The Still One
Find an accomplice
SEEDANCE 2.5 · VIDEO PROMPT · 5,997 CHARACTERS
SEEDANCE 2.5 · VIDEO PROMPT · 5,997 CHARACTERS FORMAT: 30s / 16:9 / cinematic music-video scene / one continuous shot, no cuts STYLE: Live-action music-video realism, overcast daylight, cool desaturated grade, fine 35mm film grain, one consistent look. Match the color, contrast, grain and lens texture of @Video 1. REFERENCES: @Video 1 is the timing, blocking, camera and environment master. Inherit its shot structure, camera moves, the lead performer's path and gestures, the moment the crowd starts dancing, the choreography, the crowd rows, and the location (red-brick building with arched windows behind a paved courtyard). Do NOT inherit its performers, faces, clothing, subtitles, voices or music. @Image 1 defines THE MAN, the only human: curly brown hair, short beard, black square sunglasses, burgundy wool varsity jacket with brown leather sleeves and a cream K patch, dark grey cardigan over a striped shirt, charcoal pinstripe trousers, grey sneakers. Ignore the five-panel layout, labels and studio background. No bag. He takes the place of the lead performer of @Video 1. @Image 2 defines TYPE A: a slim performer in an off-white structured long coat, trousers, boots and gloves, wearing a smooth matte BLACK multi-lobed sphere helmet with two small glowing white eyes and a small black sphere on a thin stem at the neck. Fully covered costume. @Image 3 defines TYPE B: a slim performer in a black structured long coat, trousers, boots and gloves, wearing a smooth glossy WHITE multi-lobed sphere helmet with two small black eyes and a small white sphere on a thin stem at the neck. Fully covered costume. The crowd is many identical copies of TYPE A and TYPE B only, alternating A, B, A, B along each row and offset row to row as a checkerboard. Every crowd member of @Video 1 becomes one of these types. VIDEO PROMPT: [Stage 1 - approx 0-4s] Initial state: tight close-up, head and shoulders, THE MAN centered, sunglasses on. Behind him, out of focus, rows of TYPE A and TYPE B stand motionless facing the camera. Event: he takes a slow deep breath, his breath shows as cold vapor in the chilly air and drifts up past the sunglasses; his right hand adjusts the bridge of the sunglasses, then lowers. He turns his head slightly left, then right, then back to the lens, expression flat. Camera: locked-off close-up, shallow depth of field. End state: hand lowered, he faces the lens. [Stage 2 - approx 4-9s] Continue from that frame. At 4s he turns his back to the camera over his left shoulder, head slightly lowered, and walks straight away from the lens into the crowd through a narrow gap between the figures, following the path and pace of @Video 1. The figures never move. Camera: holds position and height; focus follows him into the wall of alternating black and white helmets. End state: he stops about six rows deep, back to camera, a TYPE A on his left and a TYPE B on his right. [Stage 3 - approx 9-17s] Continue from the unchanged formation. At 13s he turns in place to face the camera, straightens the jacket collar with both hands, lowers his arms and settles into a relaxed standing pose, looking straight into the lens through the crowd. Every figure stays frozen, all helmets facing the camera. Camera: very slow drift, him centered at mid-depth, medium-wide frame. End state: total stillness; he is the calm center of a rigid checkerboard of helmets. [Stage 4 - approx 17-24s] Continue from the frozen formation. At 19s, on a single beat, every figure except THE MAN starts the dance from @Video 1, matching its choreography: each performer twists the whole torso around the vertical spine axis, shoulders rotating to face diagonally left, then diagonally right, head following the shoulders and tipping back, elbows bent and arms sweeping across the chest with the twist, feet planted. Neighboring performers move in opposite phase: when one twists to face left, the performer beside them twists to face right, so every row reads as an interlocking corkscrew of alternating diagonals, never a line of performers all facing the same way. Stylized, energetic; each performer stays in their own space. The sphere helmets stay firmly on and keep their shape through every movement. THE MAN does not move at all, only his breath vapor drifting. Camera: begins a slow, steady pull back and crane up. End state: the whole visible crowd is mid-dance, he stands motionless in the exact center. [Stage 5 - approx 24-30s] Continue the same crane. The camera keeps rising and pulling back to a high wide shot, revealing hundreds of alternating TYPE A and TYPE B performers filling the courtyard, all twisting in the same counter-phase corkscrew rhythm, alternating diagonals across every row, the red-brick building across the top of the frame. THE MAN stays still, sunglasses on, one unmoving point in the middle of the motion. Camera settles and holds. Final state: held high wide frame, crowd dancing, the man still. AUDIO: <cold overcast courtyard air, faint wind> <at 1s one slow breath in and out> <his steps on pavement> <at 19s hundreds of shoes tap the pavement together, coats swishing, a low rhythmic floor pulse> No dialogue. No music. Do not inherit the music, vocals or lyrics of @Video 1. CONSISTENCY: Keep THE MAN's face, hair, beard, sunglasses and full wardrobe unchanged in every frame; sunglasses never come off. Exactly one human. The crowd contains only TYPE A and TYPE B, alternating; every crowd performer keeps the sphere helmet on at all times. TYPE A always has a black helmet and an off-white outfit; TYPE B always has a white helmet and a black outfit; never swap. Keep location, lighting, screen direction and camera path consistent with @Video 1. NEGATIVE PROMPT: no cuts, no identity change, no face distortion, no sunglasses removal, no wardrobe change, no bag, no helmet removal, no mismatched helmet and outfit colors, no helmet deforming, no third character type, no music, no vocals, no subtitles, no on-screen text, no watermark, no cartoon, no cheap CGI, no style drift.
Tutorial
DENIZ AKKABAK · SEEDANCE 2.5 TUTORIAL · @DENIZ.AKKABAK
invideo MADE IN INVIDEO · invideo.io/i/DENIZ-ig
The Still One
Drop yourself into a music video’s choreography — one reference clip, two masked strangers, and a crowd that dances around you while you don’t.
A TUTORIAL BY DENIZ AKKABAK · @DENIZ.AKKABAK
Pipeline Seedance 2.5 in invideo · Format 30s · 16:9 · one shot · Genre Music-video character swap
Most people prompt a dance and get a puppet show. The trick here is to stop describing the dance at all: you hand the model a real music-video sequence and let it own time, camera and choreography — then you swap the bodies. The lead becomes you, the crowd becomes two masked figures on a checkerboard, and everything that actually makes the shot work — the freeze-to-eruption at nineteen seconds, the crane that reveals hundreds of dancers, the one man who never moves — arrives for free from the reference. Your prompt’s only real job is to say what not to inherit.
THE PIPELINE AT A GLANCE
- 01
Cut the reference
Find the sequence on YouTube, download it, trim exactly thirty seconds. This clip becomes the timing, camera and choreography master — the prompt defers to it on everything but identity.
- 02
Lock the bodies
One reference sheet of you, one still per masked figure type. Three images, three jobs, uploaded in a fixed order so the prompt’s numbering holds.
- 03
Generate in invideo
Seedance 2.5, thirty seconds, 16:9, video first then the three images, one prompt, one pass. Everything below is that single generation.
STEP 01 · REFERENCE CLIP — YOUTUBE + ANY EDITOR
The Reference Clip
The sequence is the finale of STORM by GENER8ION featuring Yung Lean, directed by Romain Gavras with choreography by Damien Jalet. Search YouTube for the title and either name and you will find it in one hit. Download the full video with any YouTube downloader, then cut the last act — the tiered “class photo” formation. Start your thirty seconds on the tight close-up of the lead standing at the front of the crowd and end on the high wide shot of the whole formation dancing.
Keep it exactly thirty seconds (Seedance 2.5’s ceiling) and keep it 16:9 — the final crane only reads in landscape. You do not need to strip the audio; the prompt refuses to inherit it. What you are handing the model is the shot map below, and nothing else.
WHAT THE MODEL WILL INHERIT — THE SHOT MAP
| 0–4s | Tight close-up of the lead at the front of a motionless crowd; a small hand gesture near the face, a glance left and right, back to the lens. |
| 4–9s | He turns his back to the camera and walks straight away from the lens into the crowd through a narrow gap. Nobody else moves. The camera holds. |
| 9–17s | He stops about six rows deep, turns to face the camera at thirteen seconds, fixes his collar, settles. Total stillness in a rigid formation. |
| 17–24s | At nineteen seconds, on one beat, everyone except him starts dancing — torsos twisting on the spine, neighbours in opposite phase. The camera begins to pull back. |
| 24–30s | A slow crane up and back to a high wide shot: hundreds of dancers, the red-brick building across the top of frame, the lead a single still point. Hold. |
STEP 02 · REFERENCE IMAGES — THREE STILLS
Three Images, Three Jobs
The video defines the where and the when; these three stills define the who. Upload order matters — Seedance numbers references in the order they land, so Video 1 goes in first, then Image 1, Image 2 and Image 3 exactly as captioned here. Each reference gets one job in the prompt and one line that tells the model what to ignore about it.
REFERENCE INPUTS — ATTACH IN THIS ORDER
Notice that the two masked stills are mirror images of each other — same grey studio, same pose, same silhouette. The only variables are helmet colour and outfit colour, which is exactly what the prompt locks. Give the model two figures that differ in five ways and it will invent a sixth; give it two figures that differ in two and it has nothing to improvise.
STEP 03 · GENERATION — SEEDANCE 2.5 IN INVIDEO
The Prompt
In invideo choose Seedance 2.5, set thirty seconds and 16:9 in the interface — never in the prompt, the model ignores settings written as prose — attach the video first and the three images in order, paste the prompt below and generate. One pass, one continuous shot. The prompt is long because a thirty-second clip needs five timed stages with a visible end state each; it is not long because it describes the dance.
SEEDANCE 2.5 · VIDEO PROMPT · 5,997 CHARACTERS
SEEDANCE 2.5 · VIDEO PROMPT · 5,997 CHARACTERS
30s / 16:9 / cinematic music-video scene / one continuous shot, no cuts
Live-action music-video realism, overcast daylight, cool desaturated grade, fine 35mm film grain, one consistent look. Match the color, contrast, grain and lens texture of @Video 1.
@Video 1 is the timing, blocking, camera and environment master. Inherit its shot structure, camera moves, the lead performer's path and gestures, the moment the crowd starts dancing, the choreography, the crowd rows, and the location (red-brick building with arched windows behind a paved courtyard). Do NOT inherit its performers, faces, clothing, subtitles, voices or music. @Image 1 defines THE MAN, the only human: curly brown hair, short beard, black square sunglasses, burgundy wool varsity jacket with brown leather sleeves and a cream K patch, dark grey cardigan over a striped shirt, charcoal pinstripe trousers, grey sneakers. Ignore the five-panel layout, labels and studio background. No bag. He takes the place of the lead performer of @Video 1. @Image 2 defines TYPE A: a slim performer in an off-white structured long coat, trousers, boots and gloves, wearing a smooth matte BLACK multi-lobed sphere helmet with two small glowing white eyes and a small black sphere on a thin stem at the neck. Fully covered costume. @Image 3 defines TYPE B: a slim performer in a black structured long coat, trousers, boots and gloves, wearing a smooth glossy WHITE multi-lobed sphere helmet with two small black eyes and a small white sphere on a thin stem at the neck. Fully covered costume. The crowd is many identical copies of TYPE A and TYPE B only, alternating A, B, A, B along each row and offset row to row as a checkerboard. Every crowd member of @Video 1 becomes one of these types.
[Stage 1 - approx 0-4s] Initial state: tight close-up, head and shoulders, THE MAN centered, sunglasses on. Behind him, out of focus, rows of TYPE A and TYPE B stand motionless facing the camera. Event: he takes a slow deep breath, his breath shows as cold vapor in the chilly air and drifts up past the sunglasses; his right hand adjusts the bridge of the sunglasses, then lowers. He turns his head slightly left, then right, then back to the lens, expression flat. Camera: locked-off close-up, shallow depth of field. End state: hand lowered, he faces the lens. [Stage 2 - approx 4-9s] Continue from that frame. At 4s he turns his back to the camera over his left shoulder, head slightly lowered, and walks straight away from the lens into the crowd through a narrow gap between the figures, following the path and pace of @Video 1. The figures never move. Camera: holds position and height; focus follows him into the wall of alternating black and white helmets. End state: he stops about six rows deep, back to camera, a TYPE A on his left and a TYPE B on his right. [Stage 3 - approx 9-17s] Continue from the unchanged formation. At 13s he turns in place to face the camera, straightens the jacket collar with both hands, lowers his arms and settles into a relaxed standing pose, looking straight into the lens through the crowd. Every figure stays frozen, all helmets facing the camera. Camera: very slow drift, him centered at mid-depth, medium-wide frame. End state: total stillness; he is the calm center of a rigid checkerboard of helmets. [Stage 4 - approx 17-24s] Continue from the frozen formation. At 19s, on a single beat, every figure except THE MAN starts the dance from @Video 1, matching its choreography: each performer twists the whole torso around the vertical spine axis, shoulders rotating to face diagonally left, then diagonally right, head following the shoulders and tipping back, elbows bent and arms sweeping across the chest with the twist, feet planted. Neighboring performers move in opposite phase: when one twists to face left, the performer beside them twists to face right, so every row reads as an interlocking corkscrew of alternating diagonals, never a line of performers all facing the same way. Stylized, energetic; each performer stays in their own space. The sphere helmets stay firmly on and keep their shape through every movement. THE MAN does not move at all, only his breath vapor drifting. Camera: begins a slow, steady pull back and crane up. End state: the whole visible crowd is mid-dance, he stands motionless in the exact center. [Stage 5 - approx 24-30s] Continue the same crane. The camera keeps rising and pulling back to a high wide shot, revealing hundreds of alternating TYPE A and TYPE B performers filling the courtyard, all twisting in the same counter-phase corkscrew rhythm, alternating diagonals across every row, the red-brick building across the top of the frame. THE MAN stays still, sunglasses on, one unmoving point in the middle of the motion. Camera settles and holds. Final state: held high wide frame, crowd dancing, the man still.
<cold overcast courtyard air, faint wind> <at 1s one slow breath in and out> <his steps on pavement> <at 19s hundreds of shoes tap the pavement together, coats swishing, a low rhythmic floor pulse> No dialogue. No music. Do not inherit the music, vocals or lyrics of @Video 1.
Keep THE MAN's face, hair, beard, sunglasses and full wardrobe unchanged in every frame; sunglasses never come off. Exactly one human. The crowd contains only TYPE A and TYPE B, alternating; every crowd performer keeps the sphere helmet on at all times. TYPE A always has a black helmet and an off-white outfit; TYPE B always has a white helmet and a black outfit; never swap. Keep location, lighting, screen direction and camera path consistent with @Video 1.
no cuts, no identity change, no face distortion, no sunglasses removal, no wardrobe change, no bag, no helmet removal, no mismatched helmet and outfit colors, no helmet deforming, no third character type, no music, no vocals, no subtitles, no on-screen text, no watermark, no cartoon, no cheap CGI, no style drift.
OUTPUT
STEP 04
Anatomy — Why the Crowd Dances and You Don’t
Seven decisions carry the whole prompt. Every one of them is tied to a line you can find in the box above.
- 01
The contract
“@Video 1 is the timing, blocking, camera and environment master.” This is the sentence that makes the rest possible: it demotes the prompt from director to editor’s note. Timing, blocking, camera and location are no longer things you describe — they are things you inherit.
- 02
The exclusion line
“Do NOT inherit its performers, faces, clothing, subtitles, voices or music.” Without it the original schoolboys leak back in on the wide shot. Every video reference needs a sentence that says what it is not allowed to bring.
- 03
Two types, one checkerboard
Naming the crowd TYPE A and TYPE B and writing “alternating A, B, A, B along each row and offset row to row” is what keeps helmet colour from swapping mid-shot. “The masked people” would have given you a grey blur by second twenty.
- 04
End states
Every stage closes on a frozen fact — “he stops about six rows deep, back to camera, a TYPE A on his left and a TYPE B on his right.” The next stage starts from that fact, so props don’t multiply and nobody teleports.
- 05
The corkscrew
“Neighboring performers move in opposite phase … an interlocking corkscrew of alternating diagonals.” The first render had every figure twisting the same way. Describing the phase relationship between neighbours, not the move itself, fixed it in one pass.
- 06
Filter-proofing
The original scene has a cigarette; a first version was blocked after generation. The safe rewrite keeps the silhouette and drops the trigger: “his breath shows as cold vapor”, masks become “helmet”, people become “performers”, and every school-and-uniform word is gone.
- 07
The insurance
CONSISTENCY and NEGATIVE PROMPT restate the same locks a third time — “sunglasses never come off”, “never swap”, “no third character type”. A rule stated once survives ten seconds. Stated three times, it survives thirty.
THE RESULT
Output — Thirty Seconds, One Still Point
Check it against the reference: the pull-back should begin the moment the crowd moves, not before; you should be able to find at least one row where neighbours face opposite diagonals; and the man in the varsity jacket should be the only warm colour in the frame from the first frame to the last. If all three hold, the swap worked — the video did the choreography, and the prompt did the casting.
THE CROWD DOES THE WORK
Don’t describe the dance. Borrow it — then say who is allowed to keep it.
Swap the varsity jacket for a wedding dress, the masked figures for a hundred copies of your dog, the school courtyard for a stadium — the skeleton carries any music video you can cut to thirty seconds: one timing master, one swap list, one exclusion line, and an end state at the close of every stage.
invideo
Seedance 2.5 runs inside invideo — thirty-second single passes, video plus image references in one upload, and the interface parameters that this guide keeps out of the prompt. It is the only place this whole workflow lives in one tab.
WRITTEN, DIRECTED & PROMPTED BY DENIZ AKKABAK · @DENIZ.AKKABAK