video-production（视频制作）

Video Production

Use this skill as the canonical programmable-video and automated-video production anchor for the repository.

The job is not to dump a generic storyboard or act like a manual editor tutorial. The job is to:

1. normalize the production request,
2. choose the best production mode,
3. return one implementation-ready packet,
4. leave explicit asset, QA, and handoff guidance.

Read references/production-modes.md, references/asset-and-qa-checklist.md, and references/handoff-boundaries.md before routing broad requests.

When to use this skill

• The user wants automated, repeatable, or template-driven video production
• The request involves short-form content ops, campaign variants, personalized videos, localized versions, or batched social assets
• The user mentions Remotion, rendering from code, video APIs, templates, or connecting app/product data into video output
• The workflow needs a production plan that can span generation, asset prep, QA, and a publishing handoff

When not to use this skill

• The job is purely a one-off manual edit in Premiere / After Effects / CapCut with no automation or repeatability goal
• The task is only final creative polish, color, taste-based pacing, or bespoke motion-design critique
• The user needs only transcript cleanup, clip selection, or direct publishing with no template/render layer
• The request is really an ad-strategy, content-strategy, or product-launch brief rather than a video-production workflow

Supported production modes

Use these as routing targets inside the skill:

1) Code-first programmable video

• Best when the user explicitly wants Remotion, React-based composition, dynamic data injection, or custom product-integrated rendering
• Default stack: Remotion

2) Template/API automation

• Best when the user needs speed, bulk generation, localization, or personalized variants without owning a full rendering codebase
• Typical comparators: Shotstack, Creatomate, Bannerbear

3) Hybrid clip / repurposing pipeline

• Best when the source material is long-form audio/video and the main problem is extracting, captioning, resizing, and packaging clips at volume
• Common workflow shape: transcript or clip-selection tool + template/render layer + manual QA

4) Manual-finish hybrid

• Best when automated generation exists but editorial polish, captions, timing, or final approvals still need a human pass
• Use this when the right answer is not “fully automate everything” but “automate the repeatable 80%, then define the last-mile edit pass"

Instructions

Step 1: Normalize the production brief

Capture the request in this form before choosing a mode:

video_brief:
  objective: launch | onboarding | social-clips | personalization | recap | education | ads | unknown
  primary_output: mp4 | shorts | reels | stories | multiple
  audience:
    segment: "who will watch this"
    stage: unaware | evaluating | customer | community | mixed | unknown
  source_material:
    type: script | existing-video | screenshots | product-data | slides | mixed | unknown
    assets_available:
      - logo
      - footage
      - screenshots
      - captions
      - brand-guidelines
  scale:
    volume: one-off | recurring | batch | personalized
    variants:
      - language
      - aspect-ratio
      - audience-segment
  constraints:
    stack_preference: remotion | api-platform | mixed | unknown
    timeline: immediate | this-week | this-month | longer
    quality_bar: draft-fast | production-ready | polished-final
  main_question: "what does the user need next?"

If the brief is incomplete, continue with explicit assumptions rather than stalling.

Step 2: Pick exactly one primary production mode

Choose the single mode that reduces ambiguity fastest.

• Code-first programmable video when the user explicitly wants Remotion, full code control, or deep app/data integration
• Template/API automation when speed, scale, or no-code / low-code batch generation matters most
• Hybrid clip / repurposing pipeline when the source is existing long-form media and the main job is repeatable short-form extraction
• Manual-finish hybrid when automation is useful but the real risk sits in last-mile captions, timing, approval, or platform-native polish

If more than one mode fits, choose the mode that best defines the next artifact and name the secondary handoff.

Step 3: Return one implementation-ready packet

Return one of these packet types:

• video-production brief
• render-stack recommendation
• asset + template packet
• repurposing workflow packet
• QA + handoff checklist

Do not emit multiple half-built plans. Choose the single most useful packet for the ask.

Step 4: Add explicit asset, QA, and handoff logic

Every output must include:

• the chosen production mode,
• the recommended stack or workflow shape,
• the minimum required assets,
• the main QA risks,
• the next handoff after generation (manual polish, approval, publishing, analytics, etc.).

Step 5: Handle Remotion explicitly when relevant

If the user names Remotion or clearly needs code-first composition:

• say that the request fits the code-first programmable video mode,
• structure the output around Remotion scenes/templates/components/rendering,
• optionally note that remotion-video-production is the compatibility alias for the same lane.

Step 6: Use this output structure

# Video Production Brief

## Scope
- Objective: ...
- Primary output: ...
- Audience: ...
- Source material: ...
- Confidence: high | medium | low

## Chosen mode
- Code-first programmable video | Template/API automation | Hybrid clip / repurposing pipeline | Manual-finish hybrid

## Recommended workflow
- Stack / workflow shape: ...
- Why this mode fits: ...
- What to avoid: ...

## Minimum asset packet
| Asset | Required | Notes |
|-------|----------|-------|
| ... | yes/no | ... |

## Build plan
1. ...
2. ...
3. ...

## QA risks
- ...
- ...

## Handoffs
- Next owner / tool / workflow: ...
- Why: ...

## Success check
- Render / approval / publishing signal: ...

Output format

Always return a short operator-style video production brief.

Required qualities:

• one primary production mode,
• clear asset minimums,
• explicit QA / last-mile risks,
• concrete next artifact and handoff,
• assumptions called out when context is missing.

Examples

Example 1: personalized campaign videos

Input

We need to generate personalized customer recap videos from product data and render hundreds every week.

Output sketch

• Primary mode: Template/API automation
• Packet: render-stack recommendation
• Notes dynamic data, template ownership, localization/QA, and publishing handoff

Example 2: explicit Remotion ask

Input

Build a Remotion workflow for a branded onboarding video with screenshots, captions, and 16:9 + 9:16 variants.

Output sketch

• Primary mode: Code-first programmable video
• Packet: asset + template packet
• Uses scenes/components/rendering language and mentions variant strategy

Example 3: repurposing long-form content

Input

Turn our webinar recordings into short captioned clips for Shorts and Reels every week.

Output sketch

• Primary mode: Hybrid clip / repurposing pipeline
• Packet: repurposing workflow packet
• Names transcript/clip selection, caption QA, safe-area checks, and final publish handoff

Best practices

1. Separate the scalable generation layer from the last-mile editorial polish layer.
2. Treat captions, safe areas, and aspect-ratio variants as first-class planning items, not afterthoughts.
3. Choose the simplest stack that meets the repeatability requirement.
4. Preserve a reusable asset/template packet so future variants are cheaper.
5. Make the QA checklist explicit whenever short-form/social distribution is involved.

References

• references/production-modes.md
• references/asset-and-qa-checklist.md
• references/handoff-boundaries.md