Как создать AI-видео с миниатюрными фигурками

Перевел и дополнил гайд для телеграм канала @creativeprompting – подписывайтесь

Генерация видео с миниатюрными фигурками стала новым трендом в сфере ИИ-контента, что даже KlingAI запускал конкурс на создание сцен с «микро-человечками», которые готовят еду – такие ролики цепляют детализацией и юмором.

Ниже короткий гайд для старта создания подобных видео: примеры промптов, настройки и лайфхаки. Даже новички справятся за 10 минут!

Шаг 1: Генерируем изображение в любой нейросети

Опишите сцену и действие, которое должно происходить на картинке. Далее выбираем любой генератор изображений, например, Krea, Ideogram, Midjourney или другой – и создаем картинку. Иногда может потребоваться несколько попыток, чтобы добиться нужного результата.

Ниже несколько примеров промптов, которые помогут вам создать свои изображения:

1. Tiny pastry chefs in classic white uniforms and toques are carefully decorating a massive, multi-tiered cake. Some are using miniature piping bags to create intricate frosting designs, while others are placing fresh berries, edible flowers, and chocolate decorations. A few are balancing on ladders and scaffolding to reach the top, while others carry trays of delicate sugar ornaments. The cake is beautifully textured with smooth icing, rich layers, and a luxurious finish. The setting is a warm, softly lit pastry kitchen, with scattered baking tools and ingredients adding to the cozy and enchanting atmosphere. Captured as a hyper-realistic photograph with a whimsical and elegant touch.
2. A charming miniature scene where tiny film enthusiasts, no taller than a matchstick, are setting up a vintage movie projector in a cozy, old-fashioned home theater. The projector, a classic 16mm model with a polished metal finish and intricate details, sits on a small wooden table. The tiny crew, dressed in retro outfits like suspenders, bow ties, and vintage dresses, are hard at work: one is carefully threading the film through the projector’s reels, another is adjusting the focus knob, while a third balances on a tiny ladder to align the lens with the makeshift screen—a white sheet hung on the wall. The room is filled with nostalgic charm, with scattered film reels, popcorn buckets, and vintage movie posters adorning the walls. Soft, warm light from a nearby table lamp illuminates the scene, casting a golden glow over the intricate details. The atmosphere is lively yet intimate, with the tiny figures working together in perfect harmony. Captured as a hyper-realistic photograph, the scene evokes a sense of nostalgia and whimsy, blending the meticulous setup of the projector with the enchanting magic of cinema.
3. Tiny chefs, barely the size of a hand, preparing a whole lamb on a massive grill. The chefs are dwarfed by the size of the meat, with some flipping the lamb, others brushing it with spices, all working in a chaotic yet precise manner. The grill is surrounded by miniature cooking tools, large spatulas, and oversized ingredients, making the scene resemble a tiny kitchen in the middle of a giant outdoor cooking space. The chefs are focused, working with intense detail, while the massive lamb sizzles on the grill, smoke billowing up like a giant construction site in the background
4. A whimsical miniature kitchen where tiny chefs, no larger than a finger, are preparing a giant pot of rich, aromatic octopus soup. The chefs, dressed in tiny aprons and chef hats, are bustling around the massive pot, which is suspended over a flickering flame. Some are carefully chopping vegetables with miniature knives, while others are stirring the bubbling broth with oversized wooden spoons. A group of chefs is tenderly handling a large octopus, its tentacles spilling over the edge of the pot, as they season it with herbs and spices. The scene is filled with intricate details: scattered ingredients like garlic, tomatoes, and paprika, tiny bowls of saffron threads, and a rustic wooden table cluttered with cooking utensils. The steam rising from the pot carries the scent of the sea and spices, creating a warm, inviting atmosphere. The setting is a cozy, lantern-lit kitchen with a nautical theme, complete with fishing nets and seashell decorations. Captured as a hyper-realistic photograph, the scene exudes a charming and playful vibe, blending the whimsy of miniature chefs with the rich, savory allure of the dish.

Совет

Чтобы проще и быстрее создавать подобные подсказки, загрузите приведенные примеры выше в любой ИИ-чат (DeepSeek, ChatGPT, Claude и др.) и попросите его сделать похожее описание своей идеи. Или же воспользуйтесь промптом:

I am using an image generator tool to create highly detailed images of miniature scenes. I will be describe an scene, and you are tasked to give a detailed prompt following the structure of the examples provided. I will send the subject in the next message.

Exemples:
1)Tiny pastry chefs in classic white uniforms and toques are carefully decorating a massive, multi-tiered cake. Some are using miniature piping bags to create intricate frosting designs, while others are placing fresh berries, edible flowers, and chocolate decorations. A few are balancing on ladders and scaffolding to reach the top, while others carry trays of delicate sugar ornaments. The cake is beautifully textured with smooth icing, rich layers, and a luxurious finish. The setting is a warm, softly lit pastry kitchen, with scattered baking tools and ingredients adding to the cozy and enchanting atmosphere. Captured as a hyper-realistic photograph with a whimsical and elegant touch.

2)Miniature pastry chefs in classic white uniforms and hats push a cake form into a large red oven. Some are turning it on, while others are balancing on ladders and scaffolding to load the form into the oven. The setting is a warm, softly lit bakery, where scattered baking tools and ingredients create a cozy and charming atmosphere. This is a hyper-realistic video with a whimsical and elegant touch

3) Tiny chefs, barely the size of a hand, preparing a whole lamb on a massive grill. The chefs are dwarfed by the size of the meat, with some flipping the lamb, others brushing it with spices, all working in a chaotic yet precise manner. The grill is surrounded by miniature cooking tools, large spatulas, and oversized ingredients, making the scene resemble a tiny kitchen in the middle of a giant outdoor cooking space. The chefs are focused, working with intense detail, while the massive lamb sizzles on the grill, smoke billowing up like a giant construction site in the background

4) A whimsical miniature kitchen where tiny chefs, no larger than a finger, are preparing a giant pot of rich, aromatic octopus soup. The chefs, dressed in tiny aprons and chef hats, are bustling around the massive pot, which is suspended over a flickering flame. Some are carefully chopping vegetables with miniature knives, while others are stirring the bubbling broth with oversized wooden spoons. A group of chefs is tenderly handling a large octopus, its tentacles spilling over the edge of the pot, as they season it with herbs and spices. The scene is filled with intricate details: scattered ingredients like garlic, tomatoes, and paprika, tiny bowls of saffron threads, and a rustic wooden table cluttered with cooking utensils. The steam rising from the pot carries the scent of the sea and spices, creating a warm, inviting atmosphere. The setting is a cozy, lantern-lit kitchen with a nautical theme, complete with fishing nets and seashell decorations. Captured as a hyper-realistic photograph, the scene exudes a charming and playful vibe, blending the whimsy of miniature chefs with the rich, savory allure of the dish.

Шаг 2: Создаем видео с помощью Image-to-Video модели

Когда вы создали подходящее изображение самое время оживить его. Выберите удобный инструмент, который умеет генерировать видео на основе картинки, например, KlingAI, Hailuoai, RunwayML или локальные варианты вроде Wan.

Для теста я выбрал Wan 2.1 через бесплатный бот Wan 2.1 Discord от Remade, потому что 16 кадров в секунду идеально подходит для покадровой съемки миниатюрных людей и мне было интересно попробовать этот генератор (потребуется VPN)

После выбора инструмента загрузите свою картинку и промпт, который использовали для её создания. Не забудьте немного подкорректировать запрос: замените ключевые слова вроде photograph на video. Осталось немного подождать – и ваше видео готово!

Бонус: локальная альтернатива

Вы также можете настроить Wan 2.1 img2vid локально с помощью ComfyUI либо на своем компьютере, либо через облачные сервисы такие как Runpod, immers.cloud и другие.

Для запуска потребуется от 16 ГБ видеопамяти, а лучше от 24 ГБ. Каждый 5-секундный клип занимает около 10-15 минут генерации.

Если хотите попробовать, вы можете найти рабочий процесс здесь. Также вам понадобятся модели с HuggingFace, которые нужно положить в эти папки:

ComfyUI/models/text_encoders

ComfyUI/models/diffusion_models

ComfyUI/models/vae

Подробнее про локальную установку можно прочитать здесь

Надеюсь, этот гайд вам поможет – удачи в генерациях!