Clipchamp to CLI

One sentence beat a cloud upload

The Problem

A great demo recording, padded with dead air

A 2:59 screen recording of an Amplifier demo — good content, but stretched by long idle stretches where the AI sat thinking. The boring parts just needed speeding up.

Ordinary problem. The result was anything but ordinary.

2:59

Input duration

178.7 seconds

99.5 MB

Input file size

The Result

One sentence made it 66% shorter and 81% smaller

A single request to Amplifier turned the recording from 2:59 into 1:01 and from 99.5 MB into 19.2 MB — every demo frame intact.

The obvious tool would have started somewhere very different.

1:01

Output duration

from 2:59 · −66%

19.2 MB

Output file size

from 99.5 MB · −81%

The Obvious Tool

Clipchamp won't even open a local file

A GUI editor like Clipchamp is built for creative editing. Step one isn't editing — it's uploading all 99.5 MB to OneDrive before you can start, with no local-file support.

But this was never a creative-editing problem. It was algorithmic.

Requires uploading the file to OneDrive before editing
No local-file support in the GUI editor
Amplifier operates locally via ffmpeg — no cloud dependency

Mechanism · Detect

freezedetect scanned locally and flagged 30 idle sections

Instead of scrubbing by hand, one local ffmpeg pass ran the freezedetect filter to find the dead air automatically.

Thirty raw detections aren't yet something worth acting on.

# find the dead air, no manual scrubbing
ffmpeg -i input.mp4 \
  -vf "freezedetect=n=-60dB:d=2"

# result
freezedetect: 30 idle sections

Mechanism · Merge

A smart merge turned 30 micro-freezes into 10 clean regions

The judgment step: consolidate the raw detections into idle regions actually worth speeding up — merging gaps under 1.5s and keeping regions of 4s or more.

Now the edit itself could run in a single pass.

30 → 10

Merge micro-freezes with a gap under 1.5s
Keep idle regions of 4s and longer

Mechanism · Execute

One 21-segment pass sped idle 16x, kept demo frames at 1x

A single ffmpeg filter_complex pipeline fast-forwarded the idle regions at 16x while keeping every demo frame at 1x — one local pass, no intermediate files.

And it finished before the cloud editor could even begin.

21-segment filter_complex pipeline
16x on idle regions
1x on every demo frame
One pass, no intermediate files

The Payoff

Done before the upload had even completed

2:59 → 1:01 99.5 MB → 19.2 MB

Amplifier's local pass finished — shorter, smaller, every demo frame intact — while Clipchamp was still uploading the untouched original to OneDrive. Wall-clock (~2 min) and the incomplete upload are the user's recall / observation, not logged measurements.

The Takeaway

Algorithmic work deserves a sentence and a local pipeline — not an editor and an upload

This was never a creative-editing problem. The right tool was a local CLI pipeline (freezedetect → merge → filter_complex), driven by one natural-language sentence — no GUI editor, no cloud upload.

Status: Active

Sources

Sources & Research Methodology

Data as of: March 13, 2026

Feature status: Active — story deck live in docs/ (not linked from docs/index.html).

Capability: General Amplifier + local tool use (bash/ffmpeg) — not a dedicated video feature.

Research performed:

Git history: git log --format='%h | %ai | %an | %s' -- docs/video-editing-story.html
Original commit: git show 56c8578 --stat (Samuel Lee, 1010 insertions; Co-Authored-By: Amplifier)
Metric checks: grep -c '178.7', grep -c '99.5 MB', grep -c '19.2 MB', grep -c '1:01'
Pipeline details: grep -io 'freezedetect=n=-60dB:d=2', grep -io 'Pipeline:[^<]*', grep -i 'freezedetect: 30 sections'
Derived: python3 -c 'print((178.7-61)/178.7*100)' → 65.9%; print((99.5-19.2)/99.5*100) → 80.7%
Index check: grep -c 'video-editing' docs/index.html → 0 (not in index)
Reproducibility: which ffmpeg → exit 1 (not installed; ffmpeg-runtime claims verified as documentary)

Measured: 2:59/178.7s & 99.5 MB input, 1:01/19.2 MB output, −66% duration, −81% size (derived from durations/sizes).

Documentary (deck + commit 56c8578, ffmpeg not re-run): 30 sections → 10 merged, 21-segment filter_complex at 16x idle / 1x content, freezedetect=n=-60dB:d=2, Clipchamp OneDrive-upload requirement.

Estimated, not measured: the "~2 min" Amplifier wall-clock (user's recall) and the "15–20 min" Clipchamp effort (user's estimate). The upload was observed incomplete when Amplifier finished; exact % was not recorded.

Contributors: Sam (Sam Schillace, requested the edit; identity mapping inferred), Amplifier (analysis/merge/pipeline/execution, commit co-author), Samuel Lee (committed the original deck).