🎬 VideoEditor

A local rough-cut editor for talking-head videos. It removes silences, stumbles and repeated takes on your own computer, and leaves the doubtful decisions for you to review.

Em português: a interface é em português do Brasil e a transcrição vem configurada para português. Veja Como usar (pt-BR) mais abaixo.

---

What it does

You record yourself talking. VideoEditor transcribes the recording locally with Whisper, finds the dead air, the stumbles and the takes you repeated, and builds a virtual cut you can watch instantly in the browser. It only cuts on its own what is safe (silence, an empty chair, an exact retake); anything that might remove real content becomes a decision for you, with the evidence next to it. When you are done, it exports a clean MP4 + SRT subtitles, ready to finish in CapCut, Premiere or any other editor.

• 🔒 Runs on your computer. Video and audio never leave it. Your original file is never modified.
• ✂️ Few decisions, not hundreds. A 20-minute recording usually leaves around ten things to review: repeated takes ("the last one stays"), an empty chair, a false start.
• 🎞️ Review on a timeline. Waveform, cuts, pending decisions, zoom, keyboard shortcuts.
• 📤 Finish elsewhere. MP4 + SRT + a list of markers for your editor of choice.
• 🤖 Optional AI notes. If you turn it on and explicitly allow it, the text of the transcript (not the audio or video) is sent to Google Gemini's free tier for editorial suggestions. It is off by default.

Status

This is a young project used daily by its author. On Windows it installs like any other program; on macOS and Linux it runs from the source code.

	Works	Notes
Windows + NVIDIA GPU	✅ best experience	Installer; NVENC export; 10 min of audio transcribed in ~40 s on an RTX 2060
Windows without GPU	✅ slower	Installer ("sem placa NVIDIA" option); transcription on the CPU
Linux	✅ from source	CPU or NVIDIA
macOS	⚠️ untested	From source, CPU only
Docker	✅ for trying it out	CPU only in the provided image

---

💾 Windows: install

1. Download VideoEditor-Setup-x.y.z.exe from the latest release.
2. Run it. It installs for your user only (no administrator password) and brings its own Python and FFmpeg; nothing else is needed. Keep "Completa" if you have an NVIDIA card.
3. Open VideoEditor from the Start menu. The Studio opens in your browser and, the first time, offers to download the speech model (~1.5 GB, once).

To close it, use Configurações → Encerrar o estúdio or the Start menu entry Encerrar o VideoEditor. Your cuts stay in %LOCALAPPDATA%\VideoEditor, also after uninstalling.

"Windows protected your PC"? The installer is not code-signed yet, so SmartScreen warns the first time. Click More info → Run anyway. The .exe is built by GitHub Actions from this repository (workflow).

---

🚀 Quick start (from source)

You need: Python 3.11+, FFmpeg in your PATH, and about 2 GB free for the speech model. An NVIDIA GPU is optional but much faster.

git clone https://github.com/otavioxfran/VideoEditor.git
cd VideoEditor
pip install -r requirements.txt

# Download the speech model once (~1.5 GB)
python cli.py download-model --model large-v3-turbo

# Open the Studio (it lists ./videos by default; you can pick any file or folder in it)
python cli.py studio

The Studio opens at http://127.0.0.1:8765. On the first run it checks FFmpeg, your graphics card and the speech model, and offers to download the model with a progress bar (without an NVIDIA card it picks a lighter model). Click Novo corte → Escolher vídeo no computador…, pick any recording, then review the decisions and export.

On Windows you can also double-click start-studio.cmd. The folder listed in the Studio can be changed in Configurações → Pasta de vídeos. When installed in a read-only program folder, cuts and settings go to %LOCALAPPDATA%\VideoEditor; VIDEOEDITOR_DATA_DIR overrides it.

Docker

mkdir -p videos jobs        # put your recordings in ./videos
docker compose up -d        # then open http://localhost:8765

The image runs on the CPU, so transcription is slow on long videos. It is a good way to try the tool; for daily use, run it locally.

Command line

python cli.py process "./videos/my_recording.mp4" --automatic --render-now
python cli.py report JOB_ID      # naturalness metrics of the current cut

---

🇧🇷 Como usar (pt-BR)

No Windows, sem instalar mais nada:

1. Baixe o VideoEditor-Setup-x.y.z.exe na página da última versão.
2. Abra o arquivo e siga o instalador (não pede senha de administrador). Se tiver placa NVIDIA, deixe a opção Completa.
   • Se aparecer "O Windows protegeu o computador", clique em Mais informações → Executar assim mesmo. O aviso aparece porque o instalador ainda não tem assinatura digital.
3. Abra o VideoEditor pelo menu Iniciar. O navegador abre o Studio.
4. Na primeira vez, clique em Baixar para trazer o modelo de transcrição (~1,5 GB, uma vez só).
5. Clique em Novo corte → Escolher vídeo no computador…, revise as poucas decisões que aparecerem e clique em Exportar MP4 + SRT.
6. Para fechar: Configurações → Encerrar o estúdio (ou Encerrar o VideoEditor no menu Iniciar).

Pelo código-fonte (Windows, Linux ou macOS): instale o Python 3.11+ e o FFmpeg, baixe este projeto, rode pip install -r requirements.txt e abra com start-studio.cmd (Windows) ou python cli.py studio.

Quer que a lista mostre outra pasta? Configurações → Pasta de vídeos → Trocar pasta…. As opções avançadas (modelo, edição experimental, briefing) ficam escondidas até você ligar Configurações → Interface → Mostrar opções avançadas.

Quer que a assistência editorial conheça o seu estilo? Copie PERSONA.example.md para PERSONA.md e escreva sobre o seu canal. Esse arquivo fica só no seu computador.

---

🛠️ How it works

flowchart LR
    A["Raw video (MP4/MOV/MKV)"] --> B["Acoustic analysis"]
    A --> C["Local Whisper transcription"]
    A --> V["Visual presence (empty chair)"]
    B & C & V --> D["Cut policy: silence, stumbles, repeated takes"]
    D --> E["Virtual timeline + review queue"]
    E --> F["Studio (browser)"]
    E --> G["FFmpeg renderer"]
    G --> H["MP4 + SRT + markers"]

Loading

The interface follows ui/DESIGN.md.

🧪 Tests

pip install -r requirements-dev.txt
python -m pytest -q

🗺️ Roadmap

• Code-signed installer (no SmartScreen warning) and a macOS package.
• Transcription language selector and an English interface.

🤝 Contributing

Issues and pull requests are welcome. Read CONTRIBUTING.md and the code of conduct. Security issues: see SECURITY.md.

📄 License

MIT. Third-party components and their licenses are listed in THIRD_PARTY_NOTICES.md.