微表情（电影感面部特写） · Open Skill 공개 스킬

微表情（电影感面部特写）

微表情（电影感面部特写）은 Digen의 공개 Open Skill입니다. 열어 보면 이 워크플로가 무엇을 만드는지 알 수 있습니다.

@371133686

Film

0 0 0

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4203025/2026-10-13/kdq27gf2.png

# 微表情（电影感面部特写）

## When to use
Use when the user wants a portrait (9:16) cinematic close-up of a person's face in which the emotion transitions subtly and gradually over ~10 seconds — e.g. from a faint smile to a slowly gathering tear — with film-grade soft window light, shallow depth of field, realistic skin texture, absolutely no cutting/angle change, and a face that stays strictly consistent.

Use it for micro-expression shots: smile → smile fading → eyes slowly welling up → a single tear rolling down; or similar slow restrained emotional turns.

## Steps
1. Clarify the character: gender, age, hair, top/clothing, hair/eye color, setting (e.g. sitting by a window in soft backlight), and the exact emotional arc the user wants. If they gave a full scene like the sample, use it.
2. Generate the character still with `generate_image`:
   - `model='t2i'` (no reference yet), `orientation='portrait'` (9:16), `resolution='1080P'`.
   - Prompt in English describing the locked face and frame: cinematic face close-up, described character, soft window backlight, shallow depth of field, realistic detailed skin texture, calm neutral expression ready to shift, vertical 9:16 composition.
   - `name='角色定妆-<角色简称>'`, `entity_type='characters'`.
3. Generate the ~10s micro-expression video with `generate_video`:
   - Use the character still from step 2 as the source: pass `input_urls=[<该图 asset_id>]`.
   - For maximum facial consistency use `model='ref2v'` with the still as a reference image (portrait, `orientation='portrait'`, `resolution='720P'`, `duration=10`). If ref2v is unavailable, fall back to `model='i2v'` with the still as the start frame (first-frame mode).
   - Prompt in English, slow restrained single-take, no cuts, no camera re-framing (only a slow gentle push-in). Spell out the staged micro-expression arc in order and explicitly forbid mutation: e.g. "single continuous cinematic close-up of the same face, no cuts, slow gentle push-in; she starts with a faint subtle smile, then the smile slowly fades, her eyes gradually well up with tears, lips part slightly, and one tear rolls down leaving a shine trail on the cheek; stays photorealistic, soft window light, shallow depth of field, realistic skin texture; the person's face must stay identical throughout, no morphing, no change of identity."
   - `name='微表情-<空拍名>'`, `entity_type='final_videos'`.
4. On success, reply in natural language with the video (asset appears on canvas). If the face warped or identity changed, honestly say so and offer to re-render (or regenerate the still and retry).

## Prompt rules (mandatory)
- Emotional arc must be staged and delicate: 微笑 → 笑意收敛 → 眼眶湿润 → 泪水滑落 (or the arc the user asked for). Never jump straight to crying.
- Film-grade light: soft window/cinematic light, shallow depth of field, realistic detailed skin texture.
- ~10 seconds, slow restrained pacing, no cut, no camera switch.
- Face must stay consistent — explicitly forbid mutation/identity change in the prompt.

## Notes
- Report failures honestly, never fabricate URLs.
- Video generation does not support NSFW/adult content — decline such requests politely.
- If the user reuses a previously-generated character (asset already on canvas), skip step 2 and pass that existing still's asset_id directly to step 3.