Export the most important moments from any transcription as MP4 clips. Two modes:
- Auto mode: AI picks the top moments by content density and segment structure
- Focused mode: you describe what to highlight (a topic, person, product, concept) and the tool finds every relevant moment using semantic search
Optionally export each highlight as an MP4 video clip via clip_export.
Example: Auto mode
Pick the 5 best moments from a podcast — no query needed.
Request:
{
"audio_path": "/Users/you/Downloads/podcast.mp4",
"top_k": 5
}
Response:
{
"mode": "auto",
"highlight_count": 5,
"highlights": [
{
"start": 342.1,
"end": 358.1,
"timestamp": "5:42",
"text": "The moment we realized our pricing was wrong changed everything...",
"score": 0.2508,
"mode": "auto",
"chapter_number": 12,
"duration": 16.0
}
]
}
Example: Focused mode
Find every moment where product recommendations are made.
Request:
{
"audio_path": "/Users/you/Downloads/podcast.mp4",
"query": "product recommendations"
}
Response:
{
"mode": "focused",
"query": "product recommendations",
"highlight_count": 5,
"highlights": [
{
"start": 638.07,
"end": 642.95,
"timestamp": "10:38",
"text": "...the one supplement I actually recommend to everyone is...",
"score": 0.721,
"mode": "focused"
}
]
}
Example: Export as clips
Add clip: true to export each highlight as an MP4 video clip. Requires the audio to have been downloaded from a URL.
{
"audio_path": "/Users/you/Downloads/podcast.mp4",
"query": "funniest moments",
"top_k": 3,
"clip": true,
"clip_padding": 10
}
Each clip is exported to ~/Desktop as a separate MP4 file.
Parameters
| Parameter | Required | Default | Description |
|---|
audio_path | Yes | — | Path to the audio/video file (must be transcribed already) |
query | No | — | What to highlight. Omit for auto mode. Provide a topic, person, or concept for focused mode |
top_k | No | 5 | Number of highlights to return |
model_size | No | tiny | Whisper model size for transcription |
clip | No | false | Export each highlight as an MP4 video clip |
clip_padding | No | 15 | Seconds of padding around each highlight when exporting clips |
context_words | No | 40 | Words of context around each highlight in focused mode |
Focused mode ideas
The focused mode is where this tool shines. Some examples:
"product recommendations": every time a product is recommended
"heated debate": the most contentious moments
"life advice": personal wisdom and life lessons
"technical explanation": deep technical breakdowns
"funny moments": humor and jokes
"[person name]": every mention of a specific person
"pricing strategy": business strategy discussions
"controversial take": hot takes and unpopular opinions
Notes
Auto mode uses chapter detection to find topic boundaries, then ranks by content density. Works best on longer audio (10+ minutes).
Focused mode uses semantic search — it matches by meaning, not keywords. “pricing strategy” will find discussions about cost, revenue, and margins even if those exact words aren’t used.
Combine with clip: true to instantly create social media-ready clips from long podcasts or streams.
YouTube links are automatically added to each highlight when the source is a YouTube video.