Extract best moments from video
Automatically create highlight reels by selecting the best moments using AI analysis. Step-by-step.
Turn long footage into a tight highlight reel. The AI analyzes audio energy, visual interest, and selects the best moments automatically.
What you need
- Raw footage on the timeline (10+ minutes for best results)
- Main content on the main track
- About 1-2 minutes for processing
Step 1: Open the Agent panel
- Click "Agent" in the top-right header
- The agent panel opens on the left
Step 2: Request highlight extraction
Type one of these phrases:
With duration:
"Extract a 60-second highlight reel"
With count:
"Extract 5 best moments"
With analysis:
"Sample visual analysis, then extract the best moments"
Combined:
"Extract the best 3 minutes after visual analysis"
Step 3: Wait for processing
The agent will:
- Analyze audio — detect energy peaks (louder, more animated)
- Check visual analysis — use cached best-moment hints (if available)
- Score moments — rank by combined audio + visual interest
- Select winners — pick moments totaling your target duration
- Compact timeline — delete unselected clips, shift remaining together
You'll see status: "Analyzing audio..." → "Scoring moments..." → "Extracting highlights..." → "Done"
Step 4: Review the result
- Play the highlight reel — watch from start to end
- Check the flow — moments should feel connected
- Scrub through — verify pacing feels right
Step 5: Adjust if needed
If key moments missed or pacing off:
Include more from a section
Type:
"Extract more moments from the first half"
"Include more highlights between 2:00 and 5:00"
Focus on reactions
Type:
"Sample visual analysis first, then extract reaction moments"
Change duration
Undo and retry:
"Extract a 90-second highlight reel" (increase from 60s)
"Extract the best 2 minutes" (decrease from 3m)
Prerequisites for best results
Visual analysis (optional but recommended)
Before extracting, run:
"Sample visual context first"
This caches:
- Frame-by-frame scene analysis
- Best-moment labels
- Face/product/screen tags
Used to improve selection quality.
Transcript (optional alternative)
For speech-heavy content, use instead:
"Extract transcript highlights about [topic]"
Uses transcript information density rather than audio energy.
What you get
The timeline after extraction:
- Selected clips moved to the start
- Non-selected clips deleted
- Gaps closed automatically
- Overlays and audio rippled to maintain sync
Result duration:
- Close to your target (±10%)
- May be shorter if clean moments are scarce
- May be longer if many great moments found
Combining with other edits
Complete workflow:
- Import footage
- Sample visual analysis — improve selection
- Extract highlights — get best 60s
- Add captions — make accessible
- Apply template — style for platform
- Export — share
Quick TikTok:
"Apply TikTok template, then extract 45-second highlights"
Tips
- Visual analysis first — run "Sample visual context" before extraction for better results
- Combine with captions — highlights + captions = share-ready clips
- Iterate — try different durations; undo and re-run freely
- Target ±20% — a 60s target might yield 50s or 70s depending on content
- Manual fine-tune — after extraction, drag clip edges to adjust in/out points
Compare: audio vs transcript highlights
| Tool | Best for | How it works | |------|----------|--------------| | Extract best moments | Any video with energy/visuals | Audio energy + visual analysis | | Extract transcript highlights | Speech-heavy content | Information density in spoken words |
Troubleshooting
| Issue | Fix | |-------|-----| | Missed a good moment | Run visual analysis first, or manually extend clip edges after extraction | | Too many cuts | Increase target duration, or ask "extract longer moments" | | Pacing feels rushed | Try longer duration target (e.g., 90s instead of 60s) | | Visual analysis unavailable | Tool works without it; results just less precise | | Wrong clips selected | Retry after running visual analysis for better scene detection |
See also
- Extract transcript highlights — text-based highlight selection
- Sample visual context — improve selection with frame labels
- Make a TikTok — full vertical highlight workflow