raw-short-video-editor

name

raw-short-video-editor

description

Edit short videos from raw filmed clips — auto-transcribe with local Whisper, auto-cut silence/gaps (jump cuts), add TikTok-style captions (bold stroke text, word-by-word highlight), zoom effects, transitions, BGM, and sound effects using Remotion. Input is multiple raw video files in order. Output is a viral-ready 9:16 short video. Different visual style from HeyGen pipeline — uses TikTokCreator composition with stroke captions instead of backdrop-blur boxes. USE WHEN user says 'edit video thô', 'ghép video quay', 'auto cut video', 'tạo short từ clip', 'edit raw clips', 'cắt ghép video', 'video editor từ clip thô', 'chỉnh sửa video thô', 'tạo tiktok từ video quay', 'auto cut silence', 'jump cut video', 'ghép clip quay tay', 'edit clip thô thành video', 'raw video editor', 'cắt im lặng video'.

Raw Short Video Editor

Edit multiple raw filmed clips into a polished, viral-ready short video (9:16) with auto-silence-removal, TikTok-style captions, zoom effects, transitions, and music.

How this differs from HeyGen short video pipeline

This skill is for real footage you filmed (phone, screen recording, camera). No AI avatars, no AI-generated visuals. Key differences:

	HeyGen Pipeline	Raw Clip Editor
Source	AI avatar + AI visuals	Your real filmed clips
Audio	Separate MP3 voiceover	Audio from the clips themselves
Captions	Backdrop-blur boxes, 38px	Bold stroke text, 52px, no box
Silence	N/A (scripted audio)	Auto-detected and removed (jump cuts)
Composition	HeyGenShort	TikTokCreator
Feel	Polished AI presenter	Raw, energetic TikTok creator

Prerequisites

• whisper CLI — already installed at /Library/Frameworks/Python.framework/Versions/3.13/bin/whisper
• ffmpeg + ffprobe — at /opt/homebrew/bin/
• Remotion Studio — workspace/video-projects/remotion-studio/

Workflow

User provides ordered raw clips
  ↓
[Step 1] Analyze clips (ffprobe)
  ↓
[Step 2] Detect silence → trim → export segments (auto jump cuts)
  ↓
[Step 3] Transcribe trimmed audio (Whisper local, Vietnamese)
  ↓
[Step 4] Correct transcript against user notes (if provided)
  ↓
[Step 5] Plan effects — captions, zoom, transitions, BGM, SFX
  ↓
[Step 6] Build Remotion props → stage assets → preview
  ↓
[Step 7] User approves → render final MP4

Step 1: Collect Input

Ask user for:

1. Video files — paths to raw clips, in playback order
2. Script/notes (optional) — to correct Whisper transcription errors
3. Silence threshold (optional, default: -30dB, min duration 0.5s)
4. BGM preference (optional — auto-select if not specified)

Step 2: Process Raw Clips

Run the processing script:

python3 .claude/skills/raw-short-video-editor/scripts/process_raw_clips.py \
  --clips clip1.mp4 clip2.mp4 clip3.mp4 \
  --output-dir workspace/content/YYYY-MM-DD/video-short/SLUG/raw-edit/ \
  --silence-threshold -30 \
  --silence-min-duration 0.5 \
  --language vi

The script:

1. Probes each clip (duration, resolution, has_audio)
2. Detects silence gaps using ffmpeg silencedetect
3. Builds cut list — keeps non-silent segments
4. Trims clips with ffmpeg (stream copy, re-encode only if needed)
5. Concatenates trimmed audio
6. Runs Whisper with word-level timestamps
7. Outputs structured JSON files

Output structure:

output-dir/
├── clip_info.json        # Original clip metadata
├── silence_gaps.json     # Detected silence regions
├── timeline.json         # Trimmed segments with timing
├── trimmed/              # Trimmed video segments
│   ├── seg_000.mp4
│   ├── seg_001.mp4
│   └── ...
├── trimmed_audio.wav     # Combined audio for transcription
├── transcript.srt        # Whisper SRT output
└── segments.json         # Word-level timestamps

Step 3: Correct Transcript

Read transcript.srt. Whisper often mangles Vietnamese and English terms. If user provided script/notes:

1. Compare SRT text against original script
2. Replace wrong text with correct text
3. Keep all timestamps unchanged — Whisper timing is accurate, text is not
4. Save corrected SRT back

Step 4: Plan Effects (TikTok Creator Style)

Analyze the timeline and apply these TikTok creator patterns:

Caption Style

• Position: Center of screen (captionPosition: 45)
• Font: 52px bold with black text-stroke (no backdrop-blur box)
• Effect: Primarily word-by-word with bright yellow highlights
• Hook (first 3-5s): slam effect, larger text
• CTA (last segment): typewriter effect

Zoom Effects

• hookBoostSec: 5 — auto zoom boost in first 5 seconds
• Add punch zoom pulse at each scene transition
• Add slow zoom pulse on key moments (insights, reveals)
• Ken Burns drift is built-in (1.0→1.03)

Transitions

• Prefer fast, energetic transitions: whip-pan, zoom-blur, swipe-left
• Duration: 8 frames (~0.27s) — faster than default 12
• Every transition gets a zoom pulse

Sound Effects (max 4-5 total)

Select from workspace/assets/reels/MEME SOUND/:

• Transition: Whoosh sound effect on scene changes
• Hook: Sound Effect - Camera shutter or SUDDEN SUSPENSE
• Positive moment: The Price is Right Ding or tada
• Dramatic reveal: DUN DUN DUNNN
• CTA: ting

Background Music

Auto-select from workspace/assets/reels/bgm/:

Track	Energy	Best for
chase-velocity.mp3	High	Fast demos, exciting reveals
rising-star.mp3	Medium-high	Success stories, achievements
Fortitude.mp3	Medium	Tutorials, motivational
miracle.mp3	Emotional	Storytelling, transformations
aluminum.mp3	Chill	Casual tips, calm content

Settings: volume 0.08-0.12, fade in 1s, fade out 2s.

Step 5: Build Remotion Props

Build props JSON for the TikTokCreator composition.

The trimmed clip segments become clips[], captions from SRT become captions[].

{
  "clips": [
    { "videoPath": "media/raw-edit/seg_000.mp4", "durationSeconds": 3.2 },
    { "videoPath": "media/raw-edit/seg_001.mp4", "durationSeconds": 2.8 }
  ],
  "captions": [
    {
      "text": "Bạn có biết rằng AI có thể làm điều này",
      "startSec": 0.0,
      "endSec": 3.2,
      "words": [
        { "word": "Bạn", "start": 0.0, "end": 0.3 },
               
      
       
       
       
       
    
  
   
   
   
   
   
   
        
   
           
  
   
           
  
   
               
  
   
   
     
     
     
     
     
  

Step 6: Stage & Preview

1. Copy trimmed clips to workspace/video-projects/remotion-studio/public/media/raw-edit/
2. Ensure BGM/SFX files are in public/media/bgm/ and public/media/sfx/
3. Save props to workspace/video-projects/remotion-studio/props/tiktok-creator.json
4. Start Remotion Studio:

cd workspace/video-projects/remotion-studio && npx remotion studio

5. Tell user: Preview the TikTokCreator composition. Check captions, timing, effects.

Step 7: Render

After user approves:

cd workspace/video-projects/remotion-studio && npx remotion render TikTokCreator \
  --props=props/tiktok-creator.json \
  out/raw-clip-final.mp4

Output: workspace/video-projects/remotion-studio/out/raw-clip-final.mp4

Quick Reference: Caption Segment Styles

Style	When	TextEffect	Highlight Color
hook	First 3-5s	slam	#FF4444 (red)
pain	Problem/conflict	flicker	#FF6B35 (orange)
normal	Regular content	word-by-word	#FFE600 (yellow)
solution	Resolution/insight	deep-glow	#00FF88 (green)
cta	Last segment	typewriter	#A78BFA (purple)