undefined

**Requirements:**
- macOS (or Linux/Windows with `-hwaccel videotoolbox` removed from SKILL.md)
- ffmpeg with libx264
- Python 3 with numpy
- Whisper (openai-whisper)

Restart Claude Code after installation. The `/clipify` slash command will be available.

/clipify

/path/to/long-interview.mp4

Clip 1: "The worst product advice" (02:34 - 02:51)
  Reason: Reversal after awkward pause

Clip 2: "We burned $2M on this" (08:12 - 08:29)
  Reason: Audio peak + punchline

Clip 3: "My co-founder quit on Zoom" (15:03 - 15:24)
  Reason: Punchline

<source-video-dir>/clipify_out/clip_<timestamp>.mp4

undefined

**Output format (timeline.txt):**

undefined

undefined

**Arguments:**
- `--left-x`: X coordinate of left speaker's face center
- `--right-x`: X coordinate of right speaker's face center
- `--width`: Width of the 9:16 crop window (e.g., 608 for 1080p)

**Output:** ffmpeg expression string like:

undefined

undefined

**Whisper JSON format (input):**
```json
{
  "segments": [
    {
      "start": 0.5,
      "end": 2.3,
      "text": "This is the worst advice",
      "words": [
        {"word": "This", "start": 0.5, "end": 0.7},
        {"word": "is", "start": 0.7, "end": 0.85},
        {"word": "the", "start": 0.85, "end": 1.0},
        {"word": "worst", "start": 1.0, "end": 1.4},
        {"word": "advice", "start": 1.4, "end": 2.3}
      ]
    }
  ]
}

undefined

Uses audio cross-correlation. Useful for re-linking edited clips to source timestamps.

undefined

undefined

undefined

undefined

[V4+ Styles]
Format: Name, Fontname, Fontsize, PrimaryColour, SecondaryColour, OutlineColour, BackColour, Bold, Italic, Underline, StrikeOut, ScaleX, ScaleY, Spacing, Angle, BorderStyle, Outline, Shadow, Alignment, MarginL, MarginR, MarginV, Encoding
Style: Default,Impact,68,&H00FFFFFF,&H0000FFFF,&H00000000,&H00000000,-1,0,0,0,100,100,0,0,1,3,0,2,10,10,120,1

[Events]
Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
Dialogue: 0,0:00:00.50,0:00:02.30,Default,,0,0,0,,{\k70}This {\k15}is {\k15}the {\k40}worst {\k90}advice

import subprocess
import json

# Generate captions
subprocess.run([
    "python", "scripts/build_ass.py", "source.json",
    "--start", str(clip["start"]),
    "--end", str(clip["end"]),
    "--style", "opus"
], stdout=open(f"{clip['title']}.ass", "w"))

# Reframe and burn captions
subprocess.run([
    "ffmpeg", "-i", f"raw_{clip['title']}.mp4",
    "-vf", f"crop=608:1080:656:0,ass={clip['title']}.ass",
    f"{clip['title']}_final.mp4"
])

undefined

ffmpeg -hwaccel videotoolbox -i input.mp4 ...

ffmpeg -hwaccel cuda -hwaccel_output_format cuda -i input.mp4 ...

ffmpeg -hwaccel dxva2 -i input.mp4 ...

whisper video.mp4 --model tiny  # ~1GB, 10x faster

whisper video.mp4 --model medium  # ~1.5GB, 2x slower
whisper video.mp4 --model large   # ~3GB, 4x slower

ffmpeg -i input.mp4 -c:v libx264 -preset slow -crf 18 -c:a aac -b:a 192k output.mp4

ffmpeg -i input.mp4 -c:v libx264 -preset veryfast -crf 23 -c:a aac -b:a 128k output.mp4

ffmpeg -i input.mp4 -c:v libx264 -preset medium -crf 20 -c:a aac -b:a 160k output.mp4

undefined

Red = left face, blue = right face. Adjust coordinates until boxes frame each person's mouth/chin.

undefined

Or enable Whisper's word-level timestamps:
```bash
whisper video.mp4 --model base --word_timestamps True

undefined

Or add minimum duration between cuts in `build_pan.py`:

```python

undefined

which ffmpeg

undefined

undefined

undefined

undefined

Or use software decode explicitly:
```bash
ffmpeg -hwaccel none -i input.mp4 ...

undefined

undefined

undefined

for i, clip in enumerate(clips, 1):
    start_tc = frames_to_tc(int(clip["start"] * fps), fps)
    end_tc = frames_to_tc(int(clip["end"] * fps), fps)
    
    edl.append(f"{i:03d}  AX       V     C        {start_tc} {end_tc} 00:00:00:00 {end_tc}")

return "\n".join(edl)

undefined

Dialogue: 0,0:00:00.50,0:00:02.30,Default,,0,0,0,,{\fad(200,200)\move(640,1000,640,900)}This is animated text