Цвета

Свои образы в трендовом ролике. Два варианта на выбор и четыре шага.

Сначала открой бот

Все нейросети из этого гайда собраны в одном месте — не нужно заводить аккаунты и привязывать карты. Я работаю только здесь.

Открыть бот Мой готовый шаблон

Вариант А

Берёшь мои образы — меняешь лицо на своё

Самый быстрый путь. Мои пять картинок ниже, плюс своё фото — два файла в запрос. Образ остаётся мой, лицо становится твоим.

Запрос

Take the first photo as the base. Keep the outfit, the styling, the hair and the pose exactly as they are — nothing about the look changes.

Replace her face with the face from the second photo. Keep that face completely unchanged: the features, the proportions, the shape of the eyes, the nose and the lips, the skin tone and the skin texture. Do not blend the two faces. Do not average them. The result must read as the woman from the second photo wearing the look from the first.

Match the light and the shadows of the first photo. Photorealistic, not cartoon, not illustration.

Вариант Б

Или собираешь свои образы

Если хочешь свои наряды, а не мои. Тогда нужны пять своих картинок в полный рост на белом фоне — либо сгенерируй себя запросом ниже, либо возьми свои фото и приведи их к такому виду.

Белый фон и полный рост обязательны: модель должна видеть силуэт целиком, иначе додумает обувь и низ по-своему.

Запрос

Take this image. Keep the woman exactly as she is — the same face, the same hair, the same outfit, the same shoes, the same accessories. Change nothing about her appearance.

Place her against a plain pure WHITE background, completely empty, no room, no furniture, no props, no shadows on the background.

Show her FULL BODY, head to toe, standing upright, feet visible, in a simple relaxed three-quarter pose, centred in the frame with a little space above her head and below her feet.

Even soft studio light from the front. Photorealistic. Vertical 9:16, exactly 720x1280 pixels.

No text, no watermark, no border, no collage, no second person.

Дальше одинаково

Что бы ты ни выбрала

На выходе — пять картинок ровно 720×1280. Проверь каждую: если хоть одна другого размера, генерация видео не запустится и выдаст ошибку параметров.

Шаг 1

Берёшь моё референсное видео

Оно задаёт всё: движение, повороты, склейки, карточки на фоне. Открывается вместе с моим шаблоном — там же готовые настройки.

Открыть шаблон с видео

Шаг 2

Загружаешь всё в Seedance

Video references: моё видео.
Image references: свои пять образов — строго по порядку, первый образ первым.

Порядок важен: промпт обращается к ним по номерам.

Шаг 3

Вставляешь промпт

Ничего в нём менять не надо — образы модель возьмёт с твоих картинок. Запреты уже внутри, отдельное поле минус-текста заполнять не нужно.

Запрос

Use @video1 as the ABSOLUTE MASTER VIDEO for the entire output.

Use @image1, @image2, @image3, @image4 and @image5 as the exact outfit, styling and appearance references for each corresponding look.

OBJECTIVE:
Replace ONLY the person in @video1 with the woman shown in the reference images. This is a STRICT PERSON-REPLACEMENT edit. Everything in @video1 other than the person must remain unchanged.

REFERENCE MAPPING:
Look / Scene 1 → @image1
Look / Scene 2 → @image2
Look / Scene 3 → @image3
Look / Scene 4 → @image4
Look / Scene 5 → @image5

THE SAME FEMALE IDENTITY must remain consistent through all five looks — the same face, the same facial structure, the same hair colour and length, the same body proportions. Only the clothing, the styling and the colour palette change between looks. Do not generate a different woman between scenes. Do not let her face drift towards the person in @video1.

CRITICAL MOTION RULE:
The original person in @video1 is the ABSOLUTE MOTION MASTER.

Transfer her exact body position, pose, posture, arm position, hand position, leg position, foot placement, head angle, head turns, body turns, walking or stepping motion, weight shifts, movement direction, movement speed, expression timing and orientation toward camera onto the replacement.

Do NOT invent new poses. Do NOT change the choreography. Do NOT make her move differently from the original. The replacement must occupy the same position, scale, depth and perspective as the original person at every frame.

IMPORTANT: do NOT use the static poses from the reference images as the animation. The reference images define ONLY the outfit, the styling and the appearance. @video1 defines ALL motion, pose, animation, camera, timing, framing and transitions.

STRICT VIDEO PRESERVATION. DO NOT change:
- video duration
- pacing, timing, speed, frame rate
- cuts, transitions, shot order
- camera movement, zoom, rotation
- framing, crop, camera angle, camera distance, lens perspective
- subject position and scale
- the colour cards, their text, their codes and their colours
- the arrangement, angle and overlap of the cards
- background graphics and background animation
- the background itself
- shadows belonging to the environment
- composition

Treat the entire background and every colour card as LOCKED, IMMUTABLE elements from the source video. Do not recreate, reinterpret, move, recolour, resize, rotate or rewrite any card or its text.

PERSON REPLACEMENT:
Completely remove the original person and replace her entire visible body with the reference woman. Leave nothing of the original: no original face, no original hair, no original body shape, no original clothing, no original shoes. The replacement is a complete body from head to toe whenever visible. Preserve realistic anatomy while following the source motion exactly.

OUTFIT ACCURACY:

Match each reference outfit closely in colour, material, layering, fit and silhouette. If the source person turns around, show the corresponding realistic back and side views of that outfit while preserving the exact source movement.

AUDIO — GENERATE THE MUSIC:

Generate a modern underground TECHNO track. No vocals, no lyrics, no speech, no existing or copyrighted melody. This must sound like a club record, NOT like commercial or advertising background music.

STYLE: dark, driving, hypnotic techno at 128-134 BPM. A hard punchy four-on-the-floor kick with weight in the low end. Tight rolling hi-hats and off-beat shakers. Industrial metallic percussion — clangs, rim hits, short reversed metal. A deep sub bass line that pulses with the kick. Cold filtered synth stabs, a little detuned, moving through a resonant filter. A thin high arpeggio appearing and disappearing. Heavy reverb tails, tape saturation, a slightly gritty raw texture.

MOOD: confident, cold, stylish, a bit menacing. Berlin club at 2am, not a shopping centre.

THE BEAT LANDS ON THE OUTFIT CHANGES:
- choose the tempo so a strong downbeat falls EXACTLY on every outfit change
- each change is hit by the kick plus a sharp metallic accent or a synth stab
- in the beat before each change, run a short riser, a snare roll or a reversed cymbal swell
- between the changes, keep the groove hypnotic and repetitive so the hits stand out

STRUCTURE:
- open with the filtered kick and hats only, low-passed, tense
- the filter opens and the bass drops in exactly on the first outfit change
- full groove through the middle changes, each one accented
- a short break before the final change — drums drop out, only the riser
- the last change is the hardest hit of the track, everything back at full weight
- the track ends cleanly on the final frame, no fade-out, no cut mid-bar

MIX: loud, punchy and clean, tuned for phone speakers — the kick and the metallic hits must cut through on a small speaker. Tight low end, crisp top, no mud.

FINAL PRIORITY ORDER:
1. Preserve @video1 motion and timing exactly.
2. Preserve @video1 camera and framing exactly.
3. Preserve all colour cards, background and graphics exactly.
4. Replace only the person, keeping one consistent female identity.
5. Match each outfit to its corresponding reference image.
6. Generate the techno track so that its accents land exactly on the outfit changes.

The final result must look exactly as if the original video had been filmed and edited with this woman and these five outfits.

AVOID:
different woman between scenes, face drifting towards the person in the source video, blended faces, changed identity, changed hair colour, new poses, changed choreography, different movement, changed pacing, changed duration, added cuts, camera movement added, zoom, reframing, changed crop, redrawn colour cards, rewritten card text, moved cards, recoloured cards, changed background, changed background graphics, changed shadows, changed composition, static poses copied from the reference images, reference sheet layout, split screen, collage, original person still visible, original clothing, original shoes, original hair, distorted anatomy, extra limbs, commercial advertising music, cheerful pop, ukulele, stock library music, vocals, lyrics, speech, fade-out, text, captions, watermark, cartoon, illustration, beauty filter, waxy skin, low resolution

Шаг 4

Настройки и жмёшь

· Seedance, Standard
· 9:16, 720p
· длительность — ровно как у моего видео, секунда в секунду

Если не получилось

Ошибка параметров

Проверь размеры картинок — все ровно 720×1280 — и длительность: она должна совпадать с референсом.

Карточки перерисовались

Промпт вставлен не полностью — обрезался блок AVOID в самом конце.

Образ стоит в позе с картинки

Модель приняла картинку за указание к движению. Промпт из третьего шага это запрещает — значит он вставлен не полностью.

Или сразу по моему шаблону

Открывается с моим промптом, настройками и референсным видео — остаётся подставить свои картинки и нажать. Ничего копировать и настраивать не нужно.

Открыть мой шаблон

курс neuronelia

Курс «UGC + нейросети»

Делюсь своим опытом: съёмка на телефон, сценарии, монтаж в CapCut и нейросети — всё, чтобы делать ролики для брендов.

старт 1-го числа каждого месяца3 недели практикидоступ на месяцобратная связь от меня

8 готовых роликов в портфолио: нарезка продукта, два ролика на реальный продукт и пять работ на нейросетях
Каждую практику смотрю сама, работы разбираем в прямом эфире
Бартер с реальным брендом — мест ограничено, распределяю сама
Обновлённый профиль, портфолио и письмо бренду

Все подробности о курсе →

Вопросы — пиши @atopesh · neuronelia.com