Seedance 2.0 prompt library

Seedance prompts that actually work

A practical library of Seedance 2.0 prompts you can copy, run, and adapt. Each prompt is grouped by use case, paired with recommended parameters, and built on a repeatable 5-part formula so your results stop feeling random.

By Jay Yang·AI Video Technology·11 min read·

Seedance 2.0 prompt facts

Model

Seedance 2.0 (ByteDance)

Modalities

Text-to-video, Image-to-video

Recommended prompt length

12–35 words

Max single-shot duration

12 seconds

Native audio

Yes (sync audio with visuals)

Aspect ratios

16:9 · 9:16 · 1:1

Multi-shot prompts

Supported via shot list syntax

Working language for prompts

English (recommended), Chinese

Sources: ByteDance Seed launch post (Feb 12, 2026), Volcengine Ark documentation. Verified May 2026.

Writing formula

The 5-part Seedance prompt formula

A working Seedance 2.0 prompt has five slots in this order: Subject, Action, Style and mood, Camera and motion, Lighting and atmosphere. You do not need every slot every time, but skipping more than one usually hurts output stability — most weak Seedance results trace back to a missing slot.

  1. 1

    SubjectWho or what is on screen?

    Name the main subject concretely. Add one or two distinguishing visual attributes so Seedance does not default to a generic look.

    Good

    A middle-aged barista with a short grey beard and a denim apron

    Avoid

    A person making coffee

  2. 2

    ActionWhat is happening?

    Use one primary verb and a clear object. Stack actions sparingly; two short actions are stable, three or more often produce drift.

    Good

    pours steamed milk into a ceramic cup

    Avoid

    doing things in the cafe

  3. 3

    Style & moodWhat does this look and feel like?

    Anchor style with a recognizable reference (cinematic, documentary, anime, claymation) plus a mood word. Avoid stacking five style words — they cancel each other out.

    Good

    cinematic, warm, intimate

    Avoid

    beautiful epic stunning detailed 8k masterpiece

  4. 4

    Camera & motionHow does the camera move?

    Specify shot type (close-up / medium / wide) and one camera motion (slow dolly-in, handheld, static, orbit). One motion per prompt is more stable than two.

    Good

    medium close-up, slow dolly-in

    Avoid

    camera moves around fast

  5. 5

    Lighting & atmosphereWhat is lighting the shot?

    Name the light source and time of day. Atmosphere words (steam, dust, rain) act as motion seeds Seedance can animate naturally.

    Good

    golden-hour window light, soft steam rising

    Avoid

    good lighting

Putting all five together: "A middle-aged barista with a short grey beard and a denim apron pours steamed milk into a ceramic cup, cinematic warm intimate, medium close-up slow dolly-in, golden-hour window light with soft steam rising." That is a complete, runnable Seedance prompt.

Parameter cheatsheet

Seedance 2.0 generator parameters

Seedance 2.0 has five generator parameters that control the output container: duration (3–12 s), aspect ratio (16:9 / 9:16 / 1:1), motion strength (low / medium / high), seed, and resolution (720p / 1080p). Prompt text controls content; these parameters control the container. Match the parameter to the use case before iterating on the prompt — otherwise you will rewrite words to fix something the parameters should fix.

ParameterValuesWhen to changeRecommended default
Duration3s / 5s / 8s / 10s / 12sLonger clips for narrative shots; shorter for ad hooks and stock cutaways.5s
Aspect ratio16:9 · 9:16 · 1:19:16 for TikTok / Reels / Shorts; 1:1 for feed ads; 16:9 for landing pages and YouTube.9:16 for social, 16:9 for web
Motion strengthlow / medium / highLow for product macro shots and talking-head; medium for most narrative; high for action and transitions.medium
Seedinteger or randomLock a seed to reproduce a winning result while iterating on prompt wording.random for exploration, lock once a take works
Resolution720p / 1080p1080p for paid social and web hero; 720p for rapid creative testing to save credits.720p draft, 1080p final

Prompt library

30 working Seedance prompts

Below are 30 working Seedance 2.0 prompts grouped into six categories: cinematic, product, character, anime, B-roll, and image-to-video. Each prompt is written with the 5-part formula and paired with the parameters we recommend. Click "Try this" to load any prompt directly into the generator, or copy and paste the text into your own workflow.

Category 01

Cinematic & narrative

5 prompts

Single-shot cinematic prompts for hero reels, brand stories, and short-film concept frames. Favor medium-low motion and one named camera move.

Slow dolly-in on a barista pour

5s9:16motion: low

Cafe brand reel hero shot

A middle-aged barista with a short grey beard and a denim apron pours steamed milk into a ceramic cup, cinematic warm intimate, medium close-up slow dolly-in, golden-hour window light with soft steam rising.

Rain-soaked neon street, person walks past

8s16:9motion: medium

Brand atmosphere intro for fashion or tech

A figure in a charcoal trench coat walks past a wet asphalt crosswalk, cinematic noir moody, wide shot static camera, neon shop signs reflecting in puddles, light rain and faint mist in the air.

Lone hiker crests a ridge at dawn

8s16:9motion: medium

Outdoor or insurance brand opening shot

A solo hiker with a red backpack steps onto a rocky ridge as the valley opens below, cinematic epic contemplative, wide drone shot slow rise, cool dawn light with low clouds drifting across distant peaks.

Chef slices fresh herbs in a sunlit kitchen

5s9:16motion: low

Restaurant or cookware brand reel

A chef in a white linen apron slices fresh basil on a wooden board, cinematic warm tactile, overhead close-up slow pan, late-morning side light from a tall window with subtle dust motes in the air.

Driver at the wheel, city lights streaking past

8s16:9motion: medium

Automotive or fintech narrative spot

A young woman with quiet focus grips the steering wheel, city lights streaking across the windshield, cinematic moody intimate, medium close-up subtle handheld, ambient blue dashboard glow with passing amber streetlights.

Category 02

Product & e-commerce

5 prompts

Macro and turntable shots for PDP videos, ad hooks, and Shopify hero modules. Favor low motion strength and short durations to keep product features legible.

Luxury watch slow turntable

5s1:1motion: low

PDP hero loop for a watch listing

A polished stainless-steel automatic watch on a matte black turntable, premium product photography clean minimalist, extreme close-up slow 360 orbit, single key softbox from camera-left with subtle rim light.

Serum drop into glass dish

3s9:16motion: low

Skincare ad hook, first 3 seconds

A single drop of amber serum falls in slow motion into a clear glass dish, beauty commercial clean luxurious, macro close-up locked-off camera, soft top light with crisp specular highlights and gentle ripple in the liquid.

Sneaker on rotating pedestal, studio

5s1:1motion: low

DTC sneaker drop launch loop

A white-and-coral running sneaker rotates on a matte concrete pedestal, sport editorial clean energetic, low-angle medium shot 360 orbit, twin softboxes from above with a single colored gel kicker on the heel.

Headphones rise from darkness into a key light

5s9:16motion: low

Audio brand teaser or pre-order ad

A pair of matte-black over-ear headphones rises slowly out of darkness into a single beam of light, premium product cinematic dramatic, close-up slow vertical pull-up, hard key light from above with deep falloff into pure black background.

Sauce drips across a stacked burger

3s9:16motion: medium

QSR or food-delivery ad hook

Glossy bbq sauce drips slowly down the side of a stacked cheeseburger on a slate plate, food commercial appetizing rich, macro close-up subtle push-in, warm key light from camera-right with sesame seeds catching the highlight.

Category 03

Character & portrait

5 prompts

Talking-head and emotive portrait prompts. Keep motion low so faces remain stable across the clip, and lean on lighting words to set mood.

Founder talking-head, soft window light

8s16:9motion: low

About page hero or LinkedIn ad

A woman in her thirties with shoulder-length dark hair, in a charcoal sweater, speaks calmly toward camera, documentary natural sincere, medium close-up static camera, soft north-facing window light from camera-left.

Elderly man laughs at a joke off-camera

5s16:9motion: medium

Lifestyle B-roll for healthcare or retirement brands

An elderly man with a white beard and a wool flat cap laughs warmly at someone off-screen, cinematic candid joyful, medium shot slight handheld, late-afternoon backlight through autumn leaves.

Child concentrating on a watercolor painting

5s1:1motion: low

Education or family-app marketing B-roll

A six-year-old in a blue smock concentrates on a watercolor painting, lower lip caught between teeth, documentary tender quiet, top-down close-up locked-off camera, soft cool window light with paint pots scattered around.

Runner gasps for breath at the finish line

5s16:9motion: medium

Sportswear or energy-bar campaign hero

A runner in her twenties leans forward, hands on knees, sweat on her temples, breathing deeply after a hard effort, cinematic raw triumphant, medium close-up subtle push-in, harsh midday sun with track lines visible behind.

Pianist plays in a dim concert hall

8s16:9motion: low

Arts venue or streaming-service promo

A pianist in a charcoal suit plays with closed eyes at a grand piano, cinematic intimate reverent, medium close-up slow side dolly, single warm spotlight from above with the rest of the hall fading into deep black.

Category 04

Anime & stylized

5 prompts

Stylized 2D and stop-motion-look prompts. Anchor with one strong reference style word and avoid mixing too many art directions in one prompt.

Anime girl on rooftop, wind in hair

5s9:16motion: medium

Manga adaptation teaser or music video B-roll

A teenage girl in a navy school uniform stands on a rooftop, hair and skirt drifting in the wind, anime cel-shaded vibrant, medium shot subtle parallax, late-summer evening light with warm orange sky.

Claymation fox trots through forest

8s16:9motion: medium

Kids brand intro or whimsical explainer

A small orange fox figurine trots between mossy stones, claymation handcrafted whimsical, low-angle wide shot slow tracking, dappled morning sunlight filtering through pine canopy.

Cyberpunk alley with holographic signs

5s16:9motion: high

Game trailer or web3 brand intro

A hooded courier on a futuristic motorbike rides through a narrow alley lined with holographic Japanese signs, anime cyberpunk high-contrast, low wide shot tracking from the side, neon magenta and cyan light reflecting off wet pavement.

Watercolor mountain village in soft mist

8s16:9motion: low

Travel app or audiobook cover-art motion

A small mountain village with tile rooftops sits in soft morning mist, watercolor painterly gentle, wide static shot subtle parallax, pale blue and warm cream wash with brush-stroke clouds drifting across the valley.

Pixel-art character runs through a retro arcade

5s16:9motion: high

Indie game launch teaser or branded arcade event

A pixel-art character with a red cap runs left-to-right past glowing arcade cabinets, 16-bit retro vibrant playful, side scrolling medium shot, neon CRT glow and animated scanlines tinting the floor magenta and teal.

Category 05

B-roll & stock

5 prompts

Short, neutral cutaways for editing into existing footage. Favor static or single-direction camera moves so cuts are easy.

Hands typing on a mechanical keyboard

3s16:9motion: low

Editorial or SaaS marketing B-roll

Two hands type quickly on a backlit mechanical keyboard, editorial clean focused, top-down close-up locked-off, cool desk lamp light with subtle keycap reflections.

Coffee beans pour into a hopper

3s16:9motion: medium

Cafe or food-and-beverage cutaway

Dark-roast coffee beans pour from a kraft bag into a steel grinder hopper, documentary tactile rich, close-up slight handheld, warm tungsten overhead light catching the beans mid-fall.

Laptop screen glow on a face at night

3s16:9motion: low

Cybersecurity, fintech, or remote-work editorial

Cool blue laptop screen light flickers across the side of a focused face at night, editorial moody focused, side close-up locked-off camera, only the screen as light source with deep shadow on the far side of the face.

Forest stream tumbles over mossy rocks

5s16:9motion: medium

Wellness, beverage, or outdoor brand cutaway

Clear water tumbles over green moss-covered rocks in a forest stream, nature documentary calm crisp, close-up slow side tracking, dappled morning light through tall pines with fine mist hovering above the surface.

City intersection time-lapse at dusk

8s16:9motion: high

Urban editorial or real-estate marketing

A busy city intersection seen from above with cars and pedestrians flowing in patterns, urban time-lapse vibrant, top-down wide static, dusk-to-night transition with streaking taillights and warming streetlamp glow.

Category 06

Image-to-video extension

5 prompts

Prompts written to extend an uploaded reference image. Describe only what should change — the model treats unmentioned elements as fixed.

Product on shelf — reveal pull-back

5s16:9motion: low

Animate a flat product photo into a hero clip

Camera slowly pulls back from the product on the shelf to reveal a softly lit boutique aisle, premium retail polished, motion limited to camera move only, lighting and product appearance unchanged.

Portrait — subtle breathing and blink

3s1:1motion: low

Bring a static headshot to life for an About page

The subject blinks naturally and breathes gently, head remains still, expression unchanged, no camera move, lighting and background identical to the reference image.

Logo — single light sweep across surface

3s16:9motion: low

Logo sting for video intro or sponsor reveal

A single soft specular highlight sweeps slowly left-to-right across the logo surface, no other motion, logo geometry and color unchanged, background and edge treatment identical to the reference image.

Landscape photo — clouds drift across sky

8s16:9motion: low

Animate a still landscape into a hero loop

Clouds drift slowly from left to right across the sky, foreground terrain and lighting unchanged, no camera move, ground elements completely static, atmosphere and palette identical to the reference image.

Hot dish — soft steam rises from the surface

3s1:1motion: low

Bring a flat food photo to life for menu or DoorDash

Thin wisps of steam rise gently from the surface of the dish, plate and ingredients completely static, no camera move, lighting and color tone identical to the reference image.

Troubleshooting

Common Seedance prompt failure modes — and the fix

Seven recurring causes account for most weak Seedance 2.0 output: subject morphing, generic look, jittery camera, flat lighting, conflicting style words, image-to-video drift, and mistimed audio. Diagnose first, rewrite second — the diagnostic table below maps each symptom to its cause and fix.

Symptom

Subject morphs or face shifts mid-clip

Cause

Prompt stacks two or more conflicting actions, or motion strength is too high.

Fix

Cut to one primary action. Drop motion strength one notch. Lock the seed and re-run.

Symptom

Generic or "stock" looking output

Cause

Subject described too abstractly (e.g. "a person", "a product").

Fix

Add two visual attributes to the subject (age, clothing, material, color).

Symptom

Camera move feels jittery or random

Cause

No camera move named, or two moves named in one prompt.

Fix

Pick exactly one move (slow dolly-in / static / handheld / orbit) and remove the rest.

Symptom

Lighting looks flat or muddy

Cause

No light source named, or contradictory directions ("bright moody dark sunlit").

Fix

Name one light source and one time of day. Remove competing mood words.

Symptom

Style words cancel each other

Cause

Prompt contains five or more style adjectives ("epic stunning detailed cinematic 8k").

Fix

Keep two style anchors maximum (e.g. "cinematic, intimate"). Delete the rest.

Symptom

Image-to-video drifts away from the reference

Cause

Prompt re-describes elements already visible in the reference image.

Fix

Only describe what should change. Add "lighting and appearance unchanged".

Symptom

Audio feels off or mistimed

Cause

Prompt does not describe sound at all, so the model invents.

Fix

Add one short audio cue ("soft steam rising", "light rain on asphalt") that matches the visual.

Negative prompts

What to exclude

Seedance 2.0 reads negative prompts as soft constraints. Keep them short — long negatives often hurt subject quality more than they help.

Faces
distorted face, extra fingers, malformed hands
Text
text on screen, captions, watermarks, logos
Stability
flickering, jittery motion, frame morphing
Style noise
oversaturated, plastic skin, over-sharpened

Portability

Adapting Seedance prompts for Sora 2, Kling 2.1, and Veo 3

A working Seedance prompt usually ports to other top models with two adjustments — wording and parameters. The 5-part formula transfers cleanly.

Target modelWording changeParameter change
Sora 2 (OpenAI)Tighten language, lean on natural sentences. Sora prefers fewer comma-separated style tags.Sora caps clip length differently; reduce duration to its supported tier. No motion-strength flag.
Kling 2.1 (Kuaishou)Translate to Chinese for best results, keep camera + lighting clauses intact.Aspect ratios and durations differ slightly; default to 5–10s, audio not native.
Veo 3 (Google)Veo handles diegetic audio cues well — keep one explicit audio line.Max duration shorter (around 8s); avoid stacking long action lists.

For travel-specific benchmark rows (37 prompts, failure patterns), see Seedance travel prompts lab (2026). For deeper side-by-side comparisons, see Seedance vs Sora, vs Kling, and vs Veo. For Airbnb and short-term rental prompt patterns, see Seedance Airbnb host videos (2026).

People also ask

Quick answers

What is the best Seedance prompt format?
Subject, action, style and mood, camera and motion, lighting and atmosphere — in that order, separated by commas. Twelve to thirty-five words is the most reliable range for Seedance 2.0.
How long should a Seedance prompt be?
Twelve to thirty-five words. Shorter prompts produce generic output; prompts longer than fifty words tend to introduce contradictions that destabilize the clip.
Does Seedance 2.0 support negative prompts?
Yes. Treat them as soft constraints and keep them under ten words. Long negatives can degrade subject quality.
Can I use Chinese prompts on Seedance 2.0?
Yes. Seedance 2.0 was trained heavily on Chinese-language captions and supports Chinese prompts natively. English is also fully supported and is what most prompt libraries publish.
How many words should a Seedance prompt have?
Aim for twenty to thirty words. Below twelve produces generic output; above thirty-five introduces contradictions Seedance has to resolve, which destabilizes motion. The 30-prompt library on this page averages twenty-six words per prompt.
Can Seedance generate text or captions in videos?
Not reliably. Seedance 2.0 can render simple short text on signs and product labels in some cases, but in-frame typography is not its strength. For accurate on-screen text, generate the silent video on Seedance and overlay text in your editor.
What style words work best with Seedance 2.0?
Reference-based style anchors: cinematic, documentary, anime, claymation, watercolor, editorial, food commercial, beauty commercial. Two anchors per prompt is the upper limit. Generic adjectives like "epic", "stunning", "8k", "masterpiece" are noise and reduce quality.
Do Seedance prompts need to be in English?
No. Seedance 2.0 understands prompts in Chinese and English natively, with partial support for other major languages. For consistency with prompt libraries published online, English is the most portable choice. Chinese is best for culturally specific subjects.

FAQ

Frequently asked questions

Are these prompts officially endorsed by ByteDance?
No. Seedance2Video is an independent operator built on the public Seedance 2.0 API. These prompts were tested on the same API endpoint that powers our generator, but are not authored or reviewed by ByteDance.
Why does the same prompt produce different results each run?
Generative video models sample randomly unless you lock the seed. To reproduce a winning take, copy the seed from the result you liked and reuse it on the next run.
How do I write a Seedance prompt for a TikTok ad?
Use 9:16 aspect, 3–5 second duration, and put the visual hook in the first three seconds of the prompt. Lead with the subject and one strong action, then add style and lighting.
Do Seedance prompts work for image-to-video?
Yes, with one rule: only describe what should change. Anything you do not mention is treated as fixed from the reference image. See the image-to-video category for runnable examples.
Where can I learn more before writing my own prompt?
Read the 5-part formula and parameter cheatsheet at the top of this page, then start by copying the closest prompt from the library and changing one slot at a time.
How do I prompt Seedance for slow motion?
Two levers. First, in the prompt clause for action, name the speed explicitly — for example "in slow motion" or "moves slowly". Second, lower motion strength one notch from your usual setting. Combining the two reads more cinematic than either alone. Avoid "ultra slow motion" or "bullet time" — Seedance does not have those as recognized cinematic vocabulary, and the prompt becomes noise.
Can Seedance follow a multi-shot script in a single prompt?
Up to about three shots, yes. Use a numbered shot list — "Shot 1: ... Shot 2: ... Shot 3: ..." — and keep the subject, lighting, and style consistent across shots so the model has anchors. Beyond three shots, generate each one separately and stitch them in your editor; longer single-prompt scripts increase the chance of identity drift between shots.
How do I keep characters consistent across multiple Seedance clips?
Three techniques. First, write a stable subject description block (age, hair, clothing, distinguishing feature) and reuse the exact wording across prompts. Second, lock the seed once you find a take you like and reuse it. Third, for higher fidelity, generate a key reference frame, then run image-to-video using that frame as the reference — this maintains identity better than text-only prompts.
Why is my Seedance output flickering or jittery?
Flicker usually traces to one of three causes: motion strength too high for the scene, two contradictory style anchors fighting each other (e.g. "cinematic" plus "anime"), or two camera moves named in one prompt. Drop motion one notch, keep two style anchors max, and pick one camera move per prompt. Lock the seed and re-run to confirm the fix.
Can I copyright a video generated from a Seedance prompt?
Copyright law on AI output varies by jurisdiction and is evolving rapidly through 2026. In the US, the Copyright Office has held that purely AI-generated work is not copyrightable; substantial human authorship in the editing or compositing process can confer copyright on the resulting work. In the EU and UK, similar principles apply. For commercial use, our paid plans grant a usage license — see the pricing page — but commercial license is distinct from copyright. Consult counsel for high-stakes uses.
How do I prompt Seedance for a specific camera lens or shot type?
Seedance recognizes common shot vocabulary: extreme close-up, close-up, medium close-up, medium, wide, establishing wide, low-angle, high-angle, top-down, over-the-shoulder. For lens feel, "shallow depth of field" or "macro lens" works better than "85mm lens" — the model understands optical effects more reliably than focal-length numerics. Pair the shot type with one camera move (static, slow dolly-in, handheld, orbit, tracking) for stability.
Will these prompts work on older Seedance 1.5 Pro?
Most will work but with reduced quality. The 5-part formula and parameter logic transfer cleanly to Seedance 1.5 Pro, but you should expect lower duration (capped at 8 seconds), no native audio, and weaker multi-shot coherence. Drop motion strength one notch when porting from 2.0 to 1.5 — the 1.5 model is more sensitive to high motion settings.
How are these prompts tested?
Each prompt in this library was run on the same Seedance 2.0 API endpoint that powers the generator embedded on this page. Prompts that produced unstable output across three test seeds were rewritten or replaced before publication. The library is re-tested whenever ByteDance ships a model update, with the dateModified field on the page reflecting the most recent verification pass.

Pick a prompt, run it, iterate.

Copy any prompt above, open the generator, and change one slot at a time. The fastest way to learn Seedance 2.0 is to run a working prompt and watch what changes when you swap one variable.

Seedance2Video is an independent operator built on the public Seedance 2.0 API by ByteDance. Prompts on this page were tested on the same API endpoint that powers our generator and are not authored or reviewed by ByteDance.