Seedance 2.5: 30s One-Take Futsal Video from a Single Selfie Prompt Technique

AI creator TechHalla (@techhalla) has published a production-ready prompt engineering technique for ByteDance's Seedance 2.5 video foundation model, generating a high-energy 30-second continuous one-take street futsal match in a Brazilian favela while locking character identity using only a single reference selfie.

Image source: @techhalla / X

In contemporary AI video generation workflows, maintaining character consistency across shots longer than 15 seconds remains a notorious hurdle. Models routinely suffer from identity bleed, gradual facial deformation, or face morphing into nearby extras, especially in crowded, fast-paced sports scenes where camera angles shift dynamically. TechHalla tackled this constraint by anchoring facial features to a single strict image reference and pairing it with authentic smartphone camera artifacts alongside second-by-second chronological directives.

Single Photo Identity Lock and 9:16 Handheld Continuity

The primary technical breakthrough of this workflow lies in how it maximizes Seedance 2.5's image reference capabilities while deliberately neutralizing artificial "AI gloss" through authentic mobile video grammar.

To preserve facial identity across the entire clip, the prompt designates the uploaded still photo as a strict lock for the protagonist's appearance: a bald head, thick dark salt-and-pepper beard, identical eyes, facial bone structure, and physical build. Crucially, the opposing defender is explicitly described as a distinct athletic 20-something local youth rather than an ambiguous face, preventing the model from accidentally generating clones or merging characters during fast physical collisions.

Rather than relying on smooth, cinematic crane sweeps or multiple cuts, the sequence is locked into an unbroken 30-second 9:16 vertical handheld take. The prompt intentionally introduces realistic smartphone camera imperfections:

• Raw Mobile Video Artifacts: Shaky handheld motion, film grain, compression macroblocks, and slight overexposure against the dusk sky.
• Physical Lens Dynamics: Visible lens dirt, autofocus hunting when the subject turns, and motion blur across fast footwork and the ball.
• Absence of Post-Processing Clichés: Explicit bans on cinematic color grading, slow-motion replays, artificial cut transitions, overlaid text, or social media stickers.
• Synchronized Environmental Audio: Concrete ball thuds, squeaking sneaker rubber, shouts in Portuguese, rattling chain-link fences, and explosive crowd cheering upon scoring.

These intentional imperfections anchor the model in physical realism, convincing the viewer that the footage was captured live on a handheld mobile phone rather than rendered in a studio.

Modular Prompt Architecture: Separating Style, Reference, and Environment

TechHalla structured the prompt into five isolated modules—[STYLE + CAMERA + ATMOSPHERE], [IMAGE REFERENCES], [PLACE], [TIMELINE], and [STYLE & QUALITY BOOSTERS]—allowing Seedance 2.5 to parse visual priorities without conflicting instructions.

The atmosphere module defines the tactile feel of an enclosed concrete pick-up court ("quadra") nestled within a hillside favela at dusk, complete with buzzing cheap floodlights, laundry lines, parked scooters, and local teenagers crowding the fence with phones held high.

The reference module specifies the protagonist's simple dark t-shirt, shorts, and sneakers, minimizing extraneous costume details that could shift across frames. By isolating the defender's casting and jersey styling, the prompt ensures visual contrast throughout the 1-on-1 duel.

The environment block grounds spatial consistency: a small makeshift goal with a bent metal frame and patched net at the far end, puddle stains, chalk marks, and stacked brick houses climbing the hill behind the cage. This rich spatial grounding keeps the scene anchored even as the camera jostles violently during celebrations.

Second-by-Second Timeline Breakdown and Complete Prompt

The most effective element of the prompt is its second-by-second timeline divided into 18 sequential keyframe beats, directing character movement, camera behavior, and crowd dynamics across the full 30 seconds:

• 0–3s (Opening Hook and Standoff): The take begins with the phone already rolling at the fence, capturing a center-court standoff, shoulder feints, and sneakers gripping concrete grit.
• 3–5s (The Nutmeg and Camera Whip): An explosive "caneta" pushes the ball cleanly through the defender's legs, triggering a sharp camera whip and brief focus snap onto the protagonist's back.
• 5–10s (Sprint, Strike, and Goal): Two rapid strides toward the makeshift goal under flaring floodlights, followed by a laces-driven rocket into the top corner and an instant eruption from spectators.
• 10–24s (Pitch Invasion and Celebration Chaos): Teenagers swarm through the chain-link gate, jostling the camera operator, lifting the protagonist briefly off his feet, and shouting into the microphone with phones held up.
• 24–30s (Hero Breakout and Natural Cut): The protagonist breaks from the pack for a short jog toward the lens, pointing at the camera before the video ends naturally amidst celebration noise, mirroring a real social media story upload.

Here is the complete, unedited Seedance 2.5 prompt template shared by TechHalla. Creators can adapt the character descriptions and environmental details to match their own reference photos:

[STYLE + CAMERA + ATMOSPHERE]
Raw mobile phone footage, vertical 9:16, one unbroken continuous handheld take for the full 30 seconds. Shaky cam, grain, compression blocks, slight overexposure on the dusk sky, lens dirt, autofocus hunting, motion blur on the ball and feet. No cinematic grade, no slow-mo, no cuts, no text, no stickers. Gritty concrete football court in a Brazilian favela at dusk: painted lines half worn off, chain-link fence, brick houses stacked uphill, laundry lines, scooters, warm orange sky, cheap floodlight starting to buzz. Crowd of teenagers and locals ring the cage, phones already half-up. Audio: ball thuds, sneakers squeak, Portuguese shouts, fence rattle, then full scream when it goes in.

[IMAGE REFERENCES]
Use the provided photo as the single strict lock for the protagonist’s face: bald head, thick dark salt-and-pepper beard, same eyes and bone structure for the entire take. He wears a simple dark t-shirt, shorts, and sneakers. Body type locked. He is the older bald man in the 1-on-1. The young athletic street footballer facing him is lean, early 20s, jersey and low socks, quick feet — a real kid from the court, not a clone of the protagonist.

[PLACE]
Enclosed concrete pick-up court / “quadra”: small makeshift goal with a bent metal frame and patched net at the far end, puddle stains, chalk X, spectators pressed to the fence. Dusk. Favela hillside behind.

[TIMELINE — ONE CONTINUOUS VERTICAL TAKE, SECOND BY SECOND]
0-1s: Phone already rolling, vertical, held by someone at the fence. Shaky. Bald bearded man and the young footballer face off center-court, ball at the bald man’s feet. Crowd murmur. Hook: he taps the ball forward and commits.
1-2s: He feints left with the shoulder. The kid mirrors. Concrete grit under sneakers. Camera jolts trying to keep both in frame.
2-3s: He rolls the ball onto his right foot, body open. The kid crouches, arms out, ready to block the lane.
3-4s: Lightning caneta — the bald man pushes the ball through the kid’s open legs in one clean nutmeg. Ball pops out behind the defender. Kid’s head snaps down in shock.
4-5s: The bald man is already past him, accelerating onto the loose ball. Vertical cam whips, soft focus for a frame, then snaps sharp on his back and bald head.
5-6s: First touch settles the ball ahead of him toward the small makeshift goal. The kid turns late, chasing. Crowd noise spikes.
6-7s: Two strides. He shapes for a strike. Floodlight flare hits the lens. Fence and heads blur at the edges.
7-8s: Explosive shot — full laces through the ball. Motion blur on the leg and ball. The ball rockets toward the patched net.
8-9s: Ball smashes the top corner of the tiny goal, net bulges, frame rattles. Goal. Instant scream from the fence.
9-10s: The kid freezes mid-run, hands on head. The bald man opens his arms, still moving forward, beard readable, face locked to the photo.
10-11s: Vertical cam shakes hard as the first teenagers vault or squeeze through the gate onto the concrete.
11-12s: Bodies flood the pitch from the near side. Dust and sneakers. Someone bangs the chain-link. Audio clips.
12-13s: The bald man gets swarmed — backs, arms, a leap from a kid trying to jump on him. Camera shoved by a shoulder, still rolling, still continuous.
13-15s: Chaotic celebration fill: phones up in frame filming the phones, shirts pulled, someone kicks the ball away into the fence. The young defender laughs despite himself at the edge.
15-18s: The bald man is lifted half a second by two teens, sneakers off the ground, then set down hard. He points at the goal, yelling with the crowd. Grain heavy. Autofocus hunts his face and locks.
18-21s: Camera operator gets pushed closer — tighter vertical on his bald head and beard in the middle of the pile, kids screaming into the mic. A scooter horn from the street outside the court.
21-24s: Wide-ish again as the pile breaks: people running in circles, one kid sliding on his knees on the concrete, another hanging on the fence. The makeshift goal leans. The ball sits in the net.
24-27s: The bald man breaks free enough to do a short cocky jog toward the phone, pointing at the lens, breathing hard, same locked face. Crowd still pouring behind him.
27-30s: Final hold: vertical frame packed with bodies, the bald man center, arms out, dusk flare, grain, fence rattle, celebration still peak. No cut. Recording ends mid-chaos like a real story upload.

[STYLE & QUALITY BOOSTERS]
Raw vertical continuous phone take, shaky grainy favela dusk court, face locked to the still, lightning caneta nutmeg then an explosive strike into the small makeshift goal, crowd storms the pitch, authentic street light and lens flare, motion blur on ball and feet, viral social footage energy, no cinema camera, no slow-motion, no text.

For video directors and digital creators looking to achieve character consistency in Seedance 2.5 without costly multi-shot training pipelines, this structured timing and mobile camera approach offers a highly repeatable blueprint across diverse sports and narrative action scenes.

Original source

• TechHalla (@techhalla) on X: Seedance 2.5 Single-Image Reference 30s One-Take Futsal Prompt Thread — Original publication thread detailing the modular prompt architecture and featuring the complete 30-second continuous vertical video output.
• ByteDance Seedance 2.5 Model Context: Next-generation multimodal video foundation model by ByteDance, demonstrating high facial fidelity from single-image prompts, extended shot temporal coherence, and synchronized contextual audio.