AetherCut™ auto highlights reel
Long-form video to a tight highlight reel. AI picks the moments. Honest about which steps use APIs.
AetherCut's auto-highlights tool takes a long-form video (a podcast, a livestream, an interview, a lecture) and produces a substantially shorter reel containing only the moments most worth keeping. It's the same workflow podcasters use to generate social cutdowns from a 90-minute episode, but the editing decisions happen automatically.
Honest disclosure of which steps use external APIs: the transcript is generated by OpenAI Whisper. The highlight scoring is done by Claude Sonnet 4.5 against that transcript. Everything after that — the actual razor cuts, the timeline assembly, the export — runs locally on your device.
The two-step pipeline
Step 1: Transcription. Your audio is sent to Whisper, which returns a word-level transcript with timestamps. This is the same Whisper call AetherCut's auto-caption tool makes.
Step 2: Highlight selection. The transcript is sent to Sonnet 4.5 with a prompt asking it to identify the highest-signal moments: clear thesis statements, surprising claims, emotional peaks, quotable lines, hook-worthy openers. Sonnet returns timestamps. AetherCut then assembles those segments into a new timeline you can review, re-order, and trim.
Neither your video frames nor your project state ever leaves your device. Only the audio (for Whisper) and the resulting text transcript (for Sonnet) are sent to external APIs.
How long the reel ends up
You set a target length. Common picks: 90 seconds for TikTok/Reels, 3–5 minutes for a YouTube highlights cut, 10 minutes for a long-form podcast cutdown. AetherCut works within ±15% of your target.
If you want more control, the AI returns a ranked list of candidate moments. You can manually pick which ones make the cut rather than accepting the auto-assembly.
Privacy Mode behavior
With Privacy Mode on, auto-highlights is disabled (because it depends on Whisper + Sonnet, both API calls). Manual highlight selection via the on-device Scene Detection tool — which uses frame-difference analysis with no API call — continues to work and produces decent results for clip-based highlight reels even if it can't match transcript-aware selection.
Frequently asked questions
What's actually sent to the APIs?
Only the audio of your clip (to Whisper for transcription) and the resulting text transcript (to Sonnet for highlight selection). Your video frames, your other clips, and your project state never leave your device.
Can I edit the AI's choices?
Yes. The auto-assembled timeline is fully editable — re-order segments, trim them, add transitions, mix in B-roll. The AI's pick is a starting point, not a final cut.
How does this compare to Descript's Underlord or Opus Clip?
Architecturally similar — transcript + LLM scoring. The difference is AetherCut runs everything except the two API calls on your device. Descript and Opus Clip upload your full video file to their servers; AetherCut sends only the audio for transcription and the text transcript for scoring.
Try AetherCut now — no signup required.
Open the editor and verify the no-upload claim yourself in Chrome DevTools.
Free tier · Pro $14.95/mo · $129.88/yr · Lifetime Pro one-time
Other comparisons + resources
Related searches
AetherCut answers for all of these search intents — pick the one closest to what you're looking for:
No signup. No upload. Verify privacy in DevTools in 30 seconds.