June 12

Как снять POV-видео с цунами с помощью Seedance 2.0

Я всё делал в SYNTX.AI

Использовал нейросети:

Seedance 2.0 Pro
ChatGPT Image 2.0

В этом гайде разберём видео из двух отрезков:

  1. Первый отрезок — человек идёт по высохшему морскому дну, рядом огромный танкер и вдалеке город.
  2. Второй отрезок — человек уже в вертолёте и снимает, как гигантское цунами накрывает прибрежный город.

1. Подготовка фото для проекта

Для начала нужно подготовить фото лица в сетке 3х3 на белом фоне в разных ракурсах.

Это важно, чтобы нейросеть лучше держала ваше лицо в видео и не меняла внешность во время движения камеры.

Этот файл загружаем как:

@Image1 — сетка лица 3х3


2. Подготовка референса танкера

Для первого отрезка нужно подготовить референс большого танкера.

Референс нужен не для готовой сцены, а чтобы Seedance понял форму, масштаб, грязную металлическую поверхность и индустриальные детали корабля.

Лучше использовать изображение, где танкер выглядит очень крупным, ржавым, мокрым и тяжёлым.

Этот файл загружаем как:

@Image2 — референс танкера


3. Первый отрезок: высохшее морское дно

Настройки:

Видео → Seedance 2.0 → Режим Pro → 720p → Соотношение сторон 9:16 → Длительность 7 сек.

Загружаем файлы по порядку:

  1. @Image1 — сетка лица 3х3
  2. @Image2 — референс танкера

ВАЖНЕЙШИЙ НЮАНС: когда вы скопируете промпт ниже и вставите его в окно нейросети, проверьте, чтобы каждый тег правильно соответствовал вложению.

ПРОМПТ ДЛЯ ПЕРВОГО ОТРЕЗКА:

Total: 7s / Single continuous handheld smartphone shot / 9:16 aspect ratio.

Style: raw ungraded bystander phone footage, amateur handheld smartphone camera, natural imperfections, realistic phone optics, 35mm normal lens, no stylization. Overcast weather, gray haze, damp air, bleak coastal urban atmosphere, grounded realism. This must look like accidental anxious witness footage, not cinematic AI footage.

Use @Image1 strictly for the exact facial identity of the main man only. @Image1 is a facial identity reference only. Do not copy the 3x3 grid, background, lighting, layout, or any reference-image elements from @Image1.

Use @Image2 strictly for the oil tanker only: dirty industrial hull, rust, wet metal, heavy mass, large scale, deck pipes, railings, cargo structure. Do not copy the background, lighting, framing, or environment from @Image2.

Create a completely new scene:
a vast exposed seabed basin. The entire visible area used to be underwater, but now it is exposed. No open water is visible anywhere in the frame. This must not look like a beach or shoreline.

Environment:
broad wet exposed seabed, muddy uneven ground, shallow ridges, dips, puddles, wet sand, mud, silt, seaweed, shells, ropes, buoys, scattered marine debris, a few stranded fish weakly twitching on the wet ground. The terrain is messy and physically real, not beautiful or stylized.

Far away on the dry horizon, a modern city skyline with high-rise buildings and skyscrapers is visible.

A gigantic oil tanker is stranded very close in the mid-ground. It must feel shockingly near and overwhelmingly large, like a massive wall of steel beside the camera. It is dirty, wet, rusty, heavy, slightly tilted, grounded awkwardly on the exposed seabed. It must not appear as a small ship in the distance. Part of the tanker extends beyond the frame.

[0-2s]
Handheld selfie POV. The main man with the exact facial identity from @Image1 walks carefully across the uneven exposed seabed while holding the phone toward himself. He wears ordinary dark casual outerwear suitable for damp windy weather. He does not look like he is posing. His expression is tense, confused, uneasy, and stunned. He may look slightly past the phone toward the horizon instead of directly into the lens. He does not speak.

Opening hook:
during the selfie shot, the distant city skyline must already be visible behind him. The viewer must immediately understand that he is standing far out on exposed seabed, not near a normal shore.

Camera movement: real handheld walking footage, slight bounce from footsteps, subtle micro-jitters, unstable movement from slippery uneven ground.

[2-4s]
The camera smoothly but shakily turns away from the man’s face in one continuous handheld shot. The man leaves the frame completely.

The camera reveals the gigantic oil tanker very close by. The tanker dominates a large part of the frame like a huge stranded industrial wall. Nearby people and seabed details look tiny compared to it. It feels heavy, dirty, wet, realistic, and disturbing in scale.

Several ordinary people are scattered across the exposed seabed. One films on a phone, another points toward the tanker or horizon, another walks uncertainly across the mud, another stands still in disbelief.

[4-7s]
The camera continues the same handheld movement and reveals more scale around the tanker and the exposed seabed. The distant city skyline remains visible. No open water is visible anywhere.

A stranded dolphin is visible in the mid-ground on the exposed seabed, still alive and weakly moving without water. It must not dominate the frame. It is a secondary emotional detail. A few nearby people are trying to help it in a messy improvised way: one kneels beside it, another splashes or pours water onto it, another tries to guide or drag it toward a shallow puddle. Their actions should feel brief, chaotic, and natural, not staged.

The frame shows only vast wet seabed, scattered people, marine debris, weakly moving fish, the struggling dolphin, and the enormous stranded tanker.

Audio:
diegetic sounds only — strong wind, wet footsteps on mud, distant confused voices, water dripping from mud and metal, subtle metal creaks from the tanker, soft dolphin distress sounds, uneasy silence. No music. No dialogue from the main character.

Constraints:
Use @Image1 only for facial identity.
Use @Image2 only for tanker design, material, and scale.
The man must leave the frame after the camera turns away.
The city skyline must already be visible in the opening selfie shot.
The tanker must be very close, huge, oppressive, and partially out of frame.
No open water visible anywhere.
The scene must feel like the middle of an exposed seabed basin, not a shore.
The dolphin must remain a secondary detail, not the main subject.
Keep it raw, grounded, non-cinematic, and ugly-realistic.

Negative constraints:
no beach shoreline, no visible open water, no ocean waves, no calm sea surface, no bright sunny weather, no tropical postcard look, no beautiful reef, no floating whales, no fantasy elements, no perfect clean beach, no heroic acting, no cinematic shot, no drone shot, no fisheye, no CGI look, no cartoon, no 3D, no camera cuts, no scene transitions, no copied layout from @Image1, no copied background from @Image2.

4. Второй отрезок: цунами с вертолёта

Настройки:

Видео → Seedance 2.0 → Режим Pro → 720p → Соотношение сторон 9:16 → Длительность 7 сек.

Загружаем файл:

  1. @Image1 — сетка лица 3х3

ПРОМПТ ДЛЯ ВТОРОГО ОТРЕЗКА:

Total: 7s / Single continuous tracking shot / 9:16 aspect ratio.
Style: Raw ungraded bystander phone footage, amateur handheld smartphone camera, natural imperfections, 35mm normal lens. Heavy overcast sky, cold gray light, marine haze, gloomy disaster mood.

Replace the character with the exact facial identity from @Image1. The person from @Image1 wears dark casual outerwear and aviation headphones.

[0-2s] Handheld selfie POV inside a vibrating helicopter. A man (@Image1) wearing dark outerwear and an aviation headset sits near the open side doorway. He looks at the lens with a calm, serious, stunned expression. Only the empty, dark, turbulent ocean and gloomy stormy horizon are visible behind him through the dark metal framing of the helicopter door. Unstable handheld shake and wind buffeting.

[2-4s] The camera quickly pans toward the open helicopter doorway, leaving the man entirely out of the frame. The view shifts to the disaster outside, transitioning smoothly from the empty ocean. The frame must remain one continuous real aerial view from the helicopter, not a split-screen composition. The dense coastal city and skyscrapers naturally occupy the left side of the frame along the shoreline, while the ocean occupies the right side beyond the coast. A terrifyingly gigantic, dark, turbulent tsunami wave advances from the ocean toward the city to engulf the shoreline.

[4-7s] The camera shakily holds on this continuous aerial view. The massive wave swallows the buildings on the left. Realistic dirty foam and dense mist. The tsunami rolls forward relentlessly. Extreme scale.

[Audio] Diegetic sounds only: loud helicopter rotor noise, cabin vibration, wind howling, and deep roaring water.

Constraints: Man MUST leave the frame completely after the pan. Avoid identity drift, no fisheye lens distortion, no camera cuts, no scene transitions, no 3D, no cartoon, no clean surf wave, maintain realistic physics, realistic heavy water dynamics, maintain one continuous unified frame, no split-screen, no side-by-side composition, no hard vertical seam.

5. Монтаж

После генерации сохраняем оба отрезка и соединяем их в монтаже.

Сначала ставим отрезок с высохшим морским дном, затем сразу отрезок с вертолётом и цунами.

В конце можно добавить лёгкий шум камеры, тряску и цветокор под телефонную съёмку.