← 提示词库 在 GitHub 查看源文件 ↗
中文

Worked Example — Vertical Micro-Drama (竖屏短剧)

Original worked example by jnMetaCode (MIT). Applies the 5-stage structure to the vertical micro-drama — the short-form dramatic scene format built for phones (抖音/Reels 短剧). Teaches vertical-first framing, shot-reverse-shot dialogue grammar, and the hook + cliffhanger structure that drives binge-watching.

Concept: a tense two-person confrontation with a cold-open hook and a cliffhanger cut. Swap the conflict, keep the structure. IP-free.

Further reading (inspiration, not copied — all rewritten in our structure): viral/short-form patterns in zhangchenchen/awesome_sora2_prompt (MIT).

中文版 →


Variables you need to define first

VariableThis exampleSwap for…
{{a}}a tired young woman, office wearthe protagonist — describe, no IP
{{b}}an older man, expensive suitthe antagonist / the twist character
{{hook}}she slaps a resignation letter on the deskthe shocking first-2s beat
{{conflict}}"You knew. The whole time."the line that escalates it
{{cliff}}he slides her a photo — she freezesthe unanswered turn that ends the clip

The complete prompt (copy-paste ready)

1 · Core theme

vertical micro-drama | dialogue confrontation | hook-heavy cold open | shot-reverse-shot | grounded performance, no soap-opera overacting

2 · Character & scene

Characters: {{a}} and {{b}}. Reference uploaded photos, features 100% preserved, no beautification. Real, restrained emotion — a tight jaw, a flicker in the eyes, a held breath — not theatrical overacting. Imperfections: tired eyes, a stray hair, real skin, a slightly creased collar. The realism makes the drama land.

Scene: a lived-in office at night (or your setting) — practical desk lamp, papers, a cold city window. Vertical 9:16 framing built around faces.

3 · Atmosphere & quality

Shot on ARRI Alexa with vintage primes, restrained cinematic grade, shallow focus that isolates faces. Composed for vertical (9:16): faces sit in the upper-middle, eye-lines respected for shot-reverse-shot. Film grain, motivated practical light, real shadow.

4 · Camera rules

Shot-reverse-shot dialogue grammar: a clean single on {{a}}, a reverse single on {{b}}, a tighter push as the tension rises. Vertical framing keeps each face large.

5 · Storyboard (3 beats, ~12s — built to hook + cliffhang)

0–2s · Cold open (the hook — survive the swipe)
Action: Straight in, no setup: {{hook}}. The shocking beat lands in the
        first two seconds, before anyone can scroll away.
Camera: A punch-in single on {{a}}, vertical, face large.
Sound:  The sharp diegetic sound of the hook; then dead quiet.

2–8s · Confrontation (shot-reverse-shot)
Action: {{a}}: "{{conflict}}" — cut to {{b}}'s reaction, then back. The
        tension escalates line by line, faces tightening.
Camera: Clean single / reverse single, a slow push on each as it heats up.
Sound:  Dialogue and room tone; maybe one low drone entering.

8–12s · Cliffhanger (the unanswered turn)
Action: {{cliff}} — a reveal or reversal that flips the scene, left
        unresolved.
Camera: Hold on the frozen reaction, push in slightly.
Close:  No resolution, no music sting, no "to be continued" card needed —
        cut on the frozen face mid-reaction. The unanswered question is
        what makes them tap the next episode.

Negative prompt (Seedance / Kling — paste into the dedicated field)

blurry, low resolution, watermark, text overlay, subtitles, logo, theatrical overacting, exaggerated soap-opera expressions, plastic CG skin, glossy idol render, video-game look, 3D cartoon, flat even studio lighting, horizontal letterboxing, faces too small in frame, distorted face, asymmetric eyes, extra fingers, lip-sync mismatch, melting/morphing geometry, swelling melodramatic score, frame flicker, jarring hard cuts, lifeless locked-off camera

Why it's built this way

Usage: generate the cold open (0–2s) first — if the hook lands and the acting is restrained (not soap-opera), build the confrontation and cliffhanger around it. Keep faces large; shoot/think 9:16 throughout. Sanitise any IP-recognisable wording before running.

Model: Veo 3 and Sora 2 are strongest for synced dialogue and grounded faces; Kling 3.0 supports multilingual dialogue. Keep dialogue lines short so lip-sync holds; cut away on the reverse if a take's sync drifts.