Auto-Editor makes a first pass through video or audio using rules such as audio loudness. It can help tighten a narration draft before you assemble generated scenes, captions and music. It does not generate images, understand your story or decide which pause carries emotion.
This guide includes a real local Auto-Editor run: a 12-second source with three short spoken phrases, the resulting cut and its exported timeline. You can download both files and inspect exactly which source frames remain. The narration uses deterministic eSpeak NG speech and the cards are authored graphics, not AI-generated footage.
What is Auto-Editor
The official Auto-Editor repository describes a command-line first-pass editor. An audio rule can keep louder regions and cut quieter ones; other methods cover motion and additional signals. Think of it as a proposed assembly that still needs editorial review.
For a faceless explainer, it can shorten gaps in a voice track before you commit generated visuals to exact timings. For a recorded tutorial, it can produce a candidate cut between spoken instructions. For an AI drama, silence may be a reaction beat worth keeping. Choose the source and purpose before choosing the threshold.
Install the current CLI
The official installation guide recommends platform binaries from its GitHub releases and lists Homebrew as an alternative. It explicitly says the CLI is no longer published on pip. Follow the current instructions for your operating system rather than copying an older Python installation tutorial.
auto-editor --version
auto-editor --helpOur local run used the official macOS ARM64 binary reporting version 31.8.0. It is a measured example for that version, not a claim that every operating system or future build produces identical files. Keep your original media and write a separate output.
Download the source and actual cut
The source shows FIRST IDEA, NEXT SCENE and FINAL FRAME for four seconds each. Speech begins half a second into each card. This makes the long gaps intentional and the retained sections easy to locate.
Download the original 12-second source and download the actual Auto-Editor cut. These are small local test files, not a model-quality comparison or a creator recording.
Run the audio rule with explicit margin
auto-editor auto-editor-pause-source.mp4 \
--edit audio:threshold=0.04 --margin 0.2s \
--no-open --output auto-editor-pause-cut.mp4The audio threshold determines which regions count as active. Margin retains time around them. The repository documents this padding as a pacing control. Change one setting at a time when trying your own source, and compare the spoken starts and ends after every export.
Our source contains isolated speech without a music bed. That matters: music, room noise or other continuous sound can keep sections above the threshold. A dialogue track isolated from the soundtrack is a different input from a finished film mix. This fixture does not test those harder conditions.
Inspect the actual timeline
Use the documented v1 timeline export to inspect the selected source ranges. The correct export mode here is v1, even though the contents are JSON. The mode --export json was rejected by the tested version.
auto-editor auto-editor-pause-source.mp4 \
--edit audio:threshold=0.04 --margin 0.2s --no-open \
--export v1 --output auto-editor-timeline.json| Source frames | Output frames | Result |
|---|---|---|
| 10 to 55 exclusive | 0 to 45 exclusive | First card retained |
| 129 to 172 exclusive | 45 to 88 exclusive | Second card retained |
| 250 to 293 exclusive | 88 to 131 exclusive | Third card retained |
At 30 frames per second, the retained video spans 131 frames, or about 4.367 seconds. FFprobe reports a 4.373333-second container duration including its audio packaging. The source contains 360 video frames. The 229-frame reduction describes this deliberately pause-rich fixture, not an expected reduction for your content.
Download the actual v1 timeline and retained-range CSV. The published timeline uses a relative source filename instead of the machine-specific temporary path. In v1, ranges use the declared timebase and end positions are exclusive; a cut action is represented separately from a normal-speed retained segment.
Review words and pacing before accepting
The output decoded completely with FFmpeg and contains H264 video at 1280 by 720 and 30 fps plus AAC audio. We inspected the three extracted card frames. We did not perform a listening review or browser playback test, so clear speech boundaries and comfortable pacing remain acceptance checks for you to perform.
- Listen to every phrase at normal speed and check its first and last sounds.
- Watch each new cut for an abrupt pose or background jump.
- Keep any pause that explains a diagram or gives a reaction time to land.
- Check the complete export after captions and music are added.
A technically valid export can still be a poor edit. If a word is clipped, try more margin on a copy. If noise keeps an unwanted gap, inspect the source track and rule before increasing the threshold across the whole project. Keep a small log of the setting changed and the observed result.
For a separate browser-based long-to-short route read the Vmaker AI workflow guide and its original clip-selection exercise.
Use the cut in a generative video workflow
For a narration-led sequence, approve the script and voice first, inspect the first-pass audio cut, then set shot lengths from the accepted timing. Our AI trailer workflow covers a separate narrative assembly decision. Avoid generating a long batch of clips around timing you still plan to remove.
For AI drama, preserve dialogue pauses that communicate hesitation, tension or a character response. For product demonstrations, retain the interval where the viewer needs to see the action even if nobody speaks. Audio activity alone does not know either intention.
Explore Clout AI Cinema for scene creation and Faceless Studio for narrated workflows. Auto-Editor is a separate local tool; this article does not add a native Auto-Editor integration to Clout. Generated assets and first-pass editing should meet in an editor after you have chosen the cut.
Reproduce the fixture and keep the evidence
The original fixture source builds the three cards and eSpeak NG narration. Use Bun, Sharp 0.34.5, FFmpeg and eSpeak NG, then run the Auto-Editor commands in the reproduction notes. The provenance record includes native measurements and file hashes.
The source script recreates authored media, while Auto-Editor performs the actual cut. We did not test NLE project imports, semantic removal of filler words, caption alignment or model generation. Those need their own examples rather than being inferred from this short demonstration.
