Seedance 2.5 prompt guide

How the creators getting 15 to 30 seconds out of one prompt actually write them, sorted by the shot you want. A template for each, the checklist, the mistakes, and 96 clips to learn from.

Pick an effectSend us yours

Subway carriage fight with a headphone lock

Shot-by-shot timeline · Seedance 2.5 per the author · Oct 28, 2026

One prop gets its own lock paragraph, a pair of over-ear headphones, and every timed block repeats that they stay on, the same restate-after-motion habit this guide recommends for colours behind cover. Damage builds on one object: an emergency window cracks, cracks further, then shatters and becomes the exit for two attackers. Slow motion is allowed at one impact only, and the last three seconds hold a still, alert end state.

Close combat and impactChases, racing and fast movement

Read @MissDelulu9's prompt
FORMAT: Exactly 30 seconds, 16:9 widescreen, ultra-photorealistic live-action.

FEMALE CHARACTER:
A beautiful adult Japanese woman with a delicate oval face, fair natural skin, large dark-brown eyes, straight long black hair with neat blunt bangs, and a calm but intensely focused expression.

STRICT HEADPHONE LOCK: From the very first frame to the final frame, she wears large, matte-black over-ear headphones covering both ears. Both ear cups and the headband remain clearly visible and securely positioned over her head throughout every punch, jump, kick, fall, and rapid camera movement. Her headphones never disappear, fall off, change shape, or get replaced. Her hair moves naturally around the headphones. Do not use tiny earbuds.

EXACT OUTFIT: Fitted black ribbed long-sleeve turtleneck sweater, high-waisted black pleated leather mini skirt with a wide waistband, semi-transparent black stockings, chunky black lace-up combat boots with low block heels, and a small gold hair accessory. The same face, hairstyle, headphones, and outfit remain consistent throughout the video.

ENVIRONMENT: A moving KRL subway carriage at night, with a narrow aisle, stainless-steel poles, overhead handrails, passenger seats, flickering fluorescent lights, and a cracked emergency window. Several gangsters surround her. The carriage becomes progressively damaged during the fight.

0–7 SECONDS-SUDDEN AMBUSH

The handheld camera rushes through the narrow aisle as three gangsters attack from different directions. The woman dodges a punch, counters with a fast strike, drives another attacker into a metal pole, and knocks the third onto the floor. A fourth attacker charges toward her near the emergency window. Her long black hair swings naturally, but her large over-ear headphones remain firmly on her ears.

7–14 SECONDS- WALL-REBOUND KICK

Without a visible cut, an attacker charges. She plants one foot against the subway wall, twists diagonally upward, and launches a powerful flying kick into his upper chest. Use slow motion only during the impact. The emergency window cracks further as the attacker recoils into the carriage. She lands, grabs the overhead handrail with both hands, swings forward, and kicks another attacker into the seats and metal pole. Both headphones remain visible and securely positioned.

14–21 SECONDS- SUBWAY CHAOS

The camera whip-pans as two attackers strike from opposite directions. She blocks one attack, counters with an elbow strike, and pushes him against the damaged window frame. Another attacker tackles her toward the seats, but she regains her balance and counters with a spinning kick. The emergency window shatters outward, and one attacker disappears through the opening. Maintain the same character identity and uninterrupted headphone placement.

21–27 SECONDS- FINAL FIGHT

A final gangster charges with a broken bottle. She dodges as it smashes against a metal pole. They struggle between the seats and handrails. She traps his wrist, counters with rapid body strikes, grabs an overhead handrail with one hand, and launches a powerful flying knee. The attacker crashes backward through the shattered window opening and disappears from view.

27–30 SECONDS- COLD ENDING

The camera gradually steadies inside the damaged carriage. Broken glass covers the aisle, and defeated attackers lie near the seats and bent pole. The woman stands in the center aisle, breathing heavily but remaining alert. Her long black hair is slightly messy, sweat glistens naturally on her skin, and her large matte-black over-ear headphones are still clearly visible over both ears. She lowers into a combat-ready stance and scans the dark end of the train with cold, emotionless eyes. The fluorescent lights flicker.

Original post ↗

Pick the shot you want

  1. 01 Close combat and impact Fists, blades and grapples that read as real exchanges of force rather than two people performing moves. 44 clips →
  2. 02 Energy, spells and blasts Sword waves, lightning and explosions that look like the result of an action, not a filter over it. 32 clips →
  3. 03 Chases, racing and fast movement Speed built from distance, obstacles and parallax, not from shaking the camera. 38 clips →
  4. 04 Performance and emotion Quiet scenes carried by gaze, hands, props and distance rather than by adjectives. 36 clips →
  5. 05 Sport, dance and rhythm Athletic and rhythmic movement where body phases and edit rhythm carry the energy. 10 clips →
  6. 06 Transformation and style shifts Live action turning into manga, weapons changing form, worlds being painted in, with clear boundaries. 20 clips →
  7. 07 Camera moves and transitions Match transitions, occlusion wipes, infinite zooms and drone paths that connect for a reason. 21 clips →
  8. 08 Giants, armies and epic scale Monsters, armies and cosmic journeys that feel huge because the prompt sets a human reference first. 26 clips →
  9. 09 Horror, suspense and POV Fear built by limiting what the camera can see and delaying the reveal. 8 clips →
  10. 10 Product, cars and trailers Trailers and product spots where each cut has one job and the ending sells the subject. 6 clips →
  11. 11 Atmosphere and realism Documentary, found-footage and naturalistic looks built from light, material and camera behaviour. 41 clips →

Seven rules that hold for every shot

  1. If you do not write the motion, it will not move

    Seedance 2.5 animates what the prompt assigns and leaves the rest close to static. Give every layer a job: the lead, the opponent, the extras in the background, the environment (wind, water, debris) and the camera. Crowds and armies need described behaviour, varied looks and a direction of travel, not just a headcount. The strongest prompts in the library assign motion to background enemies, falling debris and cloth as deliberately as they do to the hero.

  2. Lock identity once, at the top, and say which reference does what

    Put the full description of each character, weapon and location in one block before the timeline, then use short names in the beats. When you attach images, say which image controls the face and outfit, which controls the set, and which only sets the first frame. Forbid the usual reference leaks: turnaround sheets, grey studio backdrops, borders, text. Identify characters by their appearance, never by which side of the frame they stand on, because positions change the moment they circle each other.

  3. Write cause and effect, not effect words

    A beat that works answers a short chain: starting position, motive, path, contact or evasion, the other side's reaction, what the environment keeps, the new position. Two fighters clash fiercely tells the model nothing. A steps in, B slips left, A's staff breaks the railing behind B, B ends up on A's right gives it something to stage. Shake, sparks and cracks belong after contact, never before.

  4. Give the camera a subject and a reason

    Listing ten camera terms does less than stating what the lens is following (a face, a blade tip, the contact point, an escape route), why it moves (the opponent advances, a new threat appears), the physical path it takes around obstacles, and where attention lands at the end of the beat. Match the camera to the genre: fast orbits suit duels, while a locked surveillance frame or a trembling flashlight POV is the right choice for horror.

  5. Pick the structure that fits the job

    Timeline prompts (0-3s, 3-6s, often in three-second blocks) suit fights, trailers and anything with many scheduled beats. Treat the timecodes as a plan, not a frame-accurate contract. A continuous action chain, written as flowing prose, suits one space and one unbroken cause and effect: a long duel, a transformation, a frozen-time walk. A short JSON block works when the shot is one situation with few variables, such as a POV horror walk. JSON is a writing style, not a special input format the model requires.

  6. Keep a state ledger and give the ending its own time

    Long takes fail when the world resets. Whatever breaks stays broken. Enemy counts are either finite (and can be cleared) or endless (and come in layers), never both. A prop that changes shape is the same object changing, not a new one appearing. The final state (who stands where, sword sheathed or not) needs its own one to three seconds. Several sampled cases hit the overall shape and still missed the planned ending.

  7. Budget effects and pace by genre

    Grade abilities so the finale has room to escalate; if every swing is a nuke, the climax has nothing left. Keep silhouettes, weapon paths and contact visible through the effects. Allow one slow-motion window at most and say exactly where it is. Choose density by genre: nonstop action suits a duel, but a breakup scene, a found-footage reveal or a documentary needs space, and the model handles them well when you give it.

Newest in the guide

Each plays from the creator's own post on X. Our note sits under it; the prompt belongs to them.

Toddler mischief in three rooms

scene list · Seedance 2.5 per the author · Oct 28, 2026

Seven short scenes escalate one joke (water in the bathroom, a spilled glass in the kitchen, a puddle outside) and end on a freeze-frame payoff. Each character's look is written out once in a consistency block, down to shirt pattern and hair, and the mother's reaction gets its own close-up scene. There are no timecodes or stated length, so pacing is left to the model.

Performance and emotion

Read @Zarnab_with_Ai's prompt
Create a high-quality 3D animated family comedy video featuring an adorable mischievous toddler boy and his young mother. Use a polished Pixar-inspired animation style, expressive facial features, soft cinematic lighting, realistic water physics, detailed environments, and smooth character animation. Vertical 9:16 aspect ratio, 4K UHD quality, 24fps.

Scene 1 – Bathroom Mischief: A cute toddler boy with messy brown hair, huge expressive brown eyes, a yellow T-shirt with small white cloud patterns, and navy-blue shorts is playing with water in a modern bathroom. The floor is wet and covered with splashes. His young mother, wearing a dusty rose T-shirt and light gray shorts, enters carrying a laundry basket and discovers the mess. The toddler looks innocent and mischievous.

Scene 2 – Mother's Reaction: Show a close-up of the mother's expressive face as she raises her eyebrows in surprise and looks at her naughty little son. Use cinematic close-up framing and exaggerated but natural animated expressions.

Scene 3 – Cleaning Up: The mother gently picks up the toddler and tries to clean him up near the toilet. The little boy squirms playfully, making it difficult for her to manage him. Show realistic body movement, affectionate parenting, and humorous timing.

Scene 4 – The Glass of Water: Cut to a warm, sunlit kitchen with wooden countertops, large windows, and soft curtains. The mother places a transparent glass filled with water on the kitchen table and steps away briefly. The toddler notices the glass and curiously approaches it.

Scene 5 – Unexpected Splash: The toddler climbs toward the table, accidentally spills the water, and slides onto the wet tabletop. Water spreads across the surface as he crawls and giggles. Use detailed water splashes, realistic reflections, and playful cartoon physics.

Scene 6 – Mom Rushes In: The mother hurries into the kitchen, lifts her mischievous toddler into her arms, and carries him toward the door. Her expression shows exhaustion, surprise, and disbelief at his endless mischief.

Scene 7 – Outdoor Chaos: The mother brings the toddler outside. He suddenly runs toward a shallow puddle on the wet ground. She reaches out to catch him, but he jumps into the water with a huge joyful splash. Water droplets fly dramatically in every direction.

Final Shot: Capture the toddler laughing happily in the puddle while his mother stands behind him, completely shocked, with her hair standing wildly on end from the enormous splash. The toddler remains delighted and innocent. End with a funny cinematic freeze-frame.

Visual Style: Premium 3D animated family film, Pixar-inspired character design, oversized expressive eyes, soft skin shading, detailed hair, warm natural lighting, cinematic depth of field, realistic water simulation, smooth camera transitions, wide shots mixed with expressive close-ups, comedic timing, and consistent character design.

Character Consistency: Maintain the exact same toddler face, hairstyle, yellow cloud-pattern shirt, navy shorts, and body proportions in every scene. Keep the mother’s face, brown hair in a messy bun, dusty rose T-shirt, and gray shorts consistent throughout the video.

Audio: Playful instrumental background music, water splashing sounds, toddler giggles, and subtle comedic sound effects. No dialogue, no subtitles, and no text overlays.

Original post ↗

Phone travel vlog in Switzerland

Shot-by-shot timeline · Seedance 2.5 per the author · Oct 28, 2026

The prompt says what the footage is not (a travel commercial, a cinematic film) before any scene, then alternates selfie mode and rear-camera clips across six timed scenes. Phone behaviour is listed as specific faults: small framing corrections, autofocus changes, exposure shifts and simple cuts as if edited on a phone. The three spoken lines sit at the start, middle and end, so the speaker frames the clip.

Atmosphere and realismPerformance and emotion

Read @AIwithNatalia's prompt
Create a 30-second realistic iPhone travel vlog featuring the same girl exploring Switzerland. The entire video should feel like a casual vlog filmed by a real person on an iPhone, not a professional travel commercial or a cinematic film.

Scene 1 (0–5 seconds): A casual selfie video of the girl standing on a beautiful Swiss street with mountains in the background. She smiles naturally at the camera and says, “Hey guys! I’m finally in Switzerland, and look at this view!”

Scene 2 (5–10 seconds): Handheld iPhone footage as she walks through a charming Swiss village. Show cute houses, flowers on balconies, people walking in the background, and little spontaneous camera movements. She briefly turns the camera toward the scenery.

Scene 3 (10–15 seconds): She films a beautiful lake with mountains behind it, then switches back to selfie mode and says, “Guys, this place literally doesn’t look real!” Let a little wind and natural background noise remain in the audio.

Scene 4 (15–21 seconds): A casual café moment showing her holding a coffee, taking a bite of Swiss chocolate, and laughing softly. Include slightly imperfect framing and natural movements, like a friend casually recording her.

Scene 5 (21–26 seconds): Quick handheld clips from a scenic train ride, capturing green fields, little houses, and mountains through the window. The footage should feel spontaneous, as if she is recording whatever catches her eye.

Scene 6 (26–30 seconds): Back to selfie mode outdoors. She smiles at the camera and says, “I swear, I could stay here forever. Okay, guys, see you in the next vlog!”

Camera and realism instructions:

Shoot entirely in a natural iPhone vlog style, with handheld selfie footage and casual rear-camera clips.
Use realistic iPhone exposure, natural colors, subtle HDR, authentic skin texture, and ordinary smartphone sharpness.
Include slight hand movements, tiny framing adjustments, natural autofocus changes, and occasional exposure shifts.
Use natural daylight, realistic shadows, ambient street sounds, gentle wind, and café background noise.
Keep the girl's facial features, hairstyle, outfit, and appearance consistent across every scene.
Make her expressions, gestures, walking, and speech feel spontaneous and unscripted.
Use simple cuts between clips, like a real travel vlog edited on a phone.
Avoid dramatic cinematic lighting, drone shots, artificial slow motion, excessive color grading, perfect camera movements, beauty filters, and overly polished commercial aesthetics.

Format: Vertical 9:16, 30 seconds, realistic smartphone footage, natural conversational voice, believable lip-sync, and authentic travel-vlog energy.

Original post ↗

One seated person, one room, four eras

Shot-by-shot timeline · Seedance 2.5 per the author · Oct 27, 2026

Everything that must not change is fixed before the timeline: one person, one chair, and one room whose chair, window and table keep their positions in every era. Each era change is timed to the moment the orbiting camera passes behind the sitter, so the subject's own body hides the swap. The closing rules also ban extra people and an empty-chair ending, naming the likely failures in advance.

Transformation and style shiftsCamera moves and transitionsAtmosphere and realism

Five worlds through one train window

Shot-by-shot timeline · Seedance 2.5 per the author · Oct 27, 2026

The camera is fixed on a window that fills most of the frame, so every location change happens outside the glass while the carriage and passenger stay the same. Each of the four changes runs through a cloud bank that fills the window, timed to a climb with a physical jolt, which gives the transition both a cause and an occluder. Parallax is spelled out for each landscape (near objects fast, far objects slow); the post is one step of a tutorial thread whose earlier steps make the reference images.

Camera moves and transitionsGiants, armies and epic scaleAtmosphere and realism

How we chose the clips

The core set comes from a research library of public posts by four creators on X: Arvin010717, lansenai, CharaspowerAI and xIrissy. A post from that library became a case only if its video could be matched to a readable prompt and it passed a version check: either the author states the clip was made with Seedance 2.5, or the clip runs longer than 15 seconds and a continuity check supports 2.5. That gave 82 cases from Arvin010717, lansenai and CharaspowerAI. xIrissy's posts shaped the workflow notes, but none of their videos came with a matched prompt and a confirmed version. A smaller set was sampled frame by frame against its prompt; where the result drifted from the plan, the case note says so. A second batch came from a search of public posts made in September and October 2026: 14 posts in which the author names Seedance 2.5 for the video and shares the prompt. These were not sampled, so their notes describe only what the prompt does on paper. That brings the guide to 96 cases from 13 creators. Version labels come from the authors' own posts, not from platform logs, and nothing here is a measured success rate.

Every case plays from the author's own post on X, and every prompt belongs to its author. Where an author wrote the prompt in English and we could match our copy word for word against their post, we quote it in full with credit and a link; otherwise we link to the post rather than translate it. The case notes, templates and checklists are our own analysis of what the prompts do structurally, so you can learn the method and write your own.

By creator: @CharaspowerAI (43), @Arvin010717 (31), @lansenai (11), @HeyAbhishek (2), @Zarnab_with_Ai (1), @AIwithNatalia (1), @MissDelulu9 (1), @ImaStudio_ai (1), @itxabdullaa (1), @aimikoda (1), @doctorwasif (1), @harboriis (1), @itsSaira_1 (1). Examples checked Sep 23, 2026.

If you are one of these creators and want a post left out or credited differently, write to [email protected] and we will change it the same day. Made a shot that worked? Send us the prompt and the video it made.

The week's best new dramas and AI movies, in one email.

We are putting together a weekly email: the new dramas and AI movies worth your time, and any free-coin deals we find. Leave your address and you will get the first issue when it starts. No spam.