TalkCut

Owned and maintained by Rajan Gupta

Free, offline Descript alternative. Edit video by editing text, remove ums in one click, Studio Sound, 4K export, no watermark.

Watch/download the full demo — MP4, 1080p, 1:33 · Direct MP4 download

Recorded from the real app using the bundled sample. Hear Studio Sound Before/After in the full video, then see dark mode, resizable panels and a completed 1080p export. The export wait is visibly labeled as accelerated. Recording details.

Download

Platform	Latest release
Windows x64 (NSIS)	Download .exe
macOS Intel / Apple Silicon with Rosetta	Download .dmg
Linux x86_64	Download AppImage

All releases, SHA-256 checksums and update metadata. Installers are unsigned; verify the SHA-256 checksums attached to each release. macOS/Linux packages must be smoke-tested on their target platforms. For AppImage, make the downloaded file executable before launching.

This app bundles FFmpeg/ffprobe, including GPL-enabled builds. Third-party notices and source/build notes describe their separate terms. TalkCut's own code is MIT licensed.

Features

• Edit the words, edit the video: cut, correct and rearrange a word-timestamped transcript. Preserve originals with non-destructive edits and Undo/Redo.
• Offline transcription: whisper.cpp with base/small/medium models. Download once, transcribe locally thereafter.
• One-click cleanup: preview filler words, long gaps and adjacent repeated takes; apply as one undoable edit and review removed text/times.
• Studio Sound: DeepFilterNet voice cleanup plus FFmpeg loudness normalization, with Before/After preview after local preparation.
• No watermark: MP4 through 4K, MP3, SRT, WebVTT and text export. Quality depends on the source and hardware.
• Captions, media/text/audio layers, recording, local projects, a bundled sample, a dismissible tour and light/dark themes.
• Draggable dividers resize the sidebar, Script/Preview, tools and timeline. Sizes persist; double-click resets; arrow keys adjust.
• Optional AI: OpenAI, fal, ElevenLabs and Pexels are off by default under Settings → Optional AI. Core editing needs no account or keys.

Recognition and cleanup need review. Retake detection conservatively targets adjacent repeated phrases, not every possible mistake.

Screenshots

How it works

1. Import or drop a recording, or select Try with a sample.
2. Download a model in Settings and transcribe. Select words and press Delete to cut their media intervals.
3. Use Clean up to preview fillers, silence and retakes, then Apply or Undo.
4. Prepare Studio Sound locally and compare Before/After at the same playhead.
5. Add captions and export. FFmpeg renders a new file; source media remains intact.

Space plays/pauses, Del cuts, Ctrl+Z undoes, and Ctrl+E opens Export. See tooltips for shortcuts.

Privacy

Core transcription, editing, cleanup and export run locally. No account, API key or built-in telemetry is required. Files stay in your local application-data directory.

Model downloads and manual update checks contact their download services. Explicitly enabling and using Optional AI sends selected media/text to the chosen provider under its policies. Keys entered in Settings stay in memory for the session. Recording uses OS media permissions.

System requirements

• Windows 10/11 x64, macOS 12+ on Intel (Apple Silicon requires Rosetta 2 for the current x64 package), or a modern Electron-compatible x86_64 Linux desktop.
• 8 GB RAM recommended for base/small; 16 GB+ for medium and 4K. These are recommendations, not measured minimum guarantees.
• Disk space for the app, originals, processing caches and exports. Models use about 148 MB / 488 MB / 1.53 GB.
• A CPU compatible with the native helpers. The default CPU workflow needs no GPU; speed varies by hardware.
• Internet for installation/model downloads, not for core editing afterwards. The bundled sample works immediately.

Build from source

Use Node.js 24, npm and Git. macOS/Linux native builds need CMake, a C++ toolchain and platform SDKs. Windows uses prebuilt x64 helpers. Build on the target OS.

git clone https://github.com/iamspyder/talkcut.git
cd talkcut/app
npm ci
npm run native:setup
npm run lint
npm test
npm run build
npm run desktop

Installers: npm run dist:win, npm run dist:mac, npm run dist:linux. Local commands never publish. The release workflow builds x64 macOS packages, using Rosetta on ARM hosts. See release instructions, development details and verification scope and successful GitHub build verification.

Roadmap

• Target-platform smoke tests, Windows signing and macOS signing/notarization.
• Native Apple Silicon packaging and broader Linux/CPU validation.
• Improved multilingual transcript correction and retake previews.
• Accessible timeline editing, long-recording performance and project portability.

These are plans, not delivery-date commitments.

Contributing

Read CONTRIBUTING.md and CODE_OF_CONDUCT.md. Use the bug/feature templates; report vulnerabilities through SECURITY.md. Rajan Gupta is the sole project owner and maintainer. Contributors retain copyright in their contributions.

Credits and license

TalkCut code and original branding: Copyright © 2026 Rajan Gupta, MIT. Projects studied: autoEdit_2, auto-editor, DeepFilterNet and whisper.cpp.

Dependencies retain their own authorship and licenses. See THIRD_PARTY_NOTICES.md, the full dependency inventory and app/LICENSES/. Descript is a third-party product name; TalkCut is independent and unaffiliated.

FAQ

Is it really free? Yes: local editing and exports have no watermark. Optional cloud providers may charge.

Can I work offline? Yes, after downloading the app and one transcription model. The bundled sample works before a model download. Updates and Optional AI require internet.

Why does Windows SmartScreen warn me? Initial builds are unsigned unless a certificate is configured. Verify the GitHub release and SHA-256 checksum. If you trust the download, choose More info → Run anyway. Do not disable SmartScreen globally. Checksums verify integrity, not publisher identity or safety.

What about macOS Gatekeeper and updates? Unsigned/unnotarized releases may be blocked. Prefer signed/notarized builds or build from source. Automatic macOS updates require signing; the updater cannot bypass OS checks.

Is every commercial-editor feature included? No. v1 focuses on offline transcription, text editing, cleanup, Studio Sound and export.

Is Studio Sound instant? The sample includes processed audio. Your own recordings need one local enhancement pass, then A/B uses cached media.

Will 4K export improve a low-resolution recording? It can render a 3840×2160 landscape canvas but cannot recover missing source detail.

Where are my files? In Electron's per-user TalkCut application-data directory, under media-library. JSON project backups contain edits, so keep original media too.