MOMENT FINDER

Hours of footage. Seconds of
brilliance.

Six AI signals converge on every half-second of your video. LLM intelligence, audio energy, speech patterns, visual cuts, structural cues, and dramatic pauses — fused into a single score that surfaces your best moments.

Product demo video
SIX SIGNALS, ONE SCORE

Every signal reinforces the others

Each signal is normalized per-video. A loud moment in a loud video scores differently than the same loudness in a quiet video. Context is everything.

30%
LLM highlight

The primary brain

Mistral scores batches of 15 transcript segments. Returns a 1-10 rating, rich description, and category suggestion for each. Not a classifier — a critic.

25%
Audio energy

Relative loudness

RMS energy extracted every 100ms. Per-video z-score normalization means the same dB level reads differently in a podcast vs. a concert recording.

20%
Speech rate

Velocity of words

Words-per-second compared against video mean. Sudden acceleration signals excitement. Deceleration signals emphasis or gravity.

15%
Structural cues

Hook phrases

Detects "here's the thing," "what people don't realize," "the real question is" — structural markers that signal high-value content is coming.

10%
Visual change

Scene boundaries

FFmpeg histogram analysis detects shot transitions and scene cuts. A new visual context often marks a new moment of interest.

+10%
Pause proximity

Dramatic silence

Moments near 1.5-6 second pauses get a proximity bonus. The silence before a revelation carries as much weight as the words after it.

Signal fusion at work
Semantic vector search

Search with meaning

Type “emotional turning points” and Moment Finder embeds your query into a 1024-dimensional vector, then compares it against every transcript segment. Results ranked by semantic closeness, not keyword matching.

talking about the economy
0.825
"The economy's going way down, that's what people don't realize"
2:34 — key_moment
0.754
"Out of Syria, they're militants, and the cost keeps rising"
5:12 — topic_shift
0.670
"I've been shocked as I go through it, the numbers are staggering"
8:45 — emotion_peak
Zero manual work

Upload. Wait. Discover.

Moment Finder runs automatically after transcription completes. By the time you open the editor, hundreds of moments are already catalogued, scored, and searchable. No buttons to click.

REAL NUMBERS

From a single 17-minute video

317
Moments found
9
Categories active
78%
LLM-scored
<5m
Analysis time
“We shot four hours of interviews. Moment Finder surfaced the 90 seconds we needed for the trailer — clips I would have missed manually.”
Senior Producer, News Division

Find what matters

Let six signals do the watching. You do the creating.

Start for free