Out of Office
Find an accomplice
SEEDANCE 2.5 · VIDEO PROMPT
FORMAT: 20s / 9:16 vertical / viral comedy, candid iPhone beach footage / one continuous handheld take, no cuts STYLE: Realistic smartphone video look, natural overcast daylight, handheld shake from the filmer, real world physics, no cinematic grading. REFERENCES: @Image 1 defines the hero: curly dark wet hair, short beard, black square framed sunglasses that stay ON in every frame, shirtless, grey laptop, paper coffee cup. Identity and props only. @Image 2 defines the same single man's face and hair only; ignore its clothing, food and background. VIDEO PROMPT: [Stage 1 — approx 0–2s] Initial state: handheld phone view from a sandy beach, choppy sea, a few ordinary swimmers near the shore, one man visible very far out. Event: the unseen woman filming notices him. Camera: quick punch zoom toward him at 2s. End state: the man fills the frame in a long lens look. [Stage 2 — approx 2–8s] Continue from the zoomed view. The man stands chest deep in rolling waves, typing on the open laptop on one hand, lifting it above each incoming wave at the last second, sipping coffee between waves. A dark dorsal fin slowly circles him. Camera: handheld zoomed follow, micro shake. End state: the fin passes close in front of him; he freezes mid sip. [Stage 3 — approx 8–15s] Continue. He downs the last coffee, the empty cup drifts away, he snaps the laptop shut and holds it like a paddle. Each time the shark lunges near the surface he swats its snout once with the flat closed laptop while wading toward shore. The shark keeps following. Camera: the filmer tracks him along the waterline, frame bouncing with her steps. End state: he is waist deep, shark beside him. [Stage 4 — approx 15–20s] Continue. One final firm swat; the shark goes limp and dazed at the surface, unharmed. He tucks the limp shark under his right arm, closed laptop in his left hand, and walks out of the surf toward camera, deadpan, sunglasses on, water streaming off him. Camera: slow zoom out. Final state: held frame as he passes the camera, shark under his arm. AUDIO: <realistic beach ambience: rolling waves, light wind over the phone mic, distant kids, seagulls> No music. English, natural American accent, woman filming, amused, off camera: at 1s {Oh my god you guys look at this.} at 5s {Is he literally working out there?} at 9s, alarmed then laughing {There is a shark. There is an actual shark!} Male bystander, American accent, off camera at 13s {He is fighting it with the laptop bro!} Woman at 18s, laughing {No way. He is taking it home.} CONSISTENCY: Exactly one man in focus, one shark, one laptop, one cup. Sunglasses remain on in every frame. Keep face, hair and props unchanged. Camera never enters the water. No other swimmer interacts with him. NEGATIVE PROMPT: no face distortion, no identity change, no sunglasses removal, no injuries, no bite contact, no cartoon shark, no cheap CGI, no cinematic grading, no captions, no on screen text, no music, no extra sharks, no duplicate man.
Other prompts in this tutorial
GPT IMAGE 2 · IMAGE PROMPT
Use Image 1 only for the person's facial features, identity, and hair — preserve them exactly. A candid photorealistic photograph of that person standing chest-deep in a rough, cold-looking sea, mid-distance from the camera. Small breaking waves and white foam splash around their chest and shoulders. In one raised hand they hold an open space-gray MacBook laptop above the water, calmly looking at its screen; in the other raised hand they hold [a takeaway coffee cup / an iced drink in a clear plastic cup]. Their expression is completely deadpan and focused, as if working normally, unbothered by the waves. Hair is wet and clinging naturally to the head and neck, water droplets on the skin and forearms. Overcast gray daytime sky, flat diffused natural light, cold blue-gray sea with layered whitecaps stretching to a hazy horizon. No beach, no other people, no boats — only open water. Shot like a real editorial stock photograph on a telephoto lens from the shore, roughly 135mm, eye-level, vertical framing, the subject centered in the lower half of the frame with waves filling the rest. Deep focus, everything sharp. Muted desaturated palette, realistic sea foam physics, fine natural grain, natural skin texture with visible pores. No CGI look, no fantasy glow, no studio lighting, no smooth plastic skin, no text, no watermark, no extra hands, no distorted laptop keyboard.
Tutorial
DENIZ AKKABAK · SEEDANCE 2.5 TUTORIAL · @DENIZ.AKKABAK
MADE IN INVIDEO · invideo.io/i/DENIZ-ig
Out of Office
How one reference still and one staged Seedance prompt turn a shark encounter into a deadpan remote-work comedy.
A TUTORIAL BY DENIZ AKKABAK · @DENIZ.AKKABAK
PIPELINE GPT IMAGE 2 → SEEDANCE 2.5 · FORMAT 20S · 9:16 VERTICAL · GENRE VIRAL COMEDY
The joke lands because nothing in the frame admits it is a joke. A man answers email chest-deep in a cold sea, a dorsal fin starts circling, and the camera — a stranger's phone on the beach — treats all of it like ordinary Tuesday footage. That deadpan is engineered in two moves: first a photorealistic still that locks the absurd premise into a believable image, then a Seedance 2.5 prompt that choreographs the escalation in four timed stages while insisting, line after line, that everything stay smartphone-plain. The model never gets permission to make it cinematic, so it can only make it real — and real absurdity is what gets shared.
THE PIPELINE AT A GLANCE
- 01
GPT Image 2 — the staged still
Deniz's identity photo goes in as Image 1; the prompt builds the scene around that locked face — chest-deep sea, raised MacBook, coffee cup, deadpan focus — shot like a 135mm editorial stock photo. The output becomes the master reference for the video.
- 02
Seedance 2.5 — the continuous take
The generated still (identity + props + world) and the original photo (face insurance) go in as @Image 1 and @Image 2. A four-stage prompt with timed off-camera dialogue directs the whole 20-second shark encounter as one handheld shot.
STEP 01 · GPT IMAGE 2 — IMAGE MODEL
The Staged Still — A Believable Absurdity
Everything the video needs to keep consistent is decided here, in a single frame: the wet curls, the sunglasses, the space-gray laptop held above the waves, the coffee cup, the flat overcast light. The prompt opens by fencing off the identity — Image 1 is for face and hair only — then spends its remaining lines on realism engineering: telephoto compression, deep focus, desaturated palette, visible pores, sea-foam physics. Note the closing negative line; it is the same insurance policy the video prompt will carry, planted one step earlier.
REFERENCE INPUT — ATTACH AS IMAGE 1
GPT IMAGE 2 · IMAGE PROMPT
Use Image 1 only for the person's facial features, identity, and hair — preserve them exactly. A candid photorealistic photograph of that person standing chest-deep in a rough, cold-looking sea, mid-distance from the camera. Small breaking waves and white foam splash around their chest and shoulders. In one raised hand they hold an open space-gray MacBook laptop above the water, calmly looking at its screen; in the other raised hand they hold [a takeaway coffee cup / an iced drink in a clear plastic cup]. Their expression is completely deadpan and focused, as if working normally, unbothered by the waves. Hair is wet and clinging naturally to the head and neck, water droplets on the skin and forearms. Overcast gray daytime sky, flat diffused natural light, cold blue-gray sea with layered whitecaps stretching to a hazy horizon. No beach, no other people, no boats — only open water. Shot like a real editorial stock photograph on a telephoto lens from the shore, roughly 135mm, eye-level, vertical framing, the subject centered in the lower half of the frame with waves filling the rest. Deep focus, everything sharp. Muted desaturated palette, realistic sea foam physics, fine natural grain, natural skin texture with visible pores. No CGI look, no fantasy glow, no studio lighting, no smooth plastic skin, no text, no watermark, no extra hands, no distorted laptop keyboard.
OUTPUT
STEP 02 · SEEDANCE 2.5 — VIDEO MODEL
The Continuous Take — Four Stages, One Shot
Seedance gets two references and one strict rule per reference. @Image 1 — the still from Step 01 — carries the hero and his props: hair, beard, sunglasses, laptop, cup. @Image 2 — the original photo — re-anchors the face and nothing else. One reference, one job. From there the prompt runs the scene like a shot list: each bracketed stage states its initial state, its event, its camera move and its end state, so the model always knows where the take has been and where it must land. The dialogue is timestamped into the AUDIO line, which means the comedy beats — the gasp at 9 seconds, the laptop commentary at 13 — arrive exactly where the choreography puts the fin and the swat.
REFERENCE INPUTS — ATTACH IN THIS ORDER
SEEDANCE 2.5 · VIDEO PROMPT
20s / 9:16 vertical / viral comedy, candid iPhone beach footage / one continuous handheld take, no cuts
Realistic smartphone video look, natural overcast daylight, handheld shake from the filmer, real world physics, no cinematic grading.
@Image 1 defines the hero: curly dark wet hair, short beard, black square framed sunglasses that stay ON in every frame, shirtless, grey laptop, paper coffee cup. Identity and props only. @Image 2 defines the same single man's face and hair only; ignore its clothing, food and background.
[Stage 1 — approx 0–2s] Initial state: handheld phone view from a sandy beach, choppy sea, a few ordinary swimmers near the shore, one man visible very far out. Event: the unseen woman filming notices him. Camera: quick punch zoom toward him at 2s. End state: the man fills the frame in a long lens look. [Stage 2 — approx 2–8s] Continue from the zoomed view. The man stands chest deep in rolling waves, typing on the open laptop on one hand, lifting it above each incoming wave at the last second, sipping coffee between waves. A dark dorsal fin slowly circles him. Camera: handheld zoomed follow, micro shake. End state: the fin passes close in front of him; he freezes mid sip. [Stage 3 — approx 8–15s] Continue. He downs the last coffee, the empty cup drifts away, he snaps the laptop shut and holds it like a paddle. Each time the shark lunges near the surface he swats its snout once with the flat closed laptop while wading toward shore. The shark keeps following. Camera: the filmer tracks him along the waterline, frame bouncing with her steps. End state: he is waist deep, shark beside him. [Stage 4 — approx 15–20s] Continue. One final firm swat; the shark goes limp and dazed at the surface, unharmed. He tucks the limp shark under his right arm, closed laptop in his left hand, and walks out of the surf toward camera, deadpan, sunglasses on, water streaming off him. Camera: slow zoom out. Final state: held frame as he passes the camera, shark under his arm.
<realistic beach ambience: rolling waves, light wind over the phone mic, distant kids, seagulls> No music. English, natural American accent, woman filming, amused, off camera: at 1s {Oh my god you guys look at this.} at 5s {Is he literally working out there?} at 9s, alarmed then laughing {There is a shark. There is an actual shark!} Male bystander, American accent, off camera at 13s {He is fighting it with the laptop bro!} Woman at 18s, laughing {No way. He is taking it home.}
Exactly one man in focus, one shark, one laptop, one cup. Sunglasses remain on in every frame. Keep face, hair and props unchanged. Camera never enters the water. No other swimmer interacts with him.
no face distortion, no identity change, no sunglasses removal, no injuries, no bite contact, no cartoon shark, no cheap CGI, no cinematic grading, no captions, no on screen text, no music, no extra sharks, no duplicate man.
STEP 03
Anatomy — Why the Shark Reads as Real
Seven decisions carry this prompt. Each one is visible in the finished take.
- 1
The contract
FORMAT: 20s / 9:16 vertical / one continuous handheld take, no cuts. One sentence removes every editing escape hatch — the model must stage the whole escalation inside a single unbroken shot, which is precisely what makes found footage believable.
- 2
The realism strategy
STYLE bans the model's instincts: no cinematic grading, real world physics, handheld shake from the filmer. The camera belongs to a bystander, not a director — so nothing is allowed to look designed.
- 3
Reference locks
@Image 1 owns identity and props; @Image 2 owns face and hair only, with its clothing, food and background explicitly waived. One reference, one job — overlap between references is where identity drift starts.
- 4
Staged choreography
Four bracketed stages, each with an initial state, an event, a camera instruction and an end state. The punch zoom at 2s, the freeze mid-sip, the laptop-as-paddle — the escalation is timed, not wished for.
- 5
Sound as structure
The AUDIO line pins five off-camera lines to exact seconds. The filmer's laughing 'There is an actual shark!' at 9s lands right as the fin passes — dialogue and choreography are the same clock.
- 6
Counts and consistency
CONSISTENCY spells out the census: exactly one man, one shark, one laptop, one cup — and sunglasses that never come off. Counting the props is the cheapest way to stop duplicates before they render.
- 7
The insurance
The NEGATIVE PROMPT closes every remaining door: no bite contact, no injuries, no cartoon shark, no cheap CGI, no captions, no music. The comedy stays harmless, the footage stays plausible.
THE RESULT
Output — The Man Who Took the Shark Home
KEY FRAMES
FINAL OUTPUT
Audit the render against the prompt's own checklist: one continuous shot with the punch zoom at 2 seconds, sunglasses on in every frame, the fin circling before the freeze mid-sip, one swat per lunge, no bite contact, and the final slow zoom out as he walks past the camera with the shark tucked under his arm. If any census item doubles — two cups, a second fin — regenerate rather than repair; consistency errors compound across a 20-second take.
THE MEETING COULD NOT WAIT
One still to lock the joke. One prompt to let it swim.
Swap the shark for a seagull heist, the laptop for an espresso machine, the sea for a rooftop pool — the skeleton carries any deadpan-absurd premise: one reference still that stages the joke, one staged prompt that choreographs it, one negative prompt that keeps it looking real.
Seedance 2.5 runs natively inside invideo — upload your references, paste the prompt, and generate the whole 20-second take in one pass, audio included.
WRITTEN, DIRECTED & PROMPTED BY DENIZ AKKABAK · @DENIZ.AKKABAK