20 text-to-video prompts across four categories, cinematic, animated, international dialogue, and e-commerce, plus 5 image-to-video prompts that come with their image prompt attached. Dialogue prompts carry their lines in native script inside double quotes, which triggers lip-synced speech. Four prompt shapes recur throughout. fal runs all three Seedance 2.5 endpoints on one API key at $0.0214 per 1,000 tokens.
In this guide, I'll go over 25 prompts for Dreamina Seedance 2.5 across four categories, so you can see what good looks like in terms of prompt structure, and also see what the AI video generator is capable of doing (hint: it's really awesome).
TL;DR
20 text-to-video prompts across four categories, and 5 image-to-video prompts that come with their image prompt attached.
The dialogue prompts carry their lines in native script inside double quotes, which is what triggers lip-synced speech.
Four prompt shapes appear throughout: the one-liner, the bracketed field block, staged beats, and the timed shot list.
fal offers the best place to run Seedance 2.5, as the platform runs all three Seedance 2.5 endpoints on one API key at $0.0214 per 1,000 tokens, which we'd quote as roughly $0.4730 per second at 720p.
Where can you access Seedance 2.5?
The best place to access Seedance 2.5 is on fal, billed per token of generated output, with no plan and no minimum spend.
You can access the model on our playground, agent, Sandbox, API, MCP server and CLI.
Here are the three endpoints that you can access, separated by what you're allowed to hand over alongside the words:
bytedance/seedance-2.5/text-to-video wants the prompt and nothing else (apart from the resolution, aspect ratio, duration, and generate_audio!), which covers the first twenty prompts below.
bytedance/seedance-2.5/image-to-video wants image_url for your first frame, and will accept end_image_url when you care where the clip finishes.
Prompts 21 to 25 use it.
bytedance/seedance-2.5/reference-to-video accepts image_urls, video_urls and audio_urls to a ceiling of 50 files, each one called by its upload position.
No prompt here needs it, though editing and extension work go through there.
Here's an example text-to-video call if you want to use fal's API:
import { fal } from "@fal-ai/client";
const result = await fal.subscribe("bytedance/seedance-2.5/text-to-video", {
input: {
prompt:
"An octopus finds a football in the ocean and excitedly calls its octopus friends to come and play. Cut scene to an octopus football game under the sea.",
},
logs: true,
onQueueUpdate: (update) => {
if (update.status === "IN_PROGRESS") {
update.logs.map((log) => log.message).forEach(console.log);
}
},
});
console.log(result.data);
console.log(result.requestId);
The five image-to-video prompts each need a still generated first, and Seedream 5.0 Pro answers to the same credentials, so the image call and the video call go in one file under one invoice.
Roughly a thousand other models are reachable the same way.
Audio generates alongside the picture at no extra token cost, and output is cleared for commercial work under fal's terms.
None of it strictly requires code either, since the playground, the Sandbox and fal Agent all reach the same endpoints, as do the CLI and the MCP server.
What are some of the best Seedance 2.5 cinematic prompts?
These are the five I'd run first.
Each one is built around a single physical event, and the waiting on either side of that event does as much work as the event itself.
1. Longtail chase through the canals
Settings: text-to-video, 10 seconds, 21:9, 720p, audio on.
A low chase camera skimming a hand's width above the water in a narrow canal at dusk, two wooden longtail boats running flat out between stilt houses with laundry strung overhead. The lead boat carries one woman at the tiller in a soaked linen shirt, the second is two lengths back and closing, its long-shaft propeller throwing a rooster tail that lights up orange in the last of the sun. She looks back once, then flattens herself along the gunwale as a low concrete footbridge comes up, and her boat goes under it with about thirty centimeters of clearance. The pursuing driver ducks too late, has to haul off into the pilings, and his boat slews, catches, and keeps coming. The camera stays locked at water level tracking parallel and never rises above the gunwale line, one continuous shot with no cuts. Anamorphic, hard low sun flaring off the wake, deep shadow under the houses, heavy grain. Two unmuffled engines at full throttle, water slamming both hulls, wood hitting concrete, birds coming off the roofline, no score.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
2. Four floors down a parking spiral
Settings: text-to-video, 15 seconds, 21:9, 720p, audio on.
Stage 1: Level six of a concrete multi-storey car park at night, empty, lit by sodium strips with every third one dead. A black sedan comes around the ramp at speed with its headlights off, and a white pursuit car follows one beat later with its lights on full. Ends with both cars committed to the same descending spiral, the sedan half a turn ahead. Stage 2: They take three continuous turns of the spiral. The sedan's rear tires break traction on every corner and the car runs wide, touching the outer wall twice and leaving paint on it. The pursuit car cuts the inside line and closes to one car length. Tire smoke and concrete dust hang in the ramp behind them. Ends with the two cars level with each other entering the next turn. Stage 3: The sedan brakes hard mid-corner. The pursuit car goes past on the outside and hits the wall square. The sedan reverses out of the corner and takes the down ramp alone. Ends on the wrecked pursuit car steaming against the wall with one headlight still burning, and the sedan's taillights dropping out of frame down the ramp. Camera: mounted low and outside the sedan's rear quarter through stages one and two, then whipping to a locked wide on the corner to catch the impact in stage three. Look: anamorphic, sodium orange against bare concrete, headlights raking a low ceiling, smeared highlights on the light strips, heavy grain. Audio: two engines echoing hard off concrete, tires howling on painted floor, one long scrape of metal along wall, the impact, then an idling engine and dripping. No score.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
3. Human tower in Tarragona
Settings: text-to-video, 16 seconds, 3:4, 720p, audio on.
A 16-second continuous shot of a castell going up in the main square of Tarragona on a hot afternoon, framed tall in 3:4 so the whole tower fits from the packed base to the top. Hundreds of castellers in white trousers and one shared shirt color, black sashes wound tight around their waists, crushed into a dense base with arms locked and heads down. [0s-4s]: Camera at chest height at the edge of the crowd, looking straight up the face of the tower. Four levels are already standing. A fifth climber goes up the backs and sashes of the level below, finds a foothold on a shoulder, and locks arms with the others on his ring. [4s-8s]: The camera rises straight up alongside the tower as the sixth level forms. The structure breathes and corrects constantly, shoulders shifting a centimeter at a time, sashes taking the load and creaking. [8s-12s]: The enxaneta, a child of about eight in a helmet, comes up the outside fast, hand over hand, using the sashes as rungs. Every casteller she climbs past holds absolutely still. The camera stays with her. [12s-16s]: She reaches the top, flattens herself across the shoulders of the last pair, and raises one hand. The square erupts. The shot ends on her hand up and the tower still standing. Every casteller wears the same shirt color, the tower keeps the same geometry, and there is only ever one child at the top. No level collapses and the tower does not fall at any point. Audio: gralla and drum playing underneath and speeding up as the tower rises, the crowd dropping to almost nothing while the child climbs and then roaring at the end, sashes creaking, breathing, bare feet on shoulders.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
4. Cormorant fishing on the Nagara River
Settings: text-to-video, 12 seconds, 16:9, 720p, audio on.
【Concept】: One continuous shot of ukai cormorant fishing on the Nagara River at full dark, no dialogue. 【Style】: Feature-film realism, firelight as the only key, black water, high shadow detail, 35mm, restrained grade with no color push. 【Duration】: 12 seconds 【Scene】: A narrow wooden boat on black water. An iron fire basket hangs off the bow on a pole, burning pine, throwing orange light about four meters across the surface. The usho stands in a straw skirt and a black linen headcloth with twelve cormorant lines gathered in his left hand. Wooded banks with no lights behind him. 【Action】: He shakes the gathered lines once to work the birds forward. Two cormorants surface at the edge of the firelight, one carrying a fish. He hauls that line hand over hand, lifts the bird aboard by the body, works the fish out of its throat into a wooden box, then puts the bird straight back over the side and it goes under. 【Camera】: Handheld from a second boat running parallel and slightly behind, medium wide, drifting with the current, one slow push toward the bow as the bird comes aboard. 【Audio】: Pine burning and cracking, embers hitting water, wet rope, birds calling and splashing, the hull knocking, the river running. No BGM and no subtitles.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
5. Thirty seconds to the bulkhead
Settings: text-to-video, 30 seconds, 21:9, 720p, audio on.
A 30-second continuous handheld take escaping a flooding cross-passage under a river, real-time throughout with no slow motion anywhere. Two people: an engineer in a hi-vis vest and a hard hat with a headlamp, and a younger colleague she is dragging along with her. Water is at knee height when the shot starts and rises visibly across the whole clip. Emergency lighting every ten meters with half of it already dead. 0 to 6s: Camera behind them at chest height, following. They wade hard toward a lit doorway forty meters off. The water is at the knee and moving against them. 6 to 13s: The younger one goes down on something underwater and she hauls him up by the vest without breaking stride. Water reaches mid-thigh through this stretch. The camera closes to two meters behind them and stays there. 13 to 19s: They reach the doorway and it will not open. She braces a boot against the frame and pulls the handle twice. The water is at the waist now and pushing both of them into the door. Her headlamp is the only light on either face. 19 to 25s: It gives, and the pressure behind it drives the door and both of them through into the next section, which is dry. They go down hard on dry floor with the water coming through after them. 25 to 30s: She gets up first, puts a shoulder into the door, and closes it against the flow while he throws the wheel lock over. The water stops. The shot ends on the two of them down on the floor, soaked, with the sound of the river on the other side of the steel. The camera never cuts and never rises above chest height. Water only ever rises. Both people keep the same clothing and gear throughout, and her headlamp stays the only moving light source. Audio: water moving in a confined concrete space with hard reverb, breathing and shouting with no intelligible words, a boot on a steel frame, the door giving, the flow through it, then the door closing and a sudden drop to dripping. No score.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
falMODEL APIs
The fastest, cheapest and most reliable way to run genAI models. 1 API, 100s of models
What are some of the best Seedance 2.5 animated prompts?
Five animated set pieces at the budget level animation actually gets broadcast at.
Two anime, one comic-book, one 3D feature, one stop-motion, and the medium is specified as hard as the action because the word animated on its own gets you a house style nobody asked for.
6. Expressway coming apart
Settings: text-to-video, 12 seconds, 21:9, 720p, audio on.
Late-eighties Japanese animation cel look with hand-inked highlights and heavy film grain, a girl in a red racing jacket flat out on a low sportbike along an elevated expressway at night while the roadway behind her collapses section by section into the city below. Neon towers on both sides, wet asphalt throwing their reflections up under the wheels, red warning lights strobing along the guardrail. She never looks back. The camera runs alongside her at hub height for the first half, then falls behind and rises as one span drops away a bike length behind her rear wheel, and the last thing in frame is her taillight going into a tunnel mouth while the section she just crossed goes down. Rear tire visibly loaded in the corners, the bike squatting under power, her jacket flogging against the airstream. Anamorphic composition, blown neon highlights, deep black in the shadows, no digital gloss. Two-stroke engine screaming through the gears, concrete tearing, rebar snapping one strand at a time, an alarm somewhere below, and one synth line that arrives on the first collapse and cuts out dead when she enters the tunnel.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
7. Something in the harbour
Settings: text-to-video, 12 seconds, 16:9, 720p, audio on.
【Concept】: A kaiju surfacing in a working harbour at night, seen from a fishing boat that is far too close. Twelve seconds, no dialogue, no subtitles. 【Style】: Modern theatrical anime, painted backgrounds with 3D camera moves, hard rim light from sodium dock lamps, volumetric mist, restrained color with one cold green accent. No chibi, no comedy, no speed lines. 【Duration】: 12 seconds 【Scene】: A commercial fishing harbour at 2am. Container gantries along the far quay, a breakwater with a green channel light, a small steel trawler with two crew on deck. Flat black water with an oily sheen. 【Action】: The water level around the trawler drops about a meter and holds, and everything on deck slides toward the low side. One crewman grabs the rail. A back rises out of the harbour between the trawler and the breakwater, tall enough to put the green channel light behind it, and keeps rising. Water sheets off it in curtains. The displaced water comes back as a wall and lifts the trawler bow-first. 【Camera】: Locked low on the trawler deck at gunwale height for the drop, then a single continuous tilt up that loses the top of the creature out of frame, then whipping down as the water returns. 【Continuity】: Two crew only. The green channel light stays on the same side of frame the whole time. Water that leaves the creature has to arrive somewhere. 【Audio】: Hull plates groaning, water draining away, then an enormous displacement roar, rigging slapping, one crewman shouting with no intelligible words, the channel bell. No score until the back clears the light, then one sustained low brass note.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
8. Swing through the interchange
Settings: text-to-video, 10 seconds, 21:9, 720p, audio on.
Comic-book animation with visible halftone dots in the shadows, hand-lettered impact shapes with no readable words, chromatic fringing on the highlights and the figure animating on twos while the background runs on ones. An original masked vigilante in a navy and copper suit, no existing character and no logos, swings on a cable through a four-level highway interchange at rush hour, passing under one deck and over the next, using a light gantry to change direction. A delivery van clips a barrier below and starts to jackknife, and the vigilante drops off the cable, lands on the van's roof, and rides it as it slews to a stop across two lanes. The camera holds a wide that lets the whole interchange read for the swing, then snaps to a low three-quarter on the van as the feet land. Ink lines thicken on contact frames. Cable takes tension, the roof panel dents where the boots hit, and the van's weight transfers visibly through the slide. Cable singing, wind, tires locking up, sheet metal denting, horns, and a brass and drums cue that hits on the landing frame.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
9. Night market, three drones
Settings: text-to-video, 14 seconds, 16:9, 720p, audio on.
Stage 1: Modern 3D animated feature style with stylized proportions, soft subsurface skin and appealing character animation. A covered night market, hot practical light from food stalls, steam and fryer smoke in the air. A boy of about twelve stands on a hoverboard at the top of the main aisle with a stolen data spool under one arm. Three matte black security drones the size of dinner plates rise over the stalls behind him and lock on. Ends with the boy setting his weight forward and the drones committing. Stage 2: He goes down the aisle fast, ducking under hanging lanterns and cutting between stalls. The drones follow in single file and take the same gaps a beat late. One clips a noodle rack and goes into a fryer with a burst of steam. He passes through a curtain of hanging duck and it swings back into the second drone. Ends with one drone left and the boy running out of aisle. Stage 3: The aisle ends at a rolling shutter half down. He drops flat on the board and goes under with a hand's width of clearance. The last drone does not adjust in time and hits the shutter square. Ends outside in the rain on the empty street, the boy still on the board, the shutter behind him ringing. Look: hot orange stall light against cool blue street, deep depth on the aisle wide, shallow on the close work, steam lit from behind throughout. Audio: hoverboard whine, three drone pitches that thin out to one as they drop away, woks and fryers, a vendor shouting, the shutter impact, rain. A percussion cue that loses an instrument each time a drone goes down.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
10. Thirty seconds in the diving bell
Settings: text-to-video, 30 seconds, 21:9, 720p, audio on.
A 30-second stop-motion sequence with hand-built puppets and practical miniature water, visible fabric weave, seams at the joints and fingerprints in the clay, shot on 35mm with real depth of field. A woman in a brass diving bell lowered on a chain into a green ocean trench, one porthole, one oil lamp inside, one speaking tube. Real-time throughout, no slow motion, no cuts. Shot 1 (0-5s): Inside the bell looking past her shoulder through the porthole. The chain pays out steadily. Silt and small fish drift past. She taps the speaking tube twice with a spanner. Shot 2 (5-11s): Outside the bell in wide, the chain running up out of frame. A single arm as thick as the chain comes into frame from below and wraps the bell once. The bell stops descending. Shot 3 (11-18s): Back inside. The oil lamp swings hard and goes out, leaving only green light through the porthole. She braces both boots against the wall as the bell rotates. A second arm crosses the porthole from the other side and blacks it out completely. Shot 4 (18-25s): She unhooks the lamp, smashes the glass, and holds the bare flame against the arm across the porthole. The arm releases. The bell drops free about its own height, then the chain snaps taut and takes the weight. Shot 5 (25-30s): Outside in wide again, the bell rising fast with the chain hauling, and the creature below it retreating down into the dark without following. The shot ends on the empty green water where it was. The same bell, chain, porthole, lamp and spanner throughout. The chain is always under load or visibly slack, never both. Puppet water moves as practical miniature water and not as digital fluid. Audio: chain links under strain, the bell's steel ringing when the arm lands, breathing inside a metal space with hard reverb, glass breaking, one long low animal note from outside the hull. Score enters only for the final rise.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
What are some of the best Seedance 2.5 international dialogue prompts?
Five genre two-handers, one language each, native script inside straight double quotes because that is what the model reads as a spoken line.
Each one has two people, two lines, and something at stake in the next thirty seconds.
11. Ten more seconds
Settings: text-to-video, 14 seconds, 4:3, 720p, audio on.
A 14-second continuous take inside a small hatchback parked on a Tokyo side street in the afternoon, engine running, framed 4:3 across both front seats. A woman in her thirties at the wheel in driving gloves and a black jacket, her younger brother in the passenger seat with a canvas holdall on his lap. Through the windscreen, the glass doors of a bank forty meters ahead. Real-time performance, one take, no cuts. 0-4 seconds: Both watching the doors. Her hands stay on the wheel. His right leg is going. An alarm starts inside the building, faint through the glass. 4-8 seconds: She checks the mirror once, then the doors again, and says, "出てこない。" Her brother does not answer and does not open his mouth while she is speaking. He reaches for the door handle. 8-12 seconds: She takes his wrist off the handle without looking at him and says, "あと十秒。" Only her mouth moves. He sits back. 12-14 seconds: Neither speaks. The alarm gets louder. The shot ends on both of them watching the doors with the car still stationary. The gloves, the jacket, the holdall, the car interior and the daylight hold from first frame to last, and each mouth moves only for its own line. Audio: clean lip-synced Japanese, an idling four-cylinder, a building alarm muffled through glass, a turn indicator ticking, his leg against the door card. No subtitles and no score.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
12. Between rounds
Settings: text-to-video, 12 seconds, 16:9, 720p, audio on.
A working fight arena in Mexico City between rounds, hot overhead ring light and a black crowd beyond it, a welterweight sitting on the stool with his right eye already closing while his trainer works in front of him with a swab and an enswell. Sweat and vaseline, ropes in the foreground, a cutman's hands entering frame from the left. The fighter looks past him and says, "No veo nada de este lado." The trainer does not stop working, takes the fighter's jaw in one hand to line his eyes up, and answers, "Entonces no lo dejes llegar a ese lado." The ten-second warning sounds and the stool goes out from under him. Handheld at the fighter's eye height inside the ropes, one slow push to a two-shot as the trainer takes his jaw, no cuts. Anamorphic, hard specular highlights on wet skin, deep falloff into the crowd, heavy grain. Clean lip-synced Mexican Spanish, a corner crowd, a bucket, tape being torn, the ten-second whistle, gloves touched together. No score.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
13. Not getting off
Settings: text-to-video, 12 seconds, 21:9, 720p, audio on.
Stage 1: The vehicle deck of an overnight car ferry mid-crossing, lit by yellow deck lamps that swing with the roll of the ship. Rows of cars lashed down under chains, sea noise coming through the hull. Two men stand in the gap between two parked trucks, one in a soaked wool coat, one in a company windbreaker holding a clipboard. The deck rolls and every chain on the deck takes up its slack at the same moment. Ends with both men bracing a hand against a truck flank as the roll peaks. Stage 2: The man with the clipboard does not look up from it and says, "몇 시에 내려?" The other man's mouth stays closed. The ship rolls back the other way and one badly secured sedan behind them shifts half a meter until its chains snap taut. Ends with that sedan settled at an angle and both men still squared up to each other. Stage 3: The man in the wool coat says, "나는 안 내려." Only his mouth moves. The other man finally looks up from the clipboard. Neither of them moves after that. Ends locked on the two of them with the deck lamps swinging across their faces and the chains still ringing. Camera: locked low in the gap between the two trucks at chest height, close enough to hold both faces, and it never moves while the entire frame rolls with the ship. Continuity: the same coat, windbreaker and clipboard throughout, the same cars in the same places apart from the one sedan that shifts, and the roll never reverses direction inside a stage. Audio: clean lip-synced Korean, a hull working in a swell, dozens of chains loading and going slack together, one sedan's suspension complaining, deck lamps creaking on their mounts, an engine room a long way below. No subtitles, nothing scored.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
14. Do not stop
Settings: text-to-video, 16 seconds, 16:9, 720p, audio on.
A 16-second continuous take inside the cab of an armored cash transport on a two-lane road outside Amman at midday, heat shimmer on the asphalt ahead, dust hills either side. The driver is in his fifties with his sleeves rolled, the guard beside him in his thirties with a clipboard on his knee. Ahead, a flatbed truck is turning across both lanes and stopping. [0s-5s]: Camera on the dashboard looking back at both men, windscreen and road visible over their shoulders. Neither speaks. The driver lifts off the accelerator. [5s-9s]: The guard looks up from the clipboard at the truck, then at the driver, and says, "شو هيدا؟" The driver keeps both hands on the wheel and does not open his mouth. [9s-13s]: The driver looks at the mirror, sees a second vehicle close behind, and says, "لا توقف. لا توقف!" Only his mouth moves. His foot goes back down and the cab pitches. [13s-16s]: Neither speaks again. The clipboard slides off the guard's knee. The shot ends with the truck filling the windscreen and the transport still accelerating. Both men keep the same clothing and seat position throughout, the clipboard exists in every frame until it falls, and the truck never finishes its turn. Audio: clean lip-synced Levantine Arabic, a diesel under load, air conditioning, the clipboard hitting the floor, gravel against the underbody. No subtitles. Nothing scored.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
15. Two minutes of air
Settings: text-to-video, 18 seconds, 21:9, 720p, audio on.
【Concept】: A diver and a boat pilot on the water at night off Rio while a police helicopter searchlight works the bay. One exchange, no resolution. 【Style】: Contemporary Brazilian crime cinema, anamorphic, near-black frame cut by one moving searchlight, hard highlights on wet neoprene and painted hull, fine grain, no stylized grade. 【Duration】: 18 seconds 【Scene】: A small open boat idling in black water a few hundred meters off a headland. Rio's lights along the shore behind. A diver in a black wetsuit hanging off the gunwale with his mask pushed up and a mesh bag clipped to his harness. A helicopter searchlight crossing the water in slow passes, arriving and leaving. 【Characters】: The pilot, a woman of about forty at the tiller in a soaked windbreaker. The diver, the same age, in the water with one arm hooked over the side. 【Action】: The searchlight passes over them and they both go still and low until it leaves. She looks at the mesh bag, then at the headland, and says, "Quantos minutos você tem?" He unclips the bag and pushes it up into the boat before answering, "Dois. Se você não me deixar aqui." She does not answer. She puts a hand on his forearm and keeps it there. The searchlight starts back toward them. 【Camera】: Locked at water level from the far side of the boat so both faces and the bag stay in one frame, no coverage and no cuts. 【Continuity】: One bag, one diver, one boat. The searchlight only ever moves in one direction per pass. Each mouth moves only for its own line. 【Audio】: Clean lip-synced Brazilian Portuguese, an idling outboard, water against a hull, rotor wash arriving and fading, a regulator venting once. No subtitles and no score.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
What are some of the best Seedance 2.5 e-commerce prompts?
Five spots written the way a broadcast ad is written, which means one idea, one payoff, and no product feature list anywhere in the prompt.
16. The gun and then nothing
Settings: text-to-video, 12 seconds, 16:9, 720p, audio on.
A 12-second athletics spot, one continuous take, real-time, no slow motion at any point. A sprinter in an unbranded charcoal kit in the blocks of lane four, a full stadium at night under white light. No logo and no readable text anywhere in frame. 0-3s: Camera on the track surface in front of the blocks, low enough that her knuckles are at lens height. Full stadium noise: crowd, a PA echo, camera shutters, a whistle. 3-6s: The starter's gun fires. On that exact frame every sound in the stadium cuts out and only her breathing remains, close and dry, as though the microphone were inside her chest. She comes out of the blocks and takes three strides. 6-9s: The camera pulls back down the track faster than she accelerates so the gap between her and the lens opens. Her spikes bite, the track compresses under each plant, and her shadow runs beside her. The stadium is still visibly roaring and still inaudible. 9-12s: On the fourth stride the crowd comes back all at once at full volume. The camera stops and lets her run through frame and out of it. The shot ends on empty track with the blocks still in it. Same athlete, kit, lane and light throughout. Spikes make contact with the surface on every plant. No product appears in the frame at any point. Audio: full stadium for three seconds, then breath alone with a low room tone under it, then the crowd returning on one frame. No music and no voiceover.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
17. Storm behind, water below
Settings: text-to-video, 14 seconds, 21:9, 720p, audio on.
Stage 1: An unbranded electric SUV in matte slate climbing a flooded single-track mountain road at dusk, water running across the surface a hand's depth deep. A black storm front is already over the pass behind it, lightning inside the cloud, the last clear light ahead. No badge, no logo, no readable text on the vehicle. Ends with the SUV entering the deepest section and the front wheels pushing a bow wave. Stage 2: The road steepens and the water gets deeper. Each wheel finds a different level, the body stays level, and the suspension works visibly at all four corners. Water sheets off the front tires and up the flanks. The storm front closes and the light drops. Ends with the SUV at the deepest point, water at hub height, still moving. Stage 3: It comes out onto the crest into clear air with the storm now behind and below it. Water pours off the underbody onto dry rock. It stops. The camera keeps going and leaves it small in a huge frame. Ends wide with the SUV on the crest, the storm filling everything behind, and dry road ahead. Camera: a low tracking arm level with the water for stage one, a rising crane through stage two, and one continuous pull back to an extreme wide for stage three. Never cuts. Look: anamorphic, cold storm blue against one warm shaft of remaining sun, no lens flares, no gloss, real weight in the water. Audio: water displacing around a moving body, tires on submerged gravel, a restrained electric whine, thunder that arrives late, then wind on the crest and water draining off metal. One low sustained cue that enters on the crest.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
18. Out of the ice
Settings: text-to-video, 8 seconds, 1:1, 720p, audio on.
An extreme macro square-frame drinks spot, hard single-source light against black. A hand comes into frame and pulls an unbranded aluminum can straight up out of a bed of crushed ice, ice grinding and collapsing into the cavity behind it, meltwater running off the base in two threads. The can is beaded with condensation and the beads slide and merge as it lifts. A thumb goes under the tab and cracks it open, and the pressure release throws a fine mist off the opening that catches the light. The can tilts and pours over a glass of ice, and the liquid climbs the glass while the foam head rises and settles just under the rim. Camera locked, one slow push in, focus only ever on the can and then only on the pour, no logos and no text anywhere on the can or the glass. Audio is the whole ad: ice grinding, meltwater, the tab cracking, the gas release, liquid onto ice, and the ice shifting once as the glass fills. Nothing else, no score.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
19. Touchdown on the lagoon
Settings: text-to-video, 16 seconds, 21:9, 720p, audio on.
【Concept】: A single-engine floatplane landing on a lagoon at golden hour and taxiing to a timber jetty. Sixteen seconds, no people visible until the last beat, no dialogue. 【Style】: Airline brand film, wide anamorphic, warm low sun, real aerial haze, no color push and no lens flares. No livery, no registration marks, no readable text on the aircraft. 【Duration】: 16 seconds 【Scene】: A shallow turquoise lagoon inside a reef at about 5pm. A timber jetty on stilts running out from a palm shoreline. Flat water with the reef line visible as a color change. Low sun from camera left. 【Action】: The floatplane comes in low over the reef with the sun behind it, holds a shallow descent, and touches down on the rear of both floats first. Spray comes off in two long plumes and the nose settles as the front of the floats take the water. It slows, the plumes shorten to nothing, and it turns onto the step toward the jetty. On the last beat one figure walks out along the jetty to meet it. 【Camera】: One continuous move that starts high and wide on the reef, descends and swings to follow the aircraft down onto the water, then settles low at water level as it taxis toward the jetty. 【Continuity】: One aircraft, two floats, one jetty. The water surface stays flat outside the aircraft's own wake, and the wake persists behind it for the whole taxi. 【Audio】: A radial engine throttling back, both floats hitting water, spray, water under the hull during the taxi, reef surf a long way off, one bird. A warm sustained string cue from touchdown onward. No voiceover and no subtitles.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
20. Twelve meters of silk
Settings: text-to-video, 12 seconds, 9:16, 720p, audio on.
Stage 1: A vertical 9:16 luxury fashion spot on a black stage with a polished black floor, one hard key from high camera left and nothing else lit. A model stands center frame in a floor-length gown with about twelve meters of unlined silk in the train, completely still, the fabric hanging dead. No brand, no logo, no text. Ends on the still figure with the silk at rest. Stage 2: A wind machine comes up from below and behind her. The train lifts off the floor in one continuous movement and takes flight above her head, filling the top two thirds of the vertical frame, edges lit hard against black. She holds position and lets it move. Ends with the silk at full extension overhead and her still centred. Stage 3: She turns a slow half revolution on the spot. The silk wraps the turn a beat behind her, crosses the key light so it goes momentarily translucent, then trails out on the far side. The wind machine drops off and the fabric comes back down around her over about two seconds. Ends with the train settled on the black floor in a new shape and the model facing away. Camera: locked vertical, tight enough that the silk exits the top of frame at full extension, never moves. Continuity: one gown, one model, one light source. The silk behaves as unlined silk with real drag and real weight, never as digital cloth, and it never passes through her body. Audio: a wind machine coming up and dropping off, silk moving against itself, the hem across a hard floor. One cello note held under the whole thing and nothing percussive.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
What are some of the best Seedance 2.5 image-to-video prompts?
These five come in pairs.
The image prompt builds the first frame, and the video prompt only describes what changes after it.
Aspect ratio on this endpoint follows the input image, so the frame shape gets decided in the image prompt.
21. Forty floors up, no window cleaner
Settings: image-to-video, 12 seconds, 720p, audio on, aspect ratio inherited from the still.
Image prompt: A cinematic film still, wide 21:9 frame, a woman in a climbing harness and a charcoal suit standing on the platform of a window cleaning rig forty floors up the glass facade of a tower at sunset. The city drops away below and behind her, reflected upside down in the glass beside her. She has a suction cup handle in her left hand and a glass cutter in her right, and there is no cleaning equipment on the rig at all. Wind is visibly moving her jacket and hair. Hard low sun from camera right blowing out one half of the glass, the other half in deep blue shadow. Anamorphic, heavy grain, no text or signage anywhere in frame.
Generated using Seedream 5.0 Pro on fal, an AI model from ByteDance.
Video prompt: She sets the suction handle against the glass, leans her weight onto it, and starts the cutter in one continuous arc. The rig drops about a meter without warning and stops hard on the cable, and she goes down with it, keeps the suction handle planted, and hangs off it with both boots off the platform for a moment before she gets a foot back on. The cutter stays in her hand the whole time. She looks up once at the cable, then puts the cutter straight back on the glass and finishes the arc. Hold her exact suit, harness, hair and both tools from the still, keep the camera locked in the same wide position outside the rig for the whole clip, and never show the top or bottom of the building. Wind across the microphone, the cutter on glass, the rig cable snapping taut and ringing, the platform grating under her boots, city traffic a very long way down. No score.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
22. The hand still works
Settings: image-to-video, 10 seconds, 720p, audio on, aspect ratio inherited from the still.
Image prompt: A theatrical anime film still in a 16:9 frame, painted background with a photographic sense of scale. A young pilot in a torn flight suit stands on the shoulder plate of a downed humanoid mech, half submerged in the flooded street of a ruined city at dawn. The mech lies face up in brown water, one arm above the surface, the head canted back with its eye lens dark and cracked. Water to the pilot's ankles on the shoulder plate. Broken towers behind, mist on the water, cold blue dawn with one warm break in the cloud. No logos and no readable text.
Generated using Seedream 5.0 Pro on fal, an AI model from ByteDance.
Video prompt: Stage 1: She crouches and wipes silt off the eye lens with her forearm. Nothing happens. Water moves gently around the shoulder plate. Ends with her hand flat on the lens. Stage 2: The eye lens comes on from the center outward in a deep amber and holds, throwing her shadow up onto the mist behind her. She takes her hand back and stands. Ends with the lens fully lit and her upright on the shoulder. Stage 3: The mech's raised arm closes its fingers once, slowly, and the movement pushes a wave down the flooded street away from the camera. She grabs a cable to stay on her feet. Ends with the wave travelling away, the fingers closed, and the lens still lit. Hold her exact flight suit, hair and position from the image, keep the mech's design and the city behind it unchanged, and keep the camera locked in the same framing throughout. Water always moves away from whatever displaced it. Silt on metal, water against a steel hull, a deep servo winding up under the surface, hydraulics taking load, her breathing. Score enters only when the lens lights.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
23. Do not look back
Settings: image-to-video, 10 seconds, 720p, audio on, aspect ratio inherited from the still.
Image prompt: A cinematic film still in a 16:9 frame, two motorbike riders stopped side by side at a railway level crossing in Hanoi at night, barriers down, red warning lamp on. The rider on the left is a woman of about thirty in a denim jacket with her visor up, the rider on the right a man the same age in a reflective delivery vest with a hard case on the back of his bike. Wet road throwing the red lamp back up at them. Behind them, one headlight of a stationary car about twenty meters back. Practical light only, deep shadow, fine grain, no readable signage.
Generated using Seedream 5.0 Pro on fal, an AI model from ByteDance.
Video prompt: A 10-second continuous take, one shot, real-time, camera locked in the same position as the still. 0 to 3s: Neither rider speaks. She checks her mirror once. The red lamp keeps flashing and a bell starts. 3 to 6s: She keeps facing forward and says, "Có người theo mình." Her visor stays up. His mouth stays closed while she is speaking. 6 to 9s: He starts to turn his head toward the car behind and she says, "Đừng quay lại." Only her mouth moves. He faces front again. 9 to 10s: Neither speaks. The barrier begins to lift. The shot ends before either bike moves. Hold both faces, the denim jacket, the courier vest, the hard case, both bikes and the red lamp from the still. The car's headlight behind them never moves or changes. Clean lip-synced northern Vietnamese, two idling engines, a crossing bell, rain off the barrier, a train a long way off. No subtitles and no score.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
24. Caught a hand's width up
Settings: image-to-video, 12 seconds, 720p, audio on, aspect ratio inherited from the still.
Image prompt: A cinematic product still in a 21:9 frame, an unbranded flagship phone in brushed titanium tipping off the edge of a marble cafe table, already past the point of balance and starting to fall. Shot from floor level so the polished marble floor fills the bottom third and the table edge and phone are against a bright window above. A hand is visible entering frame at the top, too far away and too late. Everything else in the cafe is out of focus. Hard window light, hard reflections in the marble, no logos, no screen content, no readable text.
Generated using Seedream 5.0 Pro on fal, an AI model from ByteDance.
Video prompt: 【Motion】: The phone falls, rotating about a quarter turn as it goes, and the hand comes down into frame after it, faster. The fingers close around the phone roughly a hand's width above the marble and the whole thing stops dead. The hand's momentum carries a little further and pulls up short. The phone is never touched by the floor. The hand then lifts it back up and out of the top of the frame. 【Camera】: Locked at floor level, no move at all, focus racking from the table edge down to the point of the catch and staying there. 【Hold】: The phone geometry, the titanium finish, the marble pattern, the table edge and the window light stay exactly as the still has them. One phone and one hand only. 【Physics】: The phone accelerates as it falls and does not float or slow. The rotation is caused by how it left the table. The hand decelerates the phone over a real distance and does not stop it instantly. 【Audio】: Cafe room tone that thins out to almost nothing during the fall, the small sound of skin closing on metal, one breath, then the room coming back. No impact sound at any point, because there is no impact. No score.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
25. Jet ski, phone, no case
Settings: image-to-video, 12 seconds, 720p, audio on, aspect ratio inherited from the still.
Image prompt: A vertical 9:16 phone photo with slightly imperfect framing, a woman in her late twenties sitting astride a stationary jet ski in a sheltered bay in the afternoon, wearing a life vest over a wetsuit top, hair already wet. She is holding an unbranded phone in a clear waterproof case up toward the camera at arm's length, which is how the shot is being taken. Real phone camera look: mild lens distortion, water spots on the lens, ordinary skin texture with visible pores, no retouching and no beauty filter, available light only. No text and no logos anywhere.
Generated using Seedream 5.0 Pro on fal, an AI model from ByteDance.
Video prompt: A 12-second vertical handheld take shot on the phone that is in frame, one continuous shot, natural performance, real handshake throughout. Beat 1 (0-3s): She looks into the lens with the engine idling under her and says, "Everyone keeps asking if I actually trust this thing." She turns the phone once so the case is readable as a case and not a screen. Beat 2 (3-7s): She stops talking, puts her free hand on the throttle, and accelerates hard. The bow lifts, spray comes up over both of them and across the lens, and the framing goes wild for about a second before she gets it back on her face. She is laughing and not talking. Beat 3 (7-10s): Still moving, she holds the phone out to the side at arm's length and dunks it fully under the water beside her for two seconds, then brings it back up streaming. The picture continues without interruption. Beat 4 (10-12s): She wipes the lens with her thumb, gets her face back in frame, and says, "Still filming. That's the whole review." Her mouth moves only during her own two lines. Hold the same face, hair, life vest, case and light from the still, and keep it vertical and handheld the whole way with no cuts. Clean lip-synced English, a two-stroke engine under load, water slapping the hull, spray across the microphone, the muffled underwater seconds, wind. No score.
Generated using Dreamina Seedance 2.5 on fal, an AI model from ByteDance.
Recently Added
Start building with Seedance 2.5 on fal
One API key opens all three Seedance 2.5 endpoints: nothing recurring, and the invoice only counts pixels and seconds you actually asked for (plus any reference video you send).
You can sketch a sequence from text, animate a still, pin both ends of a transformation, hold a character across a scene change with up to 50 references, or edit and continue footage you already have, all from the same integration.
Signing up for fal is free, and all of it runs in the playground before you write any code.
Seedance 2.5 prompt FAQs
Do I have to use timestamps?
No.
They're worth the typing when one take has to hold several events in a fixed order.
On a shot carrying a single idea, they add constraints for no return.
Staged beats and numbered shots work as alternatives, so pick whichever matches the way the shot is built.
How do I get lip-synced dialogue?
Put the spoken line inside straight double quotes.
That works with non-Latin scripts too, which is why the Japanese, Korean, Arabic and Vietnamese prompts above carry their lines in native script.
Which aspect ratios can I set?
On text-to-video, you can set 21:9, 16:9, 4:3, 1:1, 3:4, 9:16, or leave it on auto.
On image-to-video, the ratio follows the input image, so the still decides it.
