AI Performance Microexpression

Use this skill to write realistic Chinese AI video performance prompts. Treat every line as a relationship action, not as plain dialogue.

Core idea:

personality temperament -> expression attitude -> emotional wave -> voice rhythm -> body action -> visual parameters

The image is the result. Character motivation, relationship position, and emotional structure are the source.

Performance Principles

Write acting prompts as a time-coded emotional progression, not a single emotion label.

Prioritize:

• Who the line is spoken to, and what the speaker wants from that person.
• Personality first: calm, proud, wronged, stubborn, gentle, defensive, playful, cold, fragile, or pretending not to care.
• Hidden emotion before visible emotion.
• Eye movement before large facial movement.
• Breath, mouth corner, throat, eyelids, jaw, shoulders, and hands.
• Voice rhythm as emotional structure, not narration.
• Restrained realism over dramatic shouting.
• Continuity of identity, face shape, hair, clothing, makeup, posture, lighting, and camera position.

Avoid:

• Writing only "sad", "angry", "cute", "crying", or "coquettish" without visible behavior.
• Starting at the emotional peak.
• Big stage gestures, meme faces, exaggerated grimaces, cartoon expressions, or forced cuteness.
• Over-precise lip-sync wording that makes the performance mechanical.
• Random tears unless the user asks for tears.

Always include limits:

• Face must not collapse or deform.
• Mouth must not open unnaturally wide.
• No exaggerated mouth movement.
• No ghosting, face drift, identity drift, or body distortion.
• No extra people, limbs, reflections, or background performers.
• No subtitles or watermark unless requested.

Prompt Building Formula

Before writing the time-code, decide these six layers:

1. Personality temperament: what kind of person is performing?
2. Expression attitude: is this line a request, test, accusation, surrender, flirt, refusal, farewell, or self-protection?
3. Emotional wave: hidden -> slight leak -> pull back -> leak again -> peak -> aftertaste.
4. Voice rhythm: low frequency, high frequency, pause, breath point, stressed word, light word, tail sound.
5. Body action: only small realistic actions such as lowering head, lifting eyes, tilting head, pressing lips, touching sleeve, tightening shoulders, leaning forward slightly.
6. Visual parameters: 9:16, fixed camera, shot size, lighting, background, realism, continuity, and negative limits.

Key question:

Who is this line spoken to?

The important part is not the literal content. Decide:

• Which words should be pressed lower.
• Which words should be lighter.
• Which words cannot sound like a question.
• Which words are the emotional breath point.
• Whether the tail sound is held, dragged, trembling, softened, or dropped.

Voice rules:

• Low frequency = stable, restrained, suppressed, controlled.
• High frequency = slight emotional lift, coquettishness, grievance, urgency.
• Pause = thinking, hesitation, self-control, or something unsaid.
• Breath point = where emotion leaks out.

Standard Realistic Acting Base

Vertical 9:16, realistic human emotional performance video, fixed camera, medium close-up half-body composition, clean background, one person facing camera. Keep the same identity, face shape, facial proportion, hairstyle, clothing, and makeup throughout. Stable camera, clear face/neck/upper-body action, realistic skin texture, real eye movement, real breathing, subtle micro-expressions.

Time-Code Writing Pattern

For 6 seconds:

• 0-1s: hold the first emotional mask; eyes, lips, shoulders, and breath are restrained.
• 1-2s: emotion starts to surface through gaze, eyelids, breath, jaw, or mouth corner.
• 2-3s: body reacts slightly; head angle, weight, hand, or shoulders change.
• 3-5s: line is spoken with designed voice rhythm and emotional color.
• 5-6s: aftertaste; emotion cools, breaks, hardens, retreats, or softens.

For 15 seconds:

• 0-2s: establish posture, gaze, breath, relationship attitude, and hidden state.
• 2-4s: first small shift in eyes/head/mouth; emotion leaks but remains controlled.
• 4-6s: suspicion, pain, anger, tenderness, playfulness, or grievance becomes visible.
• 6-8s: emotion deepens, withdraws, or changes direction.
• 8-10s: self-control, self-mockery, embarrassment, recovery, or softening.
• 10-12s: look back toward light/person/camera; controlled emotional return.
• 12-15s: hold final gaze and aftertaste; no black screen, no transition unless requested.

Emotion Formulas

Before Crying / Truth Confession

Use when someone maintains dignity before admitting a painful truth.

State chain:

hurt -> denial -> excuse -> forced smile -> emotional backlash ->整理体面 -> truth

Key details:

• Red eyes but trying to smile.
• Avoid direct gaze at first.
• Breath trembles or gets stuck.
• Mouth corner holds back emotion.
• Lightly adjust clothes or posture to preserve composure.
• Truth is spoken lower and quieter than the excuse.

Restrained Crying Monologue

Use indoor natural light, fixed medium close-up, one-shot style.

Progression:

• 0-2s: head lowered, eyes down, mouth corner lowered, brows lightly gathered, red eyes holding tears, suppressed breath, throat slightly choked.
• 2-4s: head lifts slightly but gaze remains down, brows tighten, eyes wet, nose lightly twitches, body leans forward.
• 4-6s: short inhale, lips press, first question rises with a crying tone but no shouting.
• 6-8s: eyes red, gaze unsteady, shoulders tense, helplessness grows.
• 8-11s: emotional peak; small forward lean, trembling lips, controlled accusation, tail sound trembles.
• 11-13s: line ends; heavy breath, hollow eyes, fatigue after collapse.
• 13-15s: face settles, body becomes still, hand lightly wipes tears if appropriate; do not overact.

Gentle Concern / Ancient Aesthetic Look-Back

Progression:

• 0-2s: head lowered, eyes down, subtle closed-mouth smile, gentle and quiet.
• 2-4s: smile fades; mouth returns flat; eyes and head slowly lift; gaze turns right; expression becomes attentive.
• 4-6s: keep looking right; light worry and doubt enter eyes; lips slightly part as if wanting to speak; brows gather slightly but do not wrinkle hard.
• 6-8s: gaze sinks, head lowers, smile disappears, calm sadness appears; no tears, no crying.
• 8-10s: close eyes lightly, recover breath, a hidden self-mocking smile appears.
• 10-12s: slowly lift head, eyes face soft light, controlled brightness returns; mouth holds gentle委屈.
• 12-15s: hold front/right gaze, eyes soft and moist, lips move as if unsaid words remain; end in quiet, cool, slightly melancholic emotion.

Girlfriend Coquettish / Playful Request

Use when the line is intimate, soft, slightly wronged, and asking for attention.

State chain:

pretend calm -> small expectation -> soft grievance -> playful request -> waiting for response

Key details:

• Not childish, not cartoon cute.
• Eyes soften before the mouth smiles.
• Head can tilt slightly, but not too much.
• Voice is light and soft; tail sound may drag gently.
• Body leans forward a little or shoulders shrink lightly.
• Hands can touch sleeve, pillow, cup, or table edge.

Example line: "你就陪我一会儿嘛。"

Voice design:

• "你就" should be soft and slightly tentative.
• "陪我" should be warmer and closer.
• "一会儿嘛" can drag lightly with a small tail sound, not nasal or exaggerated.

One-Line Emotion Variations

When generating variations for one line such as "你有病啊" or "滚", keep the same base setup and change only emotional motivation, eyes, mouth, body, voice rhythm, and aftermath.

Angry

Not immediate shouting. Anger is pressed into the eyes first, then pushed upward, then thrown out in the line. Use hard direct gaze, tightened eyelids, clenched lips, steady shoulders, rising pressure. Keep intensity high but realistic.

For "你有病啊":

• "你" can be low and direct.
• "有病" carries the pressure.
• "啊" should not become a comic shout unless requested; it can fall coldly or rise with disbelief.

Laughing in Disbelief

Not big laughter. It is absurdity, disbelief, and light sarcasm. Eyes first show disbelief, mouth corner lifts later, body may lean back slightly, hands can open lightly as if saying "listen to yourself."

Disgusted

Not rage and not comedy. Use a slight recoil, narrowed eyes, minimal mouth movement, a rejecting gaze, and a colder voice.

Coquettish / Playful

Use a softer voice, lighter gaze, small head tilt, tiny smile, and less pressure. Keep it natural, not childish or exaggerated.

Wronged /委屈

Use wet eyes, lowered gaze, lips pressed then loosened, small breath block, quieter delivery, and a restrained after-pause.

Cold Farewell

Emotion is already past the peak. Use stillness, lowered eyelids, minimal mouth movement, low frequency voice, clean pauses, and a final gaze that does not ask for response.

Output Template

Use this structure unless the user asks for another format:

画面基础：
[9:16、镜头、人物、环境、光线、真实感、身份连续性]

台词：
"..."

人格气质：
[她/他平时如何控制情绪]

表达态度：
[这句话是说给谁听的，真实目的是什么]

声音结构：
[哪些字压低，哪些字放轻，哪里停顿，哪里吸气，尾音如何处理]

时间轴：
0-1s：...
1-2s：...
2-3s：...
3-5s：...
5-6s：...

负面限制：
[不脸崩、不夸张、不卡通、不大喊、不乱哭、不多人物、不字幕、不水印]

Output Checklist

Before finalizing, confirm:

• Time ranges add up to requested duration.
• The line has a clear listener and relationship purpose.
• Personality temperament appears before facial details.
• Every emotional beat changes something visible or audible.
• Eye path is specified: where to look, when to avoid, when to look back.
• Mouth details include mouth corner, lip line, opening size, and tail sound.
• Voice instructions specify stress, softness, breath, pitch, pauses, and emotional breath points.
• Body actions are small enough for realistic AI video.
• Negative prompt blocks face collapse, overacting, extra people, subtitles, and watermark.