Seedance Prompt Generator: Ready-to-Paste Prompts for Seedance 2.5

Seedance is the model you prompt like a production brief. ByteDance built it around references: tag an image for the face, a video for the motion, an audio file for the beat, and write what should happen. Seedance 2.5, released on July 31, 2026, pushed that to 30-second clips and 50 reference slots. This Seedance prompt generator page gives you the official prompt structure, nine prompts that use tags and timecodes properly, and notes for anyone still running Seedance 2.0 on a third-party platform.

What Seedance Is Good At Right Now (and Its Limits)

From ByteDance Seed’s launch posts and Dreamina’s documentation:

• Seedance 2.5 length: up to 30 seconds per generation, with multi-round extensions.
• Seedance 2.5 references: up to 30 images, 10 videos and 10 audio clips in one pass, plus new clay-render (untextured 3D blocking) references.
• Editing: timestamp-level edits to audio or video, green-screen background swaps, and camera-perspective edits on a generated clip.
• Audio: generated jointly with the picture, including dialogue, ambience and music.
• Resolution: launch partners list native 1080p output for 2.5.
• Seedance 2.0: still widely offered. Up to 15 seconds, and up to 9 images, 3 videos and 3 audio clips as references.
• Where to run it: Dreamina and Jimeng (ByteDance’s own apps), plus API access through BytePlus ModelArk. Availability varies by region.
• Price: Dreamina advertises Seedance pricing from about $0.046 per generated second on its best plan rates.

The big limit is faces. Dreamina’s Seedance launch note says the rollout includes safeguards against making videos from images or videos that contain real faces, so unverified portraits tend to get rejected. Use illustrated, 3D, or AI-generated characters as references. If you need text-only prompting with no reference setup at all, the Veo 3 and Hailuo pages may suit you better.

The Seedance Prompt Formula

Dreamina’s Seedance 2.5 guide lays out eight parts: format, subject, action, environment, camera, look, timing, and audio plus constraints. As a template:

[Duration, aspect ratio, shot structure].
Use @Image1 as the consistent [character / product]: [which details only].
In [environment and time], the subject [precise action with a start and finish].
Camera: [shot size + movement + lens/focus].
Look: [lighting + palette + texture + realism].
Timing: 0-Xs [beat 1]; X-Ys [beat 2]; Y-Zs [payoff].
Audio: [dialogue / ambience / SFX / music].
Preserve: [identity, wardrobe, logo, product shape].
Avoid: [unwanted cuts, extra people, on-screen text, face drift].

Instead of a long negative prompt, the guide groups exclusions into one short, concrete Avoid line at the very end. Keep it to the failures you have actually seen.

9 Copy-Ready Seedance Prompts

Doorbell camera: tiny dog, big energy (text only)

10 seconds, vertical 9:16, single continuous shot. Doorbell camera footage, fisheye lens, high angle over a sunny front porch. Timing: 0-3s a delivery driver in a brown uniform sets a box down; 3-7s a tiny chihuahua bursts through the pet flap, barking, and the driver backs down the steps with his hands up; 7-10s the chihuahua sits on the box, proud. Look: slightly overexposed daylight, compression artifacts. Audio: frantic high-pitched barking, the driver laughing nervously. Avoid: extra people, music.

Why it works: timecodes give each beat its own space, so the payoff at 7 seconds does not get rushed. More setups on the doorbell camera prompts page.

CCTV: 2 AM convenience store

12 seconds, 16:9, single shot. Static CCTV camera above the counter of a small convenience store at 2 AM, harsh fluorescent light, grainy footage. A stray tabby cat hops onto the counter, steals a sausage from the hot dog roller, and trots out the automatic door. The sleepy clerk looks up only after the door closes. Audio: fridge hum, door chime, the clerk sighing. Avoid: camera movement, on-screen text other than a timestamp.

Why it works: “static” plus “Avoid: camera movement” locks the surveillance look instead of leaving the camera behavior open. See cat video prompts for more.

Handheld phone fail: the birthday cake

8 seconds, vertical 9:16. Handheld phone video at a crowded living-room birthday party, warm lamp light, slight motion blur. A man carries a lit birthday cake toward the table, catches his foot on a rug, stumbles forward, and saves the cake at the last second while one candle falls into the frosting. Everyone gasps, then cheers. Audio: party chatter, a collective gasp, cheering. Avoid: the cake falling on the floor.

Why it works: the Avoid line steers the clip toward the near-miss ending, which reads funnier than a full crash. Browse birthday prompts for more.

30-second skincare ad with timecodes

30 seconds, vertical 9:16, one continuous take. Use @Image1 as the product: a frosted glass serum bottle with a white dropper, keep label and shape exact. A bathroom counter in soft morning light, white tiles, a small plant. Timing: 0-6s slow push-in on the bottle as light moves across the glass; 6-18s a hand squeezes the dropper, a single golden drop falls onto a fingertip in macro; 18-26s the drop spreads across the back of a hand, catching the light; 26-30s static hero frame of the bottle beside a folded towel. Look: clean, bright, soft shadows. Audio: gentle ambient music, a soft drip. Preserve: label text, bottle shape. Avoid: faces, extra products, subtitles.

Why it works: this is Dreamina’s recommended setup, demo, payoff, hero-frame rhythm, and the product-only reference sidesteps the face restriction.

Character plus product references

10 seconds, 16:9. Use @Image1 as the character: a 3D cartoon grandmother with round glasses and a purple cardigan, face and outfit only, do not use the image background. Use @Image2 as the product: a red thermos. In a snowy park at dusk, she pours tea from the thermos, blows on the cup, sips, and gives a satisfied nod to camera. Camera: medium shot, slow push-in. Audio: wind, liquid pouring. Grandmother (warm voice): "Now that's the good stuff." Preserve: thermos color and logo.

Why it works: each tag has one narrow job, which keeps @Image1’s background from leaking into the park.

Motion reference: copy a dance

12 seconds, vertical 9:16. Use @Image1 as the character: a cartoon penguin in a red scarf. Match the choreography and camera movement of @Video1 exactly. Setting: an ice rink at night under string lights. Audio: use the rhythm of @Audio1. Avoid: changing the penguin's proportions, extra dancers.

Why it works: the video carries the motion and camera, so the text only sets character and place. More in dance prompts.

Beat-synced music visual

15 seconds, 16:9, cut on the beats of @Audio1. Neon-lit Tokyo alley in the rain. Each cut shows a new close-up: a vending machine glowing, rain hitting a puddle, a cat under an umbrella, a scooter tail light streaking past. Camera: handheld, slight drift. Look: magenta and cyan neon, wet reflections, film grain. Avoid: people's faces, text.

Why it works: audio references drive pacing, so Seedance can time cuts to the music instead of guessing.

Cinematic one-take (2.5 only)

25 seconds, 2.39:1 letterbox feel inside 16:9, single unbroken tracking shot. A young cook in a white apron carries a steaming bowl of ramen from the kitchen, through a narrow crowded ramen bar, out the sliding door, and across a rainy street to an old man waiting under an awning. Timing: 0-8s kitchen; 8-16s through the bar; 16-25s across the street. Camera: steady follow shot at shoulder height. Look: warm tungsten inside, cold blue outside. Audio: kitchen clatter fading into rain. Avoid: cuts, the bowl spilling.

Why it works: 30-second single takes are the headline Seedance 2.5 feature, and three timed zones keep the camera path readable.

Short drama dialogue

12 seconds, vertical 9:16. A small office break room, fluorescent light. Two coworkers stand at the fridge. Timing: 0-4s the man opens the fridge and holds up an empty yogurt cup; 4-8s he turns: "Who ate my yogurt?"; 8-12s the woman behind him slowly hides a spoon and says, "Yogurt? What yogurt?" Camera: handheld medium shot, slight push-in on each line. Audio: fridge hum, dialogue only.

Why it works: putting each line inside a timecode tells Seedance exactly when each person speaks.

Seedance-Specific Tips

• Scope every tag. “Face and jacket only, do not use the image background” prevents bystanders and scenery from being copied.
• Timecode anything over 10 seconds. Dreamina’s own guide splits long clips into timed beats so each part gets its own direction.
• Put Preserve and Avoid last. Grouped at the end, they read as rules instead of scene details.
• Running Seedance 2.0? Cap clips at 15 seconds and references at 9 images, 3 videos and 3 audio. The tag syntax is the same. The existing Seedance prompt guide goes deeper on 2.0.
• Write aspect ratio in the first line and set it in the app. The prompt shapes composition, the setting fixes the frame.

From One Idea to 10 Seedance Prompts With LzyPrompt

The part Seedance cannot do for you is come up with the scene. LzyPrompt turns “tiny dog scares delivery guy” into 10 written variations, each with a scene title, a full prompt (camera style, setting, lighting, subject, action beats, imperfections, sound cues) and a caption. Choose a Camera Style like doorbell camera or handheld phone video, or leave Auto-detect on, and adjust Shot Type, Lighting and Aesthetic if you want.

LzyPrompt is model-agnostic, so it will not add @ tags or timecodes for you. After pasting into Seedance, put the duration and aspect ratio on the first line, split the action beats into timecodes, add your reference tags, and finish with Preserve and Avoid. Export the full set to CSV when you are running a batch. You can try LzyPrompt free for 7 days.