Seedance 2.0 Prompts: 12 Copy-Ready Prompts by Mode
Twelve Seedance 2.0 prompts you can copy today, grouped by text to video, image to video and reference to video, with the four building blocks behind them and a plain guide to the standard, Fast and Mini variants.
By Eftal Yurtseven · Nov 16, 2026 · 9 min read

Most Seedance 2.0 prompts fail in the same boring way. Someone writes a gorgeous paragraph about a place, forgets to say what moves, skips the camera, and gets a screensaver with a soundtrack. The model did what it was told. It just wasn't told much.
Seedance 2.0 is ByteDance's video model that builds picture and sound in one pass, and it rewards prompts written like a short shot list rather than a mood board in sentence form. Below are 12 copy-ready prompts grouped by mode (text to video, image to video, reference to video), the four building blocks every one of them uses, and a plain way to choose between the standard, Fast and Mini variants. Copy, swap the nouns, run.
What Seedance 2.0 Actually Takes
ByteDance Seed launched Seedance 2.0 on February 12, 2026 with four input modalities: text, image, audio and video. The launch post says you can feed it up to 9 images, 3 video clips and 3 audio clips plus your instructions, and that it can reference composition, motion, camera movement, visual effects and audio from those files. Output is multi-shot audio and video up to 15 seconds, with two-channel stereo sound and separate tracks for background music, ambient effects and character voiceover.
On each::labs that power is split into three doors. Text to video takes only a prompt. Image to video takes a start image and an optional end image. Reference to video takes the full folder: images, video clips and audio, each called by name in the prompt. Every endpoint shares the same frame: a duration of 4 to 15 seconds (or auto, which lets the model pick a length from your prompt), seven aspect ratio options from 21:9 down to 9:16, and a generate_audio switch that's on by default.

The Building Blocks of Seedance 2.0 Prompts
Every prompt below is built from four parts. You won't always need all four, but when a clip comes back wrong, one of them is almost always missing.
Subject. Who or what is on screen, described by the details that matter to the shot. "A woman" is a placeholder. "A woman in a rain-soaked wool coat holding a paper cup" is a character.
Motion. The verb the whole clip hangs on. Name the action and its pace: steam curls, a cyclist leans into a bend, a door swings shut and rattles. One clear action per shot beats three competing ones.
Camera. ByteDance says the model plans camera language from your prompt, and it listens when you're specific: slow push-in, handheld follow, locked-off wide, orbit. For more than one shot, write the shots in order and say where each cut happens.
Sound. This is the block people forget, and it's the one Seedance 2.0 is built for. Audio is generated with the picture, so describe it the way you'd brief a sound designer: room tone, foley, music style, and any spoken line in double quotes. The each::labs text to video endpoint lists English plus Japanese, Indonesian, Spanish and Portuguese. Want a silent clip for your own score? Switch generate_audio off rather than writing "no sound".

Text to Video: Four Prompts That Start From Nothing
With no image to lean on, the prompt carries everything. Write the world first, then the action, then the camera, then the audio.
1. The single-take mood piece
A night ferry crossing a dark harbor, city lights smeared on black water. A deckhand in a yellow raincoat coils a rope by the rail, breath visible in the cold. Slow handheld push-in from the stern toward him, slight sway with the boat. Sound: low diesel engine hum, water slapping the hull, a distant foghorn, wind gusting across the microphone.
2. The three-shot mini scene
Shot 1: wide, locked-off view of a small bakery at dawn, warm light inside, empty street. Shot 2: cut to a close-up of flour-dusted hands pressing dough, quick and confident. Shot 3: cut to a medium shot as the baker slides a tray into the oven and smiles at the camera. Sound: radio playing soft jazz in the background, dough thuds on the counter, oven door creaks shut.
3. The spoken line
Medium close-up of a tired lighthouse keeper in his sixties sitting at a wooden table, an oil lamp flickering beside him, storm rain streaking the window behind. He looks up from his logbook and says quietly, "Third ship this week. They never listen to the light." Static camera, shallow depth of field. Sound: rain on glass, lamp hiss, a low rumble of thunder after his line.
4. The product hero shot
A matte black wristwatch rotating slowly on a wet slate surface, single hard key light from the left, droplets catching highlights. Camera orbits 90 degrees around the watch and ends on a tight close-up of the dial. Clean studio look, deep shadows. Sound: subtle mechanical ticking, a soft low synth swell that peaks as the camera stops.
Prompt 2's "Shot 1, Shot 2, Shot 3" isn't a trick. ByteDance's launch post says the model can work from text-based storyboards directly.
Image to Video: Write the Motion, Not the Picture
Here the image already answers who, what and where. Your prompt only needs to answer when and how. The endpoint takes image_url as the start frame and an optional end_image_url for the last frame, which is the closest thing to a guaranteed ending you'll get.
5. The living portrait
She turns her head slowly toward the window as the curtain lifts in a breeze, a strand of hair drifting across her cheek. Camera holds still, then drifts gently closer. Sound: quiet room tone, curtain fabric rustling, birdsong outside.
6. The food close-up
Hot broth is poured into the bowl from the top of frame, steam rising and curling through the light, noodles shifting as the liquid settles. Slow push-in at table height. Sound: liquid pouring, a gentle sizzle, chopsticks tapping the bowl rim.
7. The start-to-end transformation (use an end image)
Time-lapse from the first frame to the last frame: morning fog burns off the valley, shadows sweep across the fields, and the town lights switch on at dusk. Camera locked off on a tripod, no movement. Sound: wind fading into evening crickets and a faraway church bell.
8. The illustration that wakes up
Keep the hand-painted watercolor style. The paper boat drifts down the stream, ripples spreading in soft brushstrokes, a frog on the lily pad blinks and hops away. Slow pan right following the boat. Sound: trickling water, a light plucked-string melody, a small splash.
Prompt 7 is where end_image_url earns its place: give the model a dusk frame and it stops guessing where the day ends. Prompt 8 opens with a style line because illustrated stills can drift toward photorealism once they move.

Reference to Video: Cast Your Files by Name
This is the mode that makes Seedance 2.0 different, and the one where vague prompts fail fastest. You upload files to image_urls (up to 9), video_urls (up to 3, combined 2 to 15 seconds) and audio_urls (up to 3, combined 15 seconds or less), with no more than 12 files in total. Then you call each one in the prompt as @Image1, @Video1, @Audio1. An uploaded file the prompt never names is a file the model is free to ignore, so give every reference a job.
Two rules from the endpoint itself: audio needs at least one image or video alongside it, and the model page asks you to use AI-generated faces rather than real people.
9. Character plus location
@Image1 is the main character. @Image2 is the location. The character walks into the courtyard of @Image2 at golden hour, stops by the fountain and looks up at the balcony. Tracking shot from behind, then the camera swings around to a medium shot of her face. Sound: footsteps on stone, fountain water, distant city traffic.
10. Borrow the camera move
Follow the camera movement and pacing of @Video1. Replace the subject with the red sports car from @Image1 driving along the coastal road in @Image2. Keep the car's design and color exactly as in @Image1. Sound: engine note rising through the corners, tires on asphalt, sea wind.
11. Cut to the beat
Use @Audio1 as the soundtrack. The dancer from @Image1 performs in the empty warehouse from @Image2, hitting a pose on every strong beat. Cut between a wide shot and a low-angle close-up on the downbeats. Hard light through high windows, dust in the air.
12. The product line-up
@Image1, @Image2 and @Image3 are three colorways of the same sneaker. Show each one on a turntable against a clean pastel backdrop, one after another, with a quick cut between them. Each shoe holds for two beats, final shot shows all three side by side. Bright soft studio light. Sound: upbeat minimal electronic track, a soft click on each cut.
Every prompt tells the model what each file is for before anything else. Assign roles first, direct second.
Write the Sound Into the Shot
Sound is where Seedance 2.0 prompts get lazy, and it's a waste, because the model generates audio whether you direct it or not. Leave it blank and you get whatever ambience feels statistically likely. Direct it and you get a scene.
Think in layers, the way ByteDance describes the model's audio: music, ambient effects and voice as separate tracks. "Rain on glass, a muffled TV in the next room, he says 'I'm fine' without looking up." Three layers, one sentence.
Keep spoken lines short. A 6-second clip can't hold a monologue, and a rushed line sounds rushed. Put the attitude before the quote: "whispers," "snaps," "says flatly."

Fast, Mini or Standard: Pick the Variant for the Phase
All three tiers read the same prompts, so the choice isn't about prompting. It's about where you are in the work.
Standard is the finishing tier. It's the only one of the three with 1080p output, and the standard image and reference endpoints also list a 4k option. Use it once the prompt is locked and you want the keeper.
Fast (text to video Fast, plus Fast versions of the image and reference endpoints) runs at 480p or 720p with the same inputs as standard. It's the drafting tier: check that the camera move reads and @Image2 is actually used, then send the winner to standard with the same prompt.
Mini (text to video, image to video and reference to video) is also capped at 720p, with the same 4 to 15 second range, audio switch and reference slots. It's built for volume, as we covered in Seedance 2.0 Mini: Cinematic Clips on Repeat.
The workflow is boring and it works. Draft on Fast or Mini, fix one thing per run, then render the final on standard. Pin the seed while you tweak wording so you're comparing prompts, not luck (the endpoint notes results stay similar, not identical, with the same seed).
If your brief needs a single take longer than 15 seconds, that's a job for Seedance 2.5, which runs 4 to 30 seconds per clip. For a straight head-to-head with another family, see Seedance 2.0 vs Kling 3.0.
Mistakes That Waste a Take
Uploading references without naming them. Three images and "a man in a city" is a coin toss. Tag each file and say what it is.
Stacking actions. Five verbs in a 5-second clip give you a blur. Pick one, or split them into shots.
Leaving the camera to chance. No camera line tends to mean a slow, polite drift. Want energy? Ask for a whip pan or a quick cut.
Forgetting the clock. A three-shot scene squeezed into 4 seconds rushes every cut, and one small action stretched to 15 seconds invents filler. When unsure, auto reads the length from your prompt.
Frequently Asked Questions
How long should Seedance 2.0 prompts be?
Aim for 50 to 90 words that cover subject, motion, camera and sound. The each::labs endpoint accepts up to roughly 1,000 words, but extra length mostly adds competing instructions. When a clip comes back wrong, add the missing block, not more adjectives.
Does Seedance 2.0 generate sound and dialogue?
Yes, and it does it in the same pass as the video. Audio generation is on by default on every each::labs endpoint, covering voices, sound effects, ambience and music. Put spoken lines in double quotes. To get a silent clip, turn generate_audio off.
How do I reference my own images or video in a prompt?
Upload them on the reference to video endpoint and call them by tag: @Image1, @Video1, @Audio1, numbered in the order you add them. You can use up to 9 images, 3 videos and 3 audio files, 12 files in total, and audio needs at least one image or video with it.
Should I use Fast, Mini or standard Seedance 2.0?
Use the phase as your guide. Fast and Mini top out at 720p and suit drafts and high-volume runs. Standard adds 1080p for finals. All three read the same prompt, so draft on one and render on the other.



