Tiny chefs, big appetites
Where did this series come from?
It started with a question I couldn't shake: what if food videos weren't about the food — but about the people making it? Not real people. Tiny ones. In hanfu. With a chicken bigger than their heads.
The whole concept of Miniature Hanfu Kitchen came from wanting to combine two things I never get tired of watching: cute miniature worlds and cooking videos. There's something about seeing effort at that scale — figures no taller than a chopstick, bustling around a kitchen, cooking food bigger than themselves. The drama of cooking, compressed into under a minute and blown up to absurd proportions.
The tool behind all of this is Dreamina — Bytedance's AI video generator. Each episode is generated as short clips, then stitched and paced in CapCut.
The food choices? They're just things I love eating.
這個系列怎麼來的
起因很簡單——我超愛看可愛的微縮料理的影片，也一直很喜歡古風的氛圍。某天突然想：如果讓穿漢服的小人來做菜呢？不是真人，是那種 比筷子還矮的迷你人物 ，在廚房裡忙進忙出，做著比人還大的食物。
用的工具是 Dreamina ，Bytedance 的 AI 影片生成器。每一集都是用 Dreamina 生成短片段，再用 CapCut 剪輯節奏。
至於為什麼選這些菜？因為都是我自己愛吃的啊。宮保雞丁、麻婆豆腐、牛肉麵、肉夾饃——全部是我的口袋名單。
Four dishes, four moods
Kung Pao Chicken — the very first episode, and a short one. The opening scene has tiny figures literally chasing a giant chicken across the courtyard. It set the tone for the whole series: not serious cooking, but joyful chaos. From the chase to the wok to the communal table.
Mapo Tofu — the one where the kitchen feeling clicked. The first episode was all outdoor courtyard action. With Mapo Tofu, I wanted the intimacy of a proper kitchen — the fire, the steam, the close-up of chili oil hitting the wok. This is where the "ancient kitchen" visual language really locked in.
Beef Noodle Soup — steam is the whole personality of this one. I wanted the viewer to almost smell it. The moment the bowls hit the table, with steam curling up past the tiny figures' faces — that's the shot I kept chasing through re-rolls.
Rou Jia Mo — my favorite episode. There's something about dough being flattened, bread being toasted golden, and meat being stuffed inside. The tiny figures pressing down on the bun with both hands, their whole body weight behind it — it looks both heroic and hilarious.
四道菜，四種畫面
宮保雞丁 是第一集，也是短的一集。開場直接讓小人追一隻比他們還大的雞，滿院子跑。這集定下了整個系列的基調：不是認真在做菜，是一群小人在 歡樂地搞事 。追雞、下鍋、圍坐吃飯。
麻婆豆腐 是我開始加上「廚房感」。第一集都在戶外庭院拍，到了麻婆豆腐我想把場景拉進室內——火光、蒸氣、辣油下鍋的特寫。古風廚房的視覺風格，是在這集才真正定下來的。
牛肉麵 的主角是蒸氣。我就是想讓看的人覺得「好像聞得到」。碗端上桌、熱氣從麵裡冒出來、飄過小人的臉——這個畫面我重跑了好幾次才抓到對的感覺。
肉夾饃 是我最愛的一集。麵團被壓扁、饃烤到金黃、肉塞進去的那個過程，本身就很療癒。小人用全身的力氣壓饃，兩隻手使勁往下推——畫面又英勇又好笑。
The Dreamina workflow
Every episode follows the same loop: prompt → generate → pick the best clips → stitch in CapCut → add sound.
The hardest part isn't generating pretty scenes — Dreamina is good at that. The hard part is consistency. These tiny figures need to look like the same characters across multiple clips. Same hanfu style, same kitchen set, same proportions. Dreamina doesn't have character memory between generations, so every clip is a fresh roll of the dice.
My workaround: keep the prompts extremely specific about costume details and kitchen elements, and be ready to generate 5-10 versions of each scene to find one that matches the rest. It's like casting the same actor for every shot, except the actor changes face every take.
CapCut handles the rest — pacing, transitions, and the sound design that makes a sub-minute video feel complete.
Dreamina 的工作流程
每一集的流程都一樣：寫 prompt → 生成圖片與動畫 → 挑最好的片段 → CapCut 剪輯 → 加音效。
最難的不是生出漂亮的畫面——Dreamina 這方面蠻厲害的。最難的是 一致性 。這些小人在不同片段裡要看起來像同一組角色：同一款漢服、同一間廚房、同一個比例。但 Dreamina 每次生成都是獨立的，沒有角色記憶，所以每個鏡頭都是重新抽卡。
我的做法是：prompt 寫得超級具體，衣服的顏色、廚房的擺設、光線的方向全部寫死，然後每個場景準備跑 5-10 次，從裡面挑一個跟其他畫面最搭的。就像每個鏡頭都要重新試鏡，偏偏演員每次都換臉。
剩下的交給 CapCut——節奏、轉場、音效。1分鐘內的影片，每一秒的節奏都要對。
Tools & Process | 創作工具
- Video Generation: Dreamina (Bytedance)
- Editing: CapCut
- Workflow: Prompt → Generate (5-10 versions per scene) → Best pick → CapCut assembly → Sound design
- Episode Length: Under 1 minute each
— TingYu's AI Art · Miniature Hanfu Kitchen · 2025 · Taiwan
