opencut

AI-powered video clipping CLI, written in TypeScript and runnable as a single spawn-friendly dist/cli.js.

opencut transcribes long-form video or audio, extracts highlight-worthy moments via an OpenAI-compatible LLM, and exports subtitles, timelines, or burned-in videos from a single CLI.

Features

• Cross-platform ASR via whisper.cpp (through smart-whisper)
• AI highlight extraction and auto-generated titles via any OpenAI-compatible API
• Translation and bilingual subtitle layouts via @vitalets/google-translate-api
• Keyword highlighting for important words inside subtitles
• Multiple export targets: srt, ass, fcpxml, video, txt, json
• Landscape and portrait output
• Transcript JSON reuse to skip ASR on reruns
• Pluggable backends — bring your own ASR, VAD, or translation provider

Requirements

Item	Requirement
Runtime	Node 20+ or Bun 1.2+ (development uses Bun)
FFmpeg	ffmpeg and ffprobe on PATH
Whisper model	A whisper.cpp GGML model (e.g. ggml-large-v3-turbo.bin)
API key	Required only for AI highlight extraction, keyword highlighting, or transcript correction

Install

bun install
bun run build

Optional native backends — install only the ones you actually use:

# Whisper.cpp in-process ASR (provides word timestamps)
bun add smart-whisper

# Silero VAD (if you want per-utterance chunking before ASR)
bun add @ricky0123/vad-node onnxruntime-node

Download a whisper.cpp model file from the ggerganov/whisper.cpp releases and either pass --asr-model /path/to/ggml-…bin or export OPENCUT_WHISPER_MODEL=/path/to/ggml-…bin (the legacy PYCUT_WHISPER_MODEL is still honored for older setups).

Configure an API key

export OPENAI_API_KEY="your_api_key_here"

Or pass --api-key per invocation. To use a compatible provider such as Gemini or DeepSeek, add --base-url and optionally --model:

# Gemini
./dist/cli.js input.mp4 \
  --api-key YOUR_KEY \
  --base-url https://generativelanguage.googleapis.com/v1beta/openai

# DeepSeek
./dist/cli.js input.mp4 \
  --api-key YOUR_KEY \
  --base-url https://api.deepseek.com \
  --model deepseek-v4-flash

Quick start

Extract AI-selected highlights and export a rendered video plus subtitles:

./dist/cli.js my_video.mp4 \
  --asr-model /models/ggml-large-v3-turbo.bin \
  --api-key YOUR_KEY \
  --format video,srt

Generate subtitles only, without highlight clipping:

./dist/cli.js my_video.mp4 \
  --asr-model /models/ggml-large-v3-turbo.bin \
  --no-clip --format srt

Create bilingual subtitles:

./dist/cli.js my_video.mp4 \
  --asr-model /models/ggml-large-v3-turbo.bin \
  --translate \
  --source-lang en \
  --target-lang zh-CN \
  --format video,srt

Export an FCPXML timeline:

./dist/cli.js my_video.mp4 \
  --asr-model /models/ggml-large-v3-turbo.bin \
  --api-key YOUR_KEY \
  --format fcpxml \
  --fcpxml-frame-rate 30

Dev workflow

bun run dev <args>      # run TS directly, no build
bun test                # run vitest-style bun:test suite
bun run typecheck       # tsc --noEmit
bun run lint            # biome check
bun run build           # bundle to dist/cli.js (Node target, shebang)

Spawn from another tool

dist/cli.js has a #!/usr/bin/env node shebang. Any process can spawn it:

import { spawn } from "node:child_process";

const child = spawn("node", ["./dist/cli.js", "video.mp4", "--format", "srt"]);

CLI usage

./dist/cli.js <video-file|directory|glob> [options]

Input expansion supports:

• A single file: video.mp4
• A directory: ./videos/
• A glob: ./recordings/*.mp4
• Multiple inputs: a.mp4 b.mp4 c.mp4

Supported media extensions:

• Video: mp4, mov, mkv, avi, m4v, webm
• Audio: wav, mp3, m4a, aac, flac, ogg

Run ./dist/cli.js --help for the full flag list.

Output formats

Format	Description
srt	Standard subtitle file
ass	Styled subtitle file with bilingual layout and highlighting
fcpxml	Timeline export for Final Cut Pro / DaVinci Resolve
video	Burned-in MP4 output
txt	Plain transcript
json	Timestamped transcript JSON reusable with --transcript

Pipeline

media
  -> audio extraction (ffmpeg)
  -> ASR + word timestamps (whisper.cpp via smart-whisper)
  -> optional LLM highlight extraction / keyword detection / transcript correction
  -> subtitle generation
  -> SRT / ASS / FCPXML / MP4 / TXT / JSON outputs

License

MIT