Video Post-Production

Use this skill when the user gives you one raw video and wants a finished
short video with:

• - readable subtitles
• highlighted key phrases
• basic SFX emphasis
• optional BGM
• a rendered export

The default workflow is single-input automation: the user provides the
video, and you handle the rest.

Inputs

Required:

• - one raw video file with spoken audio

Optional:

• - preferred subtitle color/style
• preferred BGM mood
• preferred SFX intensity
• existing BGM file
• existing SFX directory

If optional inputs are missing, proceed with defaults.

Outputs

Create a working directory next to the input video:

• - INLINECODE0

Expected output files:

• - INLINECODE1
• INLINECODE2
• INLINECODE3
• INLINECODE4

Default behavior

If the user does not specify a style:

• - subtitles: bold, high-contrast, mobile-readable
• highlights: only on high-value phrases
• SFX: light to medium density
• BGM: optional, low-volume, ducked under speech

Do not block on missing BGM or SFX. A valid delivery can still be:

• - subtitles only
• subtitles + generated simple SFX
• subtitles + BGM

Workflow

1. Check prerequisites

Confirm these are available:

• - INLINECODE5
• INLINECODE6
• INLINECODE7
• INLINECODE8 if transcription is needed
• a Chinese-capable font for ASS rendering

Use:

CODEBLOCK0

If faster-whisper is missing:

CODEBLOCK1

2. Transcribe the video

Run:

CODEBLOCK2

This produces word-level timing and segment-level timing.

3. Build a post-production plan

Read alignment.json and produce production_plan.json.

The plan must contain:

• - INLINECODE12
• INLINECODE13
• INLINECODE14

Minimum shape:

CODEBLOCK3

Planning rules

Subtitle grouping:

• - group into natural spoken phrases
• do not split a phrase awkwardly
• prefer 3-8 Chinese characters per phrase
• prefer 0.5s or longer readable display windows
• keep within mobile reading comfort

Keyword highlighting:

• - highlight emotional, contrastive, numeric, warning, CTA, and payoff phrases
• do not over-highlight
• default target: 1-3 highlighted phrases per 10 seconds

SFX placement:

• - opening transitions
• key claims
• warnings
• CTA moments
• major topic pivots

Do not place SFX on every subtitle. Keep it sparse enough to feel intentional.

BGM choice:

• - educational / explainer -> uplifting / neutral
• warning / risk -> tense / restrained
• positive / benefit-driven -> hopeful / energetic

4. Generate ASS subtitles

Run:

CODEBLOCK4

If you need deeper ASS styling guidance, read:

• - INLINECODE15

5. Prepare audio

BGM

Preferred order:

1. 1. user-provided BGM
2. existing local BGM in workspace
3. proceed without BGM

If BGM exists, prepare it:

CODEBLOCK5

SFX

Preferred order:

1. 1. user-provided SFX directory
2. existing local SFX directory
3. generate simple fallback tones/noise

If no SFX files are available, read:

• - INLINECODE16

6. Render final video

Run:

CODEBLOCK6

If no BGM exists, omit --bgm.
If no SFX directory exists, omit --sfx-dir.

Quality bar

The result should feel like a polished short-form knowledge video:

• - subtitles are easy to read on mobile
• highlighted words are meaningful, not noisy
• SFX emphasize real rhetorical beats
• BGM stays below speech and never masks intelligibility
• final export is usable without extra manual cleanup

Fallback rules

If the raw video is usable but supporting resources are missing:

• - still deliver subtitles and a rendered video
• note what was skipped
• do not block the whole workflow waiting for perfect assets

If transcription quality is poor:

• - keep subtitle grouping conservative
• prefer fewer highlights
• avoid aggressive SFX timing

Bundled resources

Scripts:

• - INLINECODE19
• INLINECODE20
• INLINECODE21

References:

• - INLINECODE22
• INLINECODE23

Final delivery checklist

Before finishing, ensure:

• - alignment.json exists
• INLINECODE25 exists
• INLINECODE26 exists
• final video renders successfully
• subtitles are visible
• speech remains clear
• SFX and BGM are not overpowering