

In-prompt editorial cuts

SHOT A → SHOT B → SHOT C within one prompt using scene break syntax. Editorial-grade cuts baked into the generation, not stitched together afterward in post.

Seedance
Others
Unique
🎵

Audio-driven motion sync

The @audio token syncs visual motion to music tempo, dialogue rhythm, or SFX beats directly from a reference file. Your video is generated to the music, not edited to it afterward.

Seedance
Others
Master prompt architecture
GLOBAL LOCK@character_ref + @location_ref — identity contract for the scene
SHOT LANGShot type · lens focal length · camera movement · frame rate
LIGHT + GRADELighting motivation · color temperature · film stock / LUT reference
TIMESTAMPS[0:00] action → [0:03] camera event → [0:06] cut or beat
AUDIO REF@audio_ref (optional) — sync motion to music tempo or dialogue

02 · Omni-Reference

Stacking @image + @video

@image locks who/what (identity, costume, props). @video locks how (motion style, camera behavior, pacing). They don't compete — they operate on different output dimensions, so you hand the model a character bible and a movement bible at once.

@mara_ref.jpg

@image locks

  • Face geometry & skin texture
  • Hair color, cut, style
  • Costume — fabric, color, cut
  • Body type & proportions
  • Lighting on subject skin
@motion_ref.mp4

@video locks

  • Camera arc & path
  • Shake, drift, micro-motion
  • Speed & ease curves
  • Focal / bokeh behavior
  • Temporal pacing & cut rhythm
TOGETHER

Unlocks

  • Character in a new environment
  • Film-accurate motion, AI subject
  • Consistent series across shots
  • Art-directed realism, not default
Ghost Lane · Omni-Reference example
[GLOBAL LOCK] @mara_ref.jpg | @ghost_lane_tunnel.jpg
[MOTION REF] @motion_ref.mp4 — extract camera behavior only, do not transfer subject or location
[SHOT] LOW ANGLE — 35mm anamorphic — Steadicam tracking follow — 24fps
[LIGHT] Sodium vapor practicals overhead — motivated fill from neon at camera left — teal shadows, amber highlights — Kodak 5219 grain
[0:00] Mara walks toward camera, hands in jacket, headlights blooming behind her
[0:04] Camera drifts left — rack focus to her eyes — streetlight catches her jaw
[0:07] Pull back wide — she stops, looks over shoulder — exhaust vapor rises
The critical instruction: — extract camera behavior only on the @video line. Without it, the model bleeds the reference clip's visual content into your output. You're scoping what to extract, not just handing it a file.

03 · Timestamp scripting

Directing by the beat

Each [M:SS] marker is a director's cue card handed to the model at that moment. Three independent channels — camera, subject, atmosphere — can each change at any beat. Timestamps are additive: whatever you set at [0:00] stays true until you explicitly change it.

Rule 01

Timestamps are additive

Whatever you set at [0:00] stays active until you change it. Only write what changes — don't repeat the full state every beat.

Rule 02

Three channels, one beat

Camera, subject, and atmosphere are independent. Move the camera without touching the subject. Layer all three for cinematic complexity.

Rule 03

3–4 second minimum gaps

Beats closer than 3 seconds compress and blur. The model needs runway to transition. For a 12s clip, 4–5 beats is optimal.

Rule 04

Declare a BASE before [0:00]

Set lens, film stock, frame rate, and grade as a BASE block above the timestamps. These become global defaults you never repeat.

Ghost Lane · Mara tunnel walk · 12s
[GLOBAL LOCK] @mara_ref.jpg | @tunnel_env.jpg — identity contract, do not drift
[MOTION REF] @ref_tracking.mp4 — extract camera physics only, not subject
[BASE] 35mm anamorphic — Steadicam — Kodak 5219 grain — teal shadows, amber highlights — 24fps
[0:00] Mara stands in tunnel mouth, back to camera — headlights silhouetting — exhaust vapor drifting — Steadicam hold, barely breathing
[0:04] She turns — slow pivot — camera drifts right to stay on face — rack focus: tunnel bg → her eyes — sodium light catches jawline
[0:07] Starts walking toward camera — Steadicam begins tracking retreat — anamorphic flare blooms left as headlights align
[0:10] Camera drops lower — Dutch tilt 8° right — she stops — looks over shoulder into tunnel dark
[0:12] Hold on her face — exhaust vapor creeps into frame from below — freeze on tension — end
The arrow syntax matters: rack focus: bg → eyes — the tells Seedance you want a transition at that beat, not a static state. Without the arrow, it just sets focus. With it, it generates the actual pull motion.

04 · Camera moves

Moves that work at every beat

Timing of when you deploy a move matters as much as the move itself. Certain behaviors work with the model at specific beat windows, and some combinations actively fight each other. Here's the full reference matrix.

MoveBest beatPair withNever combine with
Steadicam hold[0:00]Static subjectMoving subject, same beat
Steadicam retreat[0:04]–[0:07]Subject walk toward camHandheld in adjacent beat
Rack focus →[0:03]–[0:07]Subject turn / revealWhip pan (fight timing)
Dutch tilt N°[0:09]+Subject stillness / tensionOpening beat, orbital arc
Slow push in[0:04]–[0:09]Emotional close-upTracking retreat (opposing)
Whip pan[0:06]–[0:08]Scene change / cut pointSecond whip pan within 4s
Crane up reveal[0:00]–[0:04]Wide environment shotClose subject framing
Orbital arc N°AnyStatic subject anchorFull 360° under 10s
Handheld urgency[0:04]+Action / confrontationSteadicam, same 3s window
Low angle rise[0:00] or [0:09]+Character power revealDutch tilt simultaneously
Snap zoom[0:10]+Shock / climax beatMid-clip, >1 per clip
Lens breathingAny (modifier)Any primary moveDigital zoom (breaks analog)
The most powerful contrast cut: handheld after Steadicam. Start smooth and controlled, then flip to handheld urgency at [0:07]. The contrast is so jarring it reads as intentional directorial choice every time — the single most reliable technique for building tension in a clip that starts calm.

05 · GLOBAL LOCK

The anti-drift architecture

Without a lock system, each generation is stateless — the model has no memory of shot 1 when it renders shot 2. GLOBAL LOCK treats your references as persistent contracts across every generation. It's three independent binding layers you can engage in any combination per shot.

@character_ref

Identity lock

  • Face geometry
  • Skin tone & texture
  • Hair color & style
  • Costume & props
  • Body proportions
  • Light behavior on subject
@location_ref

Environment lock

  • Architecture & space
  • Practical light positions
  • Neon / sign colors
  • Ground surface type
  • Scale relationships
  • Atmosphere / haze density
[BASE] block

Grade lock

  • Film stock / grain
  • Color temperature
  • Shadow tone
  • Highlight rolloff
  • Lens character
  • Frame rate
Ghost Lane · 4-shot sequence · paste header into every generation
=== GLOBAL HEADER ===
[GLOBAL LOCK] @mara_ref.jpg | @ghost_lane_tunnel.jpg — identity + environment contract — do not drift across shots
[BASE] 35mm anamorphic — Kodak 5219 pushed 1 stop — teal shadows / amber highlights — lens breathing — 24fps

=== SHOT 01 · Tunnel mouth · EWS → WS ===
[SHOT] EWS — crane mounted — high angle 40ft above street — Mara small at tunnel mouth
[0:00] Mara motionless at entrance — back to camera — exhaust vapor at boots
[0:04] crane descends 40ft to 12ft — Mara grows in frame — scale reveals
[0:08] crane settles at chest height — WS — she turns, looks over shoulder — end

=== SHOT 02 · Close pursuit · MS → CU ===
[SHOT] MS — handheld urgency — aggressive micro-shake — camera 6ft behind Mara
[0:00] Mara running flat out — handheld follows — neon streaks peripheral
[0:05] she cuts left between cars — camera loses her one beat — rapid reframe
[0:09] she vaults hood — camera tilts to follow — exhaust cloud engulfs — end

=== SHOT 03 · Car interior · MCU → ECU ===
[GLOBAL LOCK] @mara_ref.jpg — identity lock only — environment is car interior (new space)
[BASE] 35mm anamorphic — Kodak 5219 — lit by instrument cluster only — grade held
[SHOT] TIGHT — dashboard cam POV — Mara right third — steering wheel left third
[0:00] face lit by instrument cluster — green and amber glow — jaw set
[0:04] eyes flick to rearview — tension hold — 2-second beat
[0:07] grips wheel tighter — slow push in 1 inch — end tight on eyes

=== SHOT 04 · Final reveal · tight → EWS ===
[GLOBAL LOCK] @mara_ref.jpg | @ghost_lane_tunnel.jpg — full lock restored for finale
[SHOT] CRANE UP — wide pull back — Mara and car shrink as tunnel reveals
[0:00] tight on car rear — exhaust cloud — tunnel walls press close
[0:04] crane rises 12ft — city glow at tunnel mouth ahead
[0:09] full tunnel revealed — car a point of light — Dutch tilt 8° — freeze — end
Shot 03 is the advanced move: when your character changes location, you don't abandon the lock — you drop only the layer that conflicts (environment) and hold the ones that don't (identity, grade). Partial locking is what lets you build a coherent multi-location sequence without every shot looking like a different film.

06 · Cinematic vocabulary

Words that actually fire

Some terms have strong reliable signal. Some are false friends — they sound cinematic but produce no meaningful output change. Describe the production decision, not the aesthetic result.

Lens
  • anamorphic bokeh
  • lens breathing
  • anamorphic flare
  • rack focus →
  • background compression
  • dolly zoom
Lighting
  • motivated key light
  • practicals only
  • Kelvin temperatures
  • chiaroscuro
  • hard falloff
  • shadow percentage
Camera
  • Steadicam + modifier
  • Dutch tilt N°
  • push in + distance
  • handheld urgency
  • orbital arc N°
  • snap zoom
Color grade
  • Kodak 5219
  • pushed N stops
  • teal shadows / amber highlights
  • highlight rolloff
  • crushed blacks
  • split toning
Framing
  • ECU / CU / MS / WS / EWS
  • subject left/right third
  • tight headroom
  • lead room
  • foreground bokeh element
  • negative space direction
False friends ✗
  • cinematic
  • dramatic lighting
  • photorealistic
  • high quality / 8K
  • beautiful / stunning
  • film look
AvoidWhy it failsUse instead
cinematicAttached to everything — no directional signal35mm anamorphic — Kodak 5219 — motivated key
dramatic lightingDefaults to generic studio three-point setupchiaroscuro — motivated sodium key — hard falloff
high quality / 8KResolution is platform-controlled, not prompt-controlledfine grain structure — crisp focal plane
beautiful / stunningAesthetic adjective — no photographic referentgolden hour backlight — soft highlight rolloff
slow motionAmbiguous — must declare frame rate math96fps overcranked — 4x slow playback
The one combo to remember: 35mm anamorphic — Kodak 5219 pushed 2 stops — motivated key from camera left — chiaroscuro — teal shadows / amber highlights — lens breathing. Six terms covering lens, grade, lighting motivation, contrast, color science, and texture at once. It's the Ghost Lane base.

07 · Scene break syntax

One prompt, three shots

Each shot is a self-contained block separated by a === SHOT N === divider. The global header locks identity, environment, and grade — then each shot block carries only what's unique to it. Best sequences follow a deliberate arc: wide establishing → intimate pursuit → climactic isolation.

Shot 01
Ghost Lane entrance
EWS → WS · Crane down
Shot 02
The pursuit
MS → CU · Handheld
Shot 03
Mara alone
MCU → ECU · Push in
Ghost Lane · 3-shot scene break sequence
[GLOBAL LOCK] @mara_ref.jpg | @ghost_lane.jpg — identity + environment contract — do not drift
[BASE] 35mm anamorphic — Kodak 5219 pushed 1 stop — teal shadows / amber highlights — crushed blacks — lens breathing — 24fps
[GRADE] split toning throughout — motivated practicals only — no artificial fill

=== SHOT 01 · Ghost Lane entrance · EWS → WS ===
[SHOT] EWS — crane mounted — high angle, 40ft above street — Mara small at tunnel mouth
[LIGHT] sodium gantry overhead — deep shadow between — headlight bloom from tunnel interior
[0:00] Mara motionless at tunnel entrance — back to camera — exhaust vapor pools at boots
[0:04] crane descends — 40ft to 12ft — Mara grows in frame — scale revealed
[0:08] crane settles chest height — WS — she turns to look over shoulder — hold — end

=== SHOT 02 · The pursuit · MS → CU ===
[SHOT] MS — handheld urgency — aggressive micro-shake — camera 6ft behind Mara, chasing
[LIGHT] neon strobing from headlights ahead — teal wash left — hard intermittent shadows at pillars
[0:00] Mara running flat out — handheld struggles to keep pace — neon streaks peripheral
[0:04] she cuts hard left between cars — camera loses her one beat — rapid reframe — finds her
[0:07] she stops — spins to face camera — CU on face — handheld settles — breathing visible
[0:10] rack focus: background chaos → her eyes — anamorphic flare blooms left — hold — end

=== SHOT 03 · Mara alone · MCU → ECU ===
[SHOT] MCU — Steadicam barely breathing — Mara right two-thirds — negative space left
[LIGHT] motivated key from camera left — single sodium practical — chiaroscuro — shadow side 80% black
[0:00] Mara still — jaw set — eyes tracking off-camera right — sodium catches cheekbone
[0:05] slow push in — 2-inch drift — background compression increases — she holds still
[0:08] rack focus: background → eyes — ECU lands on face — lens breathing intensifies
[0:11] Dutch tilt 8° clockwise — hold — exhaust vapor lower frame — freeze — end
Shot size is emotional register: EWS calm → MS chaos → MCU still is a visual compression of the classic dramatic shape. Wide means the world is bigger than the character. Handheld means she's losing control. Steadicam push into ECU means we're finally inside her. Every frame-size choice is a narrative statement.

08 · Audio sync

Motion choreographed to sound

The @audio token is a motion choreography tool, not background music. The model analyzes the audio's temporal structure and uses it to govern when camera events fire, how fast subjects move, and where cuts land. Three sync modes, chosen with explicit syntax.

🥁

Tempo sync

Camera events and cuts mapped to the beat grid. Bass hits trigger cuts, snares trigger camera moves, downbeats trigger subject actions. Best for music-driven sequences with a clear BPM.

tempo sync: camera on bass hits
🎙

Dialogue sync

Subject lip movement, expression shifts, and body language synced to speech cadence and emotional tone. Silence gaps become hold moments. Best for character performance and reaction shots.

dialogue sync: expression follows cadence
🌊

Atmosphere sync

Uses the audio's energy envelope (not beats) to govern motion intensity. Quiet = slow drift. Building = accelerating motion. Best for score or ambient without a rigid beat grid.

atmosphere sync: motion follows envelope
Ghost Lane · tempo + atmosphere sync · 12s
[GLOBAL LOCK] @mara_ref.jpg | @ghost_lane.jpg — identity + environment contract
[BASE] 35mm anamorphic — Kodak 5219 pushed 1 stop — teal shadows / amber highlights — 24fps — lens breathing
[AUDIO] @never_catch_me.mp3 — tempo sync: camera events on bass hits — atmosphere sync: motion follows energy envelope — extract timing only, do not add audio to output
[BPM] 120 BPM — 4/4 — downbeat at [0:00] [0:04] [0:08] [0:12]
[SYNC MAP] bass hit → camera move — snare → subject action — downbeat → major event — energy build → accelerate

[SHOT] MCU — Steadicam barely breathing — Mara right two-thirds — sodium key from camera left
[LIGHT] motivated practicals only — sodium overhead amber — teal neon fill left — chiaroscuro — no fill

[0:00] downbeat 1 — Mara completely still — exhaust vapor at jaw — Steadicam hold — sync: locked on downbeat
[0:02] bass hit — camera drifts 1 inch left — eyes move — sodium catches cheekbone — sync: drift on bass
[0:03] snare hit — jaw sets — micro-tension in shoulders — sync: subject action on snare
[0:04] downbeat 2 — rack focus: background → eyes — anamorphic flare blooms left — sync: on downbeat
[0:07] energy build — slow push in initiates — motion intensity rises — atmosphere sync kicks in
[0:08] downbeat 3 / drop — push in lands ECU — lens breathing peaks — Dutch tilt 8° — sync: major event on drop
[0:10] snare + bass coincide — head turns back to camera — eye contact — sync: contact on accent
[0:12] downbeat 4 — freeze frame — Dutch tilt holds — exhaust vapor freezes — end

[EXTRACT] use audio for motion timing only — do not render audio track to output video
[PRIORITY] beat sync takes priority over timestamp if conflict under 0.5s — snap to nearest beat
The mandatory scope instruction: extract timing only, do not add audio to output. Without it, the model may render the reference track into your final video file, or try to find beats in ambient noise. You want the timing, not the audio itself in the render.

MetricsMule · The Prompt Playbook

The ChatGPT Prompt Engineering Everything Bundle

👑 Why The Everything Bundle?

You'll never need prompts again! 
That's because you'll have prompts for every category imaginable. From AI images and video, prompt generators, content, secret tokens and so much more. PLUS you'll have lifetime access and will receive non-stop updates for life. Today 15,000+ prompts, but in the future, 50,000+ prompts as it will continue to grow. 

🧬 New databases/categories just added:
🎬 The AI Video Engine
🍌 Nano Banana Category
📚 The Content King

($389 value) But not for you.

THE EVERYTHING BUNDLE

🚀 WHAT'S INSIDE?

✅ Over 10,000 prompts (and counting)
✅ Prompts Portal
✅ AI Video Engine
✅ AI Influencer Database
✅ Freepik Custom Database
✅ MEGA Prompts Database
✅ Leonardo AI database
✅ Prompt Generators database
✅ Photoreal Ebook
✅ Nano Banana Prompts🍌
✅ JSON Prompts
✅ Unique Keywords Database
✅ Content King Database
✅ Canva Prompts Website
✅ Lifetime Access
✅ Constant Updates
✅ Immaculate Organization
✅ Bonus Items

WANT TO JOIN THE LET'S AI SKOOL COMMUNITY

It's coming soon. Join the waitlist below...

Join the waitlist.

Leverage the power of AI. Let's AI together.

    Row wavy Shape Decorative svg added to bottom
    🚀 Get it all

    Amazing Prompts Databases

    1. Prompt Generators

    ✅ 100's of Engineers
    ✅ Lifetime Access
    ✅ Nonstop Updates
    ✅ Generate for Anything
    ✅ AI Image Generators
    ✅ 10X Prompt Upgrades

    Collection of the best ai image prompts

    2.Prompts Portal

    ✅ Thousands of Prompts
    ✅ AI Images
    ✅ Every Category
    ✅ Immaculately Organized
    ✅ Nonstop Updates
    ✅ Lifetime Access

    4. Everything Bundle

    ✅ All Databases (20+ products)
    ✅ Lifetime Access
    ✅ Nonstop Updates
    ✅ Meta Tokens
    ✅ 20,000+ Prompts
    ✅ Never Need Prompts Again

    Shopping Cart
    • Your cart is empty.
    Scroll to Top