Source: the first 20 seconds of 4 high-like videos from Douyin "大师的AI小灶", analyzed on 2026-10-27. Samples see
references/samples.md. This Skill only distills the actions, rhythm, and structure; it does not include the other party's appearance, clothing, scene, or verbatim lines. Do not feed the original videos to the model as reference videos during generation.
One sentence
The main subject of the frame is the evidence (finished product, comparison, prompt, document); the real person appears full-screen only on the line where "a person needs to make a judgment", and retreats once it is said. Hands basically do not enter the frame; emotion relies on eyebrow raises and exaggerated mouth shapes.
Suitable / Not suitable
- Suitable: AI generation effect showcases, counterintuitive knowledge points, product or tool promotion (show the result first)
- Not suitable: topics with no ready-made results or materials to show
- Digital human difficulty:Low. The person's screen time is the shortest, and hands basically do not enter the frame
Camera position and composition
- Seated, head-and-shoulders close-up, camera slightly low, the person takes up a large part of the frame
- Dark background, with a green plant on one side and a lit screen on the other; a personal nameplate on the desk. Describe according to your own scene
- Light-colored top, lavalier mic
- Fixed camera position
Hand strategy
| Code | Action | Used for which type of line | Position | Duration |
|---|---|---|---|---|
| D1 | One hand rises from the bottom edge of the frame to the chest, palm up, fingers relaxed and slightly curved | Empathy or a promise: "If you're also troubled by..." "then congratulations" | Below the chest, just entering the frame | 0.5–1 second |
| D2 | Thumb and the other fingers loosely pinched together, raised near the chin | Show your own results: "These are all things I made with AI" | Chin height, slightly to one side | About 1 second |
| D3 | Both hands covering the ears, frowning with eyes closed, with a TV test pattern circular frame overlaid | Meme-style transition from the pain point to the self-introduction | — | About 0.5 second |
| — | Hands do not enter the frame, expression only | The remaining lines | — | — |
Expression and head
- Raise eyebrows, open eyes wide, exaggerated mouth shapes (the "恭喜你" moment is a surprised face)
- Head leans slightly forward
- Uses small meme-style performances as transitions
Opening structure template
| Time | Visual | Action | Line function |
|---|---|---|---|
| 0–10s | Evidence full-screen (AI failure cases, wall of finished products, prompt input, pile of documents), the person in a small box in the lower-left corner | Only expressions in the small box | Pose a counterintuitive question: "为什么……却……" |
| 10–15s | Real person full-screen insert | D1 + eyebrow-raised surprised face | "If you also... then congratulations, this video will completely..." |
| 15–18s | Cut back to the evidence | — | "The real reason is actually hidden in..." |
| 18–20s | Real person full screen or in a small window | Facial expressions | "First look at a simple example", then enter the main content |
Serialized videos insert in the middle: segment self-introduction + past episode covers fly in → "Today I'm sharing N sets of..." + key point cards fly in.
Please write the lines yourself; only learn the function of the sentences.
Digital human generation suggestions
- Put the result in the first frame, not the person. In a small window, the digital human can use a still frame plus lip-sync, which is the lowest cost
- For full-screen real person, generate only 1–2 segments, each 3–5 seconds, with only one D1
- Order for promotional videos: first show the finished piece made with the product as evidence → the digital human cuts in to state the judgment → cut back to the operation process
- Do the material segments at the editing layer, which exactly covers the seams
Prohibited actions
- Do not raise hands above the chin
- Do not use two-handed gestures
- Do not shake the head, do not sway the body left and right
Output format
## 基础姿态
## 逐句动作
| 时间 | 台词 | 画面(证据+小窗 / 真人全屏) | 手(默认画外 / D1 / D2) | 表情与头部 |
|---|---|---|---|---|
## 禁止动作
## 给视频模型的描述(只写真人全屏段)
## 证据素材清单
Generation description example (original, can be rewritten)
Close-up, the person sits directly facing the camera, head and shoulders occupy most of the frame, the camera is slightly low. Both hands are outside the frame. The camera is fixed. At second 0, the person looks at the camera, eyebrows raised high, eyes wide open, mouth shape exaggerated, as if announcing good news. At second 1, the right hand rises from the bottom edge of the frame to the chest, palm facing up, fingers relaxed, stays for about half a second, then lowers. After that, finish saying this sentence using only facial expressions.