Back to gallery

The Freeze Frame

Video
0 steals

Find an accomplice

SEEDANCE 2.5 · VIDEO PROMPT

ONE CONTINUOUS SHOT — NO CUTS, TRANSITIONS, TELEPORTATION OR HIDDEN EDITS. Use the reference images exactly: take the character's glasses-free face, eyes and facial structure from Image 2; take his curly hair, beard and the exact black square-framed sunglasses from Image 1 — the sunglasses stay ON in every frame. Ignore the clothing, backgrounds, food and table in both reference photos. Wardrobe: gray double-breasted suit, tie, briefcase and Sony Ericsson W810i. No identity, wardrobe or prop changes. Bright, colorful, sunny Midtown Manhattan, early-2000s MiniDV Handycam aesthetic: 4:3 SD, CCD softness, natural handheld shake, tape artifacts. No modern cinematic sharpness or desaturation. Character runs through the crowd while cameraman follows. Dialogue finishes within 2 seconds: "Got the prompt?" — "Not yet, I'm getting it now." — "Hurry." — "I've got it." Immediately he trips. Coffee, papers, phone and briefcase fly naturally. Sunglasses remain one intact, unbroken pair on his face. Before impact, within 2–3 seconds, TIME FREEZES COMPLETELY. Everything freezes except the camera. Absolute silence: no dialogue, voices, traffic or phone audio. Camera smoothly evolves from handheld into FPV flight, rises above the frozen Manhattan scene, flies past the character, intact sunglasses, coffee, phone and documents. End on a physical document reading exactly: "COMMENT 'PROMPT' TO GET THE FULL PROMPT" No digital text or overlays. Orbit back to the original position. Time instantly resumes with no transition. Objects continue naturally, character hits the ground, gets up, adjusts his intact sunglasses, looks at his phone, then at camera in disbelief. Crowd and traffic continue normally. Maintain realistic physics, bright colors, exact identity and continuous spatial continuity throughout.

Tutorial

DENIZ AKKABAK · SEEDANCE 2.5 TUTORIAL · @DENIZ.AKKABAK

MADE IN INVIDEO · invideo.io/i/DENIZ-ig

The finished 30-second one-shot — sprint, wipe-out, frozen time, and the flying CTA. (Open this document in Google Docs to watch it play.)
The finished 30-second one-shot — sprint, wipe-out, frozen time, and the flying CTA. (Open this document in Google Docs to watch it play.)

The Freeze Frame

How one breakfast photo and a single Seedance 2.5 prompt stop all of Midtown Manhattan mid-fall.

A TUTORIAL BY DENIZ AKKABAK · @DENIZ.AKKABAK

MODEL SEEDANCE 2.5 · FORMAT 9:16 · 30S ONE-SHOT · LOOK EARLY-2000S MINIDV

This video is a trust exercise with the model. A man in a gray double-breasted suit sprints through a sunny Midtown crowd, barks four lines of dialogue into a Sony Ericsson, trips — and just before he hits the asphalt, time stops for everyone except the camera. The handheld Handycam becomes a drone, flies through the frozen coffee and paperwork, lands on a physical page that tells viewers to comment “PROMPT”, orbits home, and lets gravity finish the joke. No cuts hide any of it — which is exactly why it reads as real. And the man in the suit? That's me, lifted out of a breakfast photo. One reference set, one prompt, one generation.

STEP 01 · GENERATE THE ONE-SHOT — SEEDANCE 2.5

One Prompt, Thirty Seconds, Zero Cuts

There is no image-model stage this time — Seedance 2.5 does everything in a single pass, working directly from photographs of a real face. The prompt names two references and gives each one exactly one job: Image 1 donates the curly hair, the beard, and the exact black square-framed sunglasses (the photo below); Image 2 is a glasses-free portrait of the same face that supplies the eyes and bone structure hiding underneath. Everything else in both photos — the croissant, the terrace, the gray t-shirt — is explicitly waived with “Ignore the clothing, backgrounds, food and table.” One reference, one job.

REFERENCE INPUTS — ATTACH IN THIS ORDER

@Image 1 — the sunglasses portrait. The prompt takes only the curly hair, beard and the exact black square frames from here; @Image 2 (not shown) is the same face without glasses, used for eyes and facial structure.
@Image 1 — the sunglasses portrait. The prompt takes only the curly hair, beard and the exact black square frames from here; @Image 2 (not shown) is the same face without glasses, used for eyes and facial structure.

SEEDANCE 2.5 · VIDEO PROMPT

ONE CONTINUOUS SHOT — NO CUTS, TRANSITIONS, TELEPORTATION OR HIDDEN EDITS. Use the reference images exactly: take the character's glasses-free face, eyes and facial structure from Image 2; take his curly hair, beard and the exact black square-framed sunglasses from Image 1 — the sunglasses stay ON in every frame. Ignore the clothing, backgrounds, food and table in both reference photos. Wardrobe: gray double-breasted suit, tie, briefcase and Sony Ericsson W810i. No identity, wardrobe or prop changes. Bright, colorful, sunny Midtown Manhattan, early-2000s MiniDV Handycam aesthetic: 4:3 SD, CCD softness, natural handheld shake, tape artifacts. No modern cinematic sharpness or desaturation. Character runs through the crowd while cameraman follows. Dialogue finishes within 2 seconds: "Got the prompt?" — "Not yet, I'm getting it now." — "Hurry." — "I've got it." Immediately he trips. Coffee, papers, phone and briefcase fly naturally. Sunglasses remain one intact, unbroken pair on his face. Before impact, within 2–3 seconds, TIME FREEZES COMPLETELY. Everything freezes except the camera. Absolute silence: no dialogue, voices, traffic or phone audio. Camera smoothly evolves from handheld into FPV flight, rises above the frozen Manhattan scene, flies past the character, intact sunglasses, coffee, phone and documents. End on a physical document reading exactly: "COMMENT 'PROMPT' TO GET THE FULL PROMPT" No digital text or overlays. Orbit back to the original position. Time instantly resumes with no transition. Objects continue naturally, character hits the ground, gets up, adjusts his intact sunglasses, looks at his phone, then at camera in disbelief. Crowd and traffic continue normally. Maintain realistic physics, bright colors, exact identity and continuous spatial continuity throughout.

Paste as one block. The quoted dialogue, the timings and the CTA text are all load-bearing — change nothing.

STEP 02

Anatomy — Why the Freeze Holds

Thirty seconds is a long time for a video model to stay honest. The prompt survives because every risky moment has a rule waiting for it before it happens:

  1. 01

    The contract

    “ONE CONTINUOUS SHOT — NO CUTS, TRANSITIONS, TELEPORTATION OR HIDDEN EDITS” opens the prompt in capitals. Everything that follows is negotiated inside this constraint; the model can't escape a hard moment by cutting away from it.

  2. 02

    The two-image identity lock

    Face geometry from Image 2, hair, beard and sunglasses from Image 1, and a blanket waiver on everything else in the photos. Splitting identity across two references keeps the likeness stable even though the glasses hide half the face.

  3. 03

    Era-locking props

    A gray double-breasted suit, a briefcase and a Sony Ericsson W810i aren't decoration — they date the frame. Paired with “No identity, wardrobe or prop changes,” they give continuity something concrete to hold onto for thirty seconds.

  4. 04

    The look is a format, not a filter

    “4:3 SD, CCD softness, natural handheld shake, tape artifacts” plus the ban on “modern cinematic sharpness or desaturation.” MiniDV grain flatters AI faces and crowds — softness reads as 2004, not as a rendering flaw.

  5. 05

    Sound as structure

    The four dialogue lines must finish “within 2 seconds,” and the freeze demands “Absolute silence: no dialogue, voices, traffic or phone audio.” The audio track is choreographed as tightly as the camera — silence is what sells the stopped clock.

  6. 06

    The camera is the only survivor

    “Everything freezes except the camera” turns the handheld cameraman into an FPV drone mid-shot, flies him to “a physical document” carrying the CTA — “No digital text or overlays” — and orbits back so time can resume from the exact same angle.

  7. 07

    The insurance

    “Sunglasses remain one intact, unbroken pair on his face” is stated twice, and the final line demands realistic physics, exact identity and “continuous spatial continuity.” The most breakable object in the scene is the one the prompt refuses to let break.

THE RESULT

Output — Midtown, Paused

KEY FRAMES

Left to right: the phone-call sprint, the frozen wipe-out mid-flight, and the physical CTA page hanging in stopped air.
Left to right: the phone-call sprint, the frozen wipe-out mid-flight, and the physical CTA page hanging in stopped air.

OUTPUT

The delivered 30-second generation, untouched. (Open in Google Docs to watch it play.)
The delivered 30-second generation, untouched. (Open in Google Docs to watch it play.)

Audit the output against the prompt's own checklist: the dialogue lands inside the first two seconds, the sunglasses come through the fall as one unbroken pair, the freeze is total except for the camera, the CTA page is a real object catching real sunlight, and time resumes with no visible seam. If your generation drops any of these, regenerate — the prompt already contains the fix.

ONE PHOTO. ONE PROMPT. ONE UNBROKEN SHOT.

When you can't cut, you have to choreograph — and the prompt is the choreography.

The skeleton carries any story you want to freeze. Swap Midtown for a Tokyo crossing, the briefcase for a skateboard, the W810i for a flip phone — keep the one-shot contract, the two-reference identity lock, the timed dialogue, the freeze, the FPV flight to a physical sign, and the clean resume. Everything else is set dressing.

Seedance 2.5 runs inside invideo — upload your references, paste the prompt, and generate without leaving the browser. Start here:

invideo.io/i/DENIZ-ig

WRITTEN, DIRECTED & PROMPTED BY DENIZ AKKABAK · @DENIZ.AKKABAK