ScriptCut

Edit videos like documents. Turn long-form recordings into polished videos and social clips, locally on your machine.

ScriptCut is an open-source, local-first desktop video production tool for YouTube creators, podcasters, educators, founders, interview creators, streamers, and anyone repurposing spoken recordings.

Its creator workflow is simple:

Choose a video → edit the transcript → review → export

ScriptCut is different from a cloud-first editor because local processing is the default path, AI helpers are optional, and project files stay creator-owned. The current product is a desktop alpha: it already supports the core editing and export workflow, while installation and platform support continue to mature.

See the Product Vision for the canonical product direction.

Start Here

The intended creator path is the ScriptCut desktop app:

1. Open the ScriptCut GitHub Releases page.
2. Choose a release explicitly identified in its notes and manifest as a self-contained macOS arm64 prerelease, then download its tagged DMG.
3. If macOS blocks the first launch, use System Settings → Privacy & Security → Open Anyway, then confirm the prompt.
4. Choose Edit full video or Create a short, then select local media.
5. Review the transcript and export when the preflight panel is ready.

For a short, non-technical walkthrough, read the First Export Guide.

If a release asset is not available, ScriptCut can be run from source using the contributor setup below. The complete release process is documented in docs/RELEASE.md.

The two primary workflows

Edit a Video

Open a video or audio file, transcribe it locally, edit the word-level transcript, preview the result, and export. You can delete, restore, mute, or hide selected words from captions. AI edit plans and filler-word suggestions are reviewable helpers; they do not replace editorial approval.

See docs/USER_GUIDE.md for the detailed creator walkthrough.

Create Clips

Open a long-form recording, find or create candidate moments, review and trim drafts, package metadata and hook-frame notes, then export approved clips. The current app supports AI clip suggestions, editable drafts, social metadata, vertical/square output, captions, readiness checks, and batch export. Discovery and packaging still depend on review, local setup, and—when selected—an available AI provider.

Self-contained public-alpha path

ScriptCut's supported self-contained public-alpha path is macOS Apple Silicon (arm64). Public alphas produced by the current release pipeline bundle the application runtime, portable Python, the core runtime pack, FFmpeg/FFprobe, and baseline Whisper transcription code. System Python and backend dependency installation are not required for that packaged baseline path.

The trusted baseline model is downloaded and verified into app-managed storage on first transcription. Subsequent baseline transcription can reuse the verified local model without model-network access. A qualifying public alpha uses an ad-hoc code signature for package integrity, but is not signed with Apple Developer ID or notarized by Apple, so macOS first launch may require System Settings → Privacy & Security → Open Anyway. Use that approval only for a DMG downloaded from the official ScriptCut Releases page. Older alpha releases may predate this self-contained runtime; check each release's notes and manifest before downloading.

See Platform Support for the current support matrix. macOS Intel preparation is available for maintainers, while Windows and Linux are source-development paths rather than verified public installers. The browser page at localhost:5173 is for development and testing; it does not provide the desktop app’s native file access or the same autosave workflow.

In browser mode, media is uploaded to the local backend for transcription and export, and finished files are offered as downloads. Use the desktop app for native file access, persistent export folders, autosave, and Finder reveal actions.

What is available today

• Word-level transcription through Parakeet TDT v3, WhisperX, or Whisper when the selected local engine is installed.
• Transcript-based editing with edited playback, undo/redo, waveform/timeline synchronization, and speaker-aware operations when speaker labels are available.
• AI-assisted edit plans, filler-word detection with review decisions, clip suggestions, editable clip drafts, clip readiness scoring, and generated clip metadata.
• Source, vertical, and square exports with creator presets, reframe controls, optional Studio Sound audio cleanup, and optional background removal.
• Word-level captions with caption styling. Depending on the FFmpeg build, captions are burned into the video or delivered as a matching .srt sidecar file; the export preflight shows the actual result.
• Approved-clip batch export with per-draft progress, retry state, and a batch manifest in the desktop workflow.
• .scriptcut project save/load, recent projects, autosave, recovery snapshots, and compatibility with legacy .aive and .cutscript files.

Some capabilities depend on optional local packages, model downloads, provider configuration, or the capabilities of the packaged FFmpeg build. The setup assistant and export preflight are the source of truth for a particular machine.

Local-first and AI behavior

The desktop app runs an Electron shell with a local React interface, a local FastAPI backend, and FFmpeg for media work. Transcription and export use local tools by default where available.

AI features can use local Ollama or configured OpenAI, Claude, or 9Router providers. Selecting an external provider can send the requested transcript or prompt context to that provider. ScriptCut does not require AI for core transcript editing or export, and this README does not make a blanket privacy claim beyond the configured workflow.

Contributor quick start

These steps are for running ScriptCut from the repository.

Prerequisites

• Node.js 18+
• Python 3.10 to 3.12
• FFmpeg in PATH for source development. Desktop release builds include a bundled FFmpeg when prepared with npm run release:ffmpeg.
• Optional: Ollama for local AI features

Python 3.11 is recommended on Apple Silicon. Python 3.13 is not supported by the current transcription dependency stack.

macOS setup

brew install ffmpeg
python3.11 -m venv .venv
source .venv/bin/activate

To force a specific interpreter:

export SCRIPTCUT_PYTHON_PATH=/absolute/path/to/python

CUTSCRIPT_PYTHON_PATH is still supported for legacy setups, but SCRIPTCUT_PYTHON_PATH is preferred.

Install and run

npm run setup
npm run doctor
npm run dev

This starts the local backend, Vite frontend, and Electron desktop app together. For local desktop packaging, run:

npm run dist:mac

That creates a macOS DMG under dist/. See docs/INSTALL.md, docs/DEVELOPMENT.md, and CONTRIBUTING.md for technical setup and contribution guidance.

To verify the backend separately:

npm run dev:backend
curl -s http://127.0.0.1:8642/health

Expected response:

{"status":"ok"}

Project structure and technical reference

scriptcut/
├── electron/   # desktop shell, IPC, and local backend lifecycle
├── frontend/   # React/Vite creator interface
├── backend/    # FastAPI transcription, AI, caption, audio, and export services
└── shared/     # project schema and shared contracts

The main technical pieces are Electron + React, a local FastAPI backend, Parakeet/WhisperX/Whisper transcription, and FFmpeg export. The local API includes health, transcription, export, job lifecycle, AI helper, caption, audio, and background-capability endpoints. See the backend routers for the current endpoint contract.

Project files are canonical JSON with schema: "scriptcut.project.v1" and version: 1. Manual saves and autosaves use the same serializer, with compatibility for legacy .aive and .cutscript files.

Local API surface

Method	Endpoint	Description
GET	/health	Health check
POST	/transcribe	Transcribe media
POST	/jobs/transcribe	Start a transcription job
POST	/export	Export edited video
POST	/jobs/export	Start an export job
GET	/jobs/{job_id}	Read job progress, logs, result, or error
POST	/jobs/{job_id}/cancel	Request job cancellation
POST	/jobs/{job_id}/retry	Retry a failed or canceled job
POST	/ai/filler-removal	Detect filler words
POST	/ai/create-clip	Suggest clips
POST	/ai/clip-metadata	Suggest title, hook, caption, and hashtags
POST	/ai/edit-plan	Create a reviewable AI edit plan
GET	/ai/ollama-models	List local Ollama models
POST	/ai/9router-models	List models exposed by 9Router
POST	/captions	Generate captions
POST	/audio/clean	Noise reduction
GET	/audio/capabilities	Audio processing availability
GET	/background/capabilities	Background removal availability

Long-running jobs use queued, running, canceling, succeeded, failed, and canceled states. A canceled job is retryable after it reaches final canceled state.

Keyboard shortcuts

Key	Action
Space	Play / Pause
J / K / L	Reverse / Pause / Forward
← / →	Seek ±5 seconds
Delete	Delete selected words
Ctrl+Z / Cmd+Z	Undo
Ctrl+Shift+Z / Cmd+Shift+Z	Redo
Ctrl+S / Cmd+S	Save project
Ctrl+E / Cmd+E	Export
?	Shortcut cheatsheet

Quality checks

For release-oriented changes, run the checks appropriate to the affected area:

npm run lint
npm run build
npm run smoke:backend
python -m compileall -q backend

Use docs/DESKTOP_QA.md for the creator workflow checklist and docs/TROUBLESHOOTING.md when setup or runtime checks fail.

Contributing

Start with docs/INSTALL.md, docs/DEVELOPMENT.md, and CONTRIBUTING.md. Keep changes focused on creator workflows, preserve support for legacy project files, and retain the original CutScript attribution.

License

ScriptCut is open-source software licensed under the GNU Affero General Public License v3.0 or later (AGPL-3.0-or-later).

Portions derived from DataAnts-AI/CutScript retain their original MIT notices. See THIRD_PARTY_NOTICES.md and LICENSES/CutScript-MIT.txt.

Organizations that need different licensing terms can see COMMERCIAL-LICENSING.md.

The software license does not grant rights to ScriptCut branding. See TRADEMARKS.md.

ScriptCut began as a fork and continuation of DataAnts-AI/CutScript. See ACKNOWLEDGEMENTS.md for the original-project attribution.