Seedance Prompts: 24 Examples for Seedance 2.0 Video
24 Seedance prompts for ByteDance's Seedance 2.5 and 2.0 models, including storyboards, reference-driven shots, dialogue and edits of existing clips. Each one explains why it works, and the page covers the syntax ByteDance documents: Shot 1 / Shot 2 blocks, @Image references, {dialogue} and <sound effects>.
★★★★★ 4.9/5 · 27,000+ users
Updated September 25, 2026.
Prompt structure
The Seedance prompt formula
ByteDance's Seedance 2.0 series prompt guide calls the model a 'multimodal AI director' that splits a prompt into a spatial layer (what is in frame) and a temporal layer (how it changes). So you first fix who is doing what, then where and in what light, then how the camera moves, and close with style, quality and constraint words. With references, you name them in the text (@Image 1, @Video 1) and bind each subject to its asset. Dialogue goes in curly braces {…} and sound effects in angle brackets <…>.
Strengths and limits
What Seedance is good at
Good at
- Audio and video generated together, with dialogue and lip sync in 11 languages (Chinese, English, Spanish, Indonesian, Malay, Thai, Arabic, Portuguese, Vietnamese, Japanese, Korean)
- Long single takes: up to 30 seconds in one generation, extendable over several rounds
- Heavy referencing: up to 50 inputs per generation (30 images, 10 videos, 10 audio clips) to lock a face, product, scene, camera move or voice
- Editing an existing clip (add, remove or replace a subject) and extending a clip forward or backward, keeping what you did not mention unchanged
- Timeline-style storyboards written as Shot 1 / Shot 2 / Shot 3
Limits
- Duration: 4-30 seconds on Seedance 2.5 (4-15 on 2.0), or -1 to let the model pick
- Resolution: 480p and 720p on 2.5, with 1080p on some platforms; 2.0 also offers 4K; 2.0 Fast and Mini top out at 720p
- Aspect ratios: 21:9, 16:9, 4:3, 1:1, 3:4, 9:16, or adaptive (required for editing, extension and first-frame tasks)
- Complex, fast motion is still weak: ByteDance names the physical plausibility of complex motions and scenes with several interacting subjects as areas to improve
- Unwanted subtitles can appear, more often in portrait; ByteDance says they cannot be prevented 100%, only reduced
- Faces that resemble celebrities are blocked in review; a weak face reference can drift toward one
Copy and paste
24 Seedance prompts to copy
UGC ad
1. Morning routine with the product
Define the woman with a dark bob and a white tank top in @Image 1 as the creator, and the frosted pink moisturiser jar in @Image 2 as the product. Shot 1: Fixed phone-height medium shot in a bright bathroom. The creator unscrews the product, dips two fingers in and smiles at the lens. Shot 2: Slow push-in to a close-up as she smooths the cream along her cheekbone. The creator says {Honestly, my skin just drinks this.} <tap running softly> Vertical phone-video look, natural window light, soft pastel tones. Keep it subtitle-free, no logo, no watermark.
Why it works: Subject labels bound to references plus one camera move per shot follow ByteDance's advanced formula directly.
UGC ad
2. Dog owner tests the chew toy
Shot 1: Handheld medium shot in a sunny living room. A man in a flannel shirt kneels and offers a bright orange rubber chew toy to a golden retriever. The dog takes it gently. Shot 2: Fixed close-up of the dog chewing, tail wagging slowly. Shot 3: Medium shot, the man holds the toy up to the camera, still intact. The man says {Two weeks. Not a scratch.} <squeaky toy, happy panting> Warm daylight, casual home-video style, HD, natural colours. No subtitles, no logo, no watermark.
Why it works: A three-shot setup, proof and payoff with slow animal motion avoids the fast action Seedance handles less well.
UGC ad
3. Desk setup tour
Shot 1: Slow lateral tracking along a tidy home-office desk: a walnut monitor stand, a green desk lamp, a small plant. Shot 2: Fixed close-up as a hand slides a slim wireless charger pad to the centre and sets a phone on it; a soft light ring appears. Shot 3: Slow pull-back to a medium shot of a young woman in a grey sweater leaning back in her chair. She says {Every cable is gone. That's the whole upgrade.} <keyboard clicks, soft lo-fi music> Cosy late-afternoon light, warm tones, clean detail. Keep it subtitle-free, no watermark.
Why it works: Desk tours are native TikTok content, and each shot carries one move and one action.
UGC ad
4. Parent's honest review
A tired but cheerful father in a baby-carrier sits on a sofa, a sleeping infant against his chest, and speaks quietly to a phone propped on the coffee table. Fixed medium shot, eye level. He slowly pats the grey carrier strap and glances down at the baby. He whispers {Forty minutes of walking and she's out. This thing earned its price.} <soft breathing, a clock ticking> Evening lamp light, warm, slightly grainy home-video texture. No subtitles, no logo, no watermark.
Why it works: Small, slow movements and a whispered line suit Seedance's strengths and read as a real testimonial.
UGC ad
5. Grocery haul voiceover
Overhead fixed shot of a kitchen counter. Hands unpack a canvas bag one item at a time: a baguette, lemons, a jar of honey with a hexagon label, a bunch of basil. The hands pause on the honey jar, turn it towards the camera, then set it in the centre. A woman's voiceover says {Weekly haul, and yes, I drove across town for this honey.} <paper bag rustling, jar set down on stone> Bright morning light, fresh natural colours, HD detail. Keep it subtitle-free, no logo.
Why it works: A fixed overhead framing keeps the scene stable while the voiceover carries the hook.
product
6. Fragrance in the fog
Reference the amber glass perfume bottle in @Image 1 as the product. Shot 1: The product stands on black volcanic rock; low fog drifts slowly across the base. The camera slowly pushes in from a medium shot. Shot 2: Fixed extreme close-up of the faceted glass as a single beam of warm light moves across it. <low wind, a soft glass chime> Dark, moody luxury style, deep blacks with amber highlights, cinematic texture, rich detail. Do not generate text, logo or watermark.
Why it works: The product is anchored to its reference, and two slow shots keep the glass geometry stable.
product
7. Running shoe rotation
Reference the white running shoe with a teal sole in @Image 1 as the product. The product floats at centre frame against a seamless light grey background and slowly rotates a full turn. The camera stays fixed at eye level. Soft studio light from both sides, gentle reflection on the floor below. <faint whoosh as the shoe turns> Clean commercial style, HD, sharp textures on the mesh and sole. Keep it subtitle-free, no logo other than the product's own, no watermark.
Why it works: A fixed camera with a single slow rotation is the most reliable product turntable setup for reference-based video.
product
8. Coffee bag pour
Shot 1: Fixed close-up of a matte black coffee bag with a kraft label. A hand tears the top open. Shot 2: Slow push-in as dark roasted beans pour from the bag into a glass grinder hopper. Shot 3: Fixed macro of the beans settling, a few rolling slowly. <paper tearing, beans rattling into glass> Warm tungsten light, brown and black palette, shallow depth of field, cinematic food commercial style. No subtitles, no watermark.
Why it works: Sound effects in angle brackets give the pour an ASMR quality that Seedance generates with the picture.
Run these prompts without five subscriptions
- Paste the prompt into Veo 3.1, Kling 3.0, Seedance 2.0, Nano Banana Pro and GPT Image 2, all in one place
- Turn the clip into a TikTok, Reel or Short with captions and a hook
- Publish straight to TikTok, Instagram and YouTube
product
9. Watch on the wrist
Define the silver watch with a navy dial in @Image 1 as the product, and the man in @Image 2 as the model. Shot 1: Medium shot, the model adjusts his shirt cuff in a hotel corridor, walking slowly towards the camera. Shot 2: Fixed close-up of his wrist as the cuff slides back to reveal the product; he glances down at it. <soft footsteps on carpet, a faint ticking> Elegant commercial style, warm corridor lighting, rich detail on the dial. Keep it subtitle-free, no watermark.
Why it works: Separate labels for model and product stop Seedance from mixing their features.
product
10. Serum drop macro
Extreme close-up, fixed camera. A glass dropper hangs above a clear pool of golden serum on a white ceramic dish. A single drop forms slowly, falls and creates a ring of ripples. <a soft, clear drip> Soft top light, warm gold and white palette, ultra-sharp detail, clean beauty commercial style. Do not generate any text, logo or watermark.
Why it works: A single slow physical event with one sound is exactly the 'slow, gentle, continuous' motion ByteDance recommends.
social hook
11. Plot twist in the fitting room
Shot 1: Fixed medium shot of a fitting-room curtain. It slides open slowly and a woman steps out in a plain beige trench coat, shrugging at the camera. Shot 2: Cut to a medium shot as she turns the coat inside out and puts it back on; it is now bright red with a checked lining. She says {Two coats. One price.} <curtain rings sliding, fabric rustle> Bright store lighting, clean fashion-video style. Keep it subtitle-free, no logo, no watermark.
Why it works: A reversible reveal delivers the hook and the product benefit in two shots.
social hook
12. Don't buy this until
A young man in a black t-shirt sits at a desk facing a phone camera and holds up a small white router. Fixed medium close-up. He raises one finger slowly, then points it at the device. He says {Don't buy a new router until you try this one setting.} <quiet room tone> Then he flips the router over and taps a small switch on the back. Natural room light, casual creator style, HD. Keep it subtitle-free, no watermark.
Why it works: A warning-style opener plus a small, precise hand action fits Seedance's preference for fine body movements.
social hook
13. Satisfying clean
Fixed top-down close-up of a stained white sneaker on a wooden bench. A soft brush with white foam slowly scrubs across the toe; the dirt lifts away in one clean stripe. The brush moves again and the whole toe turns bright white. <foam fizzing, bristles brushing> Bright daylight, high-detail texture of foam and leather, satisfying cleaning-video style. Do not generate subtitles, text or watermark.
Why it works: Cleaning reveals are proven hooks, and slow repeated strokes keep the motion plausible.
social hook
14. Street food countdown
Shot 1: Fixed medium shot at a night-market stall. A vendor in a white cap cracks an egg onto a sizzling griddle. Shot 2: Slow push-in as he folds a crepe around the egg and spring onions. Shot 3: Close-up as he hands the wrap towards the camera. The vendor says {Ninety seconds. Best breakfast in the city.} <sizzling, a spatula scraping> Warm neon and steam, lively documentary style. Keep it subtitle-free, no watermark.
Why it works: Food process shots with loud, synced sizzle make a strong opener for local-business content.
talking head
15. Pharmacist explains
Define the pharmacist in a white coat with round glasses in @Image 1 as the pharmacist; use a head-only close-up as the face reference. Fixed medium close-up behind a pharmacy counter, shelves softly blurred behind. The pharmacist slightly tilts her head, then lifts a small box of vitamins to chest height. The pharmacist says {Take this one with food, not on an empty stomach. That's where most people go wrong.} Clean, bright clinical light, natural colours, HD. Face stays stable without deformation. Keep it subtitle-free, no watermark.
Why it works: A head-only face reference plus a stability constraint follows ByteDance's own fix for identity drift.
talking head
16. Spanish-language founder
A founder in his forties with a trimmed beard and a light blue shirt sits in a small workshop full of wooden chairs, facing the camera. Fixed medium shot, eye level. He rests his hands on a half-finished chair and gently runs his thumb along the wood. He says in Spanish {Cada silla pasa por mis manos antes de salir de aquí.} <sanding in the background> Warm window light, sawdust in the air, documentary style. Keep it subtitle-free, no watermark.
Why it works: Marking the dialogue language before the braces follows the guide's rule for lines not in Chinese or English.
Run these prompts without five subscriptions
- Paste the prompt into Veo 3.1, Kling 3.0, Seedance 2.0, Nano Banana Pro and GPT Image 2, all in one place
- Turn the clip into a TikTok, Reel or Short with captions and a hook
- Publish straight to TikTok, Instagram and YouTube
talking head
17. Two-person interview
Define the woman in a green blazer as the host and the older man in a grey cardigan as the guest. Shot 1: Fixed two-shot of both seated in armchairs in a book-lined study. Shot 2: Medium close-up of the host leaning slightly forward. The host says {What would you tell yourself at twenty?} Shot 3: Medium close-up of the guest, who smiles and looks down briefly. The guest says {Stop waiting for permission.} <quiet room tone, a clock ticking> Soft warm lamp light, documentary interview style. Keep it subtitle-free, no watermark.
Why it works: Labels defined once and reused keep the two speakers from being swapped across shots.
talking head
18. Fitness coach cue
A fitness coach in a black tank top stands in a bright gym beside a kettlebell and speaks to the camera. Fixed medium shot. She slowly bends her knees into a squat position to demonstrate, keeping her back straight, then stands back up. She says {Hips back first, then knees. Feel that? That's the whole trick.} <gym music faint, weights clanking far away> Clean daylight, energetic but natural style, HD. Movements are smooth and controlled. Keep it subtitle-free, no watermark.
Why it works: A slow, controlled demonstration avoids the high-burst movements the guide warns against.
cinematic
19. Wuxia rooftop standoff
Shot 1: At dusk, the camera slowly pushes in on a swordswoman in grey robes standing on a tiled temple roof. Her sleeves stir in the wind. Shot 2: Wide shot from a high angle: a masked figure appears on the opposite roof. Both stand still. Shot 3: Fixed close-up of her hand slowly tightening on the sword hilt, eyes narrowing. <wind, distant temple bell, tiles creaking> Cinematic wuxia style, cool tones, low saturation, film grain. Faces and proportions stay stable. No subtitles, no watermark.
Why it works: Tension built through stillness and small movements plays to Seedance's strength instead of a fast fight.
cinematic
20. Deep-sea research station
Shot 1: Slow lateral tracking past the round window of an underwater research station; bioluminescent jellyfish drift outside. Shot 2: Medium shot inside: a researcher in a navy jumpsuit presses her palm to the glass, face lit blue. Shot 3: The camera slowly pulls back to reveal the station as a small glowing dome on the dark seabed. <muffled hum of machinery, a whale call far away> Science-fiction realism, deep blue and teal palette, cinematic texture, rich detail. Keep it subtitle-free, no watermark.
Why it works: The storyboard moves from detail to person to scale, each shot with its own single move.
cinematic
21. Old tailor's last suit
Shot 1: Fixed close-up of wrinkled hands sewing a button onto a charcoal jacket in a small tailor shop. Shot 2: Medium shot: the old tailor stands, slowly lifts the jacket onto a mannequin and smooths the lapel. Shot 3: The camera slowly pushes in on his face as his eyes fill with quiet pride. <a clock ticking, the soft pull of thread> Warm afternoon light through dusty windows, muted browns, 35mm film texture. Movements are slow and natural. No subtitles, no watermark.
Why it works: Emotion shown through hands and eyes, not adjectives, is exactly how the guide says to write feeling.
cinematic
22. One-take city walk
A single continuous take, 30 seconds. The camera follows a young man in a mustard coat from behind as he walks slowly through a morning flower market, then along a canal, then up a short flight of stone steps where he stops and looks out over the rooftops. Smooth steady tracking the whole time, no cuts. <market chatter fading into canal water, then quiet wind> Soft golden morning light, European old-town setting, cinematic realism. Keep it subtitle-free, no logo, no watermark.
Why it works: Seedance 2.5 is built for 30-second single takes, and a steady follow shot keeps the walk coherent.
animation
23. Clay rabbit bakery
3D claymation style with visible fingerprints on the clay. A round white clay rabbit in a tiny apron slowly lifts a tray of carrot muffins out of a small oven, sniffs them and wiggles its nose happily. Fixed medium shot, eye level. <oven door creak, a tiny happy squeak> Warm orange oven glow, pastel kitchen, soft stop-motion feel. Keep the claymation style throughout, do not become realistic. No subtitles, no watermark.
Why it works: An explicit style lock at the end prevents the drift to live action the guide lists under FAQs.
animation
24. Anime bus stop
2D Japanese anime style. Shot 1: Wide shot of a countryside bus stop in summer; cicadas, tall grass, a girl in a school uniform waiting with a straw bag. Fixed camera. Shot 2: Medium close-up: she slowly looks up as a gust of wind lifts her hair; the bus appears in the distance. <cicadas, wind, a distant engine> Bright blue sky, soft painterly backgrounds, gentle hand-drawn motion. Keep the 2D anime style throughout. No subtitles, no watermark.
Why it works: Naming the style at the start and restating it as a constraint keeps the whole clip in 2D.
From prompt to posted video
Paste the prompt, get the video, post it
Hooked Pro is $29 a month for 1,000 credits across Veo 3.1, Kling 3.0, Seedance 2.0, Nano Banana Pro and GPT Image 2, plus AI UGC ads, captions and TikTok slideshows in the same plan. Or pay $89 once for the Studio and run it on your own API keys, at the provider's price.
Better results
Tips that make a difference
Do
- Use one camera move per shot. The guide warns that asking for push, pull, pan and move together makes the image unstable.
- Write complex clips as a numbered storyboard (Shot 1, Shot 2, Shot 3) in the order events happen. Avoid exact timestamps such as 0-3s: ByteDance says precise timing support is unstable.
- Describe actions by body part, with range, speed and force ('slowly raises her right hand', 'slightly lowers his head'), and prefer slow, continuous movement over sprints, jumps and rolls.
- Show emotion through physical detail (tight jaw, eyes dropping to the floor) instead of words like 'very sad'.
- Put dialogue in curly braces, {Like this}, and sound effects in angle brackets, <a dog barks in the distance>. Mark a less common language before the line, e.g. says in Japanese {…}.
- Bind each subject to its reference every time you mention it ('the woman @Image 1'), or define a label once ('define the man in @Image 1 as the chef') and reuse it exactly.
- For faces, add a separate head-only close-up reference next to the full-body one, and list the most important reference first. Avoid multi-view character sheets, which the model can read as several people.
- Use 4-5 references, not the maximum: 1-2 character images, 1 scene image, 1 camera-move video, 1 audio clip. Too many assets confuse priorities.
- End with constraint words: 'keep it subtitle-free, no text, no logo, no watermark'. If you can, render landscape and crop to portrait, since portrait gets more stray subtitles.
- For editing or extending, refer to the clip directly ('Extend @Video 1 backward…') and do not write 'reference @Video 1', or the model treats it as a new reference task.
Common mistakes
- Stacking several camera moves in one shot
- Forcing strict per-second timings instead of letting shot order set the pace
- Describing a subject without tying it to its reference image, which causes identity drift
- Uploading every reference slot available, which leads to style conflicts and misidentified subjects
- Mixing languages in dialogue (other than proper nouns)
- Leaving out constraint words and getting subtitles, platform logos or watermarks in the output
FAQ
Frequently asked questions
What is the latest Seedance model?
Seedance 2.5, released by ByteDance on July 31, 2026. Seedance 2.0, 2.0 Fast and 2.0 Mini remain available.
How long can a Seedance video be?
Seedance 2.5 generates 4 to 30 seconds in one take, and clips can be extended over several rounds. Seedance 2.0 goes up to 15 seconds.
How do I write dialogue and sound in a Seedance prompt?
Put spoken lines in curly braces, {like this}, and sound effects in angle brackets, <a door slams>. Mark a less common language before the line.
How do I stop subtitles appearing in Seedance videos?
End the prompt with constraint words such as "keep it subtitle-free, no text, no logo, no watermark". ByteDance says this reduces stray subtitles but cannot prevent them every time, and portrait output gets them more often.
How many reference files should I use?
Up to 50 are accepted, but 4 or 5 work better: one or two character images, a scene image, a camera-move video and an audio clip. Name each one in the text, for example @Image 1.
Keep exploring
Related prompt libraries and pricing
Sources
- https://docs.byteplus.com/en/docs/ModelArk/2222480
- https://docs.byteplus.com/en/docs/ModelArk/2607688
- https://seed.bytedance.com/en/blog/one-take-creation-flexible-referencing-introducing-seedance-2-5
- https://higgsfield.ai/blog/seedance-2-5-on-higgsfield-2026
- https://higgsfield.ai/blog/seedance-prompting-guide
Two ways in · pick yours
Every template here opens in the app — 1,600+ and counting.
Pay monthly and we run the AI for you, credits included — or pay $89 once, own the app for life and bring your own keys.
Pro
Most flexibleWe run the AI for you. No API keys, nothing to set up — open the app and generate.
1,000 credits a month · cancel anytime
Secure checkout by Stripe · Cancel anytime
- About 10 UGC ads or 250 images a month
- Every video format: UGC, avatars, slideshows…
- 83 AI avatars & 76 voices in 25 languages
- Publish to TikTok, Instagram & YouTube
- 3 channels, 3 auto-channels, API included
Need more? One-time credit packs from $19. They never expire.
4.9/5 · 27,000+ users create with Hooked