Nothing here was storyboarded, composited or stitched together. Every shot is a single Seedance 2.5 run on Pixio — thirty seconds native, lip-sync driven straight from the uploaded track.
Each is bound to one image and one colour for the whole video, and the handles never move: Vega is @Image1 in every prompt, Sol is @Image5, even in a block where only two of them appear. Every prompt names her style and then bans the other four by name. Hover a portrait to put her in motion.
Tall and willowy. Platinum buzz cut, white leather trench with exaggerated shoulders worn open over a high-necked satin catsuit, chrome collar, over-knee boots.
Hard angle and line. Arm control cutting geometric shapes, wrist and elbow angles snapping into place with zero overshoot, duckwalks, catwalks, dips dropped straight onto the floor. She never flows and never lets a movement carry past its beat.
Waist-length black hair of enormous loose volume, a beauty mark below the right corner of her mouth. Crimson bias-cut gown with a high halter neck and a long train, opera gloves above the elbow.
Continuous line where Vega is angular. Spinal ripples travelling tailbone to skull, arm sweeps that carry past the beat before resolving, spiral turns that whip the train, hair throws that keep travelling after her head has stopped.
Compact and lean. Acid-green braids gathered in a high bun, structured green leather bodice with architectural hip panels, black latex trousers, green platform boots.
The most percussive vocabulary in the video and all of it driven into the ground. Heavy flat stomps, chest pops firing off the sternum, arm swings thrown from the shoulder with the whole torso behind them, buck-downs.
Very tall and slender. Jet-black razor centre part scraped into a waist-length ponytail, a thin cyan line across her left cheekbone, chrome breastplate with struts to the shoulders.
Her body moves in discrete frames rather than curves. Dime stops halting dead on the beat with zero overshoot, isolated chest pops, arm waves rippling one joint at a time, tutting. Nothing between her shapes is smooth.
Petite and lithe. Copper-gold curly afro, freckles across her nose and cheeks, cropped gold moto jacket over a high-necked chainmail top, gold metallic wide-leg trousers.
The only floor-based vocabulary in the video. Toprock, a drop from standing to the floor in one beat, six-step footwork on her hands, legs threaded through her own arms, freezes held on one hand.
Each woman is one reference image, locked to one handle for the whole video. Seedance 2.5 takes up to fifty references per run, so a cast this size never runs out of room.
Full-spectrum onset detection returned 107.7 BPM at a correlation of 0.081. A kick-band pass disagreed with itself, 112.3 against 120.84. Neither was usable. What was reliable was the energy between 30 and 120 Hz, and that is what every visual beat in the video is pinned to. Both traces are measured from the track. Click the waveform to play the film from that moment.
Prompts were written at three lengths. The long form carries the full sixteen-section structure; the short form keeps only what changes the output. Open any of them to read it in full, with a jump index and a copy button.
Six empty rooms in a line. Each room's lighting snaps on in the single frame the camera enters it, running silver, crimson, acid green, cyan and gold, so the dancers' colours are introduced before any dancer is.
Vega dances vogue on a black mirror floor while the camera completes one full orbit across three shots. At twenty-four seconds the kick lands and the camera falls straight through the mirror surface without a cut, emerging beneath to find all five dancing inverted on the underside.
Mira pops and locks inside a sealed cube of lit panels. At four seconds the room rotates ninety degrees so a wall becomes the floor. The camera is bolted to the room rather than to gravity, so the frame stays level while she steps onto the new floor without ever acknowledging it.
Sol falls through a column of rain at exactly the rain's own speed, so every drop hangs motionless beside her while she dances. The camera falls with her and orbits a full circle at the same time. Hard cut on the pre-drop hole to all five under real rain.
All five launch airborne together and freeze dead in mid-air for three seconds while the camera flies a full circle through the frozen group, threading between suspended bodies without clipping one. Time restarts and Kiki takes the front on a floor that fires a ring of light from every stomp.
Vega dances on one floating platform in a void. The camera orbits her, then pulls back to reveal hundreds more hanging at every height. On the bass slam it drops off the edge and falls, accelerating under real gravity, onto all five on the largest platform below.
One unbroken climb. The camera spirals a full circle around the group while rising, then lifts out through an opening in the ceiling, up a tower and above a night city. The music ends fourteen and a half seconds in and the dancing carries on in silence.
Alternate takes included wherever a block was generated more than once. Click any render that made the final cut to jump the film to it.
Nothing here is a summary. These are the exact strings that were pasted into Seedance, 179,180 characters of them, kept at three lengths because the model behaves differently at each. Long carries the full sixteen-section structure; Mid trims the prose but keeps every decision; Short holds only what measurably changes the output.
All twenty-two are readable in full, nothing summarised, with a copy button on each. Paste one into Pixio, swap the references for your own, and it will generate.
The scene blocks are lip-sync led. The B-roll is not. Every mouth stays closed for the whole block, because a moving mouth in a cutaway fights the lead vocal wherever the cut lands. The faces still work — eyes on the lens, chin driving with the beat. Performing faces with closed mouths, not blank ones.
| Block | Style | Space | Prompt | |
|---|---|---|---|---|
| Vega | Vogue | Seamless white infinity void | 6,232 | |
| Noor | Contemporary, waving | Dark chamber hung with drifting crimson silk | 6,763 | |
| Noor, five seconds | One phrase, one shot | The same chamber, no cut | 5,386 | |
| Kiki | Krump | Black floor scored with a green light grid | 6,955 | |
| Mira | Popping, locking | Sealed cyan light-panel cube | 7,621 | |
| Sol | Breaking | Polished stone under stacked gold light rings | 7,145 | |
| Kiki, Mira, Sol | Three ten-second scenes | A cut every ten seconds | 4,991 | |
| Vega, Noor | Two fifteen-second scenes | One cut at the halfway point | 4,984 |
A focal length means nothing without a sensor, and the model has no sensor. Every prompt names a diagonal field of view instead: 107° for the ultra-wide floor shots, 84° for the wides, 47° normal, 29° and 18° for compressed telephoto. Rectilinear is stated wherever a straight edge carries the shot, because one bowed grid line destroys the whole effect.
The casting prompts were being refused until the language changed. Naming a garment's neckline, sleeve length, panelling and finish passes. Describing how it sits on a body does not. Every character prompt now uses outerwear terms and ends by stating the look is high-fashion editorial.
Every prompt opens mid-move. No establishing frame, no fade from black, no beat of stillness. When scene 03 came back with no singing in it at all, the fix was to say her mouth is already open on a held vowel at 0.0 seconds, and that the singing is underway before frame one rather than started on it.
The vocal is not a clean lyric line. It carries non-English phrases, wordless runs, hums and ad-libs, so each is handled separately: consonant closures on sung words, vowels shaped as actually heard rather than forced into English, a riff held as one long open vowel instead of chopped into syllables, a hum with the mouth closed.
Vogue, contemporary waving, krump, popping and locking, breaking. Each prompt names its own style and then bans the other four by name. Kiki never tuts. Mira never stomps, and nothing between her shapes is smooth. Sol is the only one who touches the floor.
Identity holds far better with one woman on screen than five. The solo B-roll blocks are the most reliable renders in the project and the best source of clean close-ups. When a group shot keeps failing on faces, the fix is to drop to three.
Reference handles carry no space. It is @Image1 and @Audio1, never @Image 1. The audio brackets are typed too: < > for sound design, 【 】 for on-screen text.
Thirty seconds native per run, up to fifty references, lip-sync driven from an uploaded audio file. Every block here is a single run.
Five white-background full-length photographs, each followed by a character sheet built from that photograph as input.
Every prompt was written here, with GPT 6 Astra and Claude Fable 5.1.
themachine.myapps.aiTwo minutes forty-eight seconds, five characters, seven generations. One person, written in text. The whole thing was made on Pixio and every prompt behind it is on this page.
Go to Pixio→