Гайд: Эволюция нейросетей
Кейс & Туториал

Как изменились нейросети: 32 промпта тогда vs 1 промпт сейчас

В ноябре 2025 года мой ролик с девушкой в метро собрал сотни тысяч просмотров и принёс мне основную часть клиентов и подписчиков. Давайте сравним, сколько сил ушло на него тогда, и как этот же процесс выглядит сегодня.
Syntx logo Открыть Syntx AI
01
Ноябрь 2025

Старый подход: ручная сборка

16 кадров и 32 промпта
Тогда мне потребовалось сгенерировать 16 отдельных кадров. Сначала я писала промпты для генерации самих изображений, а потом — ещё 16 промптов для анимации каждой картинки по отдельности. Это был сложный, долгий и кропотливый процесс сборки пазла.
02
Апрель 2026

Новый подход: Без кучи промптов, без монтажа

1. Одно изображение как Character Reference
Сегодня я взяла всего одно изображение со своей героиней. Я использовала его не как стартовый кадр, а как референс персонажа. Модель сама поняла её внешность и характер.
2. Единый промпт-раскадровка
Я прописала только один промпт для Seedance 2.0 в той же раскадровке и последовательности, что и в ноябрьском видео. Нейросеть взяла на себя всю режиссуру и анимацию переходов.

💡 Главный инсайт: Пусть сгенерированный с первого раза вариант и не самый идеальный — его можно доработать, поднапрячься и сделать лучше. Но суть в том, что мы прямо сейчас наблюдаем, как порог входа падает. Нам нужно прикладывать в десятки раз меньше сил и времени, чтобы создавать крутой контент в нейросетях.

03
Пример промпта

Единый промпт для Seedance 2.0

Референс персонажа
Character Reference
@image1 - character reference only — face, outfit, identity anchor. Do not use as first frame. Apply strictly as visual reference for character consistency throughout the entire video.

Character description from @Image1: dark-haired woman, short bob, futuristic dark sunglasses, green patent leather corset top, destroyed ripped denim jeans, black patent over-the-knee heeled boots, small green chain shoulder bag. Maintain this appearance identically across every environment and every second of the video.

ONE CONTINUOUS SEAMLESS SHOT. No cuts. No edits. The camera follows the character without interruption from first to last frame. All transitions between environments happen through fluid camera movement. The camera never stops, never jumps, never cuts. 00:00–00:01 Night city street. Wide medium shot. The woman walks directly toward camera along the center of a wet road. Amber streetlights glow behind her, reflecting off the asphalt. Bokeh car lights recede into the distance. She moves with slow, deliberate confidence. Camera holds steady, slowly pushing in toward her. Her hair moves in the wind. Green bag hangs from her hand. 00:01–00:02 She reaches a manhole in the road. She crouches down, grabs the iron cover and pulls it open. Camera pushes in closer. She peers down into the darkness. Her body leans forward over the open hole. Low angle, amber light on her face. 00:02–00:02,5 Camera drops below street level — now shooting upward from inside the manhole shaft. The circular iron rim frames the night sky above. The woman crouches at the opening above, looking down into the camera. Stars visible in the dark sky behind her. She holds the green bag. The circular concrete shaft wall surrounds the frame. 00:02,5–00:03 She lets go and drops into the hole. Camera watches from below as her body falls toward the lens — arms spread wide, hair flying upward, green corset and ripped jeans visible as she descends through the circular opening. Night sky and stars visible above her through the manhole frame. 00:03–00:04 Camera rotates to overhead downward angle — now looking straight down the interior of a large circular concrete pipe shaft. The woman is seen from above, falling downward through the shaft, arms spread, body in free fall. Concrete pipe walls curve around her. She spins slightly as she falls. Motion blur on the walls, woman sharp. 00:03,5–00:04 She continues falling and spinning inside the shaft. The concrete walls rush upward past the camera. Her body tumbles in slow motion — green corset, ripped jeans, black boots all visible. Camera rotates slightly following her rotation. Deep darkness below her. 00:04–00:05,5 She crashes through the floor of a grand Soviet-style metro station — the marble floor shatters and cracks beneath her on impact, chunks of stone and dust exploding outward. She lands in a low crouch surrounded by cracked marble and rising dust clouds. Green bag lands beside her. The vaulted ceiling of the metro station arches above. She looks up slowly. Camera is at floor level, wide angle catching the full impact crater around her. 00:05,5–00:06,5 She rises and immediately breaks into a full sprint down the metro station corridor. Camera tracks her from the front in a retreating follow-shot — she runs directly toward camera, hair flying wildly behind her, boots striking the cracked marble floor hard. Vaulted arched ceiling, ornate columns, warm chandelier light, dust still rising around her. She clutches the green bag as she runs. 00:06,5–00:07 She enters the metro tunnel and grabs an overhead metal pipe with both hands, swinging up and hanging. Camera drops to extreme low angle looking up — she hangs by both arms from the pipe in the center of the dark tunnel. Her body is spread wide, boots dangling, green corset catching the sparse tunnel strip lights. Train tunnel walls rush past in heavy motion blur on both sides. She stares straight ahead, expression cold and controlled. 00:07–00:08 Camera pulls back and rises rapidly. The dark tunnel gives way to open sky — golden hour light, wide open landscape. She is now sitting on the roof of a moving metro train, centered in frame. Shot from the front of the train, locked. She sits with one leg forward, green bag dangling from her hand. Wind violently whips her dark hair. The landscape blurs past on both sides at high speed. Expression: detached, sovereign, absolute. 00:08–00:09 Split-screen vertical triptych transition: top third shows her sitting in the circular concrete opening framed by the brutalist wall, night sky behind her, green bag in hand, one booted leg extended. Middle third is an extreme close-up of her face in profile — futuristic sunglasses, city bokeh lights behind, dramatic side lighting on her jawline and cheekbone. Bottom third shows a low ground-level shot of her boots and legs walking along metro railway tracks toward camera, green bag swinging at her side. 00:09–00:10,5 Extreme close-up. She is lying face-down on the metro tracks, her face inches from the steel rail. She leans in and slowly runs her tongue along the cold metal rail. Camera is at rail level, tight on her face — futuristic transparent/clear square sunglasses, dark eyes, parted lips. The rail gleams under close light. Her expression is provocative and unhinged. Shallow depth of field, bokeh background of the tunnel. 00:10,5–00:13 Wide medium shot. She walks directly toward camera along the center of the metro railway tracks inside a dark underground tunnel. Single light source ahead illuminates her from the front. She walks slowly, deliberately, with complete confidence. Green bag in hand swinging gently. On the tunnel wall behind her the word MITINA appears in large dark green dripping graffiti letters. She walks forward as the word fills the frame behind her. Camera holds steady as she approaches. On the final beat she is close to camera, face filling frame. Fade to black.

Technical Details & Constraints:
4K ultra-high definition, sharp resolution, rich cinematic detail, natural colors, no blur on subject, no ghosting, no flickering, stable follow-cam, 24fps film look, cinematic color grade, deep blacks, silky smooth camera movement throughout.

Constraints: clear facial features, stable face, no distortion, no deformation, normal body proportions, natural structure, same character throughout all 13 seconds, consistent clothing, unchanged hairstyle, green patent corset unchanged, black over-the-knee boots unchanged, green chain bag unchanged.
Открыть Нейробазу
Made on
Tilda