The Net Below
Find an accomplice
Prompt 2
SEEDANCE 2.5 · START FRAME → 12 S SINGLE TAKE · STEP 02 OF 02 Continuous single-take handheld vertical smartphone video, 12 seconds, no cuts. Scene lock from the start frame: harsh midday sun, orange sandstone cliff edge, young woman in a black swimsuit on the lip, black climbing net stretched across the canyon far below, blonde friend sitting behind her. Keep the start frame's lighting, colors, framing and her face unchanged. 0-3s: She rocks on her toes, exhales, glances back at the friend who says "you're actually doing it". Wind moves her ponytail. The camera creeps closer to the edge, tilting down to show the drop. 3-5s: She drops off the edge feet-first, arms up, falling straight down past the rock face, the camera tilting down to follow. She hits the net dead center, the net sinking deep in a V, the anchor ropes creaking taut. 5-9s: The net snaps back and launches her upward, higher than she fell, her body straightening, arms reaching, rising up the rock face toward the camera in one clean arc. The friend screams. The camera jerks back as she rises into frame. 9-12s: At the top of the arc she grabs the exact lip of rock she jumped from with both hands, hangs for half a second with her feet against the wall, then pulls up and swings a knee over the edge, sitting down on the rock next to the friend, breathing hard, wiping chalk on her thigh. The friend grabs her arm, laughing and shouting. The net still swaying below. Camera behavior: tilt down to follow the fall, tilt back up fast, framing lost and recovered, rolling shutter wobble, sun flare at the top of the arc, autofocus hunting on the net far below. Ambient audio: desert wind, a gasp, the net creaking and thudding, the friend screaming then laughing, her breathing hard at the end. No music. No text, no overlay, no logo, no slow motion, no drone angle, no cinematic camera moves, no cuts.
Other prompts in this tutorial
Prompt 1
NANO BANANA PRO · VERTICAL STILL · STEP 01 OF 02 Vertical amateur smartphone video frame, filmed by a friend crouched on a sandstone cliff edge above a deep desert canyon, harsh midday sun, hard black shadows, orange-red rock, a thin dry riverbed far below. A young woman in a black one-piece swimsuit and bare feet stands on the very lip of the rock, toes over the edge, arms loose, looking down at a large black climbing net stretched across the canyon twenty meters below, anchored by ropes to the rock walls on both sides. A blonde friend in a white crop top sits on the rock behind her with her legs over the edge, grinning, a second friend's sneaker and a water bottle at the bottom of the frame. Chalk marks on the rock, a rope bag, a backpack with a towel hanging out. Natural skin texture, sun on her shoulders, wind in her ponytail, slight squint, an in-between expression. Composed like an iPhone 15 Pro 26mm f/1.8 shot, slightly tilted, subject off-center left, crisp sharp focus across the frame, clean HDR look, hard shadows, realistic warm but muted colors, no filters. Clear separation between the woman, the cliff edge, the net and the canyon floor. Her ponytail and the slack net are ready to move. No text, no watermark, no logo, no overlay, no extra people beyond the friends, no deformed hands, no studio polish.
Tutorial
DENIZ AKKABAK · SEEDANCE 2.5 TUTORIAL · @DENIZ.AKKABAK
MADE IN INVIDEO · invideo.io/i/DENIZ-ig

The Net Below
A cliff jump that falls twenty meters into a climbing net and bounces straight back onto the ledge it left — built from one photoreal still and one 12-second single take.
A TUTORIAL BY DENIZ AKKABAK · @DENIZ.AKKABAK
PIPELINE NANO BANANA PRO → SEEDANCE 2.5 · FORMAT 9:16 · 12 S · GENRE STUNT
Nothing in this video is possible. A climbing net doesn't throw a person back higher than she fell, and nobody catches the exact lip of rock she jumped from with both hands and a half-second hang. What makes it land is that the footage never argues with you — it looks like a friend's camera roll. Hard midday sun, chalk on the rock, a sneaker in the corner, a phone that loses the framing during the fall and hunts for focus on the net. The stunt is fiction; the camera is documentary. That is the whole trick, and it is built in two files: a still that convinces the video model it is looking at an iPhone frame, and a prompt that turns impossible physics into plain, timed mechanics.
Below: the still prompt, the Seedance 2.5 prompt, and the anatomy of why every line is doing a job.
THE PIPELINE AT A GLANCE
- 01
Nano Banana Pro — the start frame
One vertical still that carries the entire world: the light, both women, the net, the props, and the phone optics. Everything the video needs to keep is already in this frame — the prompt even tells the ponytail and the slack net to be ready to move.
- 02
Seedance 2.5 — the take
The still becomes the start frame. The prompt locks the scene to it, then choreographs 12 seconds in four timed blocks — drop, net, launch, catch — and gives the camera, the physics and the sound their own lines. Nothing is added; everything moves.
STEP 01 · NANO BANANA PRO — THE START FRAME
The Still Is the Camera
The still does the casting, the lighting and the location scouting — but its real job is to establish which camera is filming. The first line says who is holding it and where: filmed by a friend crouched on a sandstone cliff edge. The optics line names the device — Composed like an iPhone 15 Pro 26mm f/1.8 shot, slightly tilted, subject off-center left — and the look — clean HDR look, hard shadows, realistic warm but muted colors, no filters. Once the still reads as a phone photo, the video model inherits that camera instead of inventing a cinematic one.
Two details are planted for the video. The net is described as slack and the ponytail is in the wind, and the last line of the visual block says it outright: Her ponytail and the slack net are ready to move. The friend is placed behind her with her legs over the edge, grinning — she is the reaction shot and the voice you'll hear at the end. The negative line closes the frame the same way the video prompt will: no extra people beyond the friends, no deformed hands, no studio polish.
Prompt 1
NANO BANANA PRO · VERTICAL STILL · STEP 01 OF 02 Vertical amateur smartphone video frame, filmed by a friend crouched on a sandstone cliff edge above a deep desert canyon, harsh midday sun, hard black shadows, orange-red rock, a thin dry riverbed far below. A young woman in a black one-piece swimsuit and bare feet stands on the very lip of the rock, toes over the edge, arms loose, looking down at a large black climbing net stretched across the canyon twenty meters below, anchored by ropes to the rock walls on both sides. A blonde friend in a white crop top sits on the rock behind her with her legs over the edge, grinning, a second friend's sneaker and a water bottle at the bottom of the frame. Chalk marks on the rock, a rope bag, a backpack with a towel hanging out. Natural skin texture, sun on her shoulders, wind in her ponytail, slight squint, an in-between expression. Composed like an iPhone 15 Pro 26mm f/1.8 shot, slightly tilted, subject off-center left, crisp sharp focus across the frame, clean HDR look, hard shadows, realistic warm but muted colors, no filters. Clear separation between the woman, the cliff edge, the net and the canyon floor. Her ponytail and the slack net are ready to move. No text, no watermark, no logo, no overlay, no extra people beyond the friends, no deformed hands, no studio polish.
OUTPUT

STEP 02 · SEEDANCE 2.5 — THE TAKE
Twelve Seconds, Four Blocks, No Cuts
REFERENCE INPUT — ATTACH AS THE START FRAME

Attach the Step 01 still as the start frame. The prompt spends its first five lines confirming what is in it — Scene lock from the start frame — and re-listing the key elements: the sun, the sandstone lip, the swimsuit, the net, the friend. Then it fixes the part that would otherwise drift: Keep the start frame's lighting, colors, framing and her face unchanged.
From there it is choreography by timecode. Each block moves the woman, moves the camera and leaves both in a stated position before the next block begins. Vertical 9:16 and 12 seconds are declared inside the prompt itself — the model is told once at the top and reminded at the end with no cuts.
Prompt 2
SEEDANCE 2.5 · START FRAME → 12 S SINGLE TAKE · STEP 02 OF 02 Continuous single-take handheld vertical smartphone video, 12 seconds, no cuts. Scene lock from the start frame: harsh midday sun, orange sandstone cliff edge, young woman in a black swimsuit on the lip, black climbing net stretched across the canyon far below, blonde friend sitting behind her. Keep the start frame's lighting, colors, framing and her face unchanged. 0-3s: She rocks on her toes, exhales, glances back at the friend who says "you're actually doing it". Wind moves her ponytail. The camera creeps closer to the edge, tilting down to show the drop. 3-5s: She drops off the edge feet-first, arms up, falling straight down past the rock face, the camera tilting down to follow. She hits the net dead center, the net sinking deep in a V, the anchor ropes creaking taut. 5-9s: The net snaps back and launches her upward, higher than she fell, her body straightening, arms reaching, rising up the rock face toward the camera in one clean arc. The friend screams. The camera jerks back as she rises into frame. 9-12s: At the top of the arc she grabs the exact lip of rock she jumped from with both hands, hangs for half a second with her feet against the wall, then pulls up and swings a knee over the edge, sitting down on the rock next to the friend, breathing hard, wiping chalk on her thigh. The friend grabs her arm, laughing and shouting. The net still swaying below. Camera behavior: tilt down to follow the fall, tilt back up fast, framing lost and recovered, rolling shutter wobble, sun flare at the top of the arc, autofocus hunting on the net far below. Ambient audio: desert wind, a gasp, the net creaking and thudding, the friend screaming then laughing, her breathing hard at the end. No music. No text, no overlay, no logo, no slow motion, no drone angle, no cinematic camera moves, no cuts.
OUTPUT — THREE CHECKPOINTS

STEP 03
Anatomy — Why the Take Holds Together
Seven decisions in the two prompts, each tied to the lines that make it happen.
- 01
The contract comes first
The opening line — Continuous single-take handheld vertical smartphone video, 12 seconds, no cuts. — fixes format, camera type, orientation and duration before a single action is described. The last three words of the prompt, no cuts, repeat the contract so the model can't quietly edit its way out of the hard part.
- 02
Scene lock, not mood board
The still is treated as a frame to continue, not an inspiration to riff on. Scene lock from the start frame is followed by an inventory of what the frame contains, then the hardest lock of all: Keep the start frame's lighting, colors, framing and her face unchanged. Naming her face explicitly is what stops identity drift between the jump and the return.
- 03
Realism is written as flaws
The still asks for iPhone 15 Pro 26mm f/1.8, no filters and no studio polish; the video asks for rolling shutter wobble, framing lost and recovered and autofocus hunting on the net far below. The imperfections are specified as carefully as the stunt — that precision is what makes the render read as a recording.
- 04
Impossible physics, plain mechanics
Nothing is called a trick. The net sinking deep in a V, the anchor ropes creaking taut gives the launch a cause; launches her upward, higher than she fell states the exaggeration as fact; grabs the exact lip of rock she jumped from with both hands closes the loop at a specific place. Every effect has a written cause, so the model animates a mechanism instead of a miracle.
- 05
Four blocks, each with an end state
0-3s ends on the toes and the exhale, 3-5s ends in the net, 5-9s ends with her rising into frame, 9-12s ends sitting down on the rock next to the friend. The camera gets its own verb in every block — creeps, tilts, jerks back — so operator and subject move on the same clock.
- 06
Sound is the timeline
The friend's line — "you're actually doing it" — opens the take; a gasp, the net creaking and thudding, the friend screaming then laughing, her breathing hard at the end is the whole arc in audio order. No music. keeps it a phone recording. When the sound is written as a sequence, the picture tends to follow it.
- 07
The insurance
The closing negatives are each aimed at a specific failure: no slow motion (the launch would become a hero shot), no drone angle (the canyon invites one), no cinematic camera moves (the arc invites a crane). Together with the still's no text, no watermark, no logo, no overlay, they define the one thing this video must never look like — a render.
THE RESULT
Output — The Net Below

Check it against the prompt rather than against your taste. Does the net form a V and do the ropes go taut before she rises? Does she come up toward the camera and grab the same lip with both hands, hang, then swing a knee over? Is the friend still on the rock at the end, and is there sun flare at the top of the arc? If the take passes those checkpoints, the physics will feel earned even though they are invented. If it fails one, the fix is almost always in the block that owns that second, not in the prompt as a whole.
THE STUNT IS IMPOSSIBLE. THE PHONE ISN'T.
Fake the camera, not the fall.
Swap the net for a lake, the canyon for a rooftop pool, the swimsuit for a wetsuit — the skeleton carries any single-take stunt: one still that behaves like a friend's camera roll, one prompt that locks that frame, times the choreography in four blocks, gives the camera its own verbs and writes the sound before it writes the ending. Describe the imperfection as precisely as the trick and the model stops trying to make a movie.

Seedance 2.5 — the model behind this take — runs inside invideo, next to the image models that build the start frame, so the whole pipeline lives in one browser tab: generate the still, hand it to the video model, hear the net creak. Start here:
WRITTEN, DIRECTED & PROMPTED BY DENIZ AKKABAK · @DENIZ.AKKABAK