Starpop Logo
Starpop

How to write good Seedance 2.0 Prompts?

April 19, 2026
·
25 min read
·
DIDavid Ishag
·
Seedance 2.0Seedance 2.0 Prompt Engineering+5
How to write good Seedance 2.0 Prompts?

Contents

0%
Inputs at a Glance
Reference System (@Tags)
1. The Core Formula
2. The Rules That Matter
Do This
Never Do This
3. Camera Movements
4. Shot Types & Camera Pairing
5. Timeline Prompting
Timestamp Format
Beat Distribution by Duration
What Each Beat Must Contain
Transition Phrases Between Beats
Global Style Line
6. Lighting (Biggest Impact)
7. Style & Mood Keywords
By Genre
Color Grade Keywords
Director / Brand Style Anchors
8. Pacing & Motion
Motion Quality Words
Atmospheric Motion
9. Audio Design
Audio References
10. Quality Suffix
11. Constraints
Standard Constraints (use always)
Situation-Specific Constraints
13. Subject Descriptions
14. The Separation Rule
15. Dangerous Keywords
16. Full Prompt Examples
Simple: Product Beauty Shot
Simple: Lifestyle Scene
Simple: UGC Product Review
Simple: Film Noir
Timeline: Thriller Character Reveal (10s)
Timeline: Product Commercial (10s)
Timeline: Cinematic Restaurant Scene (15s)
17. Troubleshooting
18. Iteration Strategy
The Process
Efficiency Tips
19. Platform-Specific Tips
Next Steps

Hey!
Seedance 2.0, the latest AI video model, gives you cinema-level control over every shot, but only if you prompt it right. This is the full reference I use to get clean, consistent, ad-ready video out of it: the core formula, every camera move, lighting setup, audio direction, the @-tag system, and the constraints you need on every prompt.

Inputs at a Glance

Before getting into the prompting itself, here's what Seedance 2.0 will accept as input and produce as output. These limits shape every other decision in this guide.

Images
Up to 9
JPEG, PNG, WebP, BMP, TIFF, GIF — under 30MB each
Videos
Up to 3 clips
MP4, MOV — 2-15 seconds each
Audio
Up to 3 clips
MP3, WAV — under 15 seconds, under 15MB
Total files
Max 12
Across all input types combined
Output duration
4-15 seconds
Single generation length
Output resolution
Up to 2K
Full HD by default, 2K supported
Prompt length
30-200 words
60-100 words is the sweet spot

Reference System (@Tags)

Upload up to 9 images, 3 videos (15s total), and 3 audio clips, then reference them inline with @ tags. Each tag assigns a role to a specific asset, locking the model onto the appearance, movement, voice, or style you want it to use.

Image 1
@image1
Image 2
@image2
Image 3
@image3
Use CaseSyntaxWhat It Does
Character referenceThe woman in @image1 as the lead characterLocks in appearance, face, clothing
First frame@image1 as the first frameVideo starts from this exact composition
Product + SubjectProduct in @image1, subject in @image2Assigns roles to different images
Mood / styleMatch the mood and color palette of @image3Visual style reference
Camera replicationReplicate the camera movement from @video1Copies movement patterns
Action mimicryMimic the actions from @video1Copies physical movements
Background music@audio1 as background musicUses audio as soundtrack
Voice referenceUse @audio1 as voice style referenceMatches vocal tone/energy
Lip syncLip-sync to @audio1Character mouths match audio
Video extensionExtend @video1 by 5 secondsContinues existing footage

Character consistency tip: Upload a reference image and use @image1 as the character reference in your prompt. Then describe the character identically in every beat. Don't switch between "a man", "the detective", "him". Use the exact same noun.

1. The Core Formula

Every Seedance 2.0 prompt should follow this structure. The model reads left to right and gives the strongest weight to whatever comes first. Put your most important instruction at the beginning.

The Formula

Subject + Action + Scene + Camera + Style + Constraints

Subject

Who or what is in the frame. Be specific about age, hair, clothing, expression, skin tone. Use the SAME noun throughout the entire prompt. If you say "a man" at the start, don't switch to "the detective" or "he" later.

Action

One clear verb in present tense per shot. "She walks slowly toward the window" rather than "she walks, turns around, picks up the cup, looks at it, and drinks." Multiple verbs in one shot equals a blended mess.

Scene / Environment

Where it takes place, with specific details. "A modern kitchen with marble countertops, morning light through floor-to-ceiling windows, a glass of water on the counter" rather than just "a kitchen."

Camera (ONE instruction only)

This is the most common mistake. Use exactly ONE camera movement per shot. "Slow dolly push-in" or "fixed camera" rather than "dolly in while panning left and tilting up." Multiple camera moves equals jitter and incoherent output.

Style

A single strong visual anchor works better than ten adjectives. "35mm film, Kodak color palette" or "Apple keynote style" or "Wes Anderson symmetry." These give the model a clear target. "Beautiful, amazing, cinematic, epic" gives it nothing.

Constraints

What to avoid. Seedance does NOT support traditional negative prompts. Use positive constraint statements instead. "Avoid jitter. Avoid bent limbs. Face stable, no deformation." Add these to EVERY prompt.

Word count

60-100 words is the sweet spot. Under 30 lacks information. Over 200 and the model starts ignoring details. The first 2-3 instructions are followed best. After about 8 requirements, only 4-5 will land.

2. The Rules That Matter

Do This

  • One shot, one verb, one camera move
  • Use specific film terms ("slow dolly in, 1-2 feet")
  • Keep the subject noun identical throughout
  • Put the most important instructions first
  • Keep prompts to 60-100 words
  • Include constraint words in every single prompt
  • Append the quality suffix to everything
  • Start with 5-second test clips before going longer
  • Generate 2-4 variants and compare
  • Change only ONE variable when iterating
  • Use gentle motion words: slow, smooth, natural, gentle
  • Limit scenes to 1-2 characters maximum
  • Describe lighting specifically (it has the biggest impact)
  • Reference directors/brands for instant style ("Fincher framing")
  • Separate subject motion from camera motion

Never Do This

  • Multiple camera movements in one shot
  • Vague adjectives ("beautiful", "epic", "amazing", "cool")
  • Technical photography specs (f/2.8, ISO 800, 85mm)
  • Fast + complex + intricate all at once
  • Switch subject nouns mid-prompt
  • Use traditional negative prompts (not supported)
  • Skip constraint statements
  • Reference long videos (trim to key segments first)
  • Change multiple parameters when debugging
  • More than 2 characters (consistency breaks down)
  • Overload timestamps with multiple actions
  • Say "multiple angles" (causes chaos)
  • Mix subject movement and camera movement in one instruction
  • Use "lots of movement" (instant jitter)
  • Write prompts over 200 words
Priority rule

The model follows the first 2-3 instructions reliably. After about 8 specific requirements, it hits roughly 4-5. Put your most critical direction first.

3. Camera Movements

Use exactly ONE per shot. Always pair with a speed modifier (slow, medium, fast) and ideally a distance (1-2 feet). The modifier pattern is: speed + movement + distance.

MovementKeywordsWhat It DoesBest For
Push inslow dolly-in, push in, slow push forward 1-2 feetCamera moves toward subjectBuilding tension, intimacy, emphasis on details
Pull backdolly-out, pull back, slow pull outCamera retreats to reveal wider sceneEnvironmental reveals, showing scale, endings
Pangentle pan left, pan right, slow panCamera rotates horizontally on its axisScanning a scene, following horizontal action
Tilttilt up, tilt down, slow tilt up from groundCamera rotates vertically on its axisRevealing tall subjects, establishing shots
Trackingtracking shot, track left/right, lateral trackingCamera moves sideways alongside subjectFollowing a walking/running subject, action scenes
Orbitorbit, arc shot, 360 orbit, circle halfwayCamera circles around the subjectProduct showcase, portrait emphasis, hero moments
Cranecrane up, crane down, jib upCamera rises or descends on vertical axisGrand reveals, establishing shots, transitions
Aerialaerial sweep, drone shot, bird's eye viewHigh-altitude perspective looking downLandscapes, cityscapes, showing geographic scale
Handheldhandheld, handheld slight shake, organic handheldSubtle natural camera shakeDocumentary realism, urgency, intimate moments
Fixedfixed camera, static shot, locked off, tripod stableCamera does not move at allLetting action carry the scene, contemplation, products
Steadicamsteadicam, gimbal, smooth stabilizedPolished, gliding motion following subjectWalk-and-talk scenes, interior tours, polished ads
Rack focusrack focus, shift focus foreground to backgroundFocus shifts between planesDrawing attention, reveals, transitions between subjects
Dolly zoomdolly zoom, vertigo effectCamera moves while zooming opposite directionDisorientation, dramatic realization moments
Whip panwhip pan, swift horizontal panExtremely fast horizontal camera movementScene transitions, urgency, comedic timing
The #1 mistake

Using multiple camera movements in one shot. Pick ONE. If you need a different angle, make it a new timestamp beat.

4. Shot Types & Camera Pairing

Shot TypeDescriptionBest Camera PairingAvoid
Extreme close-up (ECU)Fills frame with one detail, eyes, lips, product label, textureFixed camera, micro push-inPanning (feels disorienting at this distance)
Close-up (CU)Face fills most of frame, shows emotionFixed, micro push-in, rack focusFast tracking, orbit
Medium shot (MS)Waist up, shows body language and environmentHandheld (intimate), gimbal (polished)Aerial, extreme movements
Medium wide (MW)Knees up, shows more environment contextTracking, slow pan, steadicamExtreme close movements
Wide shot (WS)Full body + environment, establishing contextSlow push-in, static, craneHandheld (too shaky for wide framing)
Extreme wide (EWS)Landscape-dominant, subject is small in frameAerial sweep, static, slow panDolly (too subtle to read at this scale)
Over-the-shoulder (OTS)Looking over one person's shoulder at anotherFixed, subtle push-inOrbit, tracking
Low angleCamera below subject looking up, creates powerFixed, slow tilt upHandheld (unstable at low angles)
High angleCamera above subject looking down, creates vulnerabilityFixed, slow crane downTracking (perspective gets confusing)
Dutch angleCamera tilted on its axis, creates uneaseFixed, slow push-inMultiple movements (amplifies disorientation too much)

5. Timeline Prompting

Timeline prompting is the technique that separates random clips from actual video sequences. You mark what should happen at each point in time using bracket timestamps. Each timestamp represents one clear beat, establish, develop, or land.

Timestamp Format

Use bracket notation with second markers. Keep each beat to 2-3 sentences, 50-75 words maximum.

[0s] First beat — establishment. Set the scene, introduce the subject.
[3s] Second beat — development. Action begins, camera moves.
[6s] Third beat — shift. Something changes, new angle or emotion.
[8s] Fourth beat — resolution. Final hold, reveal, or exit.

Beat Distribution by Duration

Video LengthNumber of BeatsTimestamp Placement
4-5 seconds2-3 beats[0s], [2s], [4s]
7 seconds3-4 beats[0s], [3s], [5s]
10 seconds4 beats[0s], [3s], [6s], [8s]
15 seconds5-6 beats[0s], [3s], [5s], [8s], [11s], [13s]

What Each Beat Must Contain

  • Shot type (Wide, Medium, Close-up, etc.)
  • Subject action (One verb describing what happens)
  • Camera movement (One instruction, or "fixed camera")
  • One atmospheric detail (Lighting, sound, texture)

Transition Phrases Between Beats

  • Cut to: for a hard cut to a new angle
  • Slow push in begins for camera starts moving
  • Camera begins tracking right for lateral motion
  • Rack focus from background to subject for a focus shift
  • Hold. for camera stops, moment lingers
Overloaded beats

Stuffing too much into one timestamp is the #1 timeline mistake. One action per beat.

Global Style Line

After your timeline beats, add one closing line that sets the overall aesthetic. This acts as a style anchor for the entire video:

Cinematic 4K, 35mm film grain, shallow depth of field, warm color grade. Avoid jitter, bent limbs, identity drift.

6. Lighting (Biggest Impact)

Lighting has the single biggest impact on output quality. One specific lighting line outperforms ten generic adjectives. Always describe it.

Lighting TypeKeywordsMood / FeelBest For
Golden hourwarm golden hour sunlightWarm, romantic, dreamyLifestyle, romance, product beauty shots
Blue hourblue twilight, cool tonesMelancholy, calm, mysteriousThriller, emotional scenes, night establishing
Natural windowsoft natural window light from the rightRealistic, casual, authenticUGC, interior scenes, morning routines
Studio soft boxclean studio lighting, soft box from leftPolished, controlled, professionalProduct ads, portraits, talking heads
Candlelightwarm candlelight, flickering, low keyIntimate, romantic, warmDinner scenes, romance, luxury
Rim lighthard rim lightDramatic, separating subject from backgroundHero shots, dramatic reveals, silhouettes
Backlitbacklit subject, silhouetteMystery, drama, high contrastCharacter introductions, noir, fashion
RembrandtRembrandt lightingClassic portrait, triangle shadow on cheekPortraits, interview setups, dramatic close-ups
Neonneon reflections on wet pavementUrban, cyberpunk, energeticNight city, sci-fi, club scenes
Volumetricvolumetric light, god rays through fogEpic, atmospheric, cinematicForest scenes, churches, abandoned spaces
Overcastovercast diffused light, no harsh shadowsEven, neutral, documentaryOutdoor scenes needing consistent lighting
Chiaroscurohard chiaroscuro, deep shadows, high contrastDramatic, noir, intenseThriller, horror, dramatic portraits
Practicalpractical tungsten lighting, desk lamp onlyRealistic, grounded, intimateHome interiors, late night scenes, realism
Temperature modifier

Adding color temperature: warm or color temperature: cool gives the model strong overall color direction with just three words.

7. Style & Mood Keywords

A single strong visual reference ("Wes Anderson symmetry") works better than a stack of vague adjectives ("beautiful cinematic epic amazing"). Anchor your style to something the model can actually target.

By Genre

GenreKeywords That Work
Cinematic / Filmcinematic texture, film grain, 35mm, anamorphic lens flare, Kodak color palette
Premium / Commercialminimalist clean, premium texture, studio photography, commercial polish
Film Noirfilm noir, chiaroscuro, high contrast, deep shadows, black and white
Sci-Fi / Cyberpunkcyberpunk, neon, dark premium, holographic, retro-futuristic
Lifestyle / Vloghealing fresh, natural, authentic, soft focus, warm tones
Dreamy / Etherealdreamy soft light, ethereal, pastel, soft focus, lens flare
Vintageretro film stock, analog, VHS grain, faded colors, 1970s aesthetic
Documentarydocumentary style, naturalistic, observational, handheld, no staging
Fashion / EditorialVogue editorial, high fashion, dramatic pose, striking lighting
Horrorunsettling, slow dread, desaturated, cold blue undertones

Color Grade Keywords

GradeKeywordsFeel
Teal and orangeteal and orange color gradeHollywood blockbuster standard
Bleach bypassbleach bypass, desaturated, grittyRaw, intense, war film
Warm shadows / cool highlightswarm shadows with cool highlightsBalanced cinematic look
Desaturated colddesaturated with cold blue tonesThriller, dystopian
High saturation warmhigh saturation, warm tones, vibrantLifestyle, summer, energy
Monochromemonochrome, black and white, silver tonesArt film, timeless, dramatic

Director / Brand Style Anchors

  • Wes Anderson symmetry for centered framing, pastel colors, whimsical
  • Fincher-level precision for dark, controlled, every frame intentional
  • Apple keynote style for clean, minimal, premium product focus
  • Euphoria color grade for neon, saturated, youthful, dramatic
  • Terrence Malick for golden hour, nature, poetic, magic hour
  • Roger Deakins lighting for masterful natural light, precise shadows

8. Pacing & Motion

Speed TierKeywordsRisk LevelWhen to Use
Ultra slowimperceptible, barely movingSafestMacro shots, product close-ups, atmospheric
Slowslow, gentle, gradual, smoothSafeMost scenes, recommended default
Mediumcontrolled, steady, natural paceModerateWalking scenes, conversation, casual movement
Fastdynamic, swift, rapidHigh riskUse sparingly. Only ONE element should be fast. Never fast camera + fast action together.

Motion Quality Words

  • silky smooth for premium, polished feel
  • organic for natural, human-like movement
  • deliberate for intentional, purposeful motion
  • floating for weightless, ethereal movement
  • mechanical for precise, robotic, inhuman

Atmospheric Motion

  • steam rising slowly, dust particles in light, fog drifting
  • rain hitting window, leaves falling, hair blown by breeze
  • candle flame flickering, water rippling, heat haze

9. Audio Design

Seedance 2.0 generates synchronized audio. You can describe the sound environment and reference uploaded audio files.

Audio TypeKeywordsEffect
Room tonereverb, echoingLarge-space resonance (cathedral, parking garage)
Muffledmuffled, behind glassEnclosed, underwater, distorted
Impactmetallic clink, glass shatterCollision, breakage, contact sounds
Texturecrunchy, gravel, fabric rustleFootsteps, clothing, surface interaction
Ambientcrackling fire, rain on window, distant trafficEnvironmental atmosphere
Tonehigh-pitched, deep rumble, sharpSpecific sound character
Digitaldigital screech, glitch artifacts, static buzzTech malfunction, sci-fi, horror
Silencecomplete silence, all sound cutsDramatic pause, tension, reveal moment

Audio References

  • @audio1 as background music uses the audio as soundtrack
  • Use @audio1 as voice style reference matches tone/energy without playing the audio
  • Lip-sync to @audio1 makes the character's mouth match the audio track

10. Quality Suffix

Append this to the end of every single prompt:

4K, Ultra HD, rich details, sharp clarity, cinematic texture, natural colors, soft lighting, no blur, no ghosting, no flickering, stable picture.

Copy-paste this. It works. Don't modify it. Don't leave it out.

11. Constraints

Seedance 2.0 does NOT support traditional negative prompts. You must use positive constraint statements instead.

Standard Constraints (use always)

Avoid jitter. Avoid bent limbs. Face stable, no deformation. Natural smooth movements. Stable picture.

Situation-Specific Constraints

ScenarioAdd This Constraint
Any videoAvoid jitter. Stable picture.
Videos with peopleAvoid bent limbs. Face stable without deformation. Normal human structure.
Videos over 10 secondsAvoid temporal flicker. Maintain consistency throughout.
Multi-shot / timelineAvoid identity drift. Consistent appearance across all beats.
No text overlaysGenerate video without subtitles or text overlays.
Product shotsLogo sharp, no warping, no distortion on product surface.
Walking / runningNatural gait, feet touching ground, no floating.

13. Subject Descriptions

Vague subject descriptions produce vague results. Specific details give the model something concrete to render.

  • Weak: "A businesswoman"

  • Strong: "A woman in her early 30s, dark navy blazer, dark hair in a loose bun, serious expression, carrying a briefcase"

  • Weak: "A perfume bottle"

  • Strong: "A glass perfume bottle with a gold cap, amber-colored liquid, clean white background, professional product photography lighting"

  • Weak: "An alley at night"

  • Strong: "A narrow cobblestone alleyway in an old European city, stone walls with climbing ivy, warm evening light from a small window above, wet pavement reflecting light"

14. The Separation Rule

Subject movement and camera movement must be described as separate, independent instructions. Mixing them causes the model to blend both into one confused motion.

  • Wrong: "Spinning camera around a dancing person"

  • Right: "The dancer spins slowly in the center of the room. Camera holds fixed framing."

  • Wrong: "Camera follows him running through the forest while zooming in"

  • Right: "A man runs through a dense forest, branches brushing past him. Tracking shot following from the side."

15. Dangerous Keywords

Dangerous WordWhy It FailsReplace With
fast (alone)Creates total chaos. Model speeds up everything simultaneouslyMake only ONE element fast. "Fast subject movement, slow camera."
cinematic (alone)Too vague, means nothing specific to the modelcinematic film tone, 35mm, warm shadows
epicModel doesn't understand this conceptDescribe the specific visual you want. wide shot, sweeping crane, dramatic lighting
beautifulProvides zero actionable guidanceDescribe the specific beauty. lighting, composition, colors
amazingNo practical meaningWhat specifically is amazing about it? The lighting? The movement? Describe that.
lots of movementInstant jitter, everything moves at onceDescribe one specific motion with a speed modifier
multiple anglesCreates chaotic, unusable outputsingle tracking shot. Use timeline beats for different angles
realistic (alone)Too broadphotorealistic, natural skin texture, practical lighting

16. Full Prompt Examples

Simple: Product Beauty Shot

Matte black luxury watch on a velvet stand. Smooth gimbal orbit right to left, waist height. Soft studio lighting, cool-toned, crisp reflections on brushed metal surface. High-end product photography, clean background. Logo sharp, no warping, tripod-stable. 4K, Ultra HD, rich details, cinematic texture, no blur, stable picture.

41 words. One camera move (orbit). One subject. One lighting direction. Constraints included.

Simple: Lifestyle Scene

A woman in a white linen dress walking slowly through a sunlit lavender field. Golden hour lighting from behind. Slow tracking shot following from the right. Shallow depth of field, warm color grade, 35mm film grain. Avoid jitter, face stable. 4K, rich details, sharp clarity, stable picture.

48 words. Subject + Action + Scene + Camera + Style + Constraints.

Simple: UGC Product Review

A woman in casual clothes standing in a bright modern kitchen, holding a green supplement pouch toward camera. She smiles and speaks naturally. Medium shot, slow dolly push-in. Natural window light from the left, iPhone selfie style, authentic energy. Face stable, no deformation. 4K, sharp clarity, stable picture.

50 words. UGC feel achieved through "iPhone style" and "authentic energy" rather than by being vague.

Simple: Film Noir

A woman in a red dress walking confidently down a rainy city street at night. Neon signs reflect in wet pavement puddles. Tracking shot following from behind. Film noir, harsh chiaroscuro, deep shadows. Maintain face consistency. 4K, cinematic texture, no jitter, stable picture.

44 words. The style is carried by "film noir, chiaroscuro", two words that do more than twenty adjectives.

Timeline: Thriller Character Reveal (10s)

A figure in a long dark coat on an empty rain-slicked city street at night. High contrast, neon reflections in puddles. 35mm film grain, desaturated cold blue tones, anamorphic lens flare from streetlights. Avoid jitter, identity drift.

[0s] Wide shot. Figure stands at the end of the street. Static camera, framed from behind. Neon signs reflect in puddles. [3s] Slow dolly forward begins. Closing in on the figure from behind. Rain in foreground, shallow depth of field, bokeh streetlights. [6s] Medium shot. Figure turns head slightly, profile barely visible. Street light catches jaw. Tension holds. [8s] Rack focus. Background city blur sharpens briefly, then returns to subject. Hold.

112 words total.

Timeline: Product Commercial (10s)

Glass bottle on marble surface. Backlit, product in sharp focus, background completely blurred. Clean, minimal. Commercial photography aesthetic, 4K, no film grain, clinical but warm. Color grade: slightly warm whites, deep clean shadows. Avoid jitter.

[0s] Extreme close-up. Product centered, backlit. Cool clean lighting. Fixed camera. [2s] Slow pull back. Framing widens to reveal full bottle and a single green branch beside it. Precise, no handheld shake. [5s] Arc shot begins right to left. Camera slowly orbits product at waist height. Light catches glass at different angles. [8s] Camera settles into medium shot. Static. Product centered. Hold.

98 words. Detail, then context, then orbit showcase, then hold.

Timeline: Cinematic Restaurant Scene (15s)

Man in dark suit and woman in black dress at candlelit restaurant. Warm amber lighting, white linen, crystal glassware, deep bokeh. 35mm film, shallow depth of field. Avoid jitter, bent limbs, identity drift.

[0s] Extreme close-up on the man's eyes. Candlelight reflected. Fixed camera. He speaks quietly: "You're different from anyone I've ever met." [3s] Slow crane pull-back reveals the full table. Two people, candles, wine glasses. She smiles. Warm golden light. [5s] Medium two-shot. He leans across the table toward her ear. Slow dolly push-in to tight profile. [8s] Close-up on her face. Her expression freezes. Head jerks sharply. Digital artifacts across her skin. Eyes flash white. [11s] Medium wide shot. He sits back. Looks at camera. Small smirk. Picks up wine glass. Sips. [13s] She freezes completely. Eyes open glowing blue. Head rotates slowly toward camera. Blue light fills frame.

148 words. 6 beats across 15 seconds.

17. Troubleshooting

ProblemRoot CauseFix
Video is completely staticNo motion describedAdd one specific movement: "steam rising slowly from a cup, subtle camera drift right"
Jittering, shaky outputMultiple camera moves or "fast" keywordUse ONE camera instruction. Add "avoid jitter". Replace "fast" with "controlled".
Character looks different across beatsInconsistent subject description or no reference imageUse @image reference + identical description in every beat. Same noun throughout.
Flat, muddy lightingNo lighting direction in promptAdd one specific light source: "soft side lighting from the left, warm tone"
Wrong composition, correct actionShot size not specifiedAdd shot type: "Close-up", "Wide shot", "Medium shot". Keep action the same.
Style/color driftToo many competing style wordsReplace with one strong visual anchor: "35mm Kodak film" instead of five adjectives.
Bent or extra limbsComplex character poses without constraintsAdd "avoid bent limbs, normal human structure". Simplify the pose.
Chaotic multi-angle output"Multiple angles" in promptReplace with "single tracking shot". Use timeline beats for angle changes.
Subject doesn't match descriptionToo abstract or conflicting languageUse concrete details: "white cylindrical object with blue LED ring" not "futuristic gadget"
Model ignoring half the instructionsPrompt too long (over 200 words)Cut to under 100 words. Front-load the 3 most important instructions.
Text/subtitles appearingModel generating text overlaysAdd "generate video without subtitles or text overlays"
Temporal flicker (long videos)Consistency breaks over timeAdd "avoid temporal flicker". Consider generating shorter clips and stitching.

18. Iteration Strategy

The Process

Start short

5-second test clip with your core concept. Validate direction before going longer.

Generate 2-4 variants

Same prompt, let the model explore different interpretations. Pick the best one.

Change ONE variable

If the lighting is wrong, fix ONLY the lighting. Changing multiple things makes it impossible to know what helped.

Front-load what matters

If the model keeps getting the lighting wrong, move the lighting instruction to the very first line.

Extend when confident

Once a 5s clip looks right, extend to 10s or 15s with the same prompt structure.

Efficiency Tips

  • Tight close-ups produce more consistent output than complex wide shots
  • Reuse successful lighting/style phrases across generations
  • 4-5 second clips often outperform stretched 10-second versions
  • Shorter videos with structured prompts beat longer vague ones every time

19. Platform-Specific Tips

PlatformAspect RatioAdd to PromptTips
TikTok9:16vertical format, 9:16Hook in first 2 seconds. Close-ups and medium shots work best. High energy.
Instagram Reels9:16vertical format, 9:16Slightly more polished than TikTok. Can be slower paced. Strong visuals.
YouTube Shorts9:16vertical format, 9:16Longer hooks OK (3-4 seconds). Can be more informational.
YouTube16:9cinematic widescreen, 16:9Default landscape. Wide shots and cinematic framing shine here.
Instagram Feed1:1 or 4:5square format, 1:1Centered compositions. Product in frame center. Clean backgrounds.
Facebook16:9 or 1:1VariesMixed formats. Slightly older audience. Subtitles recommended.

Next Steps

This guide gives you the full scaffolding: the formula, the camera vocabulary, the lighting menu, the constraint suffixes, and the @-tag system. The fastest way to get good is to copy one of the prompt examples above, swap in your own subject and references, and run a 5-second test before going longer.

When you're ready to generate, you can run Seedance 2.0 inside Starpop with all 9 image, 3 video, and 3 audio reference slots available, plus our prompt input that auto-completes the @-tag syntax for you. Join the Starpop Discord to share what you build.

Starpop

Generate viral high-converting AI ads in minutes with Starpop

Join us on Discord

Contents

0%
Inputs at a Glance
Reference System (@Tags)
1. The Core Formula
2. The Rules That Matter
Do This
Never Do This
3. Camera Movements
4. Shot Types & Camera Pairing
5. Timeline Prompting
Timestamp Format
Beat Distribution by Duration
What Each Beat Must Contain
Transition Phrases Between Beats
Global Style Line
6. Lighting (Biggest Impact)
7. Style & Mood Keywords
By Genre
Color Grade Keywords
Director / Brand Style Anchors
8. Pacing & Motion
Motion Quality Words
Atmospheric Motion
9. Audio Design
Audio References
10. Quality Suffix
11. Constraints
Standard Constraints (use always)
Situation-Specific Constraints
13. Subject Descriptions
14. The Separation Rule
15. Dangerous Keywords
16. Full Prompt Examples
Simple: Product Beauty Shot
Simple: Lifestyle Scene
Simple: UGC Product Review
Simple: Film Noir
Timeline: Thriller Character Reveal (10s)
Timeline: Product Commercial (10s)
Timeline: Cinematic Restaurant Scene (15s)
17. Troubleshooting
18. Iteration Strategy
The Process
Efficiency Tips
19. Platform-Specific Tips
Next Steps
Starpop

Generate viral high-converting AI ads in minutes with Starpop

Join us on Discord

Grow Your Business with AI Content Today.

Generate viral high-converting AI ads in minutes

Scale your content marketing effortlessly

David Ishag

David Ishag

Co-Founder

Alex Le

Alex Le

Co-Founder

Starpop helps businesses create authentic AI-generated user content that drives engagement and sales. Transform your content strategy with AI-powered UGC that actually converts.

© 2025 Starpop

Become an AffiliatePrivacy policyTerms of service

Compare

Starpop vs ArcadsStarpop vs JoggAIStarpop vs MagicUGCStarpop vs MakeUGCStarpop vs TopView

Free tools

All Free ToolsTikTok Money CalculatorInstagram Engagement CalculatorTikTok Engagement CalculatorYouTube Engagement CalculatorAspect Ratio CalculatorVideo Length CalculatorSocial Media Ad Specs

Other tools

AI Script Writer - AI Flow Chat