Four focus modes
Choose motion, visual prominence, speech or a mixed score according to the story you want.
Private video AI
Auto highlights with chronological review
Choose a 15, 30 or 60 second target and a focus. Local analysis proposes moments; you—not an opaque auto-export—approve the exact ranges that become the final clip.
Result proof
A short reel whose source ranges remain visible
01
Input
Latest local clip
02
Local pass
Scenes + motion + visual prominence + real VAD
03
Result
Reviewed MP4 or alpha WebM
How to verify it: Inspect each thumbnail and editable start/end time, confirm chronological order, then play the rendered result with sound and compare its duration against the selected target.
Editorial review
Let local signals propose the cut, then keep control of every range
Scene boundaries, measured motion, visual prominence and real audio-energy VAD produce a chronological proposal for a 15, 30 or 60 second reel.
±5 seconds
Target-duration guard
No blind auto-export
Every proposed start and end remains visible and editable before rendering.
Matched A/V trims
The accepted video and audio ranges use the same millisecond filter graph to avoid drift.
Know before you process: Automated scores do not understand narrative importance. Review the selected moments, especially in long interviews and quiet scenes.
What people use this for
Event recaps
Prioritize motion and strong scene changes, then remove moments that do not fit the event story.
Interview teasers
Use Speech focus to surface active passages, then choose the quote boundaries yourself.
Product demonstrations
Use Visual prominence or Mixed focus to find high-contrast, centrally structured sections, then review whether they actually show the product.
Social cut-downs
Build a platform-length draft without giving up chronological range control.
Limits worth knowing
- Automated scores cannot understand narrative importance, humour or consent.
- Speech focus uses voice activity, not semantic quote quality or speaker identity.
- Very long or high-resolution videos may exceed browser memory during final encoding.
- A five-second duration tolerance protects scene continuity but means the output is not always exact to the second.
How the local workflow runs
- Choose the latest clip, target duration and focus.
- Sample scenes and calculate motion, visual-prominence and optional speech-energy scores.
- Review thumbnails and numeric source ranges in chronological order.
- Adjust boundaries or remove unwanted moments; no microclips under 1.5 seconds are accepted.
- Render once, then verify picture, audio alignment, duration and transparency.
Frequently asked questions
Does Auto Highlights upload the video?
No. Scene sampling, motion and visual-prominence scoring, audio-energy VAD, range review and final encoding run in the browser.
What do the four focus modes measure?
Motion compares sampled frames, Visual prominence measures luma structure and centre contrast, Speech uses adaptive RMS voice activity, and Mixed combines those three signals. It does not identify a person or understand story meaning.
Can I change the selected moments?
Yes. Proposals are sorted chronologically and expose editable start/end times. Remove any range before rendering.
Will the result be exactly 15, 30 or 60 seconds?
The selector targets that length with an allowed five-second tolerance so a useful scene is not split into an unusable microclip.
How does the tool avoid audio drift?
Accepted video and audio ranges are trimmed with the same millisecond boundaries and concatenated in the same order in one FFmpeg filter graph.
What happens with a silent clip?
Motion, Visual prominence and Mixed remain available. Speech focus reports that no audio or speech was found and leaves the clip unchanged.
Does it preserve transparent video?
Yes. Known-alpha WebM stays alpha-capable WebM. When WebM alpha metadata is unknown, transparency preservation is enabled explicitly by default because an opaque sampled frame cannot rule out sparse or later-frame alpha. Known-opaque clips export to broadly compatible MP4.
Related tools
Auto Subtitles
Create an editable caption track for the source or final reel.
Video Editor
Combine the reel with layers, captions, grading and privacy controls.
Video Compressor
Reduce delivery weight after the editorial cut is approved.
Auto Highlights Help
Read the signal definitions, review workflow and limitations.
Ready to try it?
No model download, no upload, and no blind auto-export.
Find local highlights →