ByteDance's newest video model makes 30 seconds of video with sound in a single take, which is enough room for a whole ad to play out from hook to payoff. Every clip on this page shows the exact prompt behind it, so you can borrow what works and swap in your product.
One continuous shot, no cuts. Photoreal, 35mm, smooth gimbal tracking, dusk after monsoon rain, wet reflective surfaces, saffron and deep green palette. [0-5s] On a wet rooftop, a child's hand releases a folded paper airplane. The camera follows just behind it as it lifts into the wind. Sound: wind, distant traffic. [5-11s] The plane dives between buildings and skims over a crowded vegetable market, past hanging marigold garlands. Sound: vendors calling, oil sizzling. [11-18s] It slips through the open window of a moving commuter train. Passengers, ceiling fans spinning. Sound: rhythmic train clatter. [18-25s] It exits the far window, glides across a schoolyard and through a classroom door. Sound: chatter fading to quiet. [25-30s] It lands on a wooden desk. A hand unfolds it to reveal a crayon drawing of the same rooftop. Camera settles close. Sound: paper rustle, distant thunder.
Settings
Made with Seedance 2.5 on 10/08/26. Trimmed only, nothing else touched.
One unbroken 30-second shot following a paper airplane from a rainy rooftop, over a market, through a moving train and into a classroom, where it is unfolded to show a drawing. Made with Seedance 2.5.
Seedance 2.0 stopped at 15 seconds, so anything longer meant stitching clips that never quite matched. This one holds a single shot for the full 30, and you can watch each line of the prompt land on screen.
Click a line to jump to that moment.Tap a line to jump to that moment.
Writing an ad? Plan the 30 seconds the way you'd storyboard them: what the viewer sees first, when your product shows up, and what they're left with.
Same prompt on both models, played side by side. 2.0 fits the whole sequence into 15 seconds, and 2.5 gives it the full 30.
The prompt on bothOne continuous take, warm documentary look, small café in morning light. A barista grinds beans, tamps, pulls an espresso shot, steams milk and pours a leaf of latte art. Continuous close tracking. Sound: grinder, steam wand, café ambience.
A barista grinding beans, pulling a shot and pouring latte art in one continuous take, made with the same prompt on Seedance 2.0 and Seedance 2.5.
| Feature | Seedance 2.0 | Seedance 2.5 |
|---|---|---|
| Longest single clip | 15 seconds | 30 seconds |
| References per video | 9 images, 3 videos, 3 audio clips | 30 images, 10 videos, 10 audio clips |
| Highest resolution | 4K | 1080p on Predis.ai |
| Editing | Edit and extend clips | Adds edits to one moment by timestamp |
| Sound | Made with the video | Made with the video, always on |
Hand it a product photo, a character, a motion clip and a track, and say what each one is for. It takes up to 50 references at once, which is how an ad ends up looking like your brand instead of stock footage.
Hover a reference to see where it shows up.Tap a reference to see where it shows up.
The result
A vertical sneaker ad built from 4 references: an invented skater, a product photo, a skate-trick clip and a music track.
Point at a timestamp and say what should change. Everything before and after stays put, so a wrong jacket color no longer costs you the whole take.
Before
After
Each one is a real generation with nothing hidden. Copy it as is, or swap in your product and see what changes.
A macro product shot where the glass and the liquid carry the whole frame.
A macro shot of a perfume bottle in heavy rain, with water running down the glass.
Glossy commercial look, cool blue palette. A cut-glass perfume bottle stands on wet black stone in heavy rain. Slow push-in as drops hit the glass and run down its sides, and warm light catches the bottle at the end. Sound: steady rain, one soft chime at the end.
A two-person dialogue scene, written to test lip sync.
Two strangers talk on an empty train platform at night under orange station lights.
Neo-noir, sodium-orange station lights, light haze. A woman in a camel coat holding a coffee cup and a man in a dark overcoat talk on an empty train platform at night. Slow push-in from a wide two-shot to over his shoulder, as a train's headlights appear behind her. Sound: their conversation, distant rail hum, the train arriving.
A creator-style unboxing, talking straight to camera.
A creator unboxes a pair of headphones at a desk and talks to camera.
Phone-camera realism, soft window light. A creator in a cream knit sweater sits at her desk, opens a black box and lifts out a pair of cream over-ear headphones. She puts them on and talks to camera about her first impressions. Static selfie framing. Sound: her voice, tissue paper, room tone.
Movement cut to an uploaded track.
A dancer moves to an uploaded track in an empty parking garage under colored lights.
High-contrast neon, magenta and cyan, light haze. A dancer in a white tank top moves through an empty parking garage, hitting every beat of Audio 1, dropping low to the floor and rising toward the camera. Slow push-in from a wide shot to a low close-up. Sound: Audio 1 only.
A clay-style explainer with words on screen.
A clay-style animation of a seed growing into a tree, with labels on screen.
Claymation, earthy pastels, tabletop set with paper clouds and a sun on strings. A hand presses a clay seed into the soil, and it sprouts, grows and becomes a small tree as a bird lands on a branch. Locked-off camera. An on-screen label changes with each stage: "1. Plant", "2. Sprout", "3. Grow", "4. Tree". Sound: calm narration, soft music.
A handheld walk up to a street food grill at night.
A cook grills skewers over open flames at a night market stall and talks to camera.
Handheld, warm string lights, smoke, night. The camera moves in on a street food stall as the cook turns skewers on a flaming grill, then holds a tray up to the camera and talks about the food. Sound: sizzle, crowd chatter, the cook's voice.
No prompts in this category yet.
Every model has weak spots, and these come from ByteDance's own notes and from people using it for real work. Knowing them up front saves you credits.
Quick, complex movement can morph or wobble, and upscaling won't clean it up.
FixSlow the action down in your prompt, or try a model built for action.
With several people touching or reacting to each other, gestures can come out wrong.
FixSpell out exactly who does what to whom.
There's no 4K, and most platforms top out at 720p.
FixNeed 4K? Seedance 2.0 makes it, for clips up to 15 seconds.
It guesses a voice from your references and sometimes adds an accent you didn't ask for.
FixName the voice and accent in plain words in your prompt.
Prompts written for Seedance 2.0 can glitch here, and rapid one-second cuts tend to fall apart.
FixWrite fewer, longer beats and let each one breathe.
A good take often needs a few attempts, and each 30-second attempt costs similar credits.
FixDraft on Seedance 2.0, then switch to 2.5 for the final take.
Real people's faces, selfies included, are reportedly blocked.
FixUse invented characters.
Open the Model Studio and choose Seedance 2.5 from the model list.
Split your 30 seconds into 4 or 5 moments, and for each one say what we see, what the camera does and what we hear. Give every reference a job, like "Image 1 is the product."
If one moment is wrong, edit that moment instead of starting over.
A prompt written like a storyboard works better than one long paragraph. Here's the shape we start from.
[The look: lens, light, palette, one shot or several] 0-6s: [What we see]. Camera: [move]. Sound: [what we hear]. 6-12s: [What we see]. Camera: [move]. Sound: [what we hear]. 12-20s: [Your product appears]. Camera: [move]. Sound: [what we hear]. 20-30s: [How it ends]. Camera: [move]. Sound: [what we hear]. References: Image 1 is [the product]. Image 2 is [the character]. Audio 1 is [the track].
What is Seedance 2.5?
It's ByteDance Seed's video model, and it makes up to 30 seconds of video with sound in one take. You can guide it with up to 30 images, 10 video clips and 10 audio clips.
When did Seedance 2.5 come out?
ByteDance announced it on June 23, 2026 and released it on July 31, 2026.
Is Seedance 2.5 free?
It's a paid model on Predis.ai, and you can try it during the 3-day trial. The trial needs a card and moves to a paid plan unless you cancel.
What is Seedance 2.5 used for?
Anything that needs a full scene in one take: 30-second ads, product films, short dialogue scenes, music videos and explainers.
How do I use Seedance 2.5?
Choose it in the Predis.ai video maker, write your prompt as timed beats and add any references. The prompts on this page are a good place to start.
Does Seedance 2.5 make 4K video?
No. Most platforms, Predis.ai included, offer up to 1080p, and Seedance 2.0 is the version with 4K.
How is it different from Seedance 2.0?
It doubles the longest clip from 15 to 30 seconds and takes far more references. Seedance 2.0 still has 4K and usually costs less for short clips.
Is there a Seedance 2.5 API?
Yes, through ByteDance's BytePlus ModelArk and providers like fal and Together AI. On Predis.ai you don't need one, since it's built into the video maker.
Can I turn the sound off?
Not when you make the video, because Seedance 2.5 always generates sound with it. You can mute or replace it afterward.
Start from any prompt on this page and swap in your own product.