Common commands

Preview without rendering

Use a dry run first when you are tuning settings:

quickedit recording.mp4 --dry-run

QuickEdit writes the timeline and subtitle sidecars but skips the final video render.

Render one video

quickedit recording.mp4 -o recording-edited.mp4

If you omit -o, QuickEdit writes recording_edited.mp4 next to the input.

Batch process a folder

quickedit recordings/*.mp4 -o edited/

With more than one input, -o is treated as an output directory.

Fast speech-only trimming

quickedit lecture.mp4 --combine speech --subtitle-style none

This keeps sections with speech and skips motion analysis. Disabling subtitles also skips transcription, so the run only needs speech detection and rendering.

Screen recordings

For tutorials where silent visual activity matters, keep the default speech-or- motion behavior:

quickedit tutorial.mp4 --combine or:speech,motion

For stricter edits that only keep moments with both narration and visual activity:

quickedit demo.mp4 --combine and:speech,motion

Faster motion analysis

Analyze fewer frames when long screen recordings spend time on motion analysis:

quickedit tutorial.mp4 --motion-frame-skip 3

Higher values are faster but can miss very short screen changes.

Parallel OpenCV motion

quickedit tutorial.mp4 --motion-backend opencv-parallel --motion-workers 4

Use this when CPU cores are available and OpenCV decoding is the bottleneck.

FFmpeg-backed motion

quickedit tutorial.mp4 --motion-backend ffmpeg

This is the default backend. It decodes scaled grayscale frames with FFmpeg before computing frame differences.

Disable subtitles

quickedit recording.mp4 --subtitle-style none

Use this when you only want a rendered edit and timeline JSON.

Simple SRT subtitles

quickedit recording.mp4 --subtitle-style simple

This writes a portable .srt sidecar instead of an ASS subtitle file.

CPU-safe Whisper settings

quickedit recording.mp4 --whisper-model base --device cpu --whisper-compute-type int8

Use tiny for a faster first pass or small when your CPU has enough headroom.

CUDA Whisper settings

quickedit recording.mp4 --whisper-model small --device cuda --whisper-compute-type float16

CUDA is most useful when subtitles, word-based detection, or LLM editing requires transcription.

Conservative cuts

quickedit recording.mp4 --margin 0.4 --mincut 0.5 --minclip 0.2

Larger margins keep more context around speech or motion. Larger --mincut values avoid tiny jumps.

Overwrite intentionally

quickedit recording.mp4 -o recording-edited.mp4 --overwrite

QuickEdit only replaces an output when this flag is present.

Clear or reuse cache

quickedit recording.mp4
quickedit recording.mp4 --clear-cache
quickedit recording.mp4 --no-cache

The first command reuses cached analysis when settings match. The second clears the input cache before running. The third avoids cache reads and writes.

Optional LLM editing

export ANTHROPIC_API_KEY='...'
quickedit recording.mp4 --llm --llm-model claude-sonnet-4-20250514 --dry-run

Gemini models use GOOGLE_API_KEY or GEMINI_API_KEY:

export GEMINI_API_KEY='...'
quickedit recording.mp4 --llm --llm-model gemini-3.5-flash --prompt-template tutorial

LLM editing sends transcript data and prompt context to the selected provider. Run --dry-run first and raise --confidence for more conservative semantic cuts.