The Confession

GPT IMAGE 2 → SEEDANCE 2.5FORMAT 30S / 9:16 VERTICAL · GENRE INFLUENCER VLOG ADPhoto

0 steals

Find an accomplice

Prompt 2

SEEDANCE 2.5 · IMAGE TO VIDEO · 30S · 9:16 · 2 REFERENCES FORMAT: 30s / 9:16 vertical / influencer travel vlog ad / fast-cut montage opening, then one continuous handheld selfie take STYLE: Authentic phone-shot influencer vlog realism, natural golden-hour Mediterranean light, warm color, slight handheld micro-shake from her arm, shallow phone-camera depth, no cinematic grading, one consistent visual identity. REFERENCES: @Image 1 defines the influencer: her face, freckles, wavy sun-bleached blonde hair, black bikini, sheer orange-and-white striped kimono, layered gold chain necklaces with a small coin pendant, gold rings, AND the balcony location with white railing, string lights, terracotta floor and the open sea view with cliffside village. Preserve identity and wardrobe exactly in every cut. @Image 2 defines ONLY the white wordmark logo that appears on the black parachute canopy in Stage 5. Treat the black background of @Image 2 as the canopy surface itself. The logo appears nowhere else in the video. VIDEO PROMPT: [Stage 1 — approx 0-2s] Quick vlog cut: the same woman in the same kimono outfit walks toward camera down a narrow sunlit Italian cliffside village lane with lemon trees and pastel walls, smiling, hair moving in the breeze. Camera: handheld phone tracking backward at chest height. End state: she is mid-stride, relaxed. [Stage 2 — approx 2-4s] Quick vlog cut: outdoor cafe table, she bites into a slice of pizza, eyes closing briefly with pleasure, silent delighted reaction. Camera: handheld close-up across the table. End state: slice lowered, she chews happily. [Stage 3 — approx 4-7s] Quick vlog cut: she stands at the edge of the lane as a young dark-haired Italian man rides past on a vintage scooter and turns his head to look at her. She glances at the camera with raised eyebrows. English, natural American accent, playful teasing influencer tone: {Italian men...} End state: scooter exits frame right, she smirks at the lens. [Stage 4 — approx 7-12s] Hard cut to the balcony from @Image 1, matching its framing: selfie-style front camera held in her extended arm, sea and cliffs behind her, boats crossing the water. She talks straight into the lens with real influencer energy. English, natural American accent, confident casual delivery: {I was made with AI. My hair is not real.} She lifts a strand of hair toward the lens; the camera pushes in close so individual hair strands catch the light. End state: camera close on hair, her face partly visible behind it. [Stage 5 — approx 12-16s] Continue from the close hair shot. She lets the hair fall and lifts the coin pendant with two fingers. English, natural American accent, same delivery: {Neither is this necklace.} Camera pushes to a near-macro on the gold chain and pendant, tiny natural handheld drift. End state: pendant resting back on her chest. [Stage 6 — approx 16-23s] English, natural American accent, same delivery: {Nor is this Amalfi vacation.} She swings the phone camera away from herself toward the open sea: boats leaving white wakes, two seagulls flying past close to the lens. At approx 19s a single paraglider crosses the frame from the cliff side gliding toward the sea: pitch-black canopy and harness, the white wordmark from @Image 2 large and clean on the canopy, one distant joyful male scream: {Woohoo!} End state: paraglider small over the water, camera still on the view. [Stage 7 — approx 23-30s] She swings the camera back to her face, wind in her hair, golden light on her skin. English, natural American accent, slower, teasing, dropping to a knowing tone: {So... how was I created?} She gives a silent knowing smile directly into the lens. Final state: held selfie frame on her smile, sea behind her, no camera movement. AUDIO: <Coastal wind, soft waves below, seagull calls, fabric flutter of the kimono. Stage 1: distant village ambience and footsteps. Stage 2: cafe chatter, plate clink, one soft bite. Stage 3: scooter engine approaching and passing. Stage 5-6: wind rises slightly on the open view, faint canopy flutter as the paraglider passes.> All dialogue lines as written per stage, recorded like a real phone vlog: close, present, slightly windy. No music. CONSISTENCY: Exactly one woman, the SAME woman from @Image 1 in every cut: same face, freckles, blonde wavy hair, black bikini, striped kimono, gold necklaces and rings, unchanged across all locations. Exactly one scooter rider, visible only in Stage 3. Exactly one paraglider, visible only in Stage 6, logo only on its canopy. Balcony layout, railing, lighting direction and sea view match @Image 1. NEGATIVE PROMPT: no identity change, no face distortion, no wardrobe change, no extra people on the balcony, no second woman, no duplicate paraglider, no logo anywhere except the canopy, no logo distortion, no on-screen text, no captions, no subtitles, no watermark, no music, no cartoon look, no cinematic color grading, no plastic skin, no warped hands, no unnatural camera stabilization.

MADE WITH SEEDANCE 2.5 IN ARCADS · arcads.ai

The finished 30-second ad — she confesses everything, and it still sells. (Open this document in Google Docs to watch it play.)

The Confession

An AI influencer walks you through her Amalfi vacation — then admits, item by item, that none of it exists.

A TUTORIAL BY DENIZ AKKABAK · @DENIZ.AKKABAK

PIPELINE GPT IMAGE 2 → SEEDANCE 2.5 · FORMAT 30S / 9:16 VERTICAL · GENRE INFLUENCER VLOG AD

Every travel vlog on your feed makes the same silent promise: this happened. This one breaks the promise on camera — and that is exactly why it works as an ad. For seven seconds you get an ordinary Amalfi montage: the lane, the pizza, the scooter boy. Then she looks into her own front camera and says I was made with AI. My hair is not real. From that moment the viewer isn't watching a vacation anymore; they're auditing one. Every strand of hair and every link of the necklace becomes proof of the product. The pipeline behind it is two moves: one GPT Image 2 still that locks the girl, her wardrobe and her balcony into a single reference, and one Seedance 2.5 prompt that plays seven staged beats across it without ever breaking her identity.

THE PIPELINE AT A GLANCE

01
GPT Image 2 — the identity still
One vertical selfie generates the entire cast and set: face, freckles, kimono, layered gold jewelry, and the balcony with its railing, string lights and sea view. This single frame becomes @Image 1 — the identity contract for every cut that follows.
02
Seedance 2.5 — the 30-second take
A seven-stage prompt drives a fast-cut montage into one continuous handheld confession, pinning the woman to @Image 1 and scoping the Arcads wordmark (@Image 2) to a single paraglider canopy at second 19.

STEP 01 · GPT IMAGE 2 — STILL GENERATION

The Girl Who Doesn't Exist

The still does the heavy lifting for the whole ad, so the prompt is written like a casting call plus a location scout in one. It specifies the person (mid-20s, wavy blonde hair in the sea wind, freckles, unguarded laugh), the wardrobe down to the jewelry stack (chunky chain, coin pendant, fine long chain, stacked rings — each one will be interrogated on camera later), and the set (pastel-peach house, moody slate sky with one golden shaft, three boats, unlit string lights, the rattan chair). The final line bans everything that would betray a render: text, watermarks, plastic skin, perfect symmetry, posing.

Prompt 1

OUTPUT

The GPT Image 2 still — one frame that will serve as face, wardrobe and location reference for all seven stages.

STEP 02 · SEEDANCE 2.5 — VIDEO GENERATION

One Take, Seven Confessions

Seedance 2.5 receives exactly two references, and the order matters — attachment order defines @Image 1 and @Image 2 inside the prompt. The first is the still from Step 01, carrying the influencer's identity and the balcony. The second is the Arcads wordmark on a black field — and the prompt immediately fences it in: the logo exists only on a paraglider canopy in Stage 6, and the black background is read as the canopy fabric itself, nowhere else in the video.

REFERENCE INPUTS — ATTACH IN THIS ORDER

@Image 1 — the GPT Image 2 still: identity, wardrobe, jewelry and the balcony location, preserved in every cut.

@Image 2 — the white Arcads wordmark on black: scoped to the paraglider canopy only, the black field read as canopy fabric.

Prompt 2

SEEDANCE 2.5 · IMAGE TO VIDEO · 30S · 9:16 · 2 REFERENCES

30s / 9:16 vertical / influencer travel vlog ad / fast-cut montage opening, then one continuous handheld selfie take

Authentic phone-shot influencer vlog realism, natural golden-hour Mediterranean light, warm color, slight handheld micro-shake from her arm, shallow phone-camera depth, no cinematic grading, one consistent visual identity.

@Image 1 defines the influencer: her face, freckles, wavy sun-bleached blonde hair, black bikini, sheer orange-and-white striped kimono, layered gold chain necklaces with a small coin pendant, gold rings, AND the balcony location with white railing, string lights, terracotta floor and the open sea view with cliffside village. Preserve identity and wardrobe exactly in every cut. @Image 2 defines ONLY the white wordmark logo that appears on the black parachute canopy in Stage 5. Treat the black background of @Image 2 as the canopy surface itself. The logo appears nowhere else in the video.

[Stage 1 — approx 0-2s] Quick vlog cut: the same woman in the same kimono outfit walks toward camera down a narrow sunlit Italian cliffside village lane with lemon trees and pastel walls, smiling, hair moving in the breeze. Camera: handheld phone tracking backward at chest height. End state: she is mid-stride, relaxed. [Stage 2 — approx 2-4s] Quick vlog cut: outdoor cafe table, she bites into a slice of pizza, eyes closing briefly with pleasure, silent delighted reaction. Camera: handheld close-up across the table. End state: slice lowered, she chews happily. [Stage 3 — approx 4-7s] Quick vlog cut: she stands at the edge of the lane as a young dark-haired Italian man rides past on a vintage scooter and turns his head to look at her. She glances at the camera with raised eyebrows. English, natural American accent, playful teasing influencer tone: {Italian men...} End state: scooter exits frame right, she smirks at the lens. [Stage 4 — approx 7-12s] Hard cut to the balcony from @Image 1, matching its framing: selfie-style front camera held in her extended arm, sea and cliffs behind her, boats crossing the water. She talks straight into the lens with real influencer energy. English, natural American accent, confident casual delivery: {I was made with AI. My hair is not real.} She lifts a strand of hair toward the lens; the camera pushes in close so individual hair strands catch the light. End state: camera close on hair, her face partly visible behind it. [Stage 5 — approx 12-16s] Continue from the close hair shot. She lets the hair fall and lifts the coin pendant with two fingers. English, natural American accent, same delivery: {Neither is this necklace.} Camera pushes to a near-macro on the gold chain and pendant, tiny natural handheld drift. End state: pendant resting back on her chest. [Stage 6 — approx 16-23s] English, natural American accent, same delivery: {Nor is this Amalfi vacation.} She swings the phone camera away from herself toward the open sea: boats leaving white wakes, two seagulls flying past close to the lens. At approx 19s a single paraglider crosses the frame from the cliff side gliding toward the sea: pitch-black canopy and harness, the white wordmark from @Image 2 large and clean on the canopy, one distant joyful male scream: {Woohoo!} End state: paraglider small over the water, camera still on the view. [Stage 7 — approx 23-30s] She swings the camera back to her face, wind in her hair, golden light on her skin. English, natural American accent, slower, teasing, dropping to a knowing tone: {So... how was I created?} She gives a silent knowing smile directly into the lens. Final state: held selfie frame on her smile, sea behind her, no camera movement.

<Coastal wind, soft waves below, seagull calls, fabric flutter of the kimono. Stage 1: distant village ambience and footsteps. Stage 2: cafe chatter, plate clink, one soft bite. Stage 3: scooter engine approaching and passing. Stage 5-6: wind rises slightly on the open view, faint canopy flutter as the paraglider passes.> All dialogue lines as written per stage, recorded like a real phone vlog: close, present, slightly windy. No music.

Exactly one woman, the SAME woman from @Image 1 in every cut: same face, freckles, blonde wavy hair, black bikini, striped kimono, gold necklaces and rings, unchanged across all locations. Exactly one scooter rider, visible only in Stage 3. Exactly one paraglider, visible only in Stage 6, logo only on its canopy. Balcony layout, railing, lighting direction and sea view match @Image 1.

no identity change, no face distortion, no wardrobe change, no extra people on the balcony, no second woman, no duplicate paraglider, no logo anywhere except the canopy, no logo distortion, no on-screen text, no captions, no subtitles, no watermark, no music, no cartoon look, no cinematic color grading, no plastic skin, no warped hands, no unnatural camera stabilization.

STEP 03 · THE PLATFORM — ARCADS

Running Seedance 2.5 Inside Arcads

Everything in Step 02 runs inside Arcads — the AI ads platform built exactly for this kind of UGC-style spot, and the reason the paraglider wears its logo. Sign in at arcads.ai, open a new AI video generation and pick Seedance 2.5 as the model. Attach the two references in the order documented above — the identity still first, the logo card second — so the prompt's @Image 1 and @Image 2 tags resolve correctly, then paste the video prompt verbatim and set the output to 30 seconds, 9:16 vertical. The whole confession — dialogue, sound design and the second-19 logo flyby — generates in a single run; no editing timeline, no compositing, no post.

STEP 04 — WHY THE PROMPT WORKS

Anatomy — Why She Gets Away With It

01
The contract
FORMAT: 30s / 9:16 vertical / influencer travel vlog ad / fast-cut montage opening, then one continuous handheld selfie take — before a single scene is described, the model knows the ad's exact shape: how long, which frame, and that the edit grammar switches from montage to one unbroken take at the confession.
02
Two references, two jobs
@Image 1 owns identity, wardrobe AND the balcony location; @Image 2 is fenced with defines ONLY the white wordmark logo and The logo appears nowhere else in the video. Scoping each reference to a job is what stops the logo from leaking onto t-shirts, walls or boats.
03
Realism by refusal
The style line builds authenticity out of bans: no cinematic grading, slight handheld micro-shake from her arm, shallow phone-camera depth. The video looks true because everything that looks expensive is forbidden.
04
The confession ladder
Three escalating reveals share one grammar: {My hair is not real.} → {Neither is this necklace.} → {Nor is this Amalfi vacation.} — each paired with a camera push (hair strands, pendant macro, the swing to the sea) so the eye verifies exactly what the voice denies.
05
The logo delivery vehicle
The brand never interrupts; it flies past. At approx 19s a single paraglider crosses the frame with a pitch-black canopy carrying the wordmark, sold as scenery by one distant {Woohoo!} — the ad placement disguised as a vacation moment.
06
Sound as structure
The AUDIO block scores each stage separately — footsteps, plate clink, scooter engine, rising wind, canopy flutter — and ends with the realism keystone: No music. A soundtrack would confess the production; wind confesses nothing.
07
The insurance
The negative prompt patrols every failure mode the concept invites: no second woman, no duplicate paraglider, no logo anywhere except the canopy, no logo distortion, plus the vlog-killers no captions, no music, no unnatural camera stabilization.

THE RESULT — SEEDANCE 2.5 OUTPUT

Output — Thirty Seconds of Honest Lying

The final generation, untouched — montage, confession, flyby and close, in one Seedance 2.5 run.

Three beats of the confession — the balcony admission, the necklace macro, and the Arcads canopy crossing the frame at second 19.

Audit the output against the prompt's own checkpoints: the same face, freckles and kimono in all seven stages; the montage cutting hard at seconds 2, 4 and 7; the push into hair strands at the first confession; the pendant macro; the paraglider entering from the cliff side at second 19 with the wordmark clean on a black canopy and appearing nowhere else; and the final held selfie frame with no camera movement. If any cut swaps her jewelry or a second canopy drifts in, the identity lock or the reference scoping has been loosened — regenerate before you touch anything else.

THE BEST AD IS THE ONE THAT CONFESSES

She told you nothing was real — and you watched to the end anyway.

Swap the kimono for streetwear, the Amalfi balcony for a Tokyo rooftop, the necklace for a watch and the paraglider for a passing tram — the skeleton carries any AI-influencer ad: one still that locks the identity, one scoped logo reference, one confession ladder that turns the disclaimer into the hook.

Seedance 2.5 — the model behind every cut of this video — is ready to run inside Arcads: attach your references, paste your prompt, and generate UGC-style AI ads end to end.

arcads.ai

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

Back to gallery