Seedance prompts: the guide and the library
Seedance responds to structure, not adjective density. This guide gives you a six-part framework for writing prompts, an interactive builder that assembles a prompt from those parts, explains how references change what you should write, and provides twenty original prompts across categories, each with a button that opens it in the workspace. Everything here was written for Seedance 2.5 and 2.0 and works on 1.5 Pro with the reference sections removed.
The prompt framework
Write the prompt as six short parts in this order. You do not need every part in every prompt, but keeping the order makes prompts easier to edit and compare.
| Part | What to write | Example |
|---|---|---|
| 1. Subject | Who or what, with two or three identifying details | A courier in a yellow rain jacket on a cargo bike |
| 2. Action | One clear action with a beginning and an end | brakes at a crossing, checks her watch, pushes off again |
| 3. Environment | Where, time of day, weather, one or two set details | narrow old-town street at dusk, wet cobblestones, shop lights coming on |
| 4. Camera | One move, framing and lens feel in plain words | side tracking shot at handlebar height, shallow depth of field |
| 5. Lighting and look | Key light direction and quality, colour, grain | cool blue dusk with warm shop windows, light film grain |
| 6. Audio | Ambience, effects, dialogue in quotes; or 'no music' | tyres on wet stone, distant bells, no music |
Then add constraints: 'no cuts', 'no text on screen', 'no people in the background'. Seedance takes negative instructions seriously when they are explicit and short.
インタラクティブツール
Seedance プロンプトビルダー
必要な項目を入力すると、ビルダーが上記のフレームワークの順序で組み立てます。処理はすべてブラウザ内で行われ、ワークスペースでプロンプトを開くまでどこにも送信されません。
組み立てたプロンプト
0 / 30000 文字
Subject and action
Be concrete about the subject and modest about the action. One complete action per clip is the sweet spot for 4–8 seconds; two or three chained actions are realistic on a 10–30 second Seedance 2.5 clip if you write them in order with 'then'. Avoid actions that require precise hand-object interaction unless you are prepared to iterate.
Camera
One camera instruction per clip. Say the move, the start and end framing, and whether the subject stays centred. Plain English beats numbers: 'slow dolly in from a wide to a medium shot' works better than '24mm to 50mm push'. If you want no movement, say 'locked-off static camera'.
Lighting
Describe the key light's direction and quality and whether there is fill. Add a colour relationship if you want one (warm practicals against cold window light). Name grain, haze or bloom explicitly; the model does not add them by default.
Motion and timing
Use words that carry speed: 'slow', 'deliberate', 'sudden', 'settles'. For animation, specify cadence ('animated on twos'). For physical realism, describe secondary motion (cloth, hair, water) once; over-describing it produces exaggerated movement.
Dialogue and audio
Put spoken lines in quotation marks, name the speaker, and give tone and pace. Keep lines under ten words per clip for tight lip sync. Ambience works best as two or three named sources. Say 'no music' unless you want a bed. The native audio page goes deeper.
Reference instructions
On Seedance 2.0 and 2.5 the prompt also has to explain what each reference is for. Refer to them by role: 'the character from the reference images', 'the camera move from the reference video', 'speaks in the voice of the reference audio'. Say what to keep and what to change. Seedance 2.5 takes up to 30 images, 10 videos and 10 audio clips on Seedancee; 2.0 takes 9, 3 and 3. More references are not always better: three clean views of a character beat nine inconsistent ones.
Cinematic prompts
A courier in a yellow rain jacket on a cargo bike brakes at a crossing, checks her watch, and pushes off again. Narrow old-town street at dusk, wet cobblestones, shop lights coming on. Side tracking shot at handlebar height, shallow depth of field. Cool blue dusk with warm shop windows, light film grain. Tyres on wet stone, distant bells, no music. No cuts, no text.
- duration:
- 8s
- resolution:
- 1080p
- aspect:
- 16:9
- audio:
- on
An empty two-lane highway through red desert at first light. A single vintage pickup truck drives away from camera toward the horizon, dust trailing. Low sun from the right, long shadows, faint heat shimmer. Locked-off static wide shot from the centre of the road. Wind and a fading engine, no music. No cuts.
- duration:
- 10s
- resolution:
- 1080p
- aspect:
- 21:9
- audio:
- on
Close-up on a train window at night, rain streaks sliding sideways, city lights smearing past out of focus. A hand in a grey sleeve rests on the sill and taps twice. Reflection of the carriage interior faintly visible. Static camera, shallow depth of field, cool tones with warm reflected interior light. Rail rhythm and rain, no music.
- duration:
- 6s
- resolution:
- 1080p
- aspect:
- 16:9
- audio:
- on
A white lighthouse on a rocky headland at storm light. Slow half-orbit clockwise around the lighthouse, keeping it centred, revealing waves breaking on the rocks below. Dark grey sky with a break of gold on the horizon, spray catching the light. Aerial perspective, smooth motion, no cuts. Wind and surf, no music.
- duration:
- 10s
- resolution:
- 1080p
- aspect:
- 16:9
- audio:
- on
Product prompts
A matte black ceramic mug with a plain white interior rotates slowly 180 degrees on a white turntable. Soft studio light from above-left, one clean specular highlight tracking across the glaze, subtle contact shadow. Plain off-white background, no props, no text, no logo. Static camera, medium close-up.
- duration:
- 5s
- resolution:
- 1080p
- aspect:
- 1:1
- audio:
- off
A frosted glass bottle with a blank label stands on a wet slate surface. A single drop falls into a shallow pool in front of it and sends a ring of ripples toward the lens. Soft top light, cool tone, gentle reflections. Slow push-in, slight low angle, shallow depth of field. No text, no hands.
- duration:
- 5s
- resolution:
- 1080p
- aspect:
- 9:16
- audio:
- off
An unbranded white canvas sneaker with a gum sole sits on a raw concrete step in hard afternoon sun. Camera starts tight on the stitching and pulls back to a three-quarter view, keeping the shoe sharp. Hard key light from the right, deep shadow, slight lens flare. No logos, no text, no people.
- duration:
- 6s
- resolution:
- 1080p
- aspect:
- 4:3
- audio:
- off
Character prompts
The character from the reference images (keep face, hair and jacket exactly) walks through a crowded morning market, stops at a fruit stall, picks up an orange and smiles at the vendor. Medium shot, handheld follow from the front, natural overcast light. Market ambience, no dialogue, no music. No cuts.
- duration:
- 8s
- resolution:
- 720p
- aspect:
- 9:16
- audio:
- on
Two friends at a kitchen table, morning light from the window. Woman in a green sweater says, amused: "You burned it again." Man in a grey t-shirt looks at the toast and says, deadpan: "It's charred. On purpose." Static two-shot at eye level, warm tones. Kitchen ambience, a kettle finishing, no music.
- duration:
- 8s
- resolution:
- 720p
- aspect:
- 16:9
- audio:
- on
Use the camera motion from the reference video exactly: same speed, same path, same start and end framing. Apply it to a new scene: a potter at a wheel in a sunlit studio, hands shaping a tall vase, clay slip glistening. Keep the character from the reference images. Warm side light, dust in the air. Wheel hum and soft radio, no music.
- duration:
- 8s
- resolution:
- 1080p
- aspect:
- 16:9
- audio:
- on
Close-up of an elderly man with a white beard reading a letter by a window. His expression moves from neutral to a slow, private smile; he lowers the letter and looks outside. Soft window light from the left, no fill, shallow depth of field, warm muted palette. Static camera. Room tone only, no music.
- duration:
- 6s
- resolution:
- 1080p
- aspect:
- 4:3
- audio:
- on
Image-to-video prompts
With a first frame, describe only the motion and anything that should change. Do not re-describe the scene; the image already carries it. If you also give a last frame, describe the path between the two.
Starting from the first frame: the subject blinks, takes a slow breath, and turns their head slightly toward the light. Hair moves in a faint breeze. Nothing else in the frame changes. Very slow push-in. Keep the lighting and colour of the source exactly.
- duration:
- 5s
- resolution:
- 1080p
- aspect:
- 3:4
- audio:
- off
Starting from the first frame: clouds drift slowly left to right, the grass sways, and light shifts as a cloud passes over the hills. Water in the foreground ripples gently. Camera static. Preserve the composition and palette of the source image.
- duration:
- 8s
- resolution:
- 1080p
- aspect:
- 16:9
- audio:
- on
Move from the first frame to the last frame: the empty stage lights come up one by one, the curtain rises, and the camera dollies in until the final framing matches the last frame. Smooth, continuous motion, no cuts. Theatre ambience, a murmur that settles, no music.
- duration:
- 8s
- resolution:
- 1080p
- aspect:
- 16:9
- audio:
- on
Animation prompts
2D hand-drawn animation, animated on twos, ink outlines with slight line boil, flat cel colours, painted background. A tall thin baker with a flour-dusted apron slides a tray of bread out of a brick oven, blows on his fingers, and grins. Warm oven glow from the left. Medium shot, gentle pan. Paper texture over the frame. No 3D shading, no photorealism.
- duration:
- 6s
- resolution:
- 1080p
- aspect:
- 16:9
- audio:
- off
Stylised 3D animation, rounded shapes, soft subsurface materials. A small moss-covered stone creature with two glowing eyes wakes up on a forest floor, stretches, and shakes leaves off its back. Big soft key light through the canopy, dust motes. Slow push-in, low angle. Forest ambience, no music.
- duration:
- 6s
- resolution:
- 1080p
- aspect:
- 1:1
- audio:
- on
Social and vertical prompts
Vertical close-up: espresso pours from a portafilter into a white cup, crema forming. Warm café light, steam. Static camera, shallow depth of field. Espresso machine hiss, no music, no text.
- duration:
- 5s
- resolution:
- 720p
- aspect:
- 9:16
- audio:
- on
Vertical wide shot of a rainy city crossing at night from a high window. Umbrellas cross in every direction, traffic lights change from red to green, headlights streak. Static camera, cool tones with neon reflections. Traffic and rain, no music, no text.
- duration:
- 6s
- resolution:
- 720p
- aspect:
- 9:16
- audio:
- on
Extend and edit prompts (Seedance 2.5)
Continue the clip: the courier turns left into a narrower alley, slows under a string of lanterns, and stops at a blue door. Same camera height and tracking style, same lighting and grain, same ambience. No cuts.
- mode:
- extend
- duration:
- 8s
- audio:
- on
Edit the clip: replace the rain with light snow, add a faint breath cloud from the subject, and keep the framing, motion, lighting and colour otherwise identical.
- mode:
- edit
- audio:
- on
Mistakes to avoid
- Adjective stacking ('epic, stunning, ultra-detailed, masterpiece'): it adds nothing the model can act on.
- Two camera moves in one clip: pick one.
- Contradictions ('cinematic realism' plus 'cartoon style').
- Re-describing an image you already supplied as a first frame.
- Long monologues: split dialogue across shots.
- Requesting copyrighted characters, brands or real people; these violate the content policy and may be blocked.
- Changing several settings and the prompt at once, so you cannot tell what helped.
Open the workspace and paste a prompt
Every prompt above has a 'Use this prompt' button that pre-fills the composer.
よくある質問
What makes a good Seedance 2.0 prompt?
The same framework as 2.5: subject, action, environment, camera, lighting, audio, then constraints. Keep clips 4–15 seconds, one camera move, one action, and explain what each reference is for.
How long should a Seedance prompt be?
Usually 40–120 words. The model accepts far more, but past a few sentences per part you are adding noise. Length should grow with clip duration, not with enthusiasm.
Does Seedance support negative prompts?
There is no separate negative field; write constraints inline ('no text', 'no cuts', 'no music'). Short, explicit negatives are respected.
Can I use these prompts commercially?
Yes, they are original and free to use or adapt. What you may do with the resulting video depends on our terms and what it depicts.
How do I prompt for a specific style without naming a studio?
Describe the medium, rendering method, colour approach and motion cadence. 'Flat cel colours, ink outlines, animated on twos' communicates more than any studio name and stays within the content policy.
Do prompts work in other languages?
The models accept several languages. Our documented tests are in English; results for other languages will be published in the Lab.
Where do I find prompt results and comparisons?
Seedance Lab publishes fixed-prompt runs with dates and settings per model. Patterns that prove reliable are added to this page with a link to the run.